]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * The User Datagram Protocol (UDP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
113aa838 | 11 | * Alan Cox, <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
12 | * Hirokazu Takahashi, <taka@valinux.co.jp> |
13 | * | |
14 | * Fixes: | |
15 | * Alan Cox : verify_area() calls | |
16 | * Alan Cox : stopped close while in use off icmp | |
17 | * messages. Not a fix but a botch that | |
18 | * for udp at least is 'valid'. | |
19 | * Alan Cox : Fixed icmp handling properly | |
20 | * Alan Cox : Correct error for oversized datagrams | |
e905a9ed YH |
21 | * Alan Cox : Tidied select() semantics. |
22 | * Alan Cox : udp_err() fixed properly, also now | |
1da177e4 LT |
23 | * select and read wake correctly on errors |
24 | * Alan Cox : udp_send verify_area moved to avoid mem leak | |
25 | * Alan Cox : UDP can count its memory | |
26 | * Alan Cox : send to an unknown connection causes | |
27 | * an ECONNREFUSED off the icmp, but | |
28 | * does NOT close. | |
29 | * Alan Cox : Switched to new sk_buff handlers. No more backlog! | |
30 | * Alan Cox : Using generic datagram code. Even smaller and the PEEK | |
31 | * bug no longer crashes it. | |
32 | * Fred Van Kempen : Net2e support for sk->broadcast. | |
33 | * Alan Cox : Uses skb_free_datagram | |
34 | * Alan Cox : Added get/set sockopt support. | |
35 | * Alan Cox : Broadcasting without option set returns EACCES. | |
36 | * Alan Cox : No wakeup calls. Instead we now use the callbacks. | |
37 | * Alan Cox : Use ip_tos and ip_ttl | |
38 | * Alan Cox : SNMP Mibs | |
39 | * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. | |
40 | * Matt Dillon : UDP length checks. | |
41 | * Alan Cox : Smarter af_inet used properly. | |
42 | * Alan Cox : Use new kernel side addressing. | |
43 | * Alan Cox : Incorrect return on truncated datagram receive. | |
44 | * Arnt Gulbrandsen : New udp_send and stuff | |
45 | * Alan Cox : Cache last socket | |
46 | * Alan Cox : Route cache | |
47 | * Jon Peatfield : Minor efficiency fix to sendto(). | |
48 | * Mike Shaver : RFC1122 checks. | |
49 | * Alan Cox : Nonblocking error fix. | |
50 | * Willy Konynenberg : Transparent proxying support. | |
51 | * Mike McLagan : Routing by source | |
52 | * David S. Miller : New socket lookup architecture. | |
53 | * Last socket cache retained as it | |
54 | * does have a high hit rate. | |
55 | * Olaf Kirch : Don't linearise iovec on sendmsg. | |
56 | * Andi Kleen : Some cleanups, cache destination entry | |
e905a9ed | 57 | * for connect. |
1da177e4 LT |
58 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. |
59 | * Melvin Smith : Check msg_name not msg_namelen in sendto(), | |
60 | * return ENOTCONN for unconnected sockets (POSIX) | |
61 | * Janos Farkas : don't deliver multi/broadcasts to a different | |
62 | * bound-to-device socket | |
63 | * Hirokazu Takahashi : HW checksumming for outgoing UDP | |
64 | * datagrams. | |
65 | * Hirokazu Takahashi : sendfile() on UDP works now. | |
66 | * Arnaldo C. Melo : convert /proc/net/udp to seq_file | |
67 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
68 | * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind | |
69 | * a single port at the same time. | |
70 | * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support | |
342f0234 | 71 | * James Chapman : Add L2TP encapsulation type. |
1da177e4 LT |
72 | * |
73 | * | |
74 | * This program is free software; you can redistribute it and/or | |
75 | * modify it under the terms of the GNU General Public License | |
76 | * as published by the Free Software Foundation; either version | |
77 | * 2 of the License, or (at your option) any later version. | |
78 | */ | |
e905a9ed | 79 | |
afd46503 JP |
80 | #define pr_fmt(fmt) "UDP: " fmt |
81 | ||
7c0f6ba6 | 82 | #include <linux/uaccess.h> |
1da177e4 | 83 | #include <asm/ioctls.h> |
57c8a661 | 84 | #include <linux/memblock.h> |
8203efb3 ED |
85 | #include <linux/highmem.h> |
86 | #include <linux/swap.h> | |
1da177e4 LT |
87 | #include <linux/types.h> |
88 | #include <linux/fcntl.h> | |
89 | #include <linux/module.h> | |
90 | #include <linux/socket.h> | |
91 | #include <linux/sockios.h> | |
14c85021 | 92 | #include <linux/igmp.h> |
6e540309 | 93 | #include <linux/inetdevice.h> |
1da177e4 LT |
94 | #include <linux/in.h> |
95 | #include <linux/errno.h> | |
96 | #include <linux/timer.h> | |
97 | #include <linux/mm.h> | |
1da177e4 | 98 | #include <linux/inet.h> |
1da177e4 | 99 | #include <linux/netdevice.h> |
5a0e3ad6 | 100 | #include <linux/slab.h> |
c752f073 | 101 | #include <net/tcp_states.h> |
1da177e4 LT |
102 | #include <linux/skbuff.h> |
103 | #include <linux/proc_fs.h> | |
104 | #include <linux/seq_file.h> | |
457c4cbc | 105 | #include <net/net_namespace.h> |
1da177e4 | 106 | #include <net/icmp.h> |
421b3885 | 107 | #include <net/inet_hashtables.h> |
e7cc0824 | 108 | #include <net/ip_tunnels.h> |
1da177e4 | 109 | #include <net/route.h> |
1da177e4 LT |
110 | #include <net/checksum.h> |
111 | #include <net/xfrm.h> | |
296f7ea7 | 112 | #include <trace/events/udp.h> |
447167bf | 113 | #include <linux/static_key.h> |
22911fc5 | 114 | #include <trace/events/skb.h> |
076bb0c8 | 115 | #include <net/busy_poll.h> |
ba4e58ec | 116 | #include "udp_impl.h" |
e32ea7e7 | 117 | #include <net/sock_reuseport.h> |
217375a0 | 118 | #include <net/addrconf.h> |
60fb9567 | 119 | #include <net/udp_tunnel.h> |
1da177e4 | 120 | |
f86dcc5a | 121 | struct udp_table udp_table __read_mostly; |
645ca708 | 122 | EXPORT_SYMBOL(udp_table); |
1da177e4 | 123 | |
8d987e5c | 124 | long sysctl_udp_mem[3] __read_mostly; |
95766fff | 125 | EXPORT_SYMBOL(sysctl_udp_mem); |
c482c568 | 126 | |
8d987e5c | 127 | atomic_long_t udp_memory_allocated; |
95766fff HA |
128 | EXPORT_SYMBOL(udp_memory_allocated); |
129 | ||
f86dcc5a ED |
130 | #define MAX_UDP_PORTS 65536 |
131 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 132 | |
63a6fff3 RS |
133 | /* IPCB reference means this can not be used from early demux */ |
134 | static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) | |
135 | { | |
136 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
137 | if (!net->ipv4.sysctl_udp_l3mdev_accept && | |
138 | skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) | |
139 | return true; | |
140 | #endif | |
141 | return false; | |
142 | } | |
143 | ||
f24d43c0 | 144 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 145 | const struct udp_hslot *hslot, |
98322f22 | 146 | unsigned long *bitmap, |
fe38d2a1 | 147 | struct sock *sk, unsigned int log) |
1da177e4 | 148 | { |
f24d43c0 | 149 | struct sock *sk2; |
ba418fa3 | 150 | kuid_t uid = sock_i_uid(sk); |
25030a7f | 151 | |
ca065d0c | 152 | sk_for_each(sk2, &hslot->head) { |
9d4fb27d JP |
153 | if (net_eq(sock_net(sk2), net) && |
154 | sk2 != sk && | |
d4cada4a | 155 | (bitmap || udp_sk(sk2)->udp_port_hash == num) && |
9d4fb27d JP |
156 | (!sk2->sk_reuse || !sk->sk_reuse) && |
157 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
158 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 159 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
160 | if (sk2->sk_reuseport && sk->sk_reuseport && |
161 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
162 | uid_eq(uid, sock_i_uid(sk2))) { | |
163 | if (!bitmap) | |
164 | return 0; | |
165 | } else { | |
166 | if (!bitmap) | |
167 | return 1; | |
168 | __set_bit(udp_sk(sk2)->udp_port_hash >> log, | |
169 | bitmap); | |
170 | } | |
98322f22 | 171 | } |
4243cdc2 | 172 | } |
25030a7f GR |
173 | return 0; |
174 | } | |
175 | ||
30fff923 ED |
176 | /* |
177 | * Note: we still hold spinlock of primary hash chain, so no other writer | |
178 | * can insert/delete a socket with local_port == num | |
179 | */ | |
180 | static int udp_lib_lport_inuse2(struct net *net, __u16 num, | |
4243cdc2 | 181 | struct udp_hslot *hslot2, |
fe38d2a1 | 182 | struct sock *sk) |
30fff923 ED |
183 | { |
184 | struct sock *sk2; | |
ba418fa3 | 185 | kuid_t uid = sock_i_uid(sk); |
30fff923 ED |
186 | int res = 0; |
187 | ||
188 | spin_lock(&hslot2->lock); | |
ca065d0c | 189 | udp_portaddr_for_each_entry(sk2, &hslot2->head) { |
9d4fb27d JP |
190 | if (net_eq(sock_net(sk2), net) && |
191 | sk2 != sk && | |
192 | (udp_sk(sk2)->udp_port_hash == num) && | |
193 | (!sk2->sk_reuse || !sk->sk_reuse) && | |
194 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
195 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 196 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
197 | if (sk2->sk_reuseport && sk->sk_reuseport && |
198 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
199 | uid_eq(uid, sock_i_uid(sk2))) { | |
200 | res = 0; | |
201 | } else { | |
202 | res = 1; | |
203 | } | |
30fff923 ED |
204 | break; |
205 | } | |
4243cdc2 | 206 | } |
30fff923 ED |
207 | spin_unlock(&hslot2->lock); |
208 | return res; | |
209 | } | |
210 | ||
fe38d2a1 | 211 | static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) |
e32ea7e7 CG |
212 | { |
213 | struct net *net = sock_net(sk); | |
e32ea7e7 CG |
214 | kuid_t uid = sock_i_uid(sk); |
215 | struct sock *sk2; | |
216 | ||
ca065d0c | 217 | sk_for_each(sk2, &hslot->head) { |
e32ea7e7 CG |
218 | if (net_eq(sock_net(sk2), net) && |
219 | sk2 != sk && | |
220 | sk2->sk_family == sk->sk_family && | |
221 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
222 | (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && | |
223 | (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
224 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && | |
fe38d2a1 | 225 | inet_rcv_saddr_equal(sk, sk2, false)) { |
2dbb9b9e MKL |
226 | return reuseport_add_sock(sk, sk2, |
227 | inet_rcv_saddr_any(sk)); | |
e32ea7e7 CG |
228 | } |
229 | } | |
230 | ||
2dbb9b9e | 231 | return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); |
e32ea7e7 CG |
232 | } |
233 | ||
25030a7f | 234 | /** |
6ba5a3c5 | 235 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
236 | * |
237 | * @sk: socket struct in question | |
238 | * @snum: port number to look up | |
25985edc | 239 | * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, |
30fff923 | 240 | * with NULL address |
25030a7f | 241 | */ |
6ba5a3c5 | 242 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
30fff923 | 243 | unsigned int hash2_nulladdr) |
25030a7f | 244 | { |
512615b6 | 245 | struct udp_hslot *hslot, *hslot2; |
645ca708 | 246 | struct udp_table *udptable = sk->sk_prot->h.udp_table; |
25030a7f | 247 | int error = 1; |
3b1e0a65 | 248 | struct net *net = sock_net(sk); |
1da177e4 | 249 | |
32c1da70 | 250 | if (!snum) { |
9088c560 | 251 | int low, high, remaining; |
95c96174 | 252 | unsigned int rand; |
98322f22 ED |
253 | unsigned short first, last; |
254 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 255 | |
0bbf87d8 | 256 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 257 | remaining = (high - low) + 1; |
227b60f5 | 258 | |
63862b5b | 259 | rand = prandom_u32(); |
8fc54f68 | 260 | first = reciprocal_scale(rand, remaining) + low; |
98322f22 ED |
261 | /* |
262 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
263 | */ | |
f86dcc5a | 264 | rand = (rand | 1) * (udptable->mask + 1); |
5781b235 ED |
265 | last = first + udptable->mask + 1; |
266 | do { | |
f86dcc5a | 267 | hslot = udp_hashslot(udptable, net, first); |
98322f22 | 268 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 269 | spin_lock_bh(&hslot->lock); |
98322f22 | 270 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
fe38d2a1 | 271 | udptable->log); |
98322f22 ED |
272 | |
273 | snum = first; | |
274 | /* | |
275 | * Iterate on all possible values of snum for this hash. | |
276 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
277 | * give us randomization and full range coverage. | |
278 | */ | |
9088c560 | 279 | do { |
98322f22 | 280 | if (low <= snum && snum <= high && |
e3826f1e | 281 | !test_bit(snum >> udptable->log, bitmap) && |
122ff243 | 282 | !inet_is_local_reserved_port(net, snum)) |
98322f22 ED |
283 | goto found; |
284 | snum += rand; | |
285 | } while (snum != first); | |
286 | spin_unlock_bh(&hslot->lock); | |
df560056 | 287 | cond_resched(); |
5781b235 | 288 | } while (++first != last); |
98322f22 | 289 | goto fail; |
645ca708 | 290 | } else { |
f86dcc5a | 291 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 292 | spin_lock_bh(&hslot->lock); |
30fff923 ED |
293 | if (hslot->count > 10) { |
294 | int exist; | |
295 | unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; | |
296 | ||
297 | slot2 &= udptable->mask; | |
298 | hash2_nulladdr &= udptable->mask; | |
299 | ||
300 | hslot2 = udp_hashslot2(udptable, slot2); | |
301 | if (hslot->count < hslot2->count) | |
302 | goto scan_primary_hash; | |
303 | ||
fe38d2a1 | 304 | exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); |
30fff923 ED |
305 | if (!exist && (hash2_nulladdr != slot2)) { |
306 | hslot2 = udp_hashslot2(udptable, hash2_nulladdr); | |
307 | exist = udp_lib_lport_inuse2(net, snum, hslot2, | |
fe38d2a1 | 308 | sk); |
30fff923 ED |
309 | } |
310 | if (exist) | |
311 | goto fail_unlock; | |
312 | else | |
313 | goto found; | |
314 | } | |
315 | scan_primary_hash: | |
fe38d2a1 | 316 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) |
645ca708 ED |
317 | goto fail_unlock; |
318 | } | |
98322f22 | 319 | found: |
c720c7e8 | 320 | inet_sk(sk)->inet_num = snum; |
d4cada4a ED |
321 | udp_sk(sk)->udp_port_hash = snum; |
322 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
1da177e4 | 323 | if (sk_unhashed(sk)) { |
e32ea7e7 | 324 | if (sk->sk_reuseport && |
fe38d2a1 | 325 | udp_reuseport_add_sock(sk, hslot)) { |
e32ea7e7 CG |
326 | inet_sk(sk)->inet_num = 0; |
327 | udp_sk(sk)->udp_port_hash = 0; | |
328 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
329 | goto fail_unlock; | |
330 | } | |
331 | ||
ca065d0c | 332 | sk_add_node_rcu(sk, &hslot->head); |
fdcc8aa9 | 333 | hslot->count++; |
c29a0bc4 | 334 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
512615b6 ED |
335 | |
336 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
337 | spin_lock(&hslot2->lock); | |
d894ba18 | 338 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
1602f49b DM |
339 | sk->sk_family == AF_INET6) |
340 | hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, | |
341 | &hslot2->head); | |
d894ba18 | 342 | else |
1602f49b DM |
343 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
344 | &hslot2->head); | |
512615b6 ED |
345 | hslot2->count++; |
346 | spin_unlock(&hslot2->lock); | |
1da177e4 | 347 | } |
ca065d0c | 348 | sock_set_flag(sk, SOCK_RCU_FREE); |
25030a7f | 349 | error = 0; |
645ca708 ED |
350 | fail_unlock: |
351 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 352 | fail: |
25030a7f GR |
353 | return error; |
354 | } | |
c482c568 | 355 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 356 | |
6ba5a3c5 | 357 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 358 | { |
30fff923 | 359 | unsigned int hash2_nulladdr = |
f0b1e64c | 360 | ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); |
30fff923 | 361 | unsigned int hash2_partial = |
f0b1e64c | 362 | ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); |
30fff923 | 363 | |
d4cada4a | 364 | /* precompute partial secondary hash */ |
30fff923 | 365 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
fe38d2a1 | 366 | return udp_lib_get_port(sk, snum, hash2_nulladdr); |
db8dac20 DM |
367 | } |
368 | ||
d1e37288 SX |
369 | static int compute_score(struct sock *sk, struct net *net, |
370 | __be32 saddr, __be16 sport, | |
fb74c277 DA |
371 | __be32 daddr, unsigned short hnum, |
372 | int dif, int sdif, bool exact_dif) | |
645ca708 | 373 | { |
60c04aec JP |
374 | int score; |
375 | struct inet_sock *inet; | |
6da5b0f0 | 376 | bool dev_match; |
645ca708 | 377 | |
60c04aec JP |
378 | if (!net_eq(sock_net(sk), net) || |
379 | udp_sk(sk)->udp_port_hash != hnum || | |
380 | ipv6_only_sock(sk)) | |
381 | return -1; | |
645ca708 | 382 | |
4cdeeee9 PO |
383 | if (sk->sk_rcv_saddr != daddr) |
384 | return -1; | |
60c04aec | 385 | |
4cdeeee9 | 386 | score = (sk->sk_family == PF_INET) ? 2 : 1; |
60c04aec | 387 | |
4cdeeee9 | 388 | inet = inet_sk(sk); |
60c04aec JP |
389 | if (inet->inet_daddr) { |
390 | if (inet->inet_daddr != saddr) | |
391 | return -1; | |
392 | score += 4; | |
393 | } | |
394 | ||
395 | if (inet->inet_dport) { | |
396 | if (inet->inet_dport != sport) | |
397 | return -1; | |
398 | score += 4; | |
399 | } | |
400 | ||
6da5b0f0 MM |
401 | dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, |
402 | dif, sdif); | |
403 | if (!dev_match) | |
404 | return -1; | |
405 | score += 4; | |
fb74c277 | 406 | |
70da268b ED |
407 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
408 | score++; | |
645ca708 ED |
409 | return score; |
410 | } | |
411 | ||
6eada011 ED |
412 | static u32 udp_ehashfn(const struct net *net, const __be32 laddr, |
413 | const __u16 lport, const __be32 faddr, | |
414 | const __be16 fport) | |
65cd8033 | 415 | { |
1bbdceef HFS |
416 | static u32 udp_ehash_secret __read_mostly; |
417 | ||
418 | net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); | |
419 | ||
65cd8033 | 420 | return __inet_ehashfn(laddr, lport, faddr, fport, |
1bbdceef | 421 | udp_ehash_secret + net_hash_mix(net)); |
65cd8033 HFS |
422 | } |
423 | ||
d1e37288 | 424 | /* called with rcu_read_lock() */ |
5051ebd2 | 425 | static struct sock *udp4_lib_lookup2(struct net *net, |
fb74c277 DA |
426 | __be32 saddr, __be16 sport, |
427 | __be32 daddr, unsigned int hnum, | |
428 | int dif, int sdif, bool exact_dif, | |
429 | struct udp_hslot *hslot2, | |
430 | struct sk_buff *skb) | |
5051ebd2 ED |
431 | { |
432 | struct sock *sk, *result; | |
e94a62f5 | 433 | int score, badness; |
ba418fa3 | 434 | u32 hash = 0; |
5051ebd2 | 435 | |
5051ebd2 | 436 | result = NULL; |
ba418fa3 | 437 | badness = 0; |
ca065d0c | 438 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
d1e37288 | 439 | score = compute_score(sk, net, saddr, sport, |
fb74c277 | 440 | daddr, hnum, dif, sdif, exact_dif); |
5051ebd2 | 441 | if (score > badness) { |
e94a62f5 | 442 | if (sk->sk_reuseport) { |
65cd8033 HFS |
443 | hash = udp_ehashfn(net, daddr, hnum, |
444 | saddr, sport); | |
ca065d0c | 445 | result = reuseport_select_sock(sk, hash, skb, |
ed0dfffd | 446 | sizeof(struct udphdr)); |
ca065d0c ED |
447 | if (result) |
448 | return result; | |
ba418fa3 | 449 | } |
ca065d0c ED |
450 | badness = score; |
451 | result = sk; | |
5051ebd2 ED |
452 | } |
453 | } | |
5051ebd2 ED |
454 | return result; |
455 | } | |
456 | ||
db8dac20 DM |
457 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
458 | * harder than this. -DaveM | |
459 | */ | |
fce82338 | 460 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, |
fb74c277 DA |
461 | __be16 sport, __be32 daddr, __be16 dport, int dif, |
462 | int sdif, struct udp_table *udptable, struct sk_buff *skb) | |
db8dac20 | 463 | { |
4cdeeee9 | 464 | struct sock *result; |
db8dac20 | 465 | unsigned short hnum = ntohs(dport); |
4cdeeee9 PO |
466 | unsigned int hash2, slot2; |
467 | struct udp_hslot *hslot2; | |
63a6fff3 | 468 | bool exact_dif = udp_lib_exact_dif_match(net, skb); |
645ca708 | 469 | |
4cdeeee9 PO |
470 | hash2 = ipv4_portaddr_hash(net, daddr, hnum); |
471 | slot2 = hash2 & udptable->mask; | |
472 | hslot2 = &udptable->hash2[slot2]; | |
473 | ||
474 | result = udp4_lib_lookup2(net, saddr, sport, | |
475 | daddr, hnum, dif, sdif, | |
476 | exact_dif, hslot2, skb); | |
477 | if (!result) { | |
478 | hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); | |
5051ebd2 ED |
479 | slot2 = hash2 & udptable->mask; |
480 | hslot2 = &udptable->hash2[slot2]; | |
5051ebd2 ED |
481 | |
482 | result = udp4_lib_lookup2(net, saddr, sport, | |
4cdeeee9 | 483 | htonl(INADDR_ANY), hnum, dif, sdif, |
63a6fff3 | 484 | exact_dif, hslot2, skb); |
db8dac20 | 485 | } |
4cdeeee9 PO |
486 | if (unlikely(IS_ERR(result))) |
487 | return NULL; | |
db8dac20 DM |
488 | return result; |
489 | } | |
fce82338 | 490 | EXPORT_SYMBOL_GPL(__udp4_lib_lookup); |
db8dac20 | 491 | |
607c4aaf KK |
492 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
493 | __be16 sport, __be16 dport, | |
645ca708 | 494 | struct udp_table *udptable) |
607c4aaf KK |
495 | { |
496 | const struct iphdr *iph = ip_hdr(skb); | |
497 | ||
ed7cbbce | 498 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, |
8afdd99a | 499 | iph->daddr, dport, inet_iif(skb), |
fb74c277 | 500 | inet_sdif(skb), udptable, skb); |
607c4aaf KK |
501 | } |
502 | ||
63058308 TH |
503 | struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, |
504 | __be16 sport, __be16 dport) | |
505 | { | |
ed7cbbce | 506 | return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); |
63058308 TH |
507 | } |
508 | EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); | |
509 | ||
ca065d0c ED |
510 | /* Must be called under rcu_read_lock(). |
511 | * Does increment socket refcount. | |
512 | */ | |
6e86000c | 513 | #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) |
bcd41303 KK |
514 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
515 | __be32 daddr, __be16 dport, int dif) | |
516 | { | |
ca065d0c ED |
517 | struct sock *sk; |
518 | ||
519 | sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, | |
fb74c277 | 520 | dif, 0, &udp_table, NULL); |
41c6d650 | 521 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) |
ca065d0c ED |
522 | sk = NULL; |
523 | return sk; | |
bcd41303 KK |
524 | } |
525 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
ca065d0c | 526 | #endif |
bcd41303 | 527 | |
421b3885 SB |
528 | static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, |
529 | __be16 loc_port, __be32 loc_addr, | |
530 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 531 | int dif, int sdif, unsigned short hnum) |
421b3885 SB |
532 | { |
533 | struct inet_sock *inet = inet_sk(sk); | |
534 | ||
535 | if (!net_eq(sock_net(sk), net) || | |
536 | udp_sk(sk)->udp_port_hash != hnum || | |
537 | (inet->inet_daddr && inet->inet_daddr != rmt_addr) || | |
538 | (inet->inet_dport != rmt_port && inet->inet_dport) || | |
539 | (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || | |
540 | ipv6_only_sock(sk) || | |
fb74c277 DA |
541 | (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && |
542 | sk->sk_bound_dev_if != sdif)) | |
421b3885 | 543 | return false; |
60d9b031 | 544 | if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) |
421b3885 SB |
545 | return false; |
546 | return true; | |
547 | } | |
548 | ||
a36e185e SB |
549 | DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); |
550 | void udp_encap_enable(void) | |
551 | { | |
9c480601 | 552 | static_branch_inc(&udp_encap_needed_key); |
a36e185e SB |
553 | } |
554 | EXPORT_SYMBOL(udp_encap_enable); | |
555 | ||
e7cc0824 SB |
556 | /* Handler for tunnels with arbitrary destination ports: no socket lookup, go |
557 | * through error handlers in encapsulations looking for a match. | |
558 | */ | |
559 | static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) | |
560 | { | |
561 | int i; | |
562 | ||
563 | for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { | |
564 | int (*handler)(struct sk_buff *skb, u32 info); | |
92b95364 | 565 | const struct ip_tunnel_encap_ops *encap; |
e7cc0824 | 566 | |
92b95364 PA |
567 | encap = rcu_dereference(iptun_encaps[i]); |
568 | if (!encap) | |
e7cc0824 | 569 | continue; |
92b95364 | 570 | handler = encap->err_handler; |
e7cc0824 SB |
571 | if (handler && !handler(skb, info)) |
572 | return 0; | |
573 | } | |
574 | ||
575 | return -ENOENT; | |
576 | } | |
577 | ||
a36e185e SB |
578 | /* Try to match ICMP errors to UDP tunnels by looking up a socket without |
579 | * reversing source and destination port: this will match tunnels that force the | |
580 | * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that | |
581 | * lwtunnels might actually break this assumption by being configured with | |
582 | * different destination ports on endpoints, in this case we won't be able to | |
583 | * trace ICMP messages back to them. | |
584 | * | |
e7cc0824 SB |
585 | * If this doesn't match any socket, probe tunnels with arbitrary destination |
586 | * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port | |
587 | * we've sent packets to won't necessarily match the local destination port. | |
588 | * | |
a36e185e SB |
589 | * Then ask the tunnel implementation to match the error against a valid |
590 | * association. | |
591 | * | |
e7cc0824 SB |
592 | * Return an error if we can't find a match, the socket if we need further |
593 | * processing, zero otherwise. | |
a36e185e SB |
594 | */ |
595 | static struct sock *__udp4_lib_err_encap(struct net *net, | |
596 | const struct iphdr *iph, | |
597 | struct udphdr *uh, | |
598 | struct udp_table *udptable, | |
e7cc0824 | 599 | struct sk_buff *skb, u32 info) |
a36e185e | 600 | { |
a36e185e | 601 | int network_offset, transport_offset; |
a36e185e SB |
602 | struct sock *sk; |
603 | ||
a36e185e SB |
604 | network_offset = skb_network_offset(skb); |
605 | transport_offset = skb_transport_offset(skb); | |
606 | ||
607 | /* Network header needs to point to the outer IPv4 header inside ICMP */ | |
608 | skb_reset_network_header(skb); | |
609 | ||
610 | /* Transport header needs to point to the UDP header */ | |
611 | skb_set_transport_header(skb, iph->ihl << 2); | |
612 | ||
e7cc0824 SB |
613 | sk = __udp4_lib_lookup(net, iph->daddr, uh->source, |
614 | iph->saddr, uh->dest, skb->dev->ifindex, 0, | |
615 | udptable, NULL); | |
616 | if (sk) { | |
617 | int (*lookup)(struct sock *sk, struct sk_buff *skb); | |
618 | struct udp_sock *up = udp_sk(sk); | |
619 | ||
620 | lookup = READ_ONCE(up->encap_err_lookup); | |
621 | if (!lookup || lookup(sk, skb)) | |
622 | sk = NULL; | |
623 | } | |
624 | ||
625 | if (!sk) | |
626 | sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); | |
a36e185e SB |
627 | |
628 | skb_set_transport_header(skb, transport_offset); | |
629 | skb_set_network_header(skb, network_offset); | |
630 | ||
631 | return sk; | |
632 | } | |
633 | ||
db8dac20 DM |
634 | /* |
635 | * This routine is called by the ICMP module when it gets some | |
636 | * sort of error condition. If err < 0 then the socket should | |
637 | * be closed and the error returned to the user. If err > 0 | |
638 | * it's just the icmp type << 8 | icmp code. | |
639 | * Header points to the ip header of the error packet. We move | |
640 | * on past this. Then (as it used to claim before adjustment) | |
641 | * header points to the first 8 bytes of the udp header. We need | |
642 | * to find the appropriate port. | |
643 | */ | |
644 | ||
32bbd879 | 645 | int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
646 | { |
647 | struct inet_sock *inet; | |
b71d1d42 | 648 | const struct iphdr *iph = (const struct iphdr *)skb->data; |
c482c568 | 649 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); |
db8dac20 DM |
650 | const int type = icmp_hdr(skb)->type; |
651 | const int code = icmp_hdr(skb)->code; | |
a36e185e | 652 | bool tunnel = false; |
db8dac20 DM |
653 | struct sock *sk; |
654 | int harderr; | |
655 | int err; | |
fd54d716 | 656 | struct net *net = dev_net(skb->dev); |
db8dac20 | 657 | |
fd54d716 | 658 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
f64bf6b8 MM |
659 | iph->saddr, uh->source, skb->dev->ifindex, |
660 | inet_sdif(skb), udptable, NULL); | |
51456b29 | 661 | if (!sk) { |
a36e185e | 662 | /* No socket for error: try tunnels before discarding */ |
e7cc0824 SB |
663 | sk = ERR_PTR(-ENOENT); |
664 | if (static_branch_unlikely(&udp_encap_needed_key)) { | |
665 | sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, | |
666 | info); | |
667 | if (!sk) | |
668 | return 0; | |
669 | } | |
a36e185e | 670 | |
e7cc0824 | 671 | if (IS_ERR(sk)) { |
a36e185e | 672 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
e7cc0824 | 673 | return PTR_ERR(sk); |
a36e185e | 674 | } |
e7cc0824 | 675 | |
a36e185e | 676 | tunnel = true; |
db8dac20 DM |
677 | } |
678 | ||
679 | err = 0; | |
680 | harderr = 0; | |
681 | inet = inet_sk(sk); | |
682 | ||
683 | switch (type) { | |
684 | default: | |
685 | case ICMP_TIME_EXCEEDED: | |
686 | err = EHOSTUNREACH; | |
687 | break; | |
688 | case ICMP_SOURCE_QUENCH: | |
689 | goto out; | |
690 | case ICMP_PARAMETERPROB: | |
691 | err = EPROTO; | |
692 | harderr = 1; | |
693 | break; | |
694 | case ICMP_DEST_UNREACH: | |
695 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
36393395 | 696 | ipv4_sk_update_pmtu(skb, sk, info); |
db8dac20 DM |
697 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { |
698 | err = EMSGSIZE; | |
699 | harderr = 1; | |
700 | break; | |
701 | } | |
702 | goto out; | |
703 | } | |
704 | err = EHOSTUNREACH; | |
705 | if (code <= NR_ICMP_UNREACH) { | |
706 | harderr = icmp_err_convert[code].fatal; | |
707 | err = icmp_err_convert[code].errno; | |
708 | } | |
709 | break; | |
55be7a9c DM |
710 | case ICMP_REDIRECT: |
711 | ipv4_sk_redirect(skb, sk); | |
1a462d18 | 712 | goto out; |
db8dac20 DM |
713 | } |
714 | ||
715 | /* | |
716 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
717 | * 4.1.3.3. | |
718 | */ | |
a36e185e SB |
719 | if (tunnel) { |
720 | /* ...not for tunnels though: we don't have a sending socket */ | |
721 | goto out; | |
722 | } | |
db8dac20 DM |
723 | if (!inet->recverr) { |
724 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
725 | goto out; | |
b1faf566 | 726 | } else |
c482c568 | 727 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
b1faf566 | 728 | |
db8dac20 DM |
729 | sk->sk_err = err; |
730 | sk->sk_error_report(sk); | |
731 | out: | |
32bbd879 | 732 | return 0; |
db8dac20 DM |
733 | } |
734 | ||
32bbd879 | 735 | int udp_err(struct sk_buff *skb, u32 info) |
db8dac20 | 736 | { |
32bbd879 | 737 | return __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
738 | } |
739 | ||
740 | /* | |
741 | * Throw away all pending data and cancel the corking. Socket is locked. | |
742 | */ | |
36d926b9 | 743 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
744 | { |
745 | struct udp_sock *up = udp_sk(sk); | |
746 | ||
747 | if (up->pending) { | |
748 | up->len = 0; | |
749 | up->pending = 0; | |
750 | ip_flush_pending_frames(sk); | |
751 | } | |
752 | } | |
36d926b9 | 753 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
754 | |
755 | /** | |
f6b9664f | 756 | * udp4_hwcsum - handle outgoing HW checksumming |
db8dac20 DM |
757 | * @skb: sk_buff containing the filled-in UDP header |
758 | * (checksum field must be zeroed out) | |
f6b9664f HX |
759 | * @src: source IP address |
760 | * @dst: destination IP address | |
db8dac20 | 761 | */ |
c26bf4a5 | 762 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) |
db8dac20 | 763 | { |
db8dac20 | 764 | struct udphdr *uh = udp_hdr(skb); |
f6b9664f HX |
765 | int offset = skb_transport_offset(skb); |
766 | int len = skb->len - offset; | |
767 | int hlen = len; | |
db8dac20 DM |
768 | __wsum csum = 0; |
769 | ||
ebbe495f | 770 | if (!skb_has_frag_list(skb)) { |
db8dac20 DM |
771 | /* |
772 | * Only one fragment on the socket. | |
773 | */ | |
774 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
775 | skb->csum_offset = offsetof(struct udphdr, check); | |
f6b9664f HX |
776 | uh->check = ~csum_tcpudp_magic(src, dst, len, |
777 | IPPROTO_UDP, 0); | |
db8dac20 | 778 | } else { |
ebbe495f WC |
779 | struct sk_buff *frags; |
780 | ||
db8dac20 DM |
781 | /* |
782 | * HW-checksum won't work as there are two or more | |
783 | * fragments on the socket so that all csums of sk_buffs | |
784 | * should be together | |
785 | */ | |
ebbe495f | 786 | skb_walk_frags(skb, frags) { |
f6b9664f HX |
787 | csum = csum_add(csum, frags->csum); |
788 | hlen -= frags->len; | |
ebbe495f | 789 | } |
db8dac20 | 790 | |
f6b9664f | 791 | csum = skb_checksum(skb, offset, hlen, csum); |
db8dac20 DM |
792 | skb->ip_summed = CHECKSUM_NONE; |
793 | ||
db8dac20 DM |
794 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); |
795 | if (uh->check == 0) | |
796 | uh->check = CSUM_MANGLED_0; | |
797 | } | |
798 | } | |
c26bf4a5 | 799 | EXPORT_SYMBOL_GPL(udp4_hwcsum); |
db8dac20 | 800 | |
af5fcba7 TH |
801 | /* Function to set UDP checksum for an IPv4 UDP packet. This is intended |
802 | * for the simple case like when setting the checksum for a UDP tunnel. | |
803 | */ | |
804 | void udp_set_csum(bool nocheck, struct sk_buff *skb, | |
805 | __be32 saddr, __be32 daddr, int len) | |
806 | { | |
807 | struct udphdr *uh = udp_hdr(skb); | |
808 | ||
179bc67f | 809 | if (nocheck) { |
af5fcba7 | 810 | uh->check = 0; |
179bc67f | 811 | } else if (skb_is_gso(skb)) { |
af5fcba7 | 812 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); |
179bc67f EC |
813 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
814 | uh->check = 0; | |
815 | uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); | |
816 | if (uh->check == 0) | |
817 | uh->check = CSUM_MANGLED_0; | |
d75f1306 | 818 | } else { |
af5fcba7 TH |
819 | skb->ip_summed = CHECKSUM_PARTIAL; |
820 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
821 | skb->csum_offset = offsetof(struct udphdr, check); | |
822 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); | |
af5fcba7 TH |
823 | } |
824 | } | |
825 | EXPORT_SYMBOL(udp_set_csum); | |
826 | ||
bec1f6f6 WB |
827 | static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, |
828 | struct inet_cork *cork) | |
db8dac20 | 829 | { |
f6b9664f | 830 | struct sock *sk = skb->sk; |
db8dac20 | 831 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
832 | struct udphdr *uh; |
833 | int err = 0; | |
834 | int is_udplite = IS_UDPLITE(sk); | |
f6b9664f HX |
835 | int offset = skb_transport_offset(skb); |
836 | int len = skb->len - offset; | |
db8dac20 DM |
837 | __wsum csum = 0; |
838 | ||
db8dac20 DM |
839 | /* |
840 | * Create a UDP header | |
841 | */ | |
842 | uh = udp_hdr(skb); | |
f6b9664f | 843 | uh->source = inet->inet_sport; |
79ab0531 | 844 | uh->dest = fl4->fl4_dport; |
f6b9664f | 845 | uh->len = htons(len); |
db8dac20 DM |
846 | uh->check = 0; |
847 | ||
bec1f6f6 WB |
848 | if (cork->gso_size) { |
849 | const int hlen = skb_network_header_len(skb) + | |
850 | sizeof(struct udphdr); | |
851 | ||
0f149c9f WB |
852 | if (hlen + cork->gso_size > cork->fragsize) { |
853 | kfree_skb(skb); | |
bec1f6f6 | 854 | return -EINVAL; |
0f149c9f WB |
855 | } |
856 | if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { | |
857 | kfree_skb(skb); | |
bec1f6f6 | 858 | return -EINVAL; |
0f149c9f WB |
859 | } |
860 | if (sk->sk_no_check_tx) { | |
861 | kfree_skb(skb); | |
a8c744a8 | 862 | return -EINVAL; |
0f149c9f | 863 | } |
ff06342c | 864 | if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || |
0f149c9f WB |
865 | dst_xfrm(skb_dst(skb))) { |
866 | kfree_skb(skb); | |
bec1f6f6 | 867 | return -EIO; |
0f149c9f | 868 | } |
bec1f6f6 WB |
869 | |
870 | skb_shinfo(skb)->gso_size = cork->gso_size; | |
871 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; | |
dfec0ee2 AD |
872 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), |
873 | cork->gso_size); | |
a8c744a8 | 874 | goto csum_partial; |
bec1f6f6 WB |
875 | } |
876 | ||
db8dac20 | 877 | if (is_udplite) /* UDP-Lite */ |
f6b9664f | 878 | csum = udplite_csum(skb); |
db8dac20 | 879 | |
ab2fb7e3 | 880 | else if (sk->sk_no_check_tx) { /* UDP csum off */ |
db8dac20 DM |
881 | |
882 | skb->ip_summed = CHECKSUM_NONE; | |
883 | goto send; | |
884 | ||
885 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
a8c744a8 | 886 | csum_partial: |
db8dac20 | 887 | |
79ab0531 | 888 | udp4_hwcsum(skb, fl4->saddr, fl4->daddr); |
db8dac20 DM |
889 | goto send; |
890 | ||
f6b9664f HX |
891 | } else |
892 | csum = udp_csum(skb); | |
db8dac20 DM |
893 | |
894 | /* add protocol-dependent pseudo-header */ | |
79ab0531 | 895 | uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, |
c482c568 | 896 | sk->sk_protocol, csum); |
db8dac20 DM |
897 | if (uh->check == 0) |
898 | uh->check = CSUM_MANGLED_0; | |
899 | ||
900 | send: | |
b5ec8eea | 901 | err = ip_send_skb(sock_net(sk), skb); |
6ce9e7b5 ED |
902 | if (err) { |
903 | if (err == -ENOBUFS && !inet->recverr) { | |
6aef70a8 ED |
904 | UDP_INC_STATS(sock_net(sk), |
905 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
6ce9e7b5 ED |
906 | err = 0; |
907 | } | |
908 | } else | |
6aef70a8 ED |
909 | UDP_INC_STATS(sock_net(sk), |
910 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
f6b9664f HX |
911 | return err; |
912 | } | |
913 | ||
914 | /* | |
915 | * Push out all pending data as one UDP datagram. Socket is locked. | |
916 | */ | |
8822b64a | 917 | int udp_push_pending_frames(struct sock *sk) |
f6b9664f HX |
918 | { |
919 | struct udp_sock *up = udp_sk(sk); | |
920 | struct inet_sock *inet = inet_sk(sk); | |
b6f21b26 | 921 | struct flowi4 *fl4 = &inet->cork.fl.u.ip4; |
f6b9664f HX |
922 | struct sk_buff *skb; |
923 | int err = 0; | |
924 | ||
77968b78 | 925 | skb = ip_finish_skb(sk, fl4); |
f6b9664f HX |
926 | if (!skb) |
927 | goto out; | |
928 | ||
bec1f6f6 | 929 | err = udp_send_skb(skb, fl4, &inet->cork.base); |
f6b9664f | 930 | |
db8dac20 DM |
931 | out: |
932 | up->len = 0; | |
933 | up->pending = 0; | |
db8dac20 DM |
934 | return err; |
935 | } | |
8822b64a | 936 | EXPORT_SYMBOL(udp_push_pending_frames); |
db8dac20 | 937 | |
2e8de857 WB |
938 | static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) |
939 | { | |
940 | switch (cmsg->cmsg_type) { | |
941 | case UDP_SEGMENT: | |
942 | if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) | |
943 | return -EINVAL; | |
944 | *gso_size = *(__u16 *)CMSG_DATA(cmsg); | |
945 | return 0; | |
946 | default: | |
947 | return -EINVAL; | |
948 | } | |
949 | } | |
950 | ||
951 | int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) | |
952 | { | |
953 | struct cmsghdr *cmsg; | |
954 | bool need_ip = false; | |
955 | int err; | |
956 | ||
957 | for_each_cmsghdr(cmsg, msg) { | |
958 | if (!CMSG_OK(msg, cmsg)) | |
959 | return -EINVAL; | |
960 | ||
961 | if (cmsg->cmsg_level != SOL_UDP) { | |
962 | need_ip = true; | |
963 | continue; | |
964 | } | |
965 | ||
966 | err = __udp_cmsg_send(cmsg, gso_size); | |
967 | if (err) | |
968 | return err; | |
969 | } | |
970 | ||
971 | return need_ip; | |
972 | } | |
973 | EXPORT_SYMBOL_GPL(udp_cmsg_send); | |
974 | ||
1b784140 | 975 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
db8dac20 DM |
976 | { |
977 | struct inet_sock *inet = inet_sk(sk); | |
978 | struct udp_sock *up = udp_sk(sk); | |
1cedee13 | 979 | DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); |
e474995f | 980 | struct flowi4 fl4_stack; |
b6f21b26 | 981 | struct flowi4 *fl4; |
db8dac20 DM |
982 | int ulen = len; |
983 | struct ipcm_cookie ipc; | |
984 | struct rtable *rt = NULL; | |
985 | int free = 0; | |
986 | int connected = 0; | |
987 | __be32 daddr, faddr, saddr; | |
988 | __be16 dport; | |
989 | u8 tos; | |
990 | int err, is_udplite = IS_UDPLITE(sk); | |
991 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
992 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
903ab86d | 993 | struct sk_buff *skb; |
f6d8bd05 | 994 | struct ip_options_data opt_copy; |
db8dac20 DM |
995 | |
996 | if (len > 0xFFFF) | |
997 | return -EMSGSIZE; | |
998 | ||
999 | /* | |
1000 | * Check the flags. | |
1001 | */ | |
1002 | ||
c482c568 | 1003 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
1004 | return -EOPNOTSUPP; |
1005 | ||
903ab86d HX |
1006 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
1007 | ||
f5fca608 | 1008 | fl4 = &inet->cork.fl.u.ip4; |
db8dac20 DM |
1009 | if (up->pending) { |
1010 | /* | |
1011 | * There are pending frames. | |
1012 | * The socket lock must be held while it's corked. | |
1013 | */ | |
1014 | lock_sock(sk); | |
1015 | if (likely(up->pending)) { | |
1016 | if (unlikely(up->pending != AF_INET)) { | |
1017 | release_sock(sk); | |
1018 | return -EINVAL; | |
1019 | } | |
1020 | goto do_append_data; | |
1021 | } | |
1022 | release_sock(sk); | |
1023 | } | |
1024 | ulen += sizeof(struct udphdr); | |
1025 | ||
1026 | /* | |
1027 | * Get and verify the address. | |
1028 | */ | |
1cedee13 | 1029 | if (usin) { |
db8dac20 DM |
1030 | if (msg->msg_namelen < sizeof(*usin)) |
1031 | return -EINVAL; | |
1032 | if (usin->sin_family != AF_INET) { | |
1033 | if (usin->sin_family != AF_UNSPEC) | |
1034 | return -EAFNOSUPPORT; | |
1035 | } | |
1036 | ||
1037 | daddr = usin->sin_addr.s_addr; | |
1038 | dport = usin->sin_port; | |
1039 | if (dport == 0) | |
1040 | return -EINVAL; | |
1041 | } else { | |
1042 | if (sk->sk_state != TCP_ESTABLISHED) | |
1043 | return -EDESTADDRREQ; | |
c720c7e8 ED |
1044 | daddr = inet->inet_daddr; |
1045 | dport = inet->inet_dport; | |
db8dac20 DM |
1046 | /* Open fast path for connected socket. |
1047 | Route will not be used, if at least one option is set. | |
1048 | */ | |
1049 | connected = 1; | |
1050 | } | |
db8dac20 | 1051 | |
35178206 | 1052 | ipcm_init_sk(&ipc, inet); |
bec1f6f6 | 1053 | ipc.gso_size = up->gso_size; |
bf84a010 | 1054 | |
db8dac20 | 1055 | if (msg->msg_controllen) { |
2e8de857 WB |
1056 | err = udp_cmsg_send(sk, msg, &ipc.gso_size); |
1057 | if (err > 0) | |
1058 | err = ip_cmsg_send(sk, msg, &ipc, | |
1059 | sk->sk_family == AF_INET6); | |
1060 | if (unlikely(err < 0)) { | |
91948309 | 1061 | kfree(ipc.opt); |
db8dac20 | 1062 | return err; |
91948309 | 1063 | } |
db8dac20 DM |
1064 | if (ipc.opt) |
1065 | free = 1; | |
1066 | connected = 0; | |
1067 | } | |
f6d8bd05 ED |
1068 | if (!ipc.opt) { |
1069 | struct ip_options_rcu *inet_opt; | |
1070 | ||
1071 | rcu_read_lock(); | |
1072 | inet_opt = rcu_dereference(inet->inet_opt); | |
1073 | if (inet_opt) { | |
1074 | memcpy(&opt_copy, inet_opt, | |
1075 | sizeof(*inet_opt) + inet_opt->opt.optlen); | |
1076 | ipc.opt = &opt_copy.opt; | |
1077 | } | |
1078 | rcu_read_unlock(); | |
1079 | } | |
db8dac20 | 1080 | |
1cedee13 AI |
1081 | if (cgroup_bpf_enabled && !connected) { |
1082 | err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, | |
1083 | (struct sockaddr *)usin, &ipc.addr); | |
1084 | if (err) | |
1085 | goto out_free; | |
1086 | if (usin) { | |
1087 | if (usin->sin_port == 0) { | |
1088 | /* BPF program set invalid port. Reject it. */ | |
1089 | err = -EINVAL; | |
1090 | goto out_free; | |
1091 | } | |
1092 | daddr = usin->sin_addr.s_addr; | |
1093 | dport = usin->sin_port; | |
1094 | } | |
1095 | } | |
1096 | ||
db8dac20 DM |
1097 | saddr = ipc.addr; |
1098 | ipc.addr = faddr = daddr; | |
1099 | ||
f6d8bd05 | 1100 | if (ipc.opt && ipc.opt->opt.srr) { |
1b97013b AI |
1101 | if (!daddr) { |
1102 | err = -EINVAL; | |
1103 | goto out_free; | |
1104 | } | |
f6d8bd05 | 1105 | faddr = ipc.opt->opt.faddr; |
db8dac20 DM |
1106 | connected = 0; |
1107 | } | |
aa661581 | 1108 | tos = get_rttos(&ipc, inet); |
db8dac20 DM |
1109 | if (sock_flag(sk, SOCK_LOCALROUTE) || |
1110 | (msg->msg_flags & MSG_DONTROUTE) || | |
f6d8bd05 | 1111 | (ipc.opt && ipc.opt->opt.is_strictroute)) { |
db8dac20 DM |
1112 | tos |= RTO_ONLINK; |
1113 | connected = 0; | |
1114 | } | |
1115 | ||
1116 | if (ipv4_is_multicast(daddr)) { | |
854da991 | 1117 | if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) |
db8dac20 DM |
1118 | ipc.oif = inet->mc_index; |
1119 | if (!saddr) | |
1120 | saddr = inet->mc_addr; | |
1121 | connected = 0; | |
9515a2e0 | 1122 | } else if (!ipc.oif) { |
76e21053 | 1123 | ipc.oif = inet->uc_index; |
9515a2e0 DA |
1124 | } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { |
1125 | /* oif is set, packet is to local broadcast and | |
1126 | * and uc_index is set. oif is most likely set | |
1127 | * by sk_bound_dev_if. If uc_index != oif check if the | |
1128 | * oif is an L3 master and uc_index is an L3 slave. | |
1129 | * If so, we want to allow the send using the uc_index. | |
1130 | */ | |
1131 | if (ipc.oif != inet->uc_index && | |
1132 | ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), | |
1133 | inet->uc_index)) { | |
1134 | ipc.oif = inet->uc_index; | |
1135 | } | |
1136 | } | |
db8dac20 DM |
1137 | |
1138 | if (connected) | |
c482c568 | 1139 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 | 1140 | |
51456b29 | 1141 | if (!rt) { |
84a3aa00 | 1142 | struct net *net = sock_net(sk); |
9a24abfa | 1143 | __u8 flow_flags = inet_sk_flowi_flags(sk); |
84a3aa00 | 1144 | |
e474995f | 1145 | fl4 = &fl4_stack; |
9a24abfa | 1146 | |
e474995f | 1147 | flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, |
c0951cbc | 1148 | RT_SCOPE_UNIVERSE, sk->sk_protocol, |
9a24abfa | 1149 | flow_flags, |
e2d118a1 LC |
1150 | faddr, saddr, dport, inet->inet_sport, |
1151 | sk->sk_uid); | |
c0951cbc | 1152 | |
e474995f DM |
1153 | security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); |
1154 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe DM |
1155 | if (IS_ERR(rt)) { |
1156 | err = PTR_ERR(rt); | |
06dc94b1 | 1157 | rt = NULL; |
db8dac20 | 1158 | if (err == -ENETUNREACH) |
f1d8cba6 | 1159 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
1160 | goto out; |
1161 | } | |
1162 | ||
1163 | err = -EACCES; | |
1164 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
1165 | !sock_flag(sk, SOCK_BROADCAST)) | |
1166 | goto out; | |
1167 | if (connected) | |
d8d1f30b | 1168 | sk_dst_set(sk, dst_clone(&rt->dst)); |
db8dac20 DM |
1169 | } |
1170 | ||
1171 | if (msg->msg_flags&MSG_CONFIRM) | |
1172 | goto do_confirm; | |
1173 | back_from_confirm: | |
1174 | ||
e474995f | 1175 | saddr = fl4->saddr; |
db8dac20 | 1176 | if (!ipc.addr) |
e474995f | 1177 | daddr = ipc.addr = fl4->daddr; |
db8dac20 | 1178 | |
903ab86d HX |
1179 | /* Lockless fast path for the non-corking case. */ |
1180 | if (!corkreq) { | |
1cd7884d WB |
1181 | struct inet_cork cork; |
1182 | ||
f69e6d13 | 1183 | skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, |
903ab86d | 1184 | sizeof(struct udphdr), &ipc, &rt, |
1cd7884d | 1185 | &cork, msg->msg_flags); |
903ab86d | 1186 | err = PTR_ERR(skb); |
50c3a487 | 1187 | if (!IS_ERR_OR_NULL(skb)) |
bec1f6f6 | 1188 | err = udp_send_skb(skb, fl4, &cork); |
903ab86d HX |
1189 | goto out; |
1190 | } | |
1191 | ||
db8dac20 DM |
1192 | lock_sock(sk); |
1193 | if (unlikely(up->pending)) { | |
1194 | /* The socket is already corked while preparing it. */ | |
1195 | /* ... which is an evident application bug. --ANK */ | |
1196 | release_sock(sk); | |
1197 | ||
197df02c | 1198 | net_dbg_ratelimited("socket already corked\n"); |
db8dac20 DM |
1199 | err = -EINVAL; |
1200 | goto out; | |
1201 | } | |
1202 | /* | |
1203 | * Now cork the socket to pend data. | |
1204 | */ | |
b6f21b26 DM |
1205 | fl4 = &inet->cork.fl.u.ip4; |
1206 | fl4->daddr = daddr; | |
1207 | fl4->saddr = saddr; | |
9cce96df DM |
1208 | fl4->fl4_dport = dport; |
1209 | fl4->fl4_sport = inet->inet_sport; | |
db8dac20 DM |
1210 | up->pending = AF_INET; |
1211 | ||
1212 | do_append_data: | |
1213 | up->len += ulen; | |
f69e6d13 | 1214 | err = ip_append_data(sk, fl4, getfrag, msg, ulen, |
f5fca608 DM |
1215 | sizeof(struct udphdr), &ipc, &rt, |
1216 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); | |
db8dac20 DM |
1217 | if (err) |
1218 | udp_flush_pending_frames(sk); | |
1219 | else if (!corkreq) | |
1220 | err = udp_push_pending_frames(sk); | |
1221 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1222 | up->pending = 0; | |
1223 | release_sock(sk); | |
1224 | ||
1225 | out: | |
1226 | ip_rt_put(rt); | |
1b97013b | 1227 | out_free: |
db8dac20 DM |
1228 | if (free) |
1229 | kfree(ipc.opt); | |
1230 | if (!err) | |
1231 | return len; | |
1232 | /* | |
1233 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1234 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1235 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1236 | * things). We could add another new stat but at least for now that | |
1237 | * seems like overkill. | |
1238 | */ | |
1239 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
6aef70a8 ED |
1240 | UDP_INC_STATS(sock_net(sk), |
1241 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
1242 | } |
1243 | return err; | |
1244 | ||
1245 | do_confirm: | |
0dec879f JA |
1246 | if (msg->msg_flags & MSG_PROBE) |
1247 | dst_confirm_neigh(&rt->dst, &fl4->daddr); | |
db8dac20 DM |
1248 | if (!(msg->msg_flags&MSG_PROBE) || len) |
1249 | goto back_from_confirm; | |
1250 | err = 0; | |
1251 | goto out; | |
1252 | } | |
c482c568 | 1253 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
1254 | |
1255 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
1256 | size_t size, int flags) | |
1257 | { | |
f5fca608 | 1258 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
1259 | struct udp_sock *up = udp_sk(sk); |
1260 | int ret; | |
1261 | ||
d3f7d56a SL |
1262 | if (flags & MSG_SENDPAGE_NOTLAST) |
1263 | flags |= MSG_MORE; | |
1264 | ||
db8dac20 DM |
1265 | if (!up->pending) { |
1266 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
1267 | ||
1268 | /* Call udp_sendmsg to specify destination address which | |
1269 | * sendpage interface can't pass. | |
1270 | * This will succeed only when the socket is connected. | |
1271 | */ | |
1b784140 | 1272 | ret = udp_sendmsg(sk, &msg, 0); |
db8dac20 DM |
1273 | if (ret < 0) |
1274 | return ret; | |
1275 | } | |
1276 | ||
1277 | lock_sock(sk); | |
1278 | ||
1279 | if (unlikely(!up->pending)) { | |
1280 | release_sock(sk); | |
1281 | ||
197df02c | 1282 | net_dbg_ratelimited("cork failed\n"); |
db8dac20 DM |
1283 | return -EINVAL; |
1284 | } | |
1285 | ||
f5fca608 DM |
1286 | ret = ip_append_page(sk, &inet->cork.fl.u.ip4, |
1287 | page, offset, size, flags); | |
db8dac20 DM |
1288 | if (ret == -EOPNOTSUPP) { |
1289 | release_sock(sk); | |
1290 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
1291 | size, flags); | |
1292 | } | |
1293 | if (ret < 0) { | |
1294 | udp_flush_pending_frames(sk); | |
1295 | goto out; | |
1296 | } | |
1297 | ||
1298 | up->len += size; | |
1299 | if (!(up->corkflag || (flags&MSG_MORE))) | |
1300 | ret = udp_push_pending_frames(sk); | |
1301 | if (!ret) | |
1302 | ret = size; | |
1303 | out: | |
1304 | release_sock(sk); | |
1305 | return ret; | |
1306 | } | |
1307 | ||
dce4551c PA |
1308 | #define UDP_SKB_IS_STATELESS 0x80000000 |
1309 | ||
b65ac446 PA |
1310 | static void udp_set_dev_scratch(struct sk_buff *skb) |
1311 | { | |
dce4551c | 1312 | struct udp_dev_scratch *scratch = udp_skb_scratch(skb); |
b65ac446 PA |
1313 | |
1314 | BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); | |
dce4551c PA |
1315 | scratch->_tsize_state = skb->truesize; |
1316 | #if BITS_PER_LONG == 64 | |
b65ac446 PA |
1317 | scratch->len = skb->len; |
1318 | scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); | |
1319 | scratch->is_linear = !skb_is_nonlinear(skb); | |
dce4551c | 1320 | #endif |
3bdefdf9 PA |
1321 | /* all head states execept sp (dst, sk, nf) are always cleared by |
1322 | * udp_rcv() and we need to preserve secpath, if present, to eventually | |
1323 | * process IP_CMSG_PASSSEC at recvmsg() time | |
1324 | */ | |
1325 | if (likely(!skb_sec_path(skb))) | |
dce4551c | 1326 | scratch->_tsize_state |= UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1327 | } |
1328 | ||
1329 | static int udp_skb_truesize(struct sk_buff *skb) | |
1330 | { | |
dce4551c | 1331 | return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1332 | } |
1333 | ||
dce4551c | 1334 | static bool udp_skb_has_head_state(struct sk_buff *skb) |
b65ac446 | 1335 | { |
dce4551c | 1336 | return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); |
b65ac446 | 1337 | } |
b65ac446 | 1338 | |
7c13f97f | 1339 | /* fully reclaim rmem/fwd memory allocated for skb */ |
6dfb4367 PA |
1340 | static void udp_rmem_release(struct sock *sk, int size, int partial, |
1341 | bool rx_queue_lock_held) | |
f970bd9e | 1342 | { |
6b229cf7 | 1343 | struct udp_sock *up = udp_sk(sk); |
2276f58a | 1344 | struct sk_buff_head *sk_queue; |
f970bd9e PA |
1345 | int amt; |
1346 | ||
6b229cf7 ED |
1347 | if (likely(partial)) { |
1348 | up->forward_deficit += size; | |
1349 | size = up->forward_deficit; | |
0d4a6608 | 1350 | if (size < (sk->sk_rcvbuf >> 2)) |
6b229cf7 ED |
1351 | return; |
1352 | } else { | |
1353 | size += up->forward_deficit; | |
1354 | } | |
1355 | up->forward_deficit = 0; | |
1356 | ||
6dfb4367 PA |
1357 | /* acquire the sk_receive_queue for fwd allocated memory scheduling, |
1358 | * if the called don't held it already | |
1359 | */ | |
2276f58a | 1360 | sk_queue = &sk->sk_receive_queue; |
6dfb4367 PA |
1361 | if (!rx_queue_lock_held) |
1362 | spin_lock(&sk_queue->lock); | |
1363 | ||
2276f58a | 1364 | |
f970bd9e PA |
1365 | sk->sk_forward_alloc += size; |
1366 | amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); | |
1367 | sk->sk_forward_alloc -= amt; | |
f970bd9e PA |
1368 | |
1369 | if (amt) | |
1370 | __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); | |
02ab0d13 ED |
1371 | |
1372 | atomic_sub(size, &sk->sk_rmem_alloc); | |
2276f58a PA |
1373 | |
1374 | /* this can save us from acquiring the rx queue lock on next receive */ | |
1375 | skb_queue_splice_tail_init(sk_queue, &up->reader_queue); | |
1376 | ||
6dfb4367 PA |
1377 | if (!rx_queue_lock_held) |
1378 | spin_unlock(&sk_queue->lock); | |
f970bd9e PA |
1379 | } |
1380 | ||
2276f58a | 1381 | /* Note: called with reader_queue.lock held. |
c84d9490 ED |
1382 | * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch |
1383 | * This avoids a cache line miss while receive_queue lock is held. | |
1384 | * Look at __udp_enqueue_schedule_skb() to find where this copy is done. | |
1385 | */ | |
7c13f97f | 1386 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) |
f970bd9e | 1387 | { |
b65ac446 PA |
1388 | prefetch(&skb->data); |
1389 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); | |
f970bd9e | 1390 | } |
7c13f97f | 1391 | EXPORT_SYMBOL(udp_skb_destructor); |
f970bd9e | 1392 | |
6dfb4367 | 1393 | /* as above, but the caller held the rx queue lock, too */ |
64f5102d | 1394 | static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) |
6dfb4367 | 1395 | { |
b65ac446 PA |
1396 | prefetch(&skb->data); |
1397 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); | |
6dfb4367 PA |
1398 | } |
1399 | ||
4b272750 ED |
1400 | /* Idea of busylocks is to let producers grab an extra spinlock |
1401 | * to relieve pressure on the receive_queue spinlock shared by consumer. | |
1402 | * Under flood, this means that only one producer can be in line | |
1403 | * trying to acquire the receive_queue spinlock. | |
1404 | * These busylock can be allocated on a per cpu manner, instead of a | |
1405 | * per socket one (that would consume a cache line per socket) | |
1406 | */ | |
1407 | static int udp_busylocks_log __read_mostly; | |
1408 | static spinlock_t *udp_busylocks __read_mostly; | |
1409 | ||
1410 | static spinlock_t *busylock_acquire(void *ptr) | |
1411 | { | |
1412 | spinlock_t *busy; | |
1413 | ||
1414 | busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); | |
1415 | spin_lock(busy); | |
1416 | return busy; | |
1417 | } | |
1418 | ||
1419 | static void busylock_release(spinlock_t *busy) | |
1420 | { | |
1421 | if (busy) | |
1422 | spin_unlock(busy); | |
1423 | } | |
1424 | ||
f970bd9e PA |
1425 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) |
1426 | { | |
1427 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
1428 | int rmem, delta, amt, err = -ENOMEM; | |
4b272750 | 1429 | spinlock_t *busy = NULL; |
c8c8b127 | 1430 | int size; |
f970bd9e PA |
1431 | |
1432 | /* try to avoid the costly atomic add/sub pair when the receive | |
1433 | * queue is full; always allow at least a packet | |
1434 | */ | |
1435 | rmem = atomic_read(&sk->sk_rmem_alloc); | |
363dc73a | 1436 | if (rmem > sk->sk_rcvbuf) |
f970bd9e PA |
1437 | goto drop; |
1438 | ||
c8c8b127 ED |
1439 | /* Under mem pressure, it might be helpful to help udp_recvmsg() |
1440 | * having linear skbs : | |
1441 | * - Reduce memory overhead and thus increase receive queue capacity | |
1442 | * - Less cache line misses at copyout() time | |
1443 | * - Less work at consume_skb() (less alien page frag freeing) | |
1444 | */ | |
4b272750 | 1445 | if (rmem > (sk->sk_rcvbuf >> 1)) { |
c8c8b127 | 1446 | skb_condense(skb); |
4b272750 ED |
1447 | |
1448 | busy = busylock_acquire(sk); | |
1449 | } | |
c8c8b127 | 1450 | size = skb->truesize; |
b65ac446 | 1451 | udp_set_dev_scratch(skb); |
c8c8b127 | 1452 | |
f970bd9e PA |
1453 | /* we drop only if the receive buf is full and the receive |
1454 | * queue contains some other skb | |
1455 | */ | |
1456 | rmem = atomic_add_return(size, &sk->sk_rmem_alloc); | |
363dc73a | 1457 | if (rmem > (size + sk->sk_rcvbuf)) |
f970bd9e PA |
1458 | goto uncharge_drop; |
1459 | ||
1460 | spin_lock(&list->lock); | |
1461 | if (size >= sk->sk_forward_alloc) { | |
1462 | amt = sk_mem_pages(size); | |
1463 | delta = amt << SK_MEM_QUANTUM_SHIFT; | |
1464 | if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { | |
1465 | err = -ENOBUFS; | |
1466 | spin_unlock(&list->lock); | |
1467 | goto uncharge_drop; | |
1468 | } | |
1469 | ||
1470 | sk->sk_forward_alloc += delta; | |
1471 | } | |
1472 | ||
1473 | sk->sk_forward_alloc -= size; | |
1474 | ||
7c13f97f PA |
1475 | /* no need to setup a destructor, we will explicitly release the |
1476 | * forward allocated memory on dequeue | |
1477 | */ | |
f970bd9e PA |
1478 | sock_skb_set_dropcount(sk, skb); |
1479 | ||
1480 | __skb_queue_tail(list, skb); | |
1481 | spin_unlock(&list->lock); | |
1482 | ||
1483 | if (!sock_flag(sk, SOCK_DEAD)) | |
1484 | sk->sk_data_ready(sk); | |
1485 | ||
4b272750 | 1486 | busylock_release(busy); |
f970bd9e PA |
1487 | return 0; |
1488 | ||
1489 | uncharge_drop: | |
1490 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); | |
1491 | ||
1492 | drop: | |
1493 | atomic_inc(&sk->sk_drops); | |
4b272750 | 1494 | busylock_release(busy); |
f970bd9e PA |
1495 | return err; |
1496 | } | |
1497 | EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); | |
1498 | ||
c915fe13 | 1499 | void udp_destruct_sock(struct sock *sk) |
f970bd9e PA |
1500 | { |
1501 | /* reclaim completely the forward allocated memory */ | |
2276f58a | 1502 | struct udp_sock *up = udp_sk(sk); |
7c13f97f PA |
1503 | unsigned int total = 0; |
1504 | struct sk_buff *skb; | |
1505 | ||
2276f58a PA |
1506 | skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); |
1507 | while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { | |
7c13f97f PA |
1508 | total += skb->truesize; |
1509 | kfree_skb(skb); | |
1510 | } | |
6dfb4367 | 1511 | udp_rmem_release(sk, total, 0, true); |
7c13f97f | 1512 | |
f970bd9e PA |
1513 | inet_sock_destruct(sk); |
1514 | } | |
c915fe13 | 1515 | EXPORT_SYMBOL_GPL(udp_destruct_sock); |
f970bd9e PA |
1516 | |
1517 | int udp_init_sock(struct sock *sk) | |
1518 | { | |
2276f58a | 1519 | skb_queue_head_init(&udp_sk(sk)->reader_queue); |
f970bd9e PA |
1520 | sk->sk_destruct = udp_destruct_sock; |
1521 | return 0; | |
1522 | } | |
1523 | EXPORT_SYMBOL_GPL(udp_init_sock); | |
1524 | ||
1525 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) | |
1526 | { | |
1527 | if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { | |
1528 | bool slow = lock_sock_fast(sk); | |
1529 | ||
1530 | sk_peek_offset_bwd(sk, len); | |
1531 | unlock_sock_fast(sk, slow); | |
1532 | } | |
0a463c78 | 1533 | |
ca2c1418 PA |
1534 | if (!skb_unref(skb)) |
1535 | return; | |
1536 | ||
dce4551c PA |
1537 | /* In the more common cases we cleared the head states previously, |
1538 | * see __udp_queue_rcv_skb(). | |
0ddf3fb2 | 1539 | */ |
dce4551c | 1540 | if (unlikely(udp_skb_has_head_state(skb))) |
0ddf3fb2 | 1541 | skb_release_head_state(skb); |
ca2c1418 | 1542 | __consume_stateless_skb(skb); |
f970bd9e PA |
1543 | } |
1544 | EXPORT_SYMBOL_GPL(skb_consume_udp); | |
1545 | ||
2276f58a PA |
1546 | static struct sk_buff *__first_packet_length(struct sock *sk, |
1547 | struct sk_buff_head *rcvq, | |
1548 | int *total) | |
1549 | { | |
1550 | struct sk_buff *skb; | |
1551 | ||
9bd780f5 PA |
1552 | while ((skb = skb_peek(rcvq)) != NULL) { |
1553 | if (udp_lib_checksum_complete(skb)) { | |
1554 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1555 | IS_UDPLITE(sk)); | |
1556 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1557 | IS_UDPLITE(sk)); | |
1558 | atomic_inc(&sk->sk_drops); | |
1559 | __skb_unlink(skb, rcvq); | |
1560 | *total += skb->truesize; | |
1561 | kfree_skb(skb); | |
1562 | } else { | |
1563 | /* the csum related bits could be changed, refresh | |
1564 | * the scratch area | |
1565 | */ | |
1566 | udp_set_dev_scratch(skb); | |
1567 | break; | |
1568 | } | |
2276f58a PA |
1569 | } |
1570 | return skb; | |
1571 | } | |
1572 | ||
85584672 ED |
1573 | /** |
1574 | * first_packet_length - return length of first packet in receive queue | |
1575 | * @sk: socket | |
1576 | * | |
1577 | * Drops all bad checksum frames, until a valid one is found. | |
e83c6744 | 1578 | * Returns the length of found skb, or -1 if none is found. |
85584672 | 1579 | */ |
e83c6744 | 1580 | static int first_packet_length(struct sock *sk) |
85584672 | 1581 | { |
2276f58a PA |
1582 | struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; |
1583 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
85584672 | 1584 | struct sk_buff *skb; |
7c13f97f | 1585 | int total = 0; |
e83c6744 | 1586 | int res; |
85584672 | 1587 | |
85584672 | 1588 | spin_lock_bh(&rcvq->lock); |
2276f58a PA |
1589 | skb = __first_packet_length(sk, rcvq, &total); |
1590 | if (!skb && !skb_queue_empty(sk_queue)) { | |
1591 | spin_lock(&sk_queue->lock); | |
1592 | skb_queue_splice_tail_init(sk_queue, rcvq); | |
1593 | spin_unlock(&sk_queue->lock); | |
1594 | ||
1595 | skb = __first_packet_length(sk, rcvq, &total); | |
85584672 | 1596 | } |
e83c6744 | 1597 | res = skb ? skb->len : -1; |
7c13f97f | 1598 | if (total) |
6dfb4367 | 1599 | udp_rmem_release(sk, total, 1, false); |
85584672 | 1600 | spin_unlock_bh(&rcvq->lock); |
85584672 ED |
1601 | return res; |
1602 | } | |
1603 | ||
1da177e4 LT |
1604 | /* |
1605 | * IOCTL requests applicable to the UDP protocol | |
1606 | */ | |
e905a9ed | 1607 | |
1da177e4 LT |
1608 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
1609 | { | |
6516c655 SH |
1610 | switch (cmd) { |
1611 | case SIOCOUTQ: | |
1da177e4 | 1612 | { |
31e6d363 ED |
1613 | int amount = sk_wmem_alloc_get(sk); |
1614 | ||
6516c655 SH |
1615 | return put_user(amount, (int __user *)arg); |
1616 | } | |
1da177e4 | 1617 | |
6516c655 SH |
1618 | case SIOCINQ: |
1619 | { | |
e83c6744 | 1620 | int amount = max_t(int, 0, first_packet_length(sk)); |
6516c655 | 1621 | |
6516c655 SH |
1622 | return put_user(amount, (int __user *)arg); |
1623 | } | |
1da177e4 | 1624 | |
6516c655 SH |
1625 | default: |
1626 | return -ENOIOCTLCMD; | |
1da177e4 | 1627 | } |
6516c655 SH |
1628 | |
1629 | return 0; | |
1da177e4 | 1630 | } |
c482c568 | 1631 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 1632 | |
2276f58a | 1633 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, |
fd69c399 | 1634 | int noblock, int *off, int *err) |
2276f58a PA |
1635 | { |
1636 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
1637 | struct sk_buff_head *queue; | |
1638 | struct sk_buff *last; | |
1639 | long timeo; | |
1640 | int error; | |
1641 | ||
1642 | queue = &udp_sk(sk)->reader_queue; | |
1643 | flags |= noblock ? MSG_DONTWAIT : 0; | |
1644 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1645 | do { | |
1646 | struct sk_buff *skb; | |
1647 | ||
1648 | error = sock_error(sk); | |
1649 | if (error) | |
1650 | break; | |
1651 | ||
1652 | error = -EAGAIN; | |
2276f58a | 1653 | do { |
2276f58a PA |
1654 | spin_lock_bh(&queue->lock); |
1655 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
1656 | udp_skb_destructor, | |
fd69c399 | 1657 | off, err, &last); |
2276f58a PA |
1658 | if (skb) { |
1659 | spin_unlock_bh(&queue->lock); | |
2276f58a PA |
1660 | return skb; |
1661 | } | |
1662 | ||
1663 | if (skb_queue_empty(sk_queue)) { | |
1664 | spin_unlock_bh(&queue->lock); | |
1665 | goto busy_check; | |
1666 | } | |
1667 | ||
6dfb4367 PA |
1668 | /* refill the reader queue and walk it again |
1669 | * keep both queues locked to avoid re-acquiring | |
1670 | * the sk_receive_queue lock if fwd memory scheduling | |
1671 | * is needed. | |
1672 | */ | |
2276f58a PA |
1673 | spin_lock(&sk_queue->lock); |
1674 | skb_queue_splice_tail_init(sk_queue, queue); | |
2276f58a PA |
1675 | |
1676 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
6dfb4367 | 1677 | udp_skb_dtor_locked, |
fd69c399 | 1678 | off, err, &last); |
6dfb4367 | 1679 | spin_unlock(&sk_queue->lock); |
2276f58a | 1680 | spin_unlock_bh(&queue->lock); |
de321ed3 | 1681 | if (skb) |
2276f58a | 1682 | return skb; |
2276f58a PA |
1683 | |
1684 | busy_check: | |
1685 | if (!sk_can_busy_loop(sk)) | |
1686 | break; | |
1687 | ||
1688 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
1689 | } while (!skb_queue_empty(sk_queue)); | |
1690 | ||
1691 | /* sk_queue is empty, reader_queue may contain peeked packets */ | |
1692 | } while (timeo && | |
1693 | !__skb_wait_for_more_packets(sk, &error, &timeo, | |
1694 | (struct sk_buff *)sk_queue)); | |
1695 | ||
1696 | *err = error; | |
1697 | return NULL; | |
1698 | } | |
7e823644 | 1699 | EXPORT_SYMBOL(__skb_recv_udp); |
2276f58a | 1700 | |
db8dac20 DM |
1701 | /* |
1702 | * This should be easy, if there is something there we | |
1703 | * return it, otherwise we block. | |
1704 | */ | |
1705 | ||
1b784140 YX |
1706 | int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, |
1707 | int flags, int *addr_len) | |
db8dac20 DM |
1708 | { |
1709 | struct inet_sock *inet = inet_sk(sk); | |
342dfc30 | 1710 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
db8dac20 | 1711 | struct sk_buff *skb; |
59c2cdae | 1712 | unsigned int ulen, copied; |
fd69c399 | 1713 | int off, err, peeking = flags & MSG_PEEK; |
db8dac20 | 1714 | int is_udplite = IS_UDPLITE(sk); |
197c949e | 1715 | bool checksum_valid = false; |
db8dac20 | 1716 | |
db8dac20 | 1717 | if (flags & MSG_ERRQUEUE) |
85fbaa75 | 1718 | return ip_recv_error(sk, msg, len, addr_len); |
db8dac20 DM |
1719 | |
1720 | try_again: | |
a0917e0b | 1721 | off = sk_peek_offset(sk, flags); |
fd69c399 | 1722 | skb = __skb_recv_udp(sk, flags, noblock, &off, &err); |
db8dac20 | 1723 | if (!skb) |
627d2d6b | 1724 | return err; |
db8dac20 | 1725 | |
b65ac446 | 1726 | ulen = udp_skb_len(skb); |
59c2cdae | 1727 | copied = len; |
627d2d6b | 1728 | if (copied > ulen - off) |
1729 | copied = ulen - off; | |
59c2cdae | 1730 | else if (copied < ulen) |
db8dac20 DM |
1731 | msg->msg_flags |= MSG_TRUNC; |
1732 | ||
1733 | /* | |
1734 | * If checksum is needed at all, try to do it while copying the | |
1735 | * data. If the data is truncated, or if we only want a partial | |
1736 | * coverage checksum (UDP-Lite), do it before the copy. | |
1737 | */ | |
1738 | ||
d21dbdfe ED |
1739 | if (copied < ulen || peeking || |
1740 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
b65ac446 PA |
1741 | checksum_valid = udp_skb_csum_unnecessary(skb) || |
1742 | !__udp_lib_checksum_complete(skb); | |
197c949e | 1743 | if (!checksum_valid) |
db8dac20 DM |
1744 | goto csum_copy_err; |
1745 | } | |
1746 | ||
b65ac446 PA |
1747 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { |
1748 | if (udp_skb_is_linear(skb)) | |
1749 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
1750 | else | |
1751 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
1752 | } else { | |
627d2d6b | 1753 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); |
db8dac20 DM |
1754 | |
1755 | if (err == -EINVAL) | |
1756 | goto csum_copy_err; | |
1757 | } | |
1758 | ||
22911fc5 | 1759 | if (unlikely(err)) { |
fd69c399 | 1760 | if (!peeking) { |
979402b1 | 1761 | atomic_inc(&sk->sk_drops); |
6aef70a8 ED |
1762 | UDP_INC_STATS(sock_net(sk), |
1763 | UDP_MIB_INERRORS, is_udplite); | |
979402b1 | 1764 | } |
850cbadd | 1765 | kfree_skb(skb); |
627d2d6b | 1766 | return err; |
22911fc5 | 1767 | } |
db8dac20 | 1768 | |
fd69c399 | 1769 | if (!peeking) |
6aef70a8 ED |
1770 | UDP_INC_STATS(sock_net(sk), |
1771 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 | 1772 | |
3b885787 | 1773 | sock_recv_ts_and_drops(msg, sk, skb); |
db8dac20 DM |
1774 | |
1775 | /* Copy the address. */ | |
c482c568 | 1776 | if (sin) { |
db8dac20 DM |
1777 | sin->sin_family = AF_INET; |
1778 | sin->sin_port = udp_hdr(skb)->source; | |
1779 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
1780 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
bceaa902 | 1781 | *addr_len = sizeof(*sin); |
db8dac20 | 1782 | } |
bcd1665e PA |
1783 | |
1784 | if (udp_sk(sk)->gro_enabled) | |
1785 | udp_cmsg_recv(msg, sk, skb); | |
1786 | ||
db8dac20 | 1787 | if (inet->cmsg_flags) |
ad959036 | 1788 | ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); |
db8dac20 | 1789 | |
59c2cdae | 1790 | err = copied; |
db8dac20 DM |
1791 | if (flags & MSG_TRUNC) |
1792 | err = ulen; | |
1793 | ||
850cbadd | 1794 | skb_consume_udp(sk, skb, peeking ? -err : err); |
db8dac20 DM |
1795 | return err; |
1796 | ||
1797 | csum_copy_err: | |
2276f58a PA |
1798 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, |
1799 | udp_skb_destructor)) { | |
6aef70a8 ED |
1800 | UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
1801 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
6a5dc9e5 | 1802 | } |
850cbadd | 1803 | kfree_skb(skb); |
db8dac20 | 1804 | |
beb39db5 ED |
1805 | /* starting over for a new packet, but check if we need to yield */ |
1806 | cond_resched(); | |
9cfaa8de | 1807 | msg->msg_flags &= ~MSG_TRUNC; |
db8dac20 DM |
1808 | goto try_again; |
1809 | } | |
1810 | ||
d74bad4e AI |
1811 | int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
1812 | { | |
1813 | /* This check is replicated from __ip4_datagram_connect() and | |
1814 | * intended to prevent BPF program called below from accessing bytes | |
1815 | * that are out of the bound specified by user in addr_len. | |
1816 | */ | |
1817 | if (addr_len < sizeof(struct sockaddr_in)) | |
1818 | return -EINVAL; | |
1819 | ||
1820 | return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); | |
1821 | } | |
1822 | EXPORT_SYMBOL(udp_pre_connect); | |
1823 | ||
286c72de | 1824 | int __udp_disconnect(struct sock *sk, int flags) |
1da177e4 LT |
1825 | { |
1826 | struct inet_sock *inet = inet_sk(sk); | |
1827 | /* | |
1828 | * 1003.1g - break association. | |
1829 | */ | |
e905a9ed | 1830 | |
1da177e4 | 1831 | sk->sk_state = TCP_CLOSE; |
c720c7e8 ED |
1832 | inet->inet_daddr = 0; |
1833 | inet->inet_dport = 0; | |
bdeab991 | 1834 | sock_rps_reset_rxhash(sk); |
1da177e4 LT |
1835 | sk->sk_bound_dev_if = 0; |
1836 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
1837 | inet_reset_saddr(sk); | |
1838 | ||
1839 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1840 | sk->sk_prot->unhash(sk); | |
c720c7e8 | 1841 | inet->inet_sport = 0; |
1da177e4 LT |
1842 | } |
1843 | sk_dst_reset(sk); | |
1844 | return 0; | |
1845 | } | |
286c72de ED |
1846 | EXPORT_SYMBOL(__udp_disconnect); |
1847 | ||
1848 | int udp_disconnect(struct sock *sk, int flags) | |
1849 | { | |
1850 | lock_sock(sk); | |
1851 | __udp_disconnect(sk, flags); | |
1852 | release_sock(sk); | |
1853 | return 0; | |
1854 | } | |
c482c568 | 1855 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 1856 | |
645ca708 ED |
1857 | void udp_lib_unhash(struct sock *sk) |
1858 | { | |
723b4610 ED |
1859 | if (sk_hashed(sk)) { |
1860 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
512615b6 ED |
1861 | struct udp_hslot *hslot, *hslot2; |
1862 | ||
1863 | hslot = udp_hashslot(udptable, sock_net(sk), | |
1864 | udp_sk(sk)->udp_port_hash); | |
1865 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
645ca708 | 1866 | |
723b4610 | 1867 | spin_lock_bh(&hslot->lock); |
e32ea7e7 CG |
1868 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1869 | reuseport_detach_sock(sk); | |
ca065d0c | 1870 | if (sk_del_node_init_rcu(sk)) { |
fdcc8aa9 | 1871 | hslot->count--; |
c720c7e8 | 1872 | inet_sk(sk)->inet_num = 0; |
723b4610 | 1873 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
512615b6 ED |
1874 | |
1875 | spin_lock(&hslot2->lock); | |
ca065d0c | 1876 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
512615b6 ED |
1877 | hslot2->count--; |
1878 | spin_unlock(&hslot2->lock); | |
723b4610 ED |
1879 | } |
1880 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 1881 | } |
645ca708 ED |
1882 | } |
1883 | EXPORT_SYMBOL(udp_lib_unhash); | |
1884 | ||
719f8358 ED |
1885 | /* |
1886 | * inet_rcv_saddr was changed, we must rehash secondary hash | |
1887 | */ | |
1888 | void udp_lib_rehash(struct sock *sk, u16 newhash) | |
1889 | { | |
1890 | if (sk_hashed(sk)) { | |
1891 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
1892 | struct udp_hslot *hslot, *hslot2, *nhslot2; | |
1893 | ||
1894 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
1895 | nhslot2 = udp_hashslot2(udptable, newhash); | |
1896 | udp_sk(sk)->udp_portaddr_hash = newhash; | |
e32ea7e7 CG |
1897 | |
1898 | if (hslot2 != nhslot2 || | |
1899 | rcu_access_pointer(sk->sk_reuseport_cb)) { | |
719f8358 ED |
1900 | hslot = udp_hashslot(udptable, sock_net(sk), |
1901 | udp_sk(sk)->udp_port_hash); | |
1902 | /* we must lock primary chain too */ | |
1903 | spin_lock_bh(&hslot->lock); | |
e32ea7e7 CG |
1904 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1905 | reuseport_detach_sock(sk); | |
1906 | ||
1907 | if (hslot2 != nhslot2) { | |
1908 | spin_lock(&hslot2->lock); | |
ca065d0c | 1909 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
e32ea7e7 CG |
1910 | hslot2->count--; |
1911 | spin_unlock(&hslot2->lock); | |
1912 | ||
1913 | spin_lock(&nhslot2->lock); | |
ca065d0c | 1914 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
e32ea7e7 CG |
1915 | &nhslot2->head); |
1916 | nhslot2->count++; | |
1917 | spin_unlock(&nhslot2->lock); | |
1918 | } | |
719f8358 ED |
1919 | |
1920 | spin_unlock_bh(&hslot->lock); | |
1921 | } | |
1922 | } | |
1923 | } | |
1924 | EXPORT_SYMBOL(udp_lib_rehash); | |
1925 | ||
8f6b5392 | 1926 | void udp_v4_rehash(struct sock *sk) |
719f8358 | 1927 | { |
f0b1e64c | 1928 | u16 new_hash = ipv4_portaddr_hash(sock_net(sk), |
719f8358 ED |
1929 | inet_sk(sk)->inet_rcv_saddr, |
1930 | inet_sk(sk)->inet_num); | |
1931 | udp_lib_rehash(sk, new_hash); | |
1932 | } | |
1933 | ||
a3f96c47 | 1934 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
93821778 | 1935 | { |
fec5e652 | 1936 | int rc; |
766e9037 | 1937 | |
005ec974 | 1938 | if (inet_sk(sk)->inet_daddr) { |
bdeab991 | 1939 | sock_rps_save_rxhash(sk, skb); |
005ec974 | 1940 | sk_mark_napi_id(sk, skb); |
2c8c56e1 | 1941 | sk_incoming_cpu_update(sk); |
e68b6e50 ED |
1942 | } else { |
1943 | sk_mark_napi_id_once(sk, skb); | |
005ec974 | 1944 | } |
fec5e652 | 1945 | |
850cbadd | 1946 | rc = __udp_enqueue_schedule_skb(sk, skb); |
766e9037 ED |
1947 | if (rc < 0) { |
1948 | int is_udplite = IS_UDPLITE(sk); | |
93821778 | 1949 | |
93821778 | 1950 | /* Note that an ENOMEM error is charged twice */ |
766e9037 | 1951 | if (rc == -ENOMEM) |
e61da9e2 | 1952 | UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
02c22347 | 1953 | is_udplite); |
e61da9e2 | 1954 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
766e9037 | 1955 | kfree_skb(skb); |
296f7ea7 | 1956 | trace_udp_fail_queue_rcv_skb(rc, sk); |
766e9037 | 1957 | return -1; |
93821778 HX |
1958 | } |
1959 | ||
1960 | return 0; | |
93821778 HX |
1961 | } |
1962 | ||
db8dac20 DM |
1963 | /* returns: |
1964 | * -1: error | |
1965 | * 0: success | |
1966 | * >0: "udp encap" protocol resubmission | |
1967 | * | |
1968 | * Note that in the success and error cases, the skb is assumed to | |
1969 | * have either been requeued or freed. | |
1970 | */ | |
cf329aa4 | 1971 | static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 DM |
1972 | { |
1973 | struct udp_sock *up = udp_sk(sk); | |
db8dac20 DM |
1974 | int is_udplite = IS_UDPLITE(sk); |
1975 | ||
1976 | /* | |
1977 | * Charge it to the socket, dropping if the queue is full. | |
1978 | */ | |
1979 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1980 | goto drop; | |
1981 | nf_reset(skb); | |
1982 | ||
88ab3108 | 1983 | if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { |
0ad92ad0 ED |
1984 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
1985 | ||
db8dac20 DM |
1986 | /* |
1987 | * This is an encapsulation socket so pass the skb to | |
1988 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
1989 | * fall through and pass this up the UDP socket. | |
1990 | * up->encap_rcv() returns the following value: | |
1991 | * =0 if skb was successfully passed to the encap | |
1992 | * handler or was discarded by it. | |
1993 | * >0 if skb should be passed on to UDP. | |
1994 | * <0 if skb should be resubmitted as proto -N | |
1995 | */ | |
1996 | ||
1997 | /* if we're overly short, let UDP handle it */ | |
6aa7de05 | 1998 | encap_rcv = READ_ONCE(up->encap_rcv); |
e5aed006 | 1999 | if (encap_rcv) { |
db8dac20 DM |
2000 | int ret; |
2001 | ||
0a80966b TH |
2002 | /* Verify checksum before giving to encap */ |
2003 | if (udp_lib_checksum_complete(skb)) | |
2004 | goto csum_error; | |
2005 | ||
0ad92ad0 | 2006 | ret = encap_rcv(sk, skb); |
db8dac20 | 2007 | if (ret <= 0) { |
02c22347 ED |
2008 | __UDP_INC_STATS(sock_net(sk), |
2009 | UDP_MIB_INDATAGRAMS, | |
2010 | is_udplite); | |
db8dac20 DM |
2011 | return -ret; |
2012 | } | |
2013 | } | |
2014 | ||
2015 | /* FALLTHROUGH -- it's a UDP Packet */ | |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * UDP-Lite specific tests, ignored on UDP sockets | |
2020 | */ | |
2021 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
2022 | ||
2023 | /* | |
2024 | * MIB statistics other than incrementing the error count are | |
2025 | * disabled for the following two types of errors: these depend | |
2026 | * on the application settings, not on the functioning of the | |
2027 | * protocol stack as such. | |
2028 | * | |
2029 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
2030 | * way ... to ... at least let the receiving application block | |
2031 | * delivery of packets with coverage values less than a value | |
2032 | * provided by the application." | |
2033 | */ | |
2034 | if (up->pcrlen == 0) { /* full coverage was set */ | |
ba7a46f1 JP |
2035 | net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", |
2036 | UDP_SKB_CB(skb)->cscov, skb->len); | |
db8dac20 DM |
2037 | goto drop; |
2038 | } | |
2039 | /* The next case involves violating the min. coverage requested | |
2040 | * by the receiver. This is subtle: if receiver wants x and x is | |
2041 | * greater than the buffersize/MTU then receiver will complain | |
2042 | * that it wants x while sender emits packets of smaller size y. | |
2043 | * Therefore the above ...()->partial_cov statement is essential. | |
2044 | */ | |
2045 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
ba7a46f1 JP |
2046 | net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", |
2047 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
db8dac20 DM |
2048 | goto drop; |
2049 | } | |
2050 | } | |
2051 | ||
dd99e425 | 2052 | prefetch(&sk->sk_rmem_alloc); |
ce25d66a ED |
2053 | if (rcu_access_pointer(sk->sk_filter) && |
2054 | udp_lib_checksum_complete(skb)) | |
e6afc8ac | 2055 | goto csum_error; |
ce25d66a | 2056 | |
ba66bbe5 | 2057 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) |
a6127697 | 2058 | goto drop; |
db8dac20 | 2059 | |
e6afc8ac | 2060 | udp_csum_pull_header(skb); |
db8dac20 | 2061 | |
fbf8866d | 2062 | ipv4_pktinfo_prepare(sk, skb); |
850cbadd | 2063 | return __udp_queue_rcv_skb(sk, skb); |
db8dac20 | 2064 | |
6a5dc9e5 | 2065 | csum_error: |
02c22347 | 2066 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
db8dac20 | 2067 | drop: |
02c22347 | 2068 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
8edf19c2 | 2069 | atomic_inc(&sk->sk_drops); |
db8dac20 DM |
2070 | kfree_skb(skb); |
2071 | return -1; | |
2072 | } | |
2073 | ||
cf329aa4 PA |
2074 | static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
2075 | { | |
2076 | struct sk_buff *next, *segs; | |
2077 | int ret; | |
2078 | ||
2079 | if (likely(!udp_unexpected_gso(sk, skb))) | |
2080 | return udp_queue_rcv_one_skb(sk, skb); | |
2081 | ||
2082 | BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET); | |
2083 | __skb_push(skb, -skb_mac_offset(skb)); | |
2084 | segs = udp_rcv_segment(sk, skb, true); | |
2085 | for (skb = segs; skb; skb = next) { | |
2086 | next = skb->next; | |
2087 | __skb_pull(skb, skb_transport_offset(skb)); | |
2088 | ret = udp_queue_rcv_one_skb(sk, skb); | |
2089 | if (ret > 0) | |
2090 | ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); | |
2091 | } | |
2092 | return 0; | |
2093 | } | |
2094 | ||
97502231 | 2095 | /* For TCP sockets, sk_rx_dst is protected by socket lock |
e47eb5df | 2096 | * For UDP, we use xchg() to guard against concurrent changes. |
97502231 | 2097 | */ |
64f0f5d1 | 2098 | bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) |
421b3885 | 2099 | { |
97502231 ED |
2100 | struct dst_entry *old; |
2101 | ||
d24406c8 WW |
2102 | if (dst_hold_safe(dst)) { |
2103 | old = xchg(&sk->sk_rx_dst, dst); | |
2104 | dst_release(old); | |
64f0f5d1 | 2105 | return old != dst; |
d24406c8 | 2106 | } |
64f0f5d1 | 2107 | return false; |
421b3885 | 2108 | } |
c9f2c1ae | 2109 | EXPORT_SYMBOL(udp_sk_rx_dst_set); |
421b3885 | 2110 | |
db8dac20 DM |
2111 | /* |
2112 | * Multicasts and broadcasts go to each listener. | |
2113 | * | |
1240d137 | 2114 | * Note: called only from the BH handler context. |
db8dac20 | 2115 | */ |
e3163493 | 2116 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
2117 | struct udphdr *uh, |
2118 | __be32 saddr, __be32 daddr, | |
36cbb245 RJ |
2119 | struct udp_table *udptable, |
2120 | int proto) | |
db8dac20 | 2121 | { |
ca065d0c | 2122 | struct sock *sk, *first = NULL; |
5cf3d461 DH |
2123 | unsigned short hnum = ntohs(uh->dest); |
2124 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
2dc41cff | 2125 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); |
ca065d0c ED |
2126 | unsigned int offset = offsetof(typeof(*sk), sk_node); |
2127 | int dif = skb->dev->ifindex; | |
fb74c277 | 2128 | int sdif = inet_sdif(skb); |
ca065d0c ED |
2129 | struct hlist_node *node; |
2130 | struct sk_buff *nskb; | |
2dc41cff DH |
2131 | |
2132 | if (use_hash2) { | |
f0b1e64c | 2133 | hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & |
73e2d5e3 | 2134 | udptable->mask; |
f0b1e64c | 2135 | hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; |
2dc41cff | 2136 | start_lookup: |
73e2d5e3 | 2137 | hslot = &udptable->hash2[hash2]; |
2dc41cff DH |
2138 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); |
2139 | } | |
db8dac20 | 2140 | |
ca065d0c ED |
2141 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { |
2142 | if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, | |
fb74c277 | 2143 | uh->source, saddr, dif, sdif, hnum)) |
ca065d0c ED |
2144 | continue; |
2145 | ||
2146 | if (!first) { | |
2147 | first = sk; | |
2148 | continue; | |
1240d137 | 2149 | } |
ca065d0c | 2150 | nskb = skb_clone(skb, GFP_ATOMIC); |
1240d137 | 2151 | |
ca065d0c ED |
2152 | if (unlikely(!nskb)) { |
2153 | atomic_inc(&sk->sk_drops); | |
02c22347 ED |
2154 | __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, |
2155 | IS_UDPLITE(sk)); | |
2156 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, | |
2157 | IS_UDPLITE(sk)); | |
ca065d0c ED |
2158 | continue; |
2159 | } | |
2160 | if (udp_queue_rcv_skb(sk, nskb) > 0) | |
2161 | consume_skb(nskb); | |
2162 | } | |
1240d137 | 2163 | |
2dc41cff DH |
2164 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ |
2165 | if (use_hash2 && hash2 != hash2_any) { | |
2166 | hash2 = hash2_any; | |
2167 | goto start_lookup; | |
2168 | } | |
2169 | ||
ca065d0c ED |
2170 | if (first) { |
2171 | if (udp_queue_rcv_skb(first, skb) > 0) | |
2172 | consume_skb(skb); | |
1240d137 | 2173 | } else { |
ca065d0c | 2174 | kfree_skb(skb); |
02c22347 ED |
2175 | __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, |
2176 | proto == IPPROTO_UDPLITE); | |
1240d137 | 2177 | } |
db8dac20 DM |
2178 | return 0; |
2179 | } | |
2180 | ||
2181 | /* Initialize UDP checksum. If exited with zero value (success), | |
2182 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
2183 | * Otherwise, csum completion requires chacksumming packet body, | |
2184 | * including udp header and folding it to skb->csum. | |
2185 | */ | |
2186 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
2187 | int proto) | |
2188 | { | |
db8dac20 DM |
2189 | int err; |
2190 | ||
2191 | UDP_SKB_CB(skb)->partial_cov = 0; | |
2192 | UDP_SKB_CB(skb)->cscov = skb->len; | |
2193 | ||
2194 | if (proto == IPPROTO_UDPLITE) { | |
2195 | err = udplite_checksum_init(skb, uh); | |
2196 | if (err) | |
2197 | return err; | |
15f35d49 AK |
2198 | |
2199 | if (UDP_SKB_CB(skb)->partial_cov) { | |
2200 | skb->csum = inet_compute_pseudo(skb, proto); | |
2201 | return 0; | |
2202 | } | |
db8dac20 DM |
2203 | } |
2204 | ||
b46d9f62 HFS |
2205 | /* Note, we are only interested in != 0 or == 0, thus the |
2206 | * force to int. | |
2207 | */ | |
db4f1be3 ST |
2208 | err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, |
2209 | inet_compute_pseudo); | |
2210 | if (err) | |
2211 | return err; | |
2212 | ||
2213 | if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { | |
2214 | /* If SW calculated the value, we know it's bad */ | |
2215 | if (skb->csum_complete_sw) | |
2216 | return 1; | |
2217 | ||
2218 | /* HW says the value is bad. Let's validate that. | |
2219 | * skb->csum is no longer the full packet checksum, | |
2220 | * so don't treat it as such. | |
2221 | */ | |
2222 | skb_checksum_complete_unset(skb); | |
2223 | } | |
2224 | ||
2225 | return 0; | |
db8dac20 DM |
2226 | } |
2227 | ||
2b5a9217 PA |
2228 | /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and |
2229 | * return code conversion for ip layer consumption | |
2230 | */ | |
2231 | static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, | |
2232 | struct udphdr *uh) | |
2233 | { | |
2234 | int ret; | |
2235 | ||
2236 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) | |
2237 | skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, | |
2238 | inet_compute_pseudo); | |
2239 | ||
2240 | ret = udp_queue_rcv_skb(sk, skb); | |
2241 | ||
2242 | /* a return value > 0 means to resubmit the input, but | |
2243 | * it wants the return to be -protocol, or 0 | |
2244 | */ | |
2245 | if (ret > 0) | |
2246 | return -ret; | |
2247 | return 0; | |
2248 | } | |
2249 | ||
db8dac20 DM |
2250 | /* |
2251 | * All we need to do is get the socket, and then do a checksum. | |
2252 | */ | |
2253 | ||
645ca708 | 2254 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
2255 | int proto) |
2256 | { | |
2257 | struct sock *sk; | |
7b5e56f9 | 2258 | struct udphdr *uh; |
db8dac20 | 2259 | unsigned short ulen; |
adf30907 | 2260 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 2261 | __be32 saddr, daddr; |
0283328e | 2262 | struct net *net = dev_net(skb->dev); |
db8dac20 DM |
2263 | |
2264 | /* | |
2265 | * Validate the packet. | |
2266 | */ | |
2267 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
2268 | goto drop; /* No space for header. */ | |
2269 | ||
7b5e56f9 | 2270 | uh = udp_hdr(skb); |
db8dac20 | 2271 | ulen = ntohs(uh->len); |
ccc2d97c BM |
2272 | saddr = ip_hdr(skb)->saddr; |
2273 | daddr = ip_hdr(skb)->daddr; | |
2274 | ||
db8dac20 DM |
2275 | if (ulen > skb->len) |
2276 | goto short_packet; | |
2277 | ||
2278 | if (proto == IPPROTO_UDP) { | |
2279 | /* UDP validates ulen. */ | |
2280 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
2281 | goto short_packet; | |
2282 | uh = udp_hdr(skb); | |
2283 | } | |
2284 | ||
2285 | if (udp4_csum_init(skb, uh, proto)) | |
2286 | goto csum_error; | |
2287 | ||
8afdd99a ED |
2288 | sk = skb_steal_sock(skb); |
2289 | if (sk) { | |
97502231 | 2290 | struct dst_entry *dst = skb_dst(skb); |
421b3885 | 2291 | int ret; |
421b3885 | 2292 | |
97502231 ED |
2293 | if (unlikely(sk->sk_rx_dst != dst)) |
2294 | udp_sk_rx_dst_set(sk, dst); | |
db8dac20 | 2295 | |
2b5a9217 | 2296 | ret = udp_unicast_rcv_skb(sk, skb, uh); |
8afdd99a | 2297 | sock_put(sk); |
2b5a9217 | 2298 | return ret; |
421b3885 | 2299 | } |
db8dac20 | 2300 | |
c18450a5 FF |
2301 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
2302 | return __udp4_lib_mcast_deliver(net, skb, uh, | |
36cbb245 | 2303 | saddr, daddr, udptable, proto); |
c18450a5 FF |
2304 | |
2305 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
2b5a9217 PA |
2306 | if (sk) |
2307 | return udp_unicast_rcv_skb(sk, skb, uh); | |
db8dac20 DM |
2308 | |
2309 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2310 | goto drop; | |
2311 | nf_reset(skb); | |
2312 | ||
2313 | /* No socket. Drop packet silently, if checksum is wrong */ | |
2314 | if (udp_lib_checksum_complete(skb)) | |
2315 | goto csum_error; | |
2316 | ||
02c22347 | 2317 | __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2318 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
2319 | ||
2320 | /* | |
2321 | * Hmm. We got an UDP packet to a port to which we | |
2322 | * don't wanna listen. Ignore it. | |
2323 | */ | |
2324 | kfree_skb(skb); | |
2325 | return 0; | |
2326 | ||
2327 | short_packet: | |
ba7a46f1 JP |
2328 | net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
2329 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2330 | &saddr, ntohs(uh->source), | |
2331 | ulen, skb->len, | |
2332 | &daddr, ntohs(uh->dest)); | |
db8dac20 DM |
2333 | goto drop; |
2334 | ||
2335 | csum_error: | |
2336 | /* | |
2337 | * RFC1122: OK. Discards the bad packet silently (as far as | |
2338 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
2339 | */ | |
ba7a46f1 JP |
2340 | net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
2341 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2342 | &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), | |
2343 | ulen); | |
02c22347 | 2344 | __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 | 2345 | drop: |
02c22347 | 2346 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2347 | kfree_skb(skb); |
2348 | return 0; | |
2349 | } | |
2350 | ||
421b3885 SB |
2351 | /* We can only early demux multicast if there is a single matching socket. |
2352 | * If more than one socket found returns NULL | |
2353 | */ | |
2354 | static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, | |
2355 | __be16 loc_port, __be32 loc_addr, | |
2356 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 2357 | int dif, int sdif) |
421b3885 SB |
2358 | { |
2359 | struct sock *sk, *result; | |
421b3885 | 2360 | unsigned short hnum = ntohs(loc_port); |
ca065d0c | 2361 | unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); |
421b3885 SB |
2362 | struct udp_hslot *hslot = &udp_table.hash[slot]; |
2363 | ||
63c6f81c ED |
2364 | /* Do not bother scanning a too big list */ |
2365 | if (hslot->count > 10) | |
2366 | return NULL; | |
2367 | ||
421b3885 | 2368 | result = NULL; |
ca065d0c ED |
2369 | sk_for_each_rcu(sk, &hslot->head) { |
2370 | if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, | |
fb74c277 | 2371 | rmt_port, rmt_addr, dif, sdif, hnum)) { |
ca065d0c ED |
2372 | if (result) |
2373 | return NULL; | |
421b3885 | 2374 | result = sk; |
421b3885 SB |
2375 | } |
2376 | } | |
ca065d0c | 2377 | |
421b3885 SB |
2378 | return result; |
2379 | } | |
2380 | ||
2381 | /* For unicast we should only early demux connected sockets or we can | |
2382 | * break forwarding setups. The chains here can be long so only check | |
2383 | * if the first socket is an exact match and if not move on. | |
2384 | */ | |
2385 | static struct sock *__udp4_lib_demux_lookup(struct net *net, | |
2386 | __be16 loc_port, __be32 loc_addr, | |
2387 | __be16 rmt_port, __be32 rmt_addr, | |
3fa6f616 | 2388 | int dif, int sdif) |
421b3885 | 2389 | { |
421b3885 | 2390 | unsigned short hnum = ntohs(loc_port); |
f0b1e64c | 2391 | unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); |
421b3885 SB |
2392 | unsigned int slot2 = hash2 & udp_table.mask; |
2393 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
c7228317 | 2394 | INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); |
421b3885 | 2395 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); |
ca065d0c | 2396 | struct sock *sk; |
421b3885 | 2397 | |
ca065d0c ED |
2398 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
2399 | if (INET_MATCH(sk, net, acookie, rmt_addr, | |
3fa6f616 | 2400 | loc_addr, ports, dif, sdif)) |
ca065d0c | 2401 | return sk; |
421b3885 SB |
2402 | /* Only check first socket in chain */ |
2403 | break; | |
2404 | } | |
ca065d0c | 2405 | return NULL; |
421b3885 SB |
2406 | } |
2407 | ||
7487449c | 2408 | int udp_v4_early_demux(struct sk_buff *skb) |
421b3885 | 2409 | { |
610438b7 | 2410 | struct net *net = dev_net(skb->dev); |
bc044e8d | 2411 | struct in_device *in_dev = NULL; |
610438b7 ED |
2412 | const struct iphdr *iph; |
2413 | const struct udphdr *uh; | |
ca065d0c | 2414 | struct sock *sk = NULL; |
421b3885 | 2415 | struct dst_entry *dst; |
421b3885 | 2416 | int dif = skb->dev->ifindex; |
fb74c277 | 2417 | int sdif = inet_sdif(skb); |
6e540309 | 2418 | int ours; |
421b3885 SB |
2419 | |
2420 | /* validate the packet */ | |
2421 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) | |
7487449c | 2422 | return 0; |
421b3885 | 2423 | |
610438b7 ED |
2424 | iph = ip_hdr(skb); |
2425 | uh = udp_hdr(skb); | |
2426 | ||
996b44fc | 2427 | if (skb->pkt_type == PACKET_MULTICAST) { |
bc044e8d | 2428 | in_dev = __in_dev_get_rcu(skb->dev); |
6e540309 SB |
2429 | |
2430 | if (!in_dev) | |
7487449c | 2431 | return 0; |
6e540309 | 2432 | |
996b44fc PA |
2433 | ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, |
2434 | iph->protocol); | |
2435 | if (!ours) | |
2436 | return 0; | |
ad0ea198 | 2437 | |
421b3885 | 2438 | sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, |
fb74c277 DA |
2439 | uh->source, iph->saddr, |
2440 | dif, sdif); | |
6e540309 | 2441 | } else if (skb->pkt_type == PACKET_HOST) { |
421b3885 | 2442 | sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, |
3fa6f616 | 2443 | uh->source, iph->saddr, dif, sdif); |
6e540309 | 2444 | } |
421b3885 | 2445 | |
41c6d650 | 2446 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) |
7487449c | 2447 | return 0; |
421b3885 SB |
2448 | |
2449 | skb->sk = sk; | |
82eabd9e | 2450 | skb->destructor = sock_efree; |
10e2eb87 | 2451 | dst = READ_ONCE(sk->sk_rx_dst); |
421b3885 SB |
2452 | |
2453 | if (dst) | |
2454 | dst = dst_check(dst, 0); | |
10e2eb87 | 2455 | if (dst) { |
bc044e8d PA |
2456 | u32 itag = 0; |
2457 | ||
d24406c8 WW |
2458 | /* set noref for now. |
2459 | * any place which wants to hold dst has to call | |
2460 | * dst_hold_safe() | |
2461 | */ | |
2462 | skb_dst_set_noref(skb, dst); | |
bc044e8d PA |
2463 | |
2464 | /* for unconnected multicast sockets we need to validate | |
2465 | * the source on each packet | |
2466 | */ | |
2467 | if (!inet_sk(sk)->inet_daddr && in_dev) | |
2468 | return ip_mc_validate_source(skb, iph->daddr, | |
2469 | iph->saddr, iph->tos, | |
2470 | skb->dev, in_dev, &itag); | |
10e2eb87 | 2471 | } |
7487449c | 2472 | return 0; |
421b3885 SB |
2473 | } |
2474 | ||
db8dac20 DM |
2475 | int udp_rcv(struct sk_buff *skb) |
2476 | { | |
645ca708 | 2477 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
2478 | } |
2479 | ||
7d06b2e0 | 2480 | void udp_destroy_sock(struct sock *sk) |
db8dac20 | 2481 | { |
44046a59 | 2482 | struct udp_sock *up = udp_sk(sk); |
8a74ad60 | 2483 | bool slow = lock_sock_fast(sk); |
db8dac20 | 2484 | udp_flush_pending_frames(sk); |
8a74ad60 | 2485 | unlock_sock_fast(sk, slow); |
60fb9567 PA |
2486 | if (static_branch_unlikely(&udp_encap_needed_key)) { |
2487 | if (up->encap_type) { | |
2488 | void (*encap_destroy)(struct sock *sk); | |
2489 | encap_destroy = READ_ONCE(up->encap_destroy); | |
2490 | if (encap_destroy) | |
2491 | encap_destroy(sk); | |
2492 | } | |
2493 | if (up->encap_enabled) | |
9c480601 | 2494 | static_branch_dec(&udp_encap_needed_key); |
44046a59 | 2495 | } |
db8dac20 DM |
2496 | } |
2497 | ||
1da177e4 LT |
2498 | /* |
2499 | * Socket option code for UDP | |
2500 | */ | |
4c0a6cb0 | 2501 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2502 | char __user *optval, unsigned int optlen, |
4c0a6cb0 | 2503 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
2504 | { |
2505 | struct udp_sock *up = udp_sk(sk); | |
1c19448c | 2506 | int val, valbool; |
1da177e4 | 2507 | int err = 0; |
b2bf1e26 | 2508 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 2509 | |
c482c568 | 2510 | if (optlen < sizeof(int)) |
1da177e4 LT |
2511 | return -EINVAL; |
2512 | ||
2513 | if (get_user(val, (int __user *)optval)) | |
2514 | return -EFAULT; | |
2515 | ||
1c19448c TH |
2516 | valbool = val ? 1 : 0; |
2517 | ||
6516c655 | 2518 | switch (optname) { |
1da177e4 LT |
2519 | case UDP_CORK: |
2520 | if (val != 0) { | |
2521 | up->corkflag = 1; | |
2522 | } else { | |
2523 | up->corkflag = 0; | |
2524 | lock_sock(sk); | |
4243cdc2 | 2525 | push_pending_frames(sk); |
1da177e4 LT |
2526 | release_sock(sk); |
2527 | } | |
2528 | break; | |
e905a9ed | 2529 | |
1da177e4 LT |
2530 | case UDP_ENCAP: |
2531 | switch (val) { | |
2532 | case 0: | |
2533 | case UDP_ENCAP_ESPINUDP: | |
2534 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
067b207b JC |
2535 | up->encap_rcv = xfrm4_udp_encap_rcv; |
2536 | /* FALLTHROUGH */ | |
342f0234 | 2537 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 | 2538 | up->encap_type = val; |
60fb9567 PA |
2539 | lock_sock(sk); |
2540 | udp_tunnel_encap_enable(sk->sk_socket); | |
2541 | release_sock(sk); | |
1da177e4 LT |
2542 | break; |
2543 | default: | |
2544 | err = -ENOPROTOOPT; | |
2545 | break; | |
2546 | } | |
2547 | break; | |
2548 | ||
1c19448c TH |
2549 | case UDP_NO_CHECK6_TX: |
2550 | up->no_check6_tx = valbool; | |
2551 | break; | |
2552 | ||
2553 | case UDP_NO_CHECK6_RX: | |
2554 | up->no_check6_rx = valbool; | |
2555 | break; | |
2556 | ||
bec1f6f6 WB |
2557 | case UDP_SEGMENT: |
2558 | if (val < 0 || val > USHRT_MAX) | |
2559 | return -EINVAL; | |
2560 | up->gso_size = val; | |
2561 | break; | |
2562 | ||
e20cf8d3 PA |
2563 | case UDP_GRO: |
2564 | lock_sock(sk); | |
2565 | if (valbool) | |
2566 | udp_tunnel_encap_enable(sk->sk_socket); | |
2567 | up->gro_enabled = valbool; | |
2568 | release_sock(sk); | |
2569 | break; | |
2570 | ||
ba4e58ec GR |
2571 | /* |
2572 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
2573 | */ | |
2574 | /* The sender sets actual checksum coverage length via this option. | |
2575 | * The case coverage > packet length is handled by send module. */ | |
2576 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 2577 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2578 | return -ENOPROTOOPT; |
2579 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
2580 | val = 8; | |
4be929be AD |
2581 | else if (val > USHRT_MAX) |
2582 | val = USHRT_MAX; | |
ba4e58ec GR |
2583 | up->pcslen = val; |
2584 | up->pcflag |= UDPLITE_SEND_CC; | |
2585 | break; | |
2586 | ||
e905a9ed YH |
2587 | /* The receiver specifies a minimum checksum coverage value. To make |
2588 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
2589 | * used, this again means full checksum coverage. */ |
2590 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 2591 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2592 | return -ENOPROTOOPT; |
2593 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
2594 | val = 8; | |
4be929be AD |
2595 | else if (val > USHRT_MAX) |
2596 | val = USHRT_MAX; | |
ba4e58ec GR |
2597 | up->pcrlen = val; |
2598 | up->pcflag |= UDPLITE_RECV_CC; | |
2599 | break; | |
2600 | ||
1da177e4 LT |
2601 | default: |
2602 | err = -ENOPROTOOPT; | |
2603 | break; | |
6516c655 | 2604 | } |
1da177e4 LT |
2605 | |
2606 | return err; | |
2607 | } | |
c482c568 | 2608 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 2609 | |
db8dac20 | 2610 | int udp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2611 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2612 | { |
2613 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2614 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2615 | udp_push_pending_frames); | |
2616 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
2617 | } | |
2618 | ||
2619 | #ifdef CONFIG_COMPAT | |
2620 | int compat_udp_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 2621 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2622 | { |
2623 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2624 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2625 | udp_push_pending_frames); | |
2626 | return compat_ip_setsockopt(sk, level, optname, optval, optlen); | |
2627 | } | |
2628 | #endif | |
2629 | ||
4c0a6cb0 GR |
2630 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
2631 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
2632 | { |
2633 | struct udp_sock *up = udp_sk(sk); | |
2634 | int val, len; | |
2635 | ||
c482c568 | 2636 | if (get_user(len, optlen)) |
1da177e4 LT |
2637 | return -EFAULT; |
2638 | ||
2639 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 2640 | |
6516c655 | 2641 | if (len < 0) |
1da177e4 LT |
2642 | return -EINVAL; |
2643 | ||
6516c655 | 2644 | switch (optname) { |
1da177e4 LT |
2645 | case UDP_CORK: |
2646 | val = up->corkflag; | |
2647 | break; | |
2648 | ||
2649 | case UDP_ENCAP: | |
2650 | val = up->encap_type; | |
2651 | break; | |
2652 | ||
1c19448c TH |
2653 | case UDP_NO_CHECK6_TX: |
2654 | val = up->no_check6_tx; | |
2655 | break; | |
2656 | ||
2657 | case UDP_NO_CHECK6_RX: | |
2658 | val = up->no_check6_rx; | |
2659 | break; | |
2660 | ||
bec1f6f6 WB |
2661 | case UDP_SEGMENT: |
2662 | val = up->gso_size; | |
2663 | break; | |
2664 | ||
ba4e58ec GR |
2665 | /* The following two cannot be changed on UDP sockets, the return is |
2666 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
2667 | case UDPLITE_SEND_CSCOV: | |
2668 | val = up->pcslen; | |
2669 | break; | |
2670 | ||
2671 | case UDPLITE_RECV_CSCOV: | |
2672 | val = up->pcrlen; | |
2673 | break; | |
2674 | ||
1da177e4 LT |
2675 | default: |
2676 | return -ENOPROTOOPT; | |
6516c655 | 2677 | } |
1da177e4 | 2678 | |
6516c655 | 2679 | if (put_user(len, optlen)) |
e905a9ed | 2680 | return -EFAULT; |
c482c568 | 2681 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 2682 | return -EFAULT; |
e905a9ed | 2683 | return 0; |
1da177e4 | 2684 | } |
c482c568 | 2685 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 2686 | |
db8dac20 DM |
2687 | int udp_getsockopt(struct sock *sk, int level, int optname, |
2688 | char __user *optval, int __user *optlen) | |
2689 | { | |
2690 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2691 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2692 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
2693 | } | |
2694 | ||
2695 | #ifdef CONFIG_COMPAT | |
2696 | int compat_udp_getsockopt(struct sock *sk, int level, int optname, | |
2697 | char __user *optval, int __user *optlen) | |
2698 | { | |
2699 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2700 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2701 | return compat_ip_getsockopt(sk, level, optname, optval, optlen); | |
2702 | } | |
2703 | #endif | |
1da177e4 LT |
2704 | /** |
2705 | * udp_poll - wait for a UDP event. | |
2706 | * @file - file struct | |
2707 | * @sock - socket | |
a11e1d43 | 2708 | * @wait - poll table |
1da177e4 | 2709 | * |
e905a9ed | 2710 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
2711 | * blocking sockets. If application is using a blocking fd |
2712 | * and a packet with checksum error is in the queue; | |
2713 | * then it could get return from select indicating data available | |
2714 | * but then block when reading it. Add special case code | |
2715 | * to work around these arguably broken applications. | |
2716 | */ | |
a11e1d43 | 2717 | __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 2718 | { |
a11e1d43 | 2719 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 | 2720 | struct sock *sk = sock->sk; |
ba4e58ec | 2721 | |
2276f58a | 2722 | if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) |
a9a08845 | 2723 | mask |= EPOLLIN | EPOLLRDNORM; |
2276f58a | 2724 | |
1da177e4 | 2725 | /* Check for false positives due to checksum errors */ |
a11e1d43 | 2726 | if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && |
e83c6744 | 2727 | !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) |
a9a08845 | 2728 | mask &= ~(EPOLLIN | EPOLLRDNORM); |
1da177e4 LT |
2729 | |
2730 | return mask; | |
e905a9ed | 2731 | |
1da177e4 | 2732 | } |
a11e1d43 | 2733 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 2734 | |
5d77dca8 DA |
2735 | int udp_abort(struct sock *sk, int err) |
2736 | { | |
2737 | lock_sock(sk); | |
2738 | ||
2739 | sk->sk_err = err; | |
2740 | sk->sk_error_report(sk); | |
286c72de | 2741 | __udp_disconnect(sk, 0); |
5d77dca8 DA |
2742 | |
2743 | release_sock(sk); | |
2744 | ||
2745 | return 0; | |
2746 | } | |
2747 | EXPORT_SYMBOL_GPL(udp_abort); | |
2748 | ||
db8dac20 | 2749 | struct proto udp_prot = { |
1e802951 TZ |
2750 | .name = "UDP", |
2751 | .owner = THIS_MODULE, | |
2752 | .close = udp_lib_close, | |
d74bad4e | 2753 | .pre_connect = udp_pre_connect, |
1e802951 TZ |
2754 | .connect = ip4_datagram_connect, |
2755 | .disconnect = udp_disconnect, | |
2756 | .ioctl = udp_ioctl, | |
2757 | .init = udp_init_sock, | |
2758 | .destroy = udp_destroy_sock, | |
2759 | .setsockopt = udp_setsockopt, | |
2760 | .getsockopt = udp_getsockopt, | |
2761 | .sendmsg = udp_sendmsg, | |
2762 | .recvmsg = udp_recvmsg, | |
2763 | .sendpage = udp_sendpage, | |
2764 | .release_cb = ip4_datagram_release_cb, | |
2765 | .hash = udp_lib_hash, | |
2766 | .unhash = udp_lib_unhash, | |
2767 | .rehash = udp_v4_rehash, | |
2768 | .get_port = udp_v4_get_port, | |
2769 | .memory_allocated = &udp_memory_allocated, | |
2770 | .sysctl_mem = sysctl_udp_mem, | |
2771 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), | |
2772 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), | |
2773 | .obj_size = sizeof(struct udp_sock), | |
2774 | .h.udp_table = &udp_table, | |
db8dac20 | 2775 | #ifdef CONFIG_COMPAT |
1e802951 TZ |
2776 | .compat_setsockopt = compat_udp_setsockopt, |
2777 | .compat_getsockopt = compat_udp_getsockopt, | |
db8dac20 | 2778 | #endif |
1e802951 | 2779 | .diag_destroy = udp_abort, |
db8dac20 | 2780 | }; |
c482c568 | 2781 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
2782 | |
2783 | /* ------------------------------------------------------------------------ */ | |
2784 | #ifdef CONFIG_PROC_FS | |
2785 | ||
645ca708 | 2786 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
2787 | { |
2788 | struct sock *sk; | |
a3d2599b | 2789 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2790 | struct udp_iter_state *state = seq->private; |
6f191efe | 2791 | struct net *net = seq_file_net(seq); |
1da177e4 | 2792 | |
a3d2599b | 2793 | for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; |
f86dcc5a | 2794 | ++state->bucket) { |
a3d2599b | 2795 | struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; |
f86dcc5a | 2796 | |
ca065d0c | 2797 | if (hlist_empty(&hslot->head)) |
f86dcc5a ED |
2798 | continue; |
2799 | ||
645ca708 | 2800 | spin_lock_bh(&hslot->lock); |
ca065d0c | 2801 | sk_for_each(sk, &hslot->head) { |
878628fb | 2802 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 2803 | continue; |
a3d2599b | 2804 | if (sk->sk_family == afinfo->family) |
1da177e4 LT |
2805 | goto found; |
2806 | } | |
645ca708 | 2807 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
2808 | } |
2809 | sk = NULL; | |
2810 | found: | |
2811 | return sk; | |
2812 | } | |
2813 | ||
2814 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
2815 | { | |
a3d2599b | 2816 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2817 | struct udp_iter_state *state = seq->private; |
6f191efe | 2818 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2819 | |
2820 | do { | |
ca065d0c | 2821 | sk = sk_next(sk); |
a3d2599b | 2822 | } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); |
1da177e4 | 2823 | |
645ca708 | 2824 | if (!sk) { |
a3d2599b CH |
2825 | if (state->bucket <= afinfo->udp_table->mask) |
2826 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
645ca708 | 2827 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
2828 | } |
2829 | return sk; | |
2830 | } | |
2831 | ||
2832 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
2833 | { | |
645ca708 | 2834 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
2835 | |
2836 | if (sk) | |
6516c655 | 2837 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
2838 | --pos; |
2839 | return pos ? NULL : sk; | |
2840 | } | |
2841 | ||
a3d2599b | 2842 | void *udp_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 2843 | { |
30842f29 | 2844 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 2845 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 2846 | |
b50660f1 | 2847 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 | 2848 | } |
a3d2599b | 2849 | EXPORT_SYMBOL(udp_seq_start); |
1da177e4 | 2850 | |
a3d2599b | 2851 | void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1da177e4 LT |
2852 | { |
2853 | struct sock *sk; | |
2854 | ||
b50660f1 | 2855 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
2856 | sk = udp_get_idx(seq, 0); |
2857 | else | |
2858 | sk = udp_get_next(seq, v); | |
2859 | ||
2860 | ++*pos; | |
2861 | return sk; | |
2862 | } | |
a3d2599b | 2863 | EXPORT_SYMBOL(udp_seq_next); |
1da177e4 | 2864 | |
a3d2599b | 2865 | void udp_seq_stop(struct seq_file *seq, void *v) |
1da177e4 | 2866 | { |
a3d2599b | 2867 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
645ca708 ED |
2868 | struct udp_iter_state *state = seq->private; |
2869 | ||
a3d2599b CH |
2870 | if (state->bucket <= afinfo->udp_table->mask) |
2871 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
1da177e4 | 2872 | } |
a3d2599b | 2873 | EXPORT_SYMBOL(udp_seq_stop); |
db8dac20 DM |
2874 | |
2875 | /* ------------------------------------------------------------------------ */ | |
5e659e4c | 2876 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
652586df | 2877 | int bucket) |
db8dac20 DM |
2878 | { |
2879 | struct inet_sock *inet = inet_sk(sp); | |
c720c7e8 ED |
2880 | __be32 dest = inet->inet_daddr; |
2881 | __be32 src = inet->inet_rcv_saddr; | |
2882 | __u16 destp = ntohs(inet->inet_dport); | |
2883 | __u16 srcp = ntohs(inet->inet_sport); | |
db8dac20 | 2884 | |
f86dcc5a | 2885 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
ea9a0379 | 2886 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", |
db8dac20 | 2887 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 | 2888 | sk_wmem_alloc_get(sp), |
6c206b20 | 2889 | udp_rqueue_get(sp), |
a7cb5a49 EB |
2890 | 0, 0L, 0, |
2891 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), | |
2892 | 0, sock_i_ino(sp), | |
41c6d650 | 2893 | refcount_read(&sp->sk_refcnt), sp, |
652586df | 2894 | atomic_read(&sp->sk_drops)); |
db8dac20 DM |
2895 | } |
2896 | ||
2897 | int udp4_seq_show(struct seq_file *seq, void *v) | |
2898 | { | |
652586df | 2899 | seq_setwidth(seq, 127); |
db8dac20 | 2900 | if (v == SEQ_START_TOKEN) |
652586df | 2901 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
db8dac20 | 2902 | "rx_queue tr tm->when retrnsmt uid timeout " |
cb61cb9b | 2903 | "inode ref pointer drops"); |
db8dac20 | 2904 | else { |
db8dac20 DM |
2905 | struct udp_iter_state *state = seq->private; |
2906 | ||
652586df | 2907 | udp4_format_sock(v, seq, state->bucket); |
db8dac20 | 2908 | } |
652586df | 2909 | seq_pad(seq, '\n'); |
db8dac20 DM |
2910 | return 0; |
2911 | } | |
2912 | ||
c3506372 | 2913 | const struct seq_operations udp_seq_ops = { |
a3d2599b CH |
2914 | .start = udp_seq_start, |
2915 | .next = udp_seq_next, | |
2916 | .stop = udp_seq_stop, | |
2917 | .show = udp4_seq_show, | |
2918 | }; | |
c3506372 | 2919 | EXPORT_SYMBOL(udp_seq_ops); |
73cb88ec | 2920 | |
db8dac20 | 2921 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 | 2922 | .family = AF_INET, |
645ca708 | 2923 | .udp_table = &udp_table, |
db8dac20 DM |
2924 | }; |
2925 | ||
2c8c1e72 | 2926 | static int __net_init udp4_proc_init_net(struct net *net) |
15439feb | 2927 | { |
c3506372 CH |
2928 | if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, |
2929 | sizeof(struct udp_iter_state), &udp4_seq_afinfo)) | |
a3d2599b CH |
2930 | return -ENOMEM; |
2931 | return 0; | |
15439feb PE |
2932 | } |
2933 | ||
2c8c1e72 | 2934 | static void __net_exit udp4_proc_exit_net(struct net *net) |
15439feb | 2935 | { |
a3d2599b | 2936 | remove_proc_entry("udp", net->proc_net); |
15439feb PE |
2937 | } |
2938 | ||
2939 | static struct pernet_operations udp4_net_ops = { | |
2940 | .init = udp4_proc_init_net, | |
2941 | .exit = udp4_proc_exit_net, | |
2942 | }; | |
2943 | ||
db8dac20 DM |
2944 | int __init udp4_proc_init(void) |
2945 | { | |
15439feb | 2946 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
2947 | } |
2948 | ||
2949 | void udp4_proc_exit(void) | |
2950 | { | |
15439feb | 2951 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 2952 | } |
1da177e4 LT |
2953 | #endif /* CONFIG_PROC_FS */ |
2954 | ||
f86dcc5a ED |
2955 | static __initdata unsigned long uhash_entries; |
2956 | static int __init set_uhash_entries(char *str) | |
645ca708 | 2957 | { |
413c27d8 EZ |
2958 | ssize_t ret; |
2959 | ||
f86dcc5a ED |
2960 | if (!str) |
2961 | return 0; | |
413c27d8 EZ |
2962 | |
2963 | ret = kstrtoul(str, 0, &uhash_entries); | |
2964 | if (ret) | |
2965 | return 0; | |
2966 | ||
f86dcc5a ED |
2967 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) |
2968 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
2969 | return 1; | |
2970 | } | |
2971 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 2972 | |
f86dcc5a ED |
2973 | void __init udp_table_init(struct udp_table *table, const char *name) |
2974 | { | |
2975 | unsigned int i; | |
2976 | ||
31fe62b9 TB |
2977 | table->hash = alloc_large_system_hash(name, |
2978 | 2 * sizeof(struct udp_hslot), | |
2979 | uhash_entries, | |
2980 | 21, /* one slot per 2 MB */ | |
2981 | 0, | |
2982 | &table->log, | |
2983 | &table->mask, | |
2984 | UDP_HTABLE_SIZE_MIN, | |
2985 | 64 * 1024); | |
2986 | ||
512615b6 | 2987 | table->hash2 = table->hash + (table->mask + 1); |
f86dcc5a | 2988 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2989 | INIT_HLIST_HEAD(&table->hash[i].head); |
fdcc8aa9 | 2990 | table->hash[i].count = 0; |
645ca708 ED |
2991 | spin_lock_init(&table->hash[i].lock); |
2992 | } | |
512615b6 | 2993 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2994 | INIT_HLIST_HEAD(&table->hash2[i].head); |
512615b6 ED |
2995 | table->hash2[i].count = 0; |
2996 | spin_lock_init(&table->hash2[i].lock); | |
2997 | } | |
645ca708 ED |
2998 | } |
2999 | ||
723b8e46 TH |
3000 | u32 udp_flow_hashrnd(void) |
3001 | { | |
3002 | static u32 hashrnd __read_mostly; | |
3003 | ||
3004 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
3005 | ||
3006 | return hashrnd; | |
3007 | } | |
3008 | EXPORT_SYMBOL(udp_flow_hashrnd); | |
3009 | ||
1e802951 TZ |
3010 | static void __udp_sysctl_init(struct net *net) |
3011 | { | |
3012 | net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; | |
3013 | net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; | |
3014 | ||
3015 | #ifdef CONFIG_NET_L3_MASTER_DEV | |
3016 | net->ipv4.sysctl_udp_l3mdev_accept = 0; | |
3017 | #endif | |
3018 | } | |
3019 | ||
3020 | static int __net_init udp_sysctl_init(struct net *net) | |
3021 | { | |
3022 | __udp_sysctl_init(net); | |
3023 | return 0; | |
3024 | } | |
3025 | ||
3026 | static struct pernet_operations __net_initdata udp_sysctl_ops = { | |
fc18999e | 3027 | .init = udp_sysctl_init, |
1e802951 TZ |
3028 | }; |
3029 | ||
95766fff HA |
3030 | void __init udp_init(void) |
3031 | { | |
f03d78db | 3032 | unsigned long limit; |
4b272750 | 3033 | unsigned int i; |
95766fff | 3034 | |
f86dcc5a | 3035 | udp_table_init(&udp_table, "UDP"); |
f03d78db | 3036 | limit = nr_free_buffer_pages() / 8; |
95766fff HA |
3037 | limit = max(limit, 128UL); |
3038 | sysctl_udp_mem[0] = limit / 4 * 3; | |
3039 | sysctl_udp_mem[1] = limit; | |
3040 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
3041 | ||
1e802951 | 3042 | __udp_sysctl_init(&init_net); |
4b272750 ED |
3043 | |
3044 | /* 16 spinlocks per cpu */ | |
3045 | udp_busylocks_log = ilog2(nr_cpu_ids) + 4; | |
3046 | udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, | |
3047 | GFP_KERNEL); | |
3048 | if (!udp_busylocks) | |
3049 | panic("UDP: failed to alloc udp_busylocks\n"); | |
3050 | for (i = 0; i < (1U << udp_busylocks_log); i++) | |
3051 | spin_lock_init(udp_busylocks + i); | |
1e802951 TZ |
3052 | |
3053 | if (register_pernet_subsys(&udp_sysctl_ops)) | |
3054 | panic("UDP: failed to init sysctl parameters.\n"); | |
95766fff | 3055 | } |