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> |
95766fff | 84 | #include <linux/bootmem.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> |
1da177e4 | 108 | #include <net/route.h> |
1da177e4 LT |
109 | #include <net/checksum.h> |
110 | #include <net/xfrm.h> | |
296f7ea7 | 111 | #include <trace/events/udp.h> |
447167bf | 112 | #include <linux/static_key.h> |
22911fc5 | 113 | #include <trace/events/skb.h> |
076bb0c8 | 114 | #include <net/busy_poll.h> |
ba4e58ec | 115 | #include "udp_impl.h" |
e32ea7e7 | 116 | #include <net/sock_reuseport.h> |
217375a0 | 117 | #include <net/addrconf.h> |
1da177e4 | 118 | |
f86dcc5a | 119 | struct udp_table udp_table __read_mostly; |
645ca708 | 120 | EXPORT_SYMBOL(udp_table); |
1da177e4 | 121 | |
8d987e5c | 122 | long sysctl_udp_mem[3] __read_mostly; |
95766fff | 123 | EXPORT_SYMBOL(sysctl_udp_mem); |
c482c568 ED |
124 | |
125 | int sysctl_udp_rmem_min __read_mostly; | |
95766fff | 126 | EXPORT_SYMBOL(sysctl_udp_rmem_min); |
c482c568 ED |
127 | |
128 | int sysctl_udp_wmem_min __read_mostly; | |
95766fff HA |
129 | EXPORT_SYMBOL(sysctl_udp_wmem_min); |
130 | ||
8d987e5c | 131 | atomic_long_t udp_memory_allocated; |
95766fff HA |
132 | EXPORT_SYMBOL(udp_memory_allocated); |
133 | ||
f86dcc5a ED |
134 | #define MAX_UDP_PORTS 65536 |
135 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 136 | |
63a6fff3 RS |
137 | /* IPCB reference means this can not be used from early demux */ |
138 | static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) | |
139 | { | |
140 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
141 | if (!net->ipv4.sysctl_udp_l3mdev_accept && | |
142 | skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) | |
143 | return true; | |
144 | #endif | |
145 | return false; | |
146 | } | |
147 | ||
f24d43c0 | 148 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 149 | const struct udp_hslot *hslot, |
98322f22 | 150 | unsigned long *bitmap, |
fe38d2a1 | 151 | struct sock *sk, unsigned int log) |
1da177e4 | 152 | { |
f24d43c0 | 153 | struct sock *sk2; |
ba418fa3 | 154 | kuid_t uid = sock_i_uid(sk); |
25030a7f | 155 | |
ca065d0c | 156 | sk_for_each(sk2, &hslot->head) { |
9d4fb27d JP |
157 | if (net_eq(sock_net(sk2), net) && |
158 | sk2 != sk && | |
d4cada4a | 159 | (bitmap || udp_sk(sk2)->udp_port_hash == num) && |
9d4fb27d JP |
160 | (!sk2->sk_reuse || !sk->sk_reuse) && |
161 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
162 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 163 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
164 | if (sk2->sk_reuseport && sk->sk_reuseport && |
165 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
166 | uid_eq(uid, sock_i_uid(sk2))) { | |
167 | if (!bitmap) | |
168 | return 0; | |
169 | } else { | |
170 | if (!bitmap) | |
171 | return 1; | |
172 | __set_bit(udp_sk(sk2)->udp_port_hash >> log, | |
173 | bitmap); | |
174 | } | |
98322f22 | 175 | } |
4243cdc2 | 176 | } |
25030a7f GR |
177 | return 0; |
178 | } | |
179 | ||
30fff923 ED |
180 | /* |
181 | * Note: we still hold spinlock of primary hash chain, so no other writer | |
182 | * can insert/delete a socket with local_port == num | |
183 | */ | |
184 | static int udp_lib_lport_inuse2(struct net *net, __u16 num, | |
4243cdc2 | 185 | struct udp_hslot *hslot2, |
fe38d2a1 | 186 | struct sock *sk) |
30fff923 ED |
187 | { |
188 | struct sock *sk2; | |
ba418fa3 | 189 | kuid_t uid = sock_i_uid(sk); |
30fff923 ED |
190 | int res = 0; |
191 | ||
192 | spin_lock(&hslot2->lock); | |
ca065d0c | 193 | udp_portaddr_for_each_entry(sk2, &hslot2->head) { |
9d4fb27d JP |
194 | if (net_eq(sock_net(sk2), net) && |
195 | sk2 != sk && | |
196 | (udp_sk(sk2)->udp_port_hash == num) && | |
197 | (!sk2->sk_reuse || !sk->sk_reuse) && | |
198 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
199 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 200 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
201 | if (sk2->sk_reuseport && sk->sk_reuseport && |
202 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
203 | uid_eq(uid, sock_i_uid(sk2))) { | |
204 | res = 0; | |
205 | } else { | |
206 | res = 1; | |
207 | } | |
30fff923 ED |
208 | break; |
209 | } | |
4243cdc2 | 210 | } |
30fff923 ED |
211 | spin_unlock(&hslot2->lock); |
212 | return res; | |
213 | } | |
214 | ||
fe38d2a1 | 215 | static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) |
e32ea7e7 CG |
216 | { |
217 | struct net *net = sock_net(sk); | |
e32ea7e7 CG |
218 | kuid_t uid = sock_i_uid(sk); |
219 | struct sock *sk2; | |
220 | ||
ca065d0c | 221 | sk_for_each(sk2, &hslot->head) { |
e32ea7e7 CG |
222 | if (net_eq(sock_net(sk2), net) && |
223 | sk2 != sk && | |
224 | sk2->sk_family == sk->sk_family && | |
225 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
226 | (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && | |
227 | (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
228 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && | |
fe38d2a1 | 229 | inet_rcv_saddr_equal(sk, sk2, false)) { |
e32ea7e7 CG |
230 | return reuseport_add_sock(sk, sk2); |
231 | } | |
232 | } | |
233 | ||
234 | /* Initial allocation may have already happened via setsockopt */ | |
235 | if (!rcu_access_pointer(sk->sk_reuseport_cb)) | |
236 | return reuseport_alloc(sk); | |
237 | return 0; | |
238 | } | |
239 | ||
25030a7f | 240 | /** |
6ba5a3c5 | 241 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
242 | * |
243 | * @sk: socket struct in question | |
244 | * @snum: port number to look up | |
25985edc | 245 | * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, |
30fff923 | 246 | * with NULL address |
25030a7f | 247 | */ |
6ba5a3c5 | 248 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
30fff923 | 249 | unsigned int hash2_nulladdr) |
25030a7f | 250 | { |
512615b6 | 251 | struct udp_hslot *hslot, *hslot2; |
645ca708 | 252 | struct udp_table *udptable = sk->sk_prot->h.udp_table; |
25030a7f | 253 | int error = 1; |
3b1e0a65 | 254 | struct net *net = sock_net(sk); |
1da177e4 | 255 | |
32c1da70 | 256 | if (!snum) { |
9088c560 | 257 | int low, high, remaining; |
95c96174 | 258 | unsigned int rand; |
98322f22 ED |
259 | unsigned short first, last; |
260 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 261 | |
0bbf87d8 | 262 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 263 | remaining = (high - low) + 1; |
227b60f5 | 264 | |
63862b5b | 265 | rand = prandom_u32(); |
8fc54f68 | 266 | first = reciprocal_scale(rand, remaining) + low; |
98322f22 ED |
267 | /* |
268 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
269 | */ | |
f86dcc5a | 270 | rand = (rand | 1) * (udptable->mask + 1); |
5781b235 ED |
271 | last = first + udptable->mask + 1; |
272 | do { | |
f86dcc5a | 273 | hslot = udp_hashslot(udptable, net, first); |
98322f22 | 274 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 275 | spin_lock_bh(&hslot->lock); |
98322f22 | 276 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
fe38d2a1 | 277 | udptable->log); |
98322f22 ED |
278 | |
279 | snum = first; | |
280 | /* | |
281 | * Iterate on all possible values of snum for this hash. | |
282 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
283 | * give us randomization and full range coverage. | |
284 | */ | |
9088c560 | 285 | do { |
98322f22 | 286 | if (low <= snum && snum <= high && |
e3826f1e | 287 | !test_bit(snum >> udptable->log, bitmap) && |
122ff243 | 288 | !inet_is_local_reserved_port(net, snum)) |
98322f22 ED |
289 | goto found; |
290 | snum += rand; | |
291 | } while (snum != first); | |
292 | spin_unlock_bh(&hslot->lock); | |
df560056 | 293 | cond_resched(); |
5781b235 | 294 | } while (++first != last); |
98322f22 | 295 | goto fail; |
645ca708 | 296 | } else { |
f86dcc5a | 297 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 298 | spin_lock_bh(&hslot->lock); |
30fff923 ED |
299 | if (hslot->count > 10) { |
300 | int exist; | |
301 | unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; | |
302 | ||
303 | slot2 &= udptable->mask; | |
304 | hash2_nulladdr &= udptable->mask; | |
305 | ||
306 | hslot2 = udp_hashslot2(udptable, slot2); | |
307 | if (hslot->count < hslot2->count) | |
308 | goto scan_primary_hash; | |
309 | ||
fe38d2a1 | 310 | exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); |
30fff923 ED |
311 | if (!exist && (hash2_nulladdr != slot2)) { |
312 | hslot2 = udp_hashslot2(udptable, hash2_nulladdr); | |
313 | exist = udp_lib_lport_inuse2(net, snum, hslot2, | |
fe38d2a1 | 314 | sk); |
30fff923 ED |
315 | } |
316 | if (exist) | |
317 | goto fail_unlock; | |
318 | else | |
319 | goto found; | |
320 | } | |
321 | scan_primary_hash: | |
fe38d2a1 | 322 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) |
645ca708 ED |
323 | goto fail_unlock; |
324 | } | |
98322f22 | 325 | found: |
c720c7e8 | 326 | inet_sk(sk)->inet_num = snum; |
d4cada4a ED |
327 | udp_sk(sk)->udp_port_hash = snum; |
328 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
1da177e4 | 329 | if (sk_unhashed(sk)) { |
e32ea7e7 | 330 | if (sk->sk_reuseport && |
fe38d2a1 | 331 | udp_reuseport_add_sock(sk, hslot)) { |
e32ea7e7 CG |
332 | inet_sk(sk)->inet_num = 0; |
333 | udp_sk(sk)->udp_port_hash = 0; | |
334 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
335 | goto fail_unlock; | |
336 | } | |
337 | ||
ca065d0c | 338 | sk_add_node_rcu(sk, &hslot->head); |
fdcc8aa9 | 339 | hslot->count++; |
c29a0bc4 | 340 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
512615b6 ED |
341 | |
342 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
343 | spin_lock(&hslot2->lock); | |
d894ba18 | 344 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
1602f49b DM |
345 | sk->sk_family == AF_INET6) |
346 | hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, | |
347 | &hslot2->head); | |
d894ba18 | 348 | else |
1602f49b DM |
349 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
350 | &hslot2->head); | |
512615b6 ED |
351 | hslot2->count++; |
352 | spin_unlock(&hslot2->lock); | |
1da177e4 | 353 | } |
ca065d0c | 354 | sock_set_flag(sk, SOCK_RCU_FREE); |
25030a7f | 355 | error = 0; |
645ca708 ED |
356 | fail_unlock: |
357 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 358 | fail: |
25030a7f GR |
359 | return error; |
360 | } | |
c482c568 | 361 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 362 | |
6eada011 ED |
363 | static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, |
364 | unsigned int port) | |
d4cada4a | 365 | { |
0eae88f3 | 366 | return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; |
d4cada4a ED |
367 | } |
368 | ||
6ba5a3c5 | 369 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 370 | { |
30fff923 | 371 | unsigned int hash2_nulladdr = |
0eae88f3 | 372 | udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); |
30fff923 ED |
373 | unsigned int hash2_partial = |
374 | udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); | |
375 | ||
d4cada4a | 376 | /* precompute partial secondary hash */ |
30fff923 | 377 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
fe38d2a1 | 378 | return udp_lib_get_port(sk, snum, hash2_nulladdr); |
db8dac20 DM |
379 | } |
380 | ||
d1e37288 SX |
381 | static int compute_score(struct sock *sk, struct net *net, |
382 | __be32 saddr, __be16 sport, | |
63a6fff3 RS |
383 | __be32 daddr, unsigned short hnum, int dif, |
384 | bool exact_dif) | |
645ca708 | 385 | { |
60c04aec JP |
386 | int score; |
387 | struct inet_sock *inet; | |
645ca708 | 388 | |
60c04aec JP |
389 | if (!net_eq(sock_net(sk), net) || |
390 | udp_sk(sk)->udp_port_hash != hnum || | |
391 | ipv6_only_sock(sk)) | |
392 | return -1; | |
645ca708 | 393 | |
60c04aec JP |
394 | score = (sk->sk_family == PF_INET) ? 2 : 1; |
395 | inet = inet_sk(sk); | |
396 | ||
397 | if (inet->inet_rcv_saddr) { | |
398 | if (inet->inet_rcv_saddr != daddr) | |
399 | return -1; | |
400 | score += 4; | |
645ca708 | 401 | } |
60c04aec JP |
402 | |
403 | if (inet->inet_daddr) { | |
404 | if (inet->inet_daddr != saddr) | |
405 | return -1; | |
406 | score += 4; | |
407 | } | |
408 | ||
409 | if (inet->inet_dport) { | |
410 | if (inet->inet_dport != sport) | |
411 | return -1; | |
412 | score += 4; | |
413 | } | |
414 | ||
63a6fff3 | 415 | if (sk->sk_bound_dev_if || exact_dif) { |
60c04aec JP |
416 | if (sk->sk_bound_dev_if != dif) |
417 | return -1; | |
418 | score += 4; | |
419 | } | |
70da268b ED |
420 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
421 | score++; | |
645ca708 ED |
422 | return score; |
423 | } | |
424 | ||
6eada011 ED |
425 | static u32 udp_ehashfn(const struct net *net, const __be32 laddr, |
426 | const __u16 lport, const __be32 faddr, | |
427 | const __be16 fport) | |
65cd8033 | 428 | { |
1bbdceef HFS |
429 | static u32 udp_ehash_secret __read_mostly; |
430 | ||
431 | net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); | |
432 | ||
65cd8033 | 433 | return __inet_ehashfn(laddr, lport, faddr, fport, |
1bbdceef | 434 | udp_ehash_secret + net_hash_mix(net)); |
65cd8033 HFS |
435 | } |
436 | ||
d1e37288 | 437 | /* called with rcu_read_lock() */ |
5051ebd2 ED |
438 | static struct sock *udp4_lib_lookup2(struct net *net, |
439 | __be32 saddr, __be16 sport, | |
63a6fff3 | 440 | __be32 daddr, unsigned int hnum, int dif, bool exact_dif, |
d1e37288 | 441 | struct udp_hslot *hslot2, |
1134158b | 442 | struct sk_buff *skb) |
5051ebd2 ED |
443 | { |
444 | struct sock *sk, *result; | |
ba418fa3 TH |
445 | int score, badness, matches = 0, reuseport = 0; |
446 | u32 hash = 0; | |
5051ebd2 | 447 | |
5051ebd2 | 448 | result = NULL; |
ba418fa3 | 449 | badness = 0; |
ca065d0c | 450 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
d1e37288 | 451 | score = compute_score(sk, net, saddr, sport, |
63a6fff3 | 452 | daddr, hnum, dif, exact_dif); |
5051ebd2 | 453 | if (score > badness) { |
ba418fa3 TH |
454 | reuseport = sk->sk_reuseport; |
455 | if (reuseport) { | |
65cd8033 HFS |
456 | hash = udp_ehashfn(net, daddr, hnum, |
457 | saddr, sport); | |
ca065d0c | 458 | result = reuseport_select_sock(sk, hash, skb, |
ed0dfffd | 459 | sizeof(struct udphdr)); |
ca065d0c ED |
460 | if (result) |
461 | return result; | |
ba418fa3 TH |
462 | matches = 1; |
463 | } | |
ca065d0c ED |
464 | badness = score; |
465 | result = sk; | |
ba418fa3 TH |
466 | } else if (score == badness && reuseport) { |
467 | matches++; | |
8fc54f68 | 468 | if (reciprocal_scale(hash, matches) == 0) |
ba418fa3 TH |
469 | result = sk; |
470 | hash = next_pseudo_random32(hash); | |
5051ebd2 ED |
471 | } |
472 | } | |
5051ebd2 ED |
473 | return result; |
474 | } | |
475 | ||
db8dac20 DM |
476 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
477 | * harder than this. -DaveM | |
478 | */ | |
fce82338 | 479 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, |
db8dac20 | 480 | __be16 sport, __be32 daddr, __be16 dport, |
538950a1 | 481 | int dif, struct udp_table *udptable, struct sk_buff *skb) |
db8dac20 | 482 | { |
271b72c7 | 483 | struct sock *sk, *result; |
db8dac20 | 484 | unsigned short hnum = ntohs(dport); |
5051ebd2 ED |
485 | unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); |
486 | struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; | |
63a6fff3 | 487 | bool exact_dif = udp_lib_exact_dif_match(net, skb); |
ba418fa3 TH |
488 | int score, badness, matches = 0, reuseport = 0; |
489 | u32 hash = 0; | |
645ca708 | 490 | |
5051ebd2 ED |
491 | if (hslot->count > 10) { |
492 | hash2 = udp4_portaddr_hash(net, daddr, hnum); | |
493 | slot2 = hash2 & udptable->mask; | |
494 | hslot2 = &udptable->hash2[slot2]; | |
495 | if (hslot->count < hslot2->count) | |
496 | goto begin; | |
497 | ||
498 | result = udp4_lib_lookup2(net, saddr, sport, | |
499 | daddr, hnum, dif, | |
63a6fff3 | 500 | exact_dif, hslot2, skb); |
5051ebd2 | 501 | if (!result) { |
d1e37288 | 502 | unsigned int old_slot2 = slot2; |
0eae88f3 | 503 | hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); |
5051ebd2 | 504 | slot2 = hash2 & udptable->mask; |
d1e37288 SX |
505 | /* avoid searching the same slot again. */ |
506 | if (unlikely(slot2 == old_slot2)) | |
507 | return result; | |
508 | ||
5051ebd2 ED |
509 | hslot2 = &udptable->hash2[slot2]; |
510 | if (hslot->count < hslot2->count) | |
511 | goto begin; | |
512 | ||
1223c67c | 513 | result = udp4_lib_lookup2(net, saddr, sport, |
d1e37288 | 514 | daddr, hnum, dif, |
63a6fff3 | 515 | exact_dif, hslot2, skb); |
5051ebd2 | 516 | } |
5051ebd2 ED |
517 | return result; |
518 | } | |
271b72c7 ED |
519 | begin: |
520 | result = NULL; | |
ba418fa3 | 521 | badness = 0; |
ca065d0c | 522 | sk_for_each_rcu(sk, &hslot->head) { |
d1e37288 | 523 | score = compute_score(sk, net, saddr, sport, |
63a6fff3 | 524 | daddr, hnum, dif, exact_dif); |
645ca708 | 525 | if (score > badness) { |
ba418fa3 TH |
526 | reuseport = sk->sk_reuseport; |
527 | if (reuseport) { | |
65cd8033 HFS |
528 | hash = udp_ehashfn(net, daddr, hnum, |
529 | saddr, sport); | |
ca065d0c | 530 | result = reuseport_select_sock(sk, hash, skb, |
538950a1 | 531 | sizeof(struct udphdr)); |
ca065d0c ED |
532 | if (result) |
533 | return result; | |
ba418fa3 TH |
534 | matches = 1; |
535 | } | |
ca065d0c ED |
536 | result = sk; |
537 | badness = score; | |
ba418fa3 TH |
538 | } else if (score == badness && reuseport) { |
539 | matches++; | |
8fc54f68 | 540 | if (reciprocal_scale(hash, matches) == 0) |
ba418fa3 TH |
541 | result = sk; |
542 | hash = next_pseudo_random32(hash); | |
db8dac20 DM |
543 | } |
544 | } | |
db8dac20 DM |
545 | return result; |
546 | } | |
fce82338 | 547 | EXPORT_SYMBOL_GPL(__udp4_lib_lookup); |
db8dac20 | 548 | |
607c4aaf KK |
549 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
550 | __be16 sport, __be16 dport, | |
645ca708 | 551 | struct udp_table *udptable) |
607c4aaf KK |
552 | { |
553 | const struct iphdr *iph = ip_hdr(skb); | |
554 | ||
ed7cbbce | 555 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, |
8afdd99a | 556 | iph->daddr, dport, inet_iif(skb), |
538950a1 | 557 | udptable, skb); |
607c4aaf KK |
558 | } |
559 | ||
63058308 TH |
560 | struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, |
561 | __be16 sport, __be16 dport) | |
562 | { | |
ed7cbbce | 563 | return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); |
63058308 TH |
564 | } |
565 | EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); | |
566 | ||
ca065d0c ED |
567 | /* Must be called under rcu_read_lock(). |
568 | * Does increment socket refcount. | |
569 | */ | |
570 | #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ | |
30f58158 AB |
571 | IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ |
572 | IS_ENABLED(CONFIG_NF_SOCKET_IPV4) | |
bcd41303 KK |
573 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
574 | __be32 daddr, __be16 dport, int dif) | |
575 | { | |
ca065d0c ED |
576 | struct sock *sk; |
577 | ||
578 | sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, | |
579 | dif, &udp_table, NULL); | |
41c6d650 | 580 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) |
ca065d0c ED |
581 | sk = NULL; |
582 | return sk; | |
bcd41303 KK |
583 | } |
584 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
ca065d0c | 585 | #endif |
bcd41303 | 586 | |
421b3885 SB |
587 | static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, |
588 | __be16 loc_port, __be32 loc_addr, | |
589 | __be16 rmt_port, __be32 rmt_addr, | |
590 | int dif, unsigned short hnum) | |
591 | { | |
592 | struct inet_sock *inet = inet_sk(sk); | |
593 | ||
594 | if (!net_eq(sock_net(sk), net) || | |
595 | udp_sk(sk)->udp_port_hash != hnum || | |
596 | (inet->inet_daddr && inet->inet_daddr != rmt_addr) || | |
597 | (inet->inet_dport != rmt_port && inet->inet_dport) || | |
598 | (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || | |
599 | ipv6_only_sock(sk) || | |
600 | (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) | |
601 | return false; | |
602 | if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) | |
603 | return false; | |
604 | return true; | |
605 | } | |
606 | ||
db8dac20 DM |
607 | /* |
608 | * This routine is called by the ICMP module when it gets some | |
609 | * sort of error condition. If err < 0 then the socket should | |
610 | * be closed and the error returned to the user. If err > 0 | |
611 | * it's just the icmp type << 8 | icmp code. | |
612 | * Header points to the ip header of the error packet. We move | |
613 | * on past this. Then (as it used to claim before adjustment) | |
614 | * header points to the first 8 bytes of the udp header. We need | |
615 | * to find the appropriate port. | |
616 | */ | |
617 | ||
645ca708 | 618 | void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
619 | { |
620 | struct inet_sock *inet; | |
b71d1d42 | 621 | const struct iphdr *iph = (const struct iphdr *)skb->data; |
c482c568 | 622 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); |
db8dac20 DM |
623 | const int type = icmp_hdr(skb)->type; |
624 | const int code = icmp_hdr(skb)->code; | |
625 | struct sock *sk; | |
626 | int harderr; | |
627 | int err; | |
fd54d716 | 628 | struct net *net = dev_net(skb->dev); |
db8dac20 | 629 | |
fd54d716 | 630 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
538950a1 CG |
631 | iph->saddr, uh->source, skb->dev->ifindex, udptable, |
632 | NULL); | |
51456b29 | 633 | if (!sk) { |
5d3848bc | 634 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
db8dac20 DM |
635 | return; /* No socket for error */ |
636 | } | |
637 | ||
638 | err = 0; | |
639 | harderr = 0; | |
640 | inet = inet_sk(sk); | |
641 | ||
642 | switch (type) { | |
643 | default: | |
644 | case ICMP_TIME_EXCEEDED: | |
645 | err = EHOSTUNREACH; | |
646 | break; | |
647 | case ICMP_SOURCE_QUENCH: | |
648 | goto out; | |
649 | case ICMP_PARAMETERPROB: | |
650 | err = EPROTO; | |
651 | harderr = 1; | |
652 | break; | |
653 | case ICMP_DEST_UNREACH: | |
654 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
36393395 | 655 | ipv4_sk_update_pmtu(skb, sk, info); |
db8dac20 DM |
656 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { |
657 | err = EMSGSIZE; | |
658 | harderr = 1; | |
659 | break; | |
660 | } | |
661 | goto out; | |
662 | } | |
663 | err = EHOSTUNREACH; | |
664 | if (code <= NR_ICMP_UNREACH) { | |
665 | harderr = icmp_err_convert[code].fatal; | |
666 | err = icmp_err_convert[code].errno; | |
667 | } | |
668 | break; | |
55be7a9c DM |
669 | case ICMP_REDIRECT: |
670 | ipv4_sk_redirect(skb, sk); | |
1a462d18 | 671 | goto out; |
db8dac20 DM |
672 | } |
673 | ||
674 | /* | |
675 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
676 | * 4.1.3.3. | |
677 | */ | |
678 | if (!inet->recverr) { | |
679 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
680 | goto out; | |
b1faf566 | 681 | } else |
c482c568 | 682 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
b1faf566 | 683 | |
db8dac20 DM |
684 | sk->sk_err = err; |
685 | sk->sk_error_report(sk); | |
686 | out: | |
ca065d0c | 687 | return; |
db8dac20 DM |
688 | } |
689 | ||
690 | void udp_err(struct sk_buff *skb, u32 info) | |
691 | { | |
645ca708 | 692 | __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
693 | } |
694 | ||
695 | /* | |
696 | * Throw away all pending data and cancel the corking. Socket is locked. | |
697 | */ | |
36d926b9 | 698 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
699 | { |
700 | struct udp_sock *up = udp_sk(sk); | |
701 | ||
702 | if (up->pending) { | |
703 | up->len = 0; | |
704 | up->pending = 0; | |
705 | ip_flush_pending_frames(sk); | |
706 | } | |
707 | } | |
36d926b9 | 708 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
709 | |
710 | /** | |
f6b9664f | 711 | * udp4_hwcsum - handle outgoing HW checksumming |
db8dac20 DM |
712 | * @skb: sk_buff containing the filled-in UDP header |
713 | * (checksum field must be zeroed out) | |
f6b9664f HX |
714 | * @src: source IP address |
715 | * @dst: destination IP address | |
db8dac20 | 716 | */ |
c26bf4a5 | 717 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) |
db8dac20 | 718 | { |
db8dac20 | 719 | struct udphdr *uh = udp_hdr(skb); |
f6b9664f HX |
720 | int offset = skb_transport_offset(skb); |
721 | int len = skb->len - offset; | |
722 | int hlen = len; | |
db8dac20 DM |
723 | __wsum csum = 0; |
724 | ||
ebbe495f | 725 | if (!skb_has_frag_list(skb)) { |
db8dac20 DM |
726 | /* |
727 | * Only one fragment on the socket. | |
728 | */ | |
729 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
730 | skb->csum_offset = offsetof(struct udphdr, check); | |
f6b9664f HX |
731 | uh->check = ~csum_tcpudp_magic(src, dst, len, |
732 | IPPROTO_UDP, 0); | |
db8dac20 | 733 | } else { |
ebbe495f WC |
734 | struct sk_buff *frags; |
735 | ||
db8dac20 DM |
736 | /* |
737 | * HW-checksum won't work as there are two or more | |
738 | * fragments on the socket so that all csums of sk_buffs | |
739 | * should be together | |
740 | */ | |
ebbe495f | 741 | skb_walk_frags(skb, frags) { |
f6b9664f HX |
742 | csum = csum_add(csum, frags->csum); |
743 | hlen -= frags->len; | |
ebbe495f | 744 | } |
db8dac20 | 745 | |
f6b9664f | 746 | csum = skb_checksum(skb, offset, hlen, csum); |
db8dac20 DM |
747 | skb->ip_summed = CHECKSUM_NONE; |
748 | ||
db8dac20 DM |
749 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); |
750 | if (uh->check == 0) | |
751 | uh->check = CSUM_MANGLED_0; | |
752 | } | |
753 | } | |
c26bf4a5 | 754 | EXPORT_SYMBOL_GPL(udp4_hwcsum); |
db8dac20 | 755 | |
af5fcba7 TH |
756 | /* Function to set UDP checksum for an IPv4 UDP packet. This is intended |
757 | * for the simple case like when setting the checksum for a UDP tunnel. | |
758 | */ | |
759 | void udp_set_csum(bool nocheck, struct sk_buff *skb, | |
760 | __be32 saddr, __be32 daddr, int len) | |
761 | { | |
762 | struct udphdr *uh = udp_hdr(skb); | |
763 | ||
179bc67f | 764 | if (nocheck) { |
af5fcba7 | 765 | uh->check = 0; |
179bc67f | 766 | } else if (skb_is_gso(skb)) { |
af5fcba7 | 767 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); |
179bc67f EC |
768 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
769 | uh->check = 0; | |
770 | uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); | |
771 | if (uh->check == 0) | |
772 | uh->check = CSUM_MANGLED_0; | |
d75f1306 | 773 | } else { |
af5fcba7 TH |
774 | skb->ip_summed = CHECKSUM_PARTIAL; |
775 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
776 | skb->csum_offset = offsetof(struct udphdr, check); | |
777 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); | |
af5fcba7 TH |
778 | } |
779 | } | |
780 | EXPORT_SYMBOL(udp_set_csum); | |
781 | ||
79ab0531 | 782 | static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) |
db8dac20 | 783 | { |
f6b9664f | 784 | struct sock *sk = skb->sk; |
db8dac20 | 785 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
786 | struct udphdr *uh; |
787 | int err = 0; | |
788 | int is_udplite = IS_UDPLITE(sk); | |
f6b9664f HX |
789 | int offset = skb_transport_offset(skb); |
790 | int len = skb->len - offset; | |
db8dac20 DM |
791 | __wsum csum = 0; |
792 | ||
db8dac20 DM |
793 | /* |
794 | * Create a UDP header | |
795 | */ | |
796 | uh = udp_hdr(skb); | |
f6b9664f | 797 | uh->source = inet->inet_sport; |
79ab0531 | 798 | uh->dest = fl4->fl4_dport; |
f6b9664f | 799 | uh->len = htons(len); |
db8dac20 DM |
800 | uh->check = 0; |
801 | ||
802 | if (is_udplite) /* UDP-Lite */ | |
f6b9664f | 803 | csum = udplite_csum(skb); |
db8dac20 | 804 | |
28448b80 | 805 | else if (sk->sk_no_check_tx) { /* UDP csum disabled */ |
db8dac20 DM |
806 | |
807 | skb->ip_summed = CHECKSUM_NONE; | |
808 | goto send; | |
809 | ||
810 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
811 | ||
79ab0531 | 812 | udp4_hwcsum(skb, fl4->saddr, fl4->daddr); |
db8dac20 DM |
813 | goto send; |
814 | ||
f6b9664f HX |
815 | } else |
816 | csum = udp_csum(skb); | |
db8dac20 DM |
817 | |
818 | /* add protocol-dependent pseudo-header */ | |
79ab0531 | 819 | uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, |
c482c568 | 820 | sk->sk_protocol, csum); |
db8dac20 DM |
821 | if (uh->check == 0) |
822 | uh->check = CSUM_MANGLED_0; | |
823 | ||
824 | send: | |
b5ec8eea | 825 | err = ip_send_skb(sock_net(sk), skb); |
6ce9e7b5 ED |
826 | if (err) { |
827 | if (err == -ENOBUFS && !inet->recverr) { | |
6aef70a8 ED |
828 | UDP_INC_STATS(sock_net(sk), |
829 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
6ce9e7b5 ED |
830 | err = 0; |
831 | } | |
832 | } else | |
6aef70a8 ED |
833 | UDP_INC_STATS(sock_net(sk), |
834 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
f6b9664f HX |
835 | return err; |
836 | } | |
837 | ||
838 | /* | |
839 | * Push out all pending data as one UDP datagram. Socket is locked. | |
840 | */ | |
8822b64a | 841 | int udp_push_pending_frames(struct sock *sk) |
f6b9664f HX |
842 | { |
843 | struct udp_sock *up = udp_sk(sk); | |
844 | struct inet_sock *inet = inet_sk(sk); | |
b6f21b26 | 845 | struct flowi4 *fl4 = &inet->cork.fl.u.ip4; |
f6b9664f HX |
846 | struct sk_buff *skb; |
847 | int err = 0; | |
848 | ||
77968b78 | 849 | skb = ip_finish_skb(sk, fl4); |
f6b9664f HX |
850 | if (!skb) |
851 | goto out; | |
852 | ||
79ab0531 | 853 | err = udp_send_skb(skb, fl4); |
f6b9664f | 854 | |
db8dac20 DM |
855 | out: |
856 | up->len = 0; | |
857 | up->pending = 0; | |
db8dac20 DM |
858 | return err; |
859 | } | |
8822b64a | 860 | EXPORT_SYMBOL(udp_push_pending_frames); |
db8dac20 | 861 | |
1b784140 | 862 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
db8dac20 DM |
863 | { |
864 | struct inet_sock *inet = inet_sk(sk); | |
865 | struct udp_sock *up = udp_sk(sk); | |
e474995f | 866 | struct flowi4 fl4_stack; |
b6f21b26 | 867 | struct flowi4 *fl4; |
db8dac20 DM |
868 | int ulen = len; |
869 | struct ipcm_cookie ipc; | |
870 | struct rtable *rt = NULL; | |
871 | int free = 0; | |
872 | int connected = 0; | |
873 | __be32 daddr, faddr, saddr; | |
874 | __be16 dport; | |
875 | u8 tos; | |
876 | int err, is_udplite = IS_UDPLITE(sk); | |
877 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
878 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
903ab86d | 879 | struct sk_buff *skb; |
f6d8bd05 | 880 | struct ip_options_data opt_copy; |
db8dac20 DM |
881 | |
882 | if (len > 0xFFFF) | |
883 | return -EMSGSIZE; | |
884 | ||
885 | /* | |
886 | * Check the flags. | |
887 | */ | |
888 | ||
c482c568 | 889 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
890 | return -EOPNOTSUPP; |
891 | ||
892 | ipc.opt = NULL; | |
2244d07b | 893 | ipc.tx_flags = 0; |
aa661581 FF |
894 | ipc.ttl = 0; |
895 | ipc.tos = -1; | |
db8dac20 | 896 | |
903ab86d HX |
897 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
898 | ||
f5fca608 | 899 | fl4 = &inet->cork.fl.u.ip4; |
db8dac20 DM |
900 | if (up->pending) { |
901 | /* | |
902 | * There are pending frames. | |
903 | * The socket lock must be held while it's corked. | |
904 | */ | |
905 | lock_sock(sk); | |
906 | if (likely(up->pending)) { | |
907 | if (unlikely(up->pending != AF_INET)) { | |
908 | release_sock(sk); | |
909 | return -EINVAL; | |
910 | } | |
911 | goto do_append_data; | |
912 | } | |
913 | release_sock(sk); | |
914 | } | |
915 | ulen += sizeof(struct udphdr); | |
916 | ||
917 | /* | |
918 | * Get and verify the address. | |
919 | */ | |
920 | if (msg->msg_name) { | |
342dfc30 | 921 | DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); |
db8dac20 DM |
922 | if (msg->msg_namelen < sizeof(*usin)) |
923 | return -EINVAL; | |
924 | if (usin->sin_family != AF_INET) { | |
925 | if (usin->sin_family != AF_UNSPEC) | |
926 | return -EAFNOSUPPORT; | |
927 | } | |
928 | ||
929 | daddr = usin->sin_addr.s_addr; | |
930 | dport = usin->sin_port; | |
931 | if (dport == 0) | |
932 | return -EINVAL; | |
933 | } else { | |
934 | if (sk->sk_state != TCP_ESTABLISHED) | |
935 | return -EDESTADDRREQ; | |
c720c7e8 ED |
936 | daddr = inet->inet_daddr; |
937 | dport = inet->inet_dport; | |
db8dac20 DM |
938 | /* Open fast path for connected socket. |
939 | Route will not be used, if at least one option is set. | |
940 | */ | |
941 | connected = 1; | |
942 | } | |
db8dac20 | 943 | |
c14ac945 SHY |
944 | ipc.sockc.tsflags = sk->sk_tsflags; |
945 | ipc.addr = inet->inet_saddr; | |
db8dac20 | 946 | ipc.oif = sk->sk_bound_dev_if; |
bf84a010 | 947 | |
db8dac20 | 948 | if (msg->msg_controllen) { |
24025c46 | 949 | err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6); |
91948309 ED |
950 | if (unlikely(err)) { |
951 | kfree(ipc.opt); | |
db8dac20 | 952 | return err; |
91948309 | 953 | } |
db8dac20 DM |
954 | if (ipc.opt) |
955 | free = 1; | |
956 | connected = 0; | |
957 | } | |
f6d8bd05 ED |
958 | if (!ipc.opt) { |
959 | struct ip_options_rcu *inet_opt; | |
960 | ||
961 | rcu_read_lock(); | |
962 | inet_opt = rcu_dereference(inet->inet_opt); | |
963 | if (inet_opt) { | |
964 | memcpy(&opt_copy, inet_opt, | |
965 | sizeof(*inet_opt) + inet_opt->opt.optlen); | |
966 | ipc.opt = &opt_copy.opt; | |
967 | } | |
968 | rcu_read_unlock(); | |
969 | } | |
db8dac20 DM |
970 | |
971 | saddr = ipc.addr; | |
972 | ipc.addr = faddr = daddr; | |
973 | ||
c14ac945 SHY |
974 | sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags); |
975 | ||
f6d8bd05 | 976 | if (ipc.opt && ipc.opt->opt.srr) { |
db8dac20 DM |
977 | if (!daddr) |
978 | return -EINVAL; | |
f6d8bd05 | 979 | faddr = ipc.opt->opt.faddr; |
db8dac20 DM |
980 | connected = 0; |
981 | } | |
aa661581 | 982 | tos = get_rttos(&ipc, inet); |
db8dac20 DM |
983 | if (sock_flag(sk, SOCK_LOCALROUTE) || |
984 | (msg->msg_flags & MSG_DONTROUTE) || | |
f6d8bd05 | 985 | (ipc.opt && ipc.opt->opt.is_strictroute)) { |
db8dac20 DM |
986 | tos |= RTO_ONLINK; |
987 | connected = 0; | |
988 | } | |
989 | ||
990 | if (ipv4_is_multicast(daddr)) { | |
991 | if (!ipc.oif) | |
992 | ipc.oif = inet->mc_index; | |
993 | if (!saddr) | |
994 | saddr = inet->mc_addr; | |
995 | connected = 0; | |
76e21053 EH |
996 | } else if (!ipc.oif) |
997 | ipc.oif = inet->uc_index; | |
db8dac20 DM |
998 | |
999 | if (connected) | |
c482c568 | 1000 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 | 1001 | |
51456b29 | 1002 | if (!rt) { |
84a3aa00 | 1003 | struct net *net = sock_net(sk); |
9a24abfa | 1004 | __u8 flow_flags = inet_sk_flowi_flags(sk); |
84a3aa00 | 1005 | |
e474995f | 1006 | fl4 = &fl4_stack; |
9a24abfa | 1007 | |
e474995f | 1008 | flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, |
c0951cbc | 1009 | RT_SCOPE_UNIVERSE, sk->sk_protocol, |
9a24abfa | 1010 | flow_flags, |
e2d118a1 LC |
1011 | faddr, saddr, dport, inet->inet_sport, |
1012 | sk->sk_uid); | |
c0951cbc | 1013 | |
e474995f DM |
1014 | security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); |
1015 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe DM |
1016 | if (IS_ERR(rt)) { |
1017 | err = PTR_ERR(rt); | |
06dc94b1 | 1018 | rt = NULL; |
db8dac20 | 1019 | if (err == -ENETUNREACH) |
f1d8cba6 | 1020 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
1021 | goto out; |
1022 | } | |
1023 | ||
1024 | err = -EACCES; | |
1025 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
1026 | !sock_flag(sk, SOCK_BROADCAST)) | |
1027 | goto out; | |
1028 | if (connected) | |
d8d1f30b | 1029 | sk_dst_set(sk, dst_clone(&rt->dst)); |
db8dac20 DM |
1030 | } |
1031 | ||
1032 | if (msg->msg_flags&MSG_CONFIRM) | |
1033 | goto do_confirm; | |
1034 | back_from_confirm: | |
1035 | ||
e474995f | 1036 | saddr = fl4->saddr; |
db8dac20 | 1037 | if (!ipc.addr) |
e474995f | 1038 | daddr = ipc.addr = fl4->daddr; |
db8dac20 | 1039 | |
903ab86d HX |
1040 | /* Lockless fast path for the non-corking case. */ |
1041 | if (!corkreq) { | |
f69e6d13 | 1042 | skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, |
903ab86d HX |
1043 | sizeof(struct udphdr), &ipc, &rt, |
1044 | msg->msg_flags); | |
1045 | err = PTR_ERR(skb); | |
50c3a487 | 1046 | if (!IS_ERR_OR_NULL(skb)) |
79ab0531 | 1047 | err = udp_send_skb(skb, fl4); |
903ab86d HX |
1048 | goto out; |
1049 | } | |
1050 | ||
db8dac20 DM |
1051 | lock_sock(sk); |
1052 | if (unlikely(up->pending)) { | |
1053 | /* The socket is already corked while preparing it. */ | |
1054 | /* ... which is an evident application bug. --ANK */ | |
1055 | release_sock(sk); | |
1056 | ||
ba7a46f1 | 1057 | net_dbg_ratelimited("cork app bug 2\n"); |
db8dac20 DM |
1058 | err = -EINVAL; |
1059 | goto out; | |
1060 | } | |
1061 | /* | |
1062 | * Now cork the socket to pend data. | |
1063 | */ | |
b6f21b26 DM |
1064 | fl4 = &inet->cork.fl.u.ip4; |
1065 | fl4->daddr = daddr; | |
1066 | fl4->saddr = saddr; | |
9cce96df DM |
1067 | fl4->fl4_dport = dport; |
1068 | fl4->fl4_sport = inet->inet_sport; | |
db8dac20 DM |
1069 | up->pending = AF_INET; |
1070 | ||
1071 | do_append_data: | |
1072 | up->len += ulen; | |
f69e6d13 | 1073 | err = ip_append_data(sk, fl4, getfrag, msg, ulen, |
f5fca608 DM |
1074 | sizeof(struct udphdr), &ipc, &rt, |
1075 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); | |
db8dac20 DM |
1076 | if (err) |
1077 | udp_flush_pending_frames(sk); | |
1078 | else if (!corkreq) | |
1079 | err = udp_push_pending_frames(sk); | |
1080 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1081 | up->pending = 0; | |
1082 | release_sock(sk); | |
1083 | ||
1084 | out: | |
1085 | ip_rt_put(rt); | |
1086 | if (free) | |
1087 | kfree(ipc.opt); | |
1088 | if (!err) | |
1089 | return len; | |
1090 | /* | |
1091 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1092 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1093 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1094 | * things). We could add another new stat but at least for now that | |
1095 | * seems like overkill. | |
1096 | */ | |
1097 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
6aef70a8 ED |
1098 | UDP_INC_STATS(sock_net(sk), |
1099 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
1100 | } |
1101 | return err; | |
1102 | ||
1103 | do_confirm: | |
0dec879f JA |
1104 | if (msg->msg_flags & MSG_PROBE) |
1105 | dst_confirm_neigh(&rt->dst, &fl4->daddr); | |
db8dac20 DM |
1106 | if (!(msg->msg_flags&MSG_PROBE) || len) |
1107 | goto back_from_confirm; | |
1108 | err = 0; | |
1109 | goto out; | |
1110 | } | |
c482c568 | 1111 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
1112 | |
1113 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
1114 | size_t size, int flags) | |
1115 | { | |
f5fca608 | 1116 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
1117 | struct udp_sock *up = udp_sk(sk); |
1118 | int ret; | |
1119 | ||
d3f7d56a SL |
1120 | if (flags & MSG_SENDPAGE_NOTLAST) |
1121 | flags |= MSG_MORE; | |
1122 | ||
db8dac20 DM |
1123 | if (!up->pending) { |
1124 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
1125 | ||
1126 | /* Call udp_sendmsg to specify destination address which | |
1127 | * sendpage interface can't pass. | |
1128 | * This will succeed only when the socket is connected. | |
1129 | */ | |
1b784140 | 1130 | ret = udp_sendmsg(sk, &msg, 0); |
db8dac20 DM |
1131 | if (ret < 0) |
1132 | return ret; | |
1133 | } | |
1134 | ||
1135 | lock_sock(sk); | |
1136 | ||
1137 | if (unlikely(!up->pending)) { | |
1138 | release_sock(sk); | |
1139 | ||
ba7a46f1 | 1140 | net_dbg_ratelimited("udp cork app bug 3\n"); |
db8dac20 DM |
1141 | return -EINVAL; |
1142 | } | |
1143 | ||
f5fca608 DM |
1144 | ret = ip_append_page(sk, &inet->cork.fl.u.ip4, |
1145 | page, offset, size, flags); | |
db8dac20 DM |
1146 | if (ret == -EOPNOTSUPP) { |
1147 | release_sock(sk); | |
1148 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
1149 | size, flags); | |
1150 | } | |
1151 | if (ret < 0) { | |
1152 | udp_flush_pending_frames(sk); | |
1153 | goto out; | |
1154 | } | |
1155 | ||
1156 | up->len += size; | |
1157 | if (!(up->corkflag || (flags&MSG_MORE))) | |
1158 | ret = udp_push_pending_frames(sk); | |
1159 | if (!ret) | |
1160 | ret = size; | |
1161 | out: | |
1162 | release_sock(sk); | |
1163 | return ret; | |
1164 | } | |
1165 | ||
dce4551c PA |
1166 | #define UDP_SKB_IS_STATELESS 0x80000000 |
1167 | ||
b65ac446 PA |
1168 | static void udp_set_dev_scratch(struct sk_buff *skb) |
1169 | { | |
dce4551c | 1170 | struct udp_dev_scratch *scratch = udp_skb_scratch(skb); |
b65ac446 PA |
1171 | |
1172 | BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); | |
dce4551c PA |
1173 | scratch->_tsize_state = skb->truesize; |
1174 | #if BITS_PER_LONG == 64 | |
b65ac446 PA |
1175 | scratch->len = skb->len; |
1176 | scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); | |
1177 | scratch->is_linear = !skb_is_nonlinear(skb); | |
dce4551c PA |
1178 | #endif |
1179 | if (likely(!skb->_skb_refdst)) | |
1180 | scratch->_tsize_state |= UDP_SKB_IS_STATELESS; | |
b65ac446 PA |
1181 | } |
1182 | ||
1183 | static int udp_skb_truesize(struct sk_buff *skb) | |
1184 | { | |
dce4551c | 1185 | return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1186 | } |
1187 | ||
dce4551c | 1188 | static bool udp_skb_has_head_state(struct sk_buff *skb) |
b65ac446 | 1189 | { |
dce4551c | 1190 | return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); |
b65ac446 | 1191 | } |
b65ac446 | 1192 | |
7c13f97f | 1193 | /* fully reclaim rmem/fwd memory allocated for skb */ |
6dfb4367 PA |
1194 | static void udp_rmem_release(struct sock *sk, int size, int partial, |
1195 | bool rx_queue_lock_held) | |
f970bd9e | 1196 | { |
6b229cf7 | 1197 | struct udp_sock *up = udp_sk(sk); |
2276f58a | 1198 | struct sk_buff_head *sk_queue; |
f970bd9e PA |
1199 | int amt; |
1200 | ||
6b229cf7 ED |
1201 | if (likely(partial)) { |
1202 | up->forward_deficit += size; | |
1203 | size = up->forward_deficit; | |
1204 | if (size < (sk->sk_rcvbuf >> 2) && | |
2276f58a | 1205 | !skb_queue_empty(&up->reader_queue)) |
6b229cf7 ED |
1206 | return; |
1207 | } else { | |
1208 | size += up->forward_deficit; | |
1209 | } | |
1210 | up->forward_deficit = 0; | |
1211 | ||
6dfb4367 PA |
1212 | /* acquire the sk_receive_queue for fwd allocated memory scheduling, |
1213 | * if the called don't held it already | |
1214 | */ | |
2276f58a | 1215 | sk_queue = &sk->sk_receive_queue; |
6dfb4367 PA |
1216 | if (!rx_queue_lock_held) |
1217 | spin_lock(&sk_queue->lock); | |
1218 | ||
2276f58a | 1219 | |
f970bd9e PA |
1220 | sk->sk_forward_alloc += size; |
1221 | amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); | |
1222 | sk->sk_forward_alloc -= amt; | |
f970bd9e PA |
1223 | |
1224 | if (amt) | |
1225 | __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); | |
02ab0d13 ED |
1226 | |
1227 | atomic_sub(size, &sk->sk_rmem_alloc); | |
2276f58a PA |
1228 | |
1229 | /* this can save us from acquiring the rx queue lock on next receive */ | |
1230 | skb_queue_splice_tail_init(sk_queue, &up->reader_queue); | |
1231 | ||
6dfb4367 PA |
1232 | if (!rx_queue_lock_held) |
1233 | spin_unlock(&sk_queue->lock); | |
f970bd9e PA |
1234 | } |
1235 | ||
2276f58a | 1236 | /* Note: called with reader_queue.lock held. |
c84d9490 ED |
1237 | * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch |
1238 | * This avoids a cache line miss while receive_queue lock is held. | |
1239 | * Look at __udp_enqueue_schedule_skb() to find where this copy is done. | |
1240 | */ | |
7c13f97f | 1241 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) |
f970bd9e | 1242 | { |
b65ac446 PA |
1243 | prefetch(&skb->data); |
1244 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); | |
f970bd9e | 1245 | } |
7c13f97f | 1246 | EXPORT_SYMBOL(udp_skb_destructor); |
f970bd9e | 1247 | |
6dfb4367 | 1248 | /* as above, but the caller held the rx queue lock, too */ |
64f5102d | 1249 | static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) |
6dfb4367 | 1250 | { |
b65ac446 PA |
1251 | prefetch(&skb->data); |
1252 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); | |
6dfb4367 PA |
1253 | } |
1254 | ||
4b272750 ED |
1255 | /* Idea of busylocks is to let producers grab an extra spinlock |
1256 | * to relieve pressure on the receive_queue spinlock shared by consumer. | |
1257 | * Under flood, this means that only one producer can be in line | |
1258 | * trying to acquire the receive_queue spinlock. | |
1259 | * These busylock can be allocated on a per cpu manner, instead of a | |
1260 | * per socket one (that would consume a cache line per socket) | |
1261 | */ | |
1262 | static int udp_busylocks_log __read_mostly; | |
1263 | static spinlock_t *udp_busylocks __read_mostly; | |
1264 | ||
1265 | static spinlock_t *busylock_acquire(void *ptr) | |
1266 | { | |
1267 | spinlock_t *busy; | |
1268 | ||
1269 | busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); | |
1270 | spin_lock(busy); | |
1271 | return busy; | |
1272 | } | |
1273 | ||
1274 | static void busylock_release(spinlock_t *busy) | |
1275 | { | |
1276 | if (busy) | |
1277 | spin_unlock(busy); | |
1278 | } | |
1279 | ||
f970bd9e PA |
1280 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) |
1281 | { | |
1282 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
1283 | int rmem, delta, amt, err = -ENOMEM; | |
4b272750 | 1284 | spinlock_t *busy = NULL; |
c8c8b127 | 1285 | int size; |
f970bd9e PA |
1286 | |
1287 | /* try to avoid the costly atomic add/sub pair when the receive | |
1288 | * queue is full; always allow at least a packet | |
1289 | */ | |
1290 | rmem = atomic_read(&sk->sk_rmem_alloc); | |
363dc73a | 1291 | if (rmem > sk->sk_rcvbuf) |
f970bd9e PA |
1292 | goto drop; |
1293 | ||
c8c8b127 ED |
1294 | /* Under mem pressure, it might be helpful to help udp_recvmsg() |
1295 | * having linear skbs : | |
1296 | * - Reduce memory overhead and thus increase receive queue capacity | |
1297 | * - Less cache line misses at copyout() time | |
1298 | * - Less work at consume_skb() (less alien page frag freeing) | |
1299 | */ | |
4b272750 | 1300 | if (rmem > (sk->sk_rcvbuf >> 1)) { |
c8c8b127 | 1301 | skb_condense(skb); |
4b272750 ED |
1302 | |
1303 | busy = busylock_acquire(sk); | |
1304 | } | |
c8c8b127 | 1305 | size = skb->truesize; |
b65ac446 | 1306 | udp_set_dev_scratch(skb); |
c8c8b127 | 1307 | |
f970bd9e PA |
1308 | /* we drop only if the receive buf is full and the receive |
1309 | * queue contains some other skb | |
1310 | */ | |
1311 | rmem = atomic_add_return(size, &sk->sk_rmem_alloc); | |
363dc73a | 1312 | if (rmem > (size + sk->sk_rcvbuf)) |
f970bd9e PA |
1313 | goto uncharge_drop; |
1314 | ||
1315 | spin_lock(&list->lock); | |
1316 | if (size >= sk->sk_forward_alloc) { | |
1317 | amt = sk_mem_pages(size); | |
1318 | delta = amt << SK_MEM_QUANTUM_SHIFT; | |
1319 | if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { | |
1320 | err = -ENOBUFS; | |
1321 | spin_unlock(&list->lock); | |
1322 | goto uncharge_drop; | |
1323 | } | |
1324 | ||
1325 | sk->sk_forward_alloc += delta; | |
1326 | } | |
1327 | ||
1328 | sk->sk_forward_alloc -= size; | |
1329 | ||
7c13f97f PA |
1330 | /* no need to setup a destructor, we will explicitly release the |
1331 | * forward allocated memory on dequeue | |
1332 | */ | |
f970bd9e PA |
1333 | sock_skb_set_dropcount(sk, skb); |
1334 | ||
1335 | __skb_queue_tail(list, skb); | |
1336 | spin_unlock(&list->lock); | |
1337 | ||
1338 | if (!sock_flag(sk, SOCK_DEAD)) | |
1339 | sk->sk_data_ready(sk); | |
1340 | ||
4b272750 | 1341 | busylock_release(busy); |
f970bd9e PA |
1342 | return 0; |
1343 | ||
1344 | uncharge_drop: | |
1345 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); | |
1346 | ||
1347 | drop: | |
1348 | atomic_inc(&sk->sk_drops); | |
4b272750 | 1349 | busylock_release(busy); |
f970bd9e PA |
1350 | return err; |
1351 | } | |
1352 | EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); | |
1353 | ||
c915fe13 | 1354 | void udp_destruct_sock(struct sock *sk) |
f970bd9e PA |
1355 | { |
1356 | /* reclaim completely the forward allocated memory */ | |
2276f58a | 1357 | struct udp_sock *up = udp_sk(sk); |
7c13f97f PA |
1358 | unsigned int total = 0; |
1359 | struct sk_buff *skb; | |
1360 | ||
2276f58a PA |
1361 | skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); |
1362 | while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { | |
7c13f97f PA |
1363 | total += skb->truesize; |
1364 | kfree_skb(skb); | |
1365 | } | |
6dfb4367 | 1366 | udp_rmem_release(sk, total, 0, true); |
7c13f97f | 1367 | |
f970bd9e PA |
1368 | inet_sock_destruct(sk); |
1369 | } | |
c915fe13 | 1370 | EXPORT_SYMBOL_GPL(udp_destruct_sock); |
f970bd9e PA |
1371 | |
1372 | int udp_init_sock(struct sock *sk) | |
1373 | { | |
2276f58a | 1374 | skb_queue_head_init(&udp_sk(sk)->reader_queue); |
f970bd9e PA |
1375 | sk->sk_destruct = udp_destruct_sock; |
1376 | return 0; | |
1377 | } | |
1378 | EXPORT_SYMBOL_GPL(udp_init_sock); | |
1379 | ||
1380 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) | |
1381 | { | |
1382 | if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { | |
1383 | bool slow = lock_sock_fast(sk); | |
1384 | ||
1385 | sk_peek_offset_bwd(sk, len); | |
1386 | unlock_sock_fast(sk, slow); | |
1387 | } | |
0a463c78 | 1388 | |
dce4551c PA |
1389 | /* In the more common cases we cleared the head states previously, |
1390 | * see __udp_queue_rcv_skb(). | |
0ddf3fb2 | 1391 | */ |
dce4551c | 1392 | if (unlikely(udp_skb_has_head_state(skb))) |
0ddf3fb2 | 1393 | skb_release_head_state(skb); |
0a463c78 | 1394 | consume_stateless_skb(skb); |
f970bd9e PA |
1395 | } |
1396 | EXPORT_SYMBOL_GPL(skb_consume_udp); | |
1397 | ||
2276f58a PA |
1398 | static struct sk_buff *__first_packet_length(struct sock *sk, |
1399 | struct sk_buff_head *rcvq, | |
1400 | int *total) | |
1401 | { | |
1402 | struct sk_buff *skb; | |
1403 | ||
9bd780f5 PA |
1404 | while ((skb = skb_peek(rcvq)) != NULL) { |
1405 | if (udp_lib_checksum_complete(skb)) { | |
1406 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1407 | IS_UDPLITE(sk)); | |
1408 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1409 | IS_UDPLITE(sk)); | |
1410 | atomic_inc(&sk->sk_drops); | |
1411 | __skb_unlink(skb, rcvq); | |
1412 | *total += skb->truesize; | |
1413 | kfree_skb(skb); | |
1414 | } else { | |
1415 | /* the csum related bits could be changed, refresh | |
1416 | * the scratch area | |
1417 | */ | |
1418 | udp_set_dev_scratch(skb); | |
1419 | break; | |
1420 | } | |
2276f58a PA |
1421 | } |
1422 | return skb; | |
1423 | } | |
1424 | ||
85584672 ED |
1425 | /** |
1426 | * first_packet_length - return length of first packet in receive queue | |
1427 | * @sk: socket | |
1428 | * | |
1429 | * Drops all bad checksum frames, until a valid one is found. | |
e83c6744 | 1430 | * Returns the length of found skb, or -1 if none is found. |
85584672 | 1431 | */ |
e83c6744 | 1432 | static int first_packet_length(struct sock *sk) |
85584672 | 1433 | { |
2276f58a PA |
1434 | struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; |
1435 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
85584672 | 1436 | struct sk_buff *skb; |
7c13f97f | 1437 | int total = 0; |
e83c6744 | 1438 | int res; |
85584672 | 1439 | |
85584672 | 1440 | spin_lock_bh(&rcvq->lock); |
2276f58a PA |
1441 | skb = __first_packet_length(sk, rcvq, &total); |
1442 | if (!skb && !skb_queue_empty(sk_queue)) { | |
1443 | spin_lock(&sk_queue->lock); | |
1444 | skb_queue_splice_tail_init(sk_queue, rcvq); | |
1445 | spin_unlock(&sk_queue->lock); | |
1446 | ||
1447 | skb = __first_packet_length(sk, rcvq, &total); | |
85584672 | 1448 | } |
e83c6744 | 1449 | res = skb ? skb->len : -1; |
7c13f97f | 1450 | if (total) |
6dfb4367 | 1451 | udp_rmem_release(sk, total, 1, false); |
85584672 | 1452 | spin_unlock_bh(&rcvq->lock); |
85584672 ED |
1453 | return res; |
1454 | } | |
1455 | ||
1da177e4 LT |
1456 | /* |
1457 | * IOCTL requests applicable to the UDP protocol | |
1458 | */ | |
e905a9ed | 1459 | |
1da177e4 LT |
1460 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
1461 | { | |
6516c655 SH |
1462 | switch (cmd) { |
1463 | case SIOCOUTQ: | |
1da177e4 | 1464 | { |
31e6d363 ED |
1465 | int amount = sk_wmem_alloc_get(sk); |
1466 | ||
6516c655 SH |
1467 | return put_user(amount, (int __user *)arg); |
1468 | } | |
1da177e4 | 1469 | |
6516c655 SH |
1470 | case SIOCINQ: |
1471 | { | |
e83c6744 | 1472 | int amount = max_t(int, 0, first_packet_length(sk)); |
6516c655 | 1473 | |
6516c655 SH |
1474 | return put_user(amount, (int __user *)arg); |
1475 | } | |
1da177e4 | 1476 | |
6516c655 SH |
1477 | default: |
1478 | return -ENOIOCTLCMD; | |
1da177e4 | 1479 | } |
6516c655 SH |
1480 | |
1481 | return 0; | |
1da177e4 | 1482 | } |
c482c568 | 1483 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 1484 | |
2276f58a PA |
1485 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, |
1486 | int noblock, int *peeked, int *off, int *err) | |
1487 | { | |
1488 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
1489 | struct sk_buff_head *queue; | |
1490 | struct sk_buff *last; | |
1491 | long timeo; | |
1492 | int error; | |
1493 | ||
1494 | queue = &udp_sk(sk)->reader_queue; | |
1495 | flags |= noblock ? MSG_DONTWAIT : 0; | |
1496 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1497 | do { | |
1498 | struct sk_buff *skb; | |
1499 | ||
1500 | error = sock_error(sk); | |
1501 | if (error) | |
1502 | break; | |
1503 | ||
1504 | error = -EAGAIN; | |
1505 | *peeked = 0; | |
1506 | do { | |
2276f58a PA |
1507 | spin_lock_bh(&queue->lock); |
1508 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
1509 | udp_skb_destructor, | |
de321ed3 | 1510 | peeked, off, err, |
2276f58a PA |
1511 | &last); |
1512 | if (skb) { | |
1513 | spin_unlock_bh(&queue->lock); | |
2276f58a PA |
1514 | return skb; |
1515 | } | |
1516 | ||
1517 | if (skb_queue_empty(sk_queue)) { | |
1518 | spin_unlock_bh(&queue->lock); | |
1519 | goto busy_check; | |
1520 | } | |
1521 | ||
6dfb4367 PA |
1522 | /* refill the reader queue and walk it again |
1523 | * keep both queues locked to avoid re-acquiring | |
1524 | * the sk_receive_queue lock if fwd memory scheduling | |
1525 | * is needed. | |
1526 | */ | |
2276f58a PA |
1527 | spin_lock(&sk_queue->lock); |
1528 | skb_queue_splice_tail_init(sk_queue, queue); | |
2276f58a PA |
1529 | |
1530 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
6dfb4367 | 1531 | udp_skb_dtor_locked, |
de321ed3 | 1532 | peeked, off, err, |
2276f58a | 1533 | &last); |
6dfb4367 | 1534 | spin_unlock(&sk_queue->lock); |
2276f58a | 1535 | spin_unlock_bh(&queue->lock); |
de321ed3 | 1536 | if (skb) |
2276f58a | 1537 | return skb; |
2276f58a PA |
1538 | |
1539 | busy_check: | |
1540 | if (!sk_can_busy_loop(sk)) | |
1541 | break; | |
1542 | ||
1543 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
1544 | } while (!skb_queue_empty(sk_queue)); | |
1545 | ||
1546 | /* sk_queue is empty, reader_queue may contain peeked packets */ | |
1547 | } while (timeo && | |
1548 | !__skb_wait_for_more_packets(sk, &error, &timeo, | |
1549 | (struct sk_buff *)sk_queue)); | |
1550 | ||
1551 | *err = error; | |
1552 | return NULL; | |
1553 | } | |
1554 | EXPORT_SYMBOL_GPL(__skb_recv_udp); | |
1555 | ||
db8dac20 DM |
1556 | /* |
1557 | * This should be easy, if there is something there we | |
1558 | * return it, otherwise we block. | |
1559 | */ | |
1560 | ||
1b784140 YX |
1561 | int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, |
1562 | int flags, int *addr_len) | |
db8dac20 DM |
1563 | { |
1564 | struct inet_sock *inet = inet_sk(sk); | |
342dfc30 | 1565 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
db8dac20 | 1566 | struct sk_buff *skb; |
59c2cdae | 1567 | unsigned int ulen, copied; |
627d2d6b | 1568 | int peeked, peeking, off; |
db8dac20 DM |
1569 | int err; |
1570 | int is_udplite = IS_UDPLITE(sk); | |
197c949e | 1571 | bool checksum_valid = false; |
db8dac20 | 1572 | |
db8dac20 | 1573 | if (flags & MSG_ERRQUEUE) |
85fbaa75 | 1574 | return ip_recv_error(sk, msg, len, addr_len); |
db8dac20 DM |
1575 | |
1576 | try_again: | |
627d2d6b | 1577 | peeking = off = sk_peek_offset(sk, flags); |
7c13f97f | 1578 | skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); |
db8dac20 | 1579 | if (!skb) |
627d2d6b | 1580 | return err; |
db8dac20 | 1581 | |
b65ac446 | 1582 | ulen = udp_skb_len(skb); |
59c2cdae | 1583 | copied = len; |
627d2d6b | 1584 | if (copied > ulen - off) |
1585 | copied = ulen - off; | |
59c2cdae | 1586 | else if (copied < ulen) |
db8dac20 DM |
1587 | msg->msg_flags |= MSG_TRUNC; |
1588 | ||
1589 | /* | |
1590 | * If checksum is needed at all, try to do it while copying the | |
1591 | * data. If the data is truncated, or if we only want a partial | |
1592 | * coverage checksum (UDP-Lite), do it before the copy. | |
1593 | */ | |
1594 | ||
d21dbdfe ED |
1595 | if (copied < ulen || peeking || |
1596 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
b65ac446 PA |
1597 | checksum_valid = udp_skb_csum_unnecessary(skb) || |
1598 | !__udp_lib_checksum_complete(skb); | |
197c949e | 1599 | if (!checksum_valid) |
db8dac20 DM |
1600 | goto csum_copy_err; |
1601 | } | |
1602 | ||
b65ac446 PA |
1603 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { |
1604 | if (udp_skb_is_linear(skb)) | |
1605 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
1606 | else | |
1607 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
1608 | } else { | |
627d2d6b | 1609 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); |
db8dac20 DM |
1610 | |
1611 | if (err == -EINVAL) | |
1612 | goto csum_copy_err; | |
1613 | } | |
1614 | ||
22911fc5 | 1615 | if (unlikely(err)) { |
979402b1 ED |
1616 | if (!peeked) { |
1617 | atomic_inc(&sk->sk_drops); | |
6aef70a8 ED |
1618 | UDP_INC_STATS(sock_net(sk), |
1619 | UDP_MIB_INERRORS, is_udplite); | |
979402b1 | 1620 | } |
850cbadd | 1621 | kfree_skb(skb); |
627d2d6b | 1622 | return err; |
22911fc5 | 1623 | } |
db8dac20 DM |
1624 | |
1625 | if (!peeked) | |
6aef70a8 ED |
1626 | UDP_INC_STATS(sock_net(sk), |
1627 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 | 1628 | |
3b885787 | 1629 | sock_recv_ts_and_drops(msg, sk, skb); |
db8dac20 DM |
1630 | |
1631 | /* Copy the address. */ | |
c482c568 | 1632 | if (sin) { |
db8dac20 DM |
1633 | sin->sin_family = AF_INET; |
1634 | sin->sin_port = udp_hdr(skb)->source; | |
1635 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
1636 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
bceaa902 | 1637 | *addr_len = sizeof(*sin); |
db8dac20 DM |
1638 | } |
1639 | if (inet->cmsg_flags) | |
ad959036 | 1640 | ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); |
db8dac20 | 1641 | |
59c2cdae | 1642 | err = copied; |
db8dac20 DM |
1643 | if (flags & MSG_TRUNC) |
1644 | err = ulen; | |
1645 | ||
850cbadd | 1646 | skb_consume_udp(sk, skb, peeking ? -err : err); |
db8dac20 DM |
1647 | return err; |
1648 | ||
1649 | csum_copy_err: | |
2276f58a PA |
1650 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, |
1651 | udp_skb_destructor)) { | |
6aef70a8 ED |
1652 | UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
1653 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
6a5dc9e5 | 1654 | } |
850cbadd | 1655 | kfree_skb(skb); |
db8dac20 | 1656 | |
beb39db5 ED |
1657 | /* starting over for a new packet, but check if we need to yield */ |
1658 | cond_resched(); | |
9cfaa8de | 1659 | msg->msg_flags &= ~MSG_TRUNC; |
db8dac20 DM |
1660 | goto try_again; |
1661 | } | |
1662 | ||
286c72de | 1663 | int __udp_disconnect(struct sock *sk, int flags) |
1da177e4 LT |
1664 | { |
1665 | struct inet_sock *inet = inet_sk(sk); | |
1666 | /* | |
1667 | * 1003.1g - break association. | |
1668 | */ | |
e905a9ed | 1669 | |
1da177e4 | 1670 | sk->sk_state = TCP_CLOSE; |
c720c7e8 ED |
1671 | inet->inet_daddr = 0; |
1672 | inet->inet_dport = 0; | |
bdeab991 | 1673 | sock_rps_reset_rxhash(sk); |
1da177e4 LT |
1674 | sk->sk_bound_dev_if = 0; |
1675 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
1676 | inet_reset_saddr(sk); | |
1677 | ||
1678 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1679 | sk->sk_prot->unhash(sk); | |
c720c7e8 | 1680 | inet->inet_sport = 0; |
1da177e4 LT |
1681 | } |
1682 | sk_dst_reset(sk); | |
1683 | return 0; | |
1684 | } | |
286c72de ED |
1685 | EXPORT_SYMBOL(__udp_disconnect); |
1686 | ||
1687 | int udp_disconnect(struct sock *sk, int flags) | |
1688 | { | |
1689 | lock_sock(sk); | |
1690 | __udp_disconnect(sk, flags); | |
1691 | release_sock(sk); | |
1692 | return 0; | |
1693 | } | |
c482c568 | 1694 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 1695 | |
645ca708 ED |
1696 | void udp_lib_unhash(struct sock *sk) |
1697 | { | |
723b4610 ED |
1698 | if (sk_hashed(sk)) { |
1699 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
512615b6 ED |
1700 | struct udp_hslot *hslot, *hslot2; |
1701 | ||
1702 | hslot = udp_hashslot(udptable, sock_net(sk), | |
1703 | udp_sk(sk)->udp_port_hash); | |
1704 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
645ca708 | 1705 | |
723b4610 | 1706 | spin_lock_bh(&hslot->lock); |
e32ea7e7 CG |
1707 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1708 | reuseport_detach_sock(sk); | |
ca065d0c | 1709 | if (sk_del_node_init_rcu(sk)) { |
fdcc8aa9 | 1710 | hslot->count--; |
c720c7e8 | 1711 | inet_sk(sk)->inet_num = 0; |
723b4610 | 1712 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
512615b6 ED |
1713 | |
1714 | spin_lock(&hslot2->lock); | |
ca065d0c | 1715 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
512615b6 ED |
1716 | hslot2->count--; |
1717 | spin_unlock(&hslot2->lock); | |
723b4610 ED |
1718 | } |
1719 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 1720 | } |
645ca708 ED |
1721 | } |
1722 | EXPORT_SYMBOL(udp_lib_unhash); | |
1723 | ||
719f8358 ED |
1724 | /* |
1725 | * inet_rcv_saddr was changed, we must rehash secondary hash | |
1726 | */ | |
1727 | void udp_lib_rehash(struct sock *sk, u16 newhash) | |
1728 | { | |
1729 | if (sk_hashed(sk)) { | |
1730 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
1731 | struct udp_hslot *hslot, *hslot2, *nhslot2; | |
1732 | ||
1733 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
1734 | nhslot2 = udp_hashslot2(udptable, newhash); | |
1735 | udp_sk(sk)->udp_portaddr_hash = newhash; | |
e32ea7e7 CG |
1736 | |
1737 | if (hslot2 != nhslot2 || | |
1738 | rcu_access_pointer(sk->sk_reuseport_cb)) { | |
719f8358 ED |
1739 | hslot = udp_hashslot(udptable, sock_net(sk), |
1740 | udp_sk(sk)->udp_port_hash); | |
1741 | /* we must lock primary chain too */ | |
1742 | spin_lock_bh(&hslot->lock); | |
e32ea7e7 CG |
1743 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1744 | reuseport_detach_sock(sk); | |
1745 | ||
1746 | if (hslot2 != nhslot2) { | |
1747 | spin_lock(&hslot2->lock); | |
ca065d0c | 1748 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
e32ea7e7 CG |
1749 | hslot2->count--; |
1750 | spin_unlock(&hslot2->lock); | |
1751 | ||
1752 | spin_lock(&nhslot2->lock); | |
ca065d0c | 1753 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
e32ea7e7 CG |
1754 | &nhslot2->head); |
1755 | nhslot2->count++; | |
1756 | spin_unlock(&nhslot2->lock); | |
1757 | } | |
719f8358 ED |
1758 | |
1759 | spin_unlock_bh(&hslot->lock); | |
1760 | } | |
1761 | } | |
1762 | } | |
1763 | EXPORT_SYMBOL(udp_lib_rehash); | |
1764 | ||
1765 | static void udp_v4_rehash(struct sock *sk) | |
1766 | { | |
1767 | u16 new_hash = udp4_portaddr_hash(sock_net(sk), | |
1768 | inet_sk(sk)->inet_rcv_saddr, | |
1769 | inet_sk(sk)->inet_num); | |
1770 | udp_lib_rehash(sk, new_hash); | |
1771 | } | |
1772 | ||
a3f96c47 | 1773 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
93821778 | 1774 | { |
fec5e652 | 1775 | int rc; |
766e9037 | 1776 | |
005ec974 | 1777 | if (inet_sk(sk)->inet_daddr) { |
bdeab991 | 1778 | sock_rps_save_rxhash(sk, skb); |
005ec974 | 1779 | sk_mark_napi_id(sk, skb); |
2c8c56e1 | 1780 | sk_incoming_cpu_update(sk); |
e68b6e50 ED |
1781 | } else { |
1782 | sk_mark_napi_id_once(sk, skb); | |
005ec974 | 1783 | } |
fec5e652 | 1784 | |
dce4551c PA |
1785 | /* At recvmsg() time we may access skb->dst or skb->sp depending on |
1786 | * the IP options and the cmsg flags, elsewhere can we clear all | |
1787 | * pending head states while they are hot in the cache | |
0ddf3fb2 | 1788 | */ |
9688f9b0 | 1789 | if (likely(IPCB(skb)->opt.optlen == 0 && !skb_sec_path(skb))) |
0ddf3fb2 | 1790 | skb_release_head_state(skb); |
0a463c78 | 1791 | |
850cbadd | 1792 | rc = __udp_enqueue_schedule_skb(sk, skb); |
766e9037 ED |
1793 | if (rc < 0) { |
1794 | int is_udplite = IS_UDPLITE(sk); | |
93821778 | 1795 | |
93821778 | 1796 | /* Note that an ENOMEM error is charged twice */ |
766e9037 | 1797 | if (rc == -ENOMEM) |
e61da9e2 | 1798 | UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
02c22347 | 1799 | is_udplite); |
e61da9e2 | 1800 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
766e9037 | 1801 | kfree_skb(skb); |
296f7ea7 | 1802 | trace_udp_fail_queue_rcv_skb(rc, sk); |
766e9037 | 1803 | return -1; |
93821778 HX |
1804 | } |
1805 | ||
1806 | return 0; | |
93821778 HX |
1807 | } |
1808 | ||
447167bf ED |
1809 | static struct static_key udp_encap_needed __read_mostly; |
1810 | void udp_encap_enable(void) | |
1811 | { | |
1812 | if (!static_key_enabled(&udp_encap_needed)) | |
1813 | static_key_slow_inc(&udp_encap_needed); | |
1814 | } | |
1815 | EXPORT_SYMBOL(udp_encap_enable); | |
1816 | ||
db8dac20 DM |
1817 | /* returns: |
1818 | * -1: error | |
1819 | * 0: success | |
1820 | * >0: "udp encap" protocol resubmission | |
1821 | * | |
1822 | * Note that in the success and error cases, the skb is assumed to | |
1823 | * have either been requeued or freed. | |
1824 | */ | |
a3f96c47 | 1825 | static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 DM |
1826 | { |
1827 | struct udp_sock *up = udp_sk(sk); | |
db8dac20 DM |
1828 | int is_udplite = IS_UDPLITE(sk); |
1829 | ||
1830 | /* | |
1831 | * Charge it to the socket, dropping if the queue is full. | |
1832 | */ | |
1833 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1834 | goto drop; | |
1835 | nf_reset(skb); | |
1836 | ||
447167bf | 1837 | if (static_key_false(&udp_encap_needed) && up->encap_type) { |
0ad92ad0 ED |
1838 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
1839 | ||
db8dac20 DM |
1840 | /* |
1841 | * This is an encapsulation socket so pass the skb to | |
1842 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
1843 | * fall through and pass this up the UDP socket. | |
1844 | * up->encap_rcv() returns the following value: | |
1845 | * =0 if skb was successfully passed to the encap | |
1846 | * handler or was discarded by it. | |
1847 | * >0 if skb should be passed on to UDP. | |
1848 | * <0 if skb should be resubmitted as proto -N | |
1849 | */ | |
1850 | ||
1851 | /* if we're overly short, let UDP handle it */ | |
0ad92ad0 | 1852 | encap_rcv = ACCESS_ONCE(up->encap_rcv); |
e5aed006 | 1853 | if (encap_rcv) { |
db8dac20 DM |
1854 | int ret; |
1855 | ||
0a80966b TH |
1856 | /* Verify checksum before giving to encap */ |
1857 | if (udp_lib_checksum_complete(skb)) | |
1858 | goto csum_error; | |
1859 | ||
0ad92ad0 | 1860 | ret = encap_rcv(sk, skb); |
db8dac20 | 1861 | if (ret <= 0) { |
02c22347 ED |
1862 | __UDP_INC_STATS(sock_net(sk), |
1863 | UDP_MIB_INDATAGRAMS, | |
1864 | is_udplite); | |
db8dac20 DM |
1865 | return -ret; |
1866 | } | |
1867 | } | |
1868 | ||
1869 | /* FALLTHROUGH -- it's a UDP Packet */ | |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * UDP-Lite specific tests, ignored on UDP sockets | |
1874 | */ | |
1875 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
1876 | ||
1877 | /* | |
1878 | * MIB statistics other than incrementing the error count are | |
1879 | * disabled for the following two types of errors: these depend | |
1880 | * on the application settings, not on the functioning of the | |
1881 | * protocol stack as such. | |
1882 | * | |
1883 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
1884 | * way ... to ... at least let the receiving application block | |
1885 | * delivery of packets with coverage values less than a value | |
1886 | * provided by the application." | |
1887 | */ | |
1888 | if (up->pcrlen == 0) { /* full coverage was set */ | |
ba7a46f1 JP |
1889 | net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", |
1890 | UDP_SKB_CB(skb)->cscov, skb->len); | |
db8dac20 DM |
1891 | goto drop; |
1892 | } | |
1893 | /* The next case involves violating the min. coverage requested | |
1894 | * by the receiver. This is subtle: if receiver wants x and x is | |
1895 | * greater than the buffersize/MTU then receiver will complain | |
1896 | * that it wants x while sender emits packets of smaller size y. | |
1897 | * Therefore the above ...()->partial_cov statement is essential. | |
1898 | */ | |
1899 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
ba7a46f1 JP |
1900 | net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", |
1901 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
db8dac20 DM |
1902 | goto drop; |
1903 | } | |
1904 | } | |
1905 | ||
dd99e425 | 1906 | prefetch(&sk->sk_rmem_alloc); |
ce25d66a ED |
1907 | if (rcu_access_pointer(sk->sk_filter) && |
1908 | udp_lib_checksum_complete(skb)) | |
e6afc8ac | 1909 | goto csum_error; |
ce25d66a | 1910 | |
ba66bbe5 | 1911 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) |
a6127697 | 1912 | goto drop; |
db8dac20 | 1913 | |
e6afc8ac | 1914 | udp_csum_pull_header(skb); |
db8dac20 | 1915 | |
fbf8866d | 1916 | ipv4_pktinfo_prepare(sk, skb); |
850cbadd | 1917 | return __udp_queue_rcv_skb(sk, skb); |
db8dac20 | 1918 | |
6a5dc9e5 | 1919 | csum_error: |
02c22347 | 1920 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
db8dac20 | 1921 | drop: |
02c22347 | 1922 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
8edf19c2 | 1923 | atomic_inc(&sk->sk_drops); |
db8dac20 DM |
1924 | kfree_skb(skb); |
1925 | return -1; | |
1926 | } | |
1927 | ||
97502231 | 1928 | /* For TCP sockets, sk_rx_dst is protected by socket lock |
e47eb5df | 1929 | * For UDP, we use xchg() to guard against concurrent changes. |
97502231 | 1930 | */ |
c9f2c1ae | 1931 | void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) |
421b3885 | 1932 | { |
97502231 ED |
1933 | struct dst_entry *old; |
1934 | ||
d24406c8 WW |
1935 | if (dst_hold_safe(dst)) { |
1936 | old = xchg(&sk->sk_rx_dst, dst); | |
1937 | dst_release(old); | |
1938 | } | |
421b3885 | 1939 | } |
c9f2c1ae | 1940 | EXPORT_SYMBOL(udp_sk_rx_dst_set); |
421b3885 | 1941 | |
db8dac20 DM |
1942 | /* |
1943 | * Multicasts and broadcasts go to each listener. | |
1944 | * | |
1240d137 | 1945 | * Note: called only from the BH handler context. |
db8dac20 | 1946 | */ |
e3163493 | 1947 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
1948 | struct udphdr *uh, |
1949 | __be32 saddr, __be32 daddr, | |
36cbb245 RJ |
1950 | struct udp_table *udptable, |
1951 | int proto) | |
db8dac20 | 1952 | { |
ca065d0c | 1953 | struct sock *sk, *first = NULL; |
5cf3d461 DH |
1954 | unsigned short hnum = ntohs(uh->dest); |
1955 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
2dc41cff | 1956 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); |
ca065d0c ED |
1957 | unsigned int offset = offsetof(typeof(*sk), sk_node); |
1958 | int dif = skb->dev->ifindex; | |
1959 | struct hlist_node *node; | |
1960 | struct sk_buff *nskb; | |
2dc41cff DH |
1961 | |
1962 | if (use_hash2) { | |
1963 | hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & | |
73e2d5e3 PN |
1964 | udptable->mask; |
1965 | hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask; | |
2dc41cff | 1966 | start_lookup: |
73e2d5e3 | 1967 | hslot = &udptable->hash2[hash2]; |
2dc41cff DH |
1968 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); |
1969 | } | |
db8dac20 | 1970 | |
ca065d0c ED |
1971 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { |
1972 | if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, | |
1973 | uh->source, saddr, dif, hnum)) | |
1974 | continue; | |
1975 | ||
1976 | if (!first) { | |
1977 | first = sk; | |
1978 | continue; | |
1240d137 | 1979 | } |
ca065d0c | 1980 | nskb = skb_clone(skb, GFP_ATOMIC); |
1240d137 | 1981 | |
ca065d0c ED |
1982 | if (unlikely(!nskb)) { |
1983 | atomic_inc(&sk->sk_drops); | |
02c22347 ED |
1984 | __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, |
1985 | IS_UDPLITE(sk)); | |
1986 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, | |
1987 | IS_UDPLITE(sk)); | |
ca065d0c ED |
1988 | continue; |
1989 | } | |
1990 | if (udp_queue_rcv_skb(sk, nskb) > 0) | |
1991 | consume_skb(nskb); | |
1992 | } | |
1240d137 | 1993 | |
2dc41cff DH |
1994 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ |
1995 | if (use_hash2 && hash2 != hash2_any) { | |
1996 | hash2 = hash2_any; | |
1997 | goto start_lookup; | |
1998 | } | |
1999 | ||
ca065d0c ED |
2000 | if (first) { |
2001 | if (udp_queue_rcv_skb(first, skb) > 0) | |
2002 | consume_skb(skb); | |
1240d137 | 2003 | } else { |
ca065d0c | 2004 | kfree_skb(skb); |
02c22347 ED |
2005 | __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, |
2006 | proto == IPPROTO_UDPLITE); | |
1240d137 | 2007 | } |
db8dac20 DM |
2008 | return 0; |
2009 | } | |
2010 | ||
2011 | /* Initialize UDP checksum. If exited with zero value (success), | |
2012 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
2013 | * Otherwise, csum completion requires chacksumming packet body, | |
2014 | * including udp header and folding it to skb->csum. | |
2015 | */ | |
2016 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
2017 | int proto) | |
2018 | { | |
db8dac20 DM |
2019 | int err; |
2020 | ||
2021 | UDP_SKB_CB(skb)->partial_cov = 0; | |
2022 | UDP_SKB_CB(skb)->cscov = skb->len; | |
2023 | ||
2024 | if (proto == IPPROTO_UDPLITE) { | |
2025 | err = udplite_checksum_init(skb, uh); | |
2026 | if (err) | |
2027 | return err; | |
2028 | } | |
2029 | ||
b46d9f62 HFS |
2030 | /* Note, we are only interested in != 0 or == 0, thus the |
2031 | * force to int. | |
2032 | */ | |
2033 | return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, | |
2034 | inet_compute_pseudo); | |
db8dac20 DM |
2035 | } |
2036 | ||
2037 | /* | |
2038 | * All we need to do is get the socket, and then do a checksum. | |
2039 | */ | |
2040 | ||
645ca708 | 2041 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
2042 | int proto) |
2043 | { | |
2044 | struct sock *sk; | |
7b5e56f9 | 2045 | struct udphdr *uh; |
db8dac20 | 2046 | unsigned short ulen; |
adf30907 | 2047 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 2048 | __be32 saddr, daddr; |
0283328e | 2049 | struct net *net = dev_net(skb->dev); |
db8dac20 DM |
2050 | |
2051 | /* | |
2052 | * Validate the packet. | |
2053 | */ | |
2054 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
2055 | goto drop; /* No space for header. */ | |
2056 | ||
7b5e56f9 | 2057 | uh = udp_hdr(skb); |
db8dac20 | 2058 | ulen = ntohs(uh->len); |
ccc2d97c BM |
2059 | saddr = ip_hdr(skb)->saddr; |
2060 | daddr = ip_hdr(skb)->daddr; | |
2061 | ||
db8dac20 DM |
2062 | if (ulen > skb->len) |
2063 | goto short_packet; | |
2064 | ||
2065 | if (proto == IPPROTO_UDP) { | |
2066 | /* UDP validates ulen. */ | |
2067 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
2068 | goto short_packet; | |
2069 | uh = udp_hdr(skb); | |
2070 | } | |
2071 | ||
2072 | if (udp4_csum_init(skb, uh, proto)) | |
2073 | goto csum_error; | |
2074 | ||
8afdd99a ED |
2075 | sk = skb_steal_sock(skb); |
2076 | if (sk) { | |
97502231 | 2077 | struct dst_entry *dst = skb_dst(skb); |
421b3885 | 2078 | int ret; |
421b3885 | 2079 | |
97502231 ED |
2080 | if (unlikely(sk->sk_rx_dst != dst)) |
2081 | udp_sk_rx_dst_set(sk, dst); | |
db8dac20 | 2082 | |
421b3885 | 2083 | ret = udp_queue_rcv_skb(sk, skb); |
8afdd99a | 2084 | sock_put(sk); |
421b3885 SB |
2085 | /* a return value > 0 means to resubmit the input, but |
2086 | * it wants the return to be -protocol, or 0 | |
2087 | */ | |
2088 | if (ret > 0) | |
2089 | return -ret; | |
2090 | return 0; | |
421b3885 | 2091 | } |
db8dac20 | 2092 | |
c18450a5 FF |
2093 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
2094 | return __udp4_lib_mcast_deliver(net, skb, uh, | |
36cbb245 | 2095 | saddr, daddr, udptable, proto); |
c18450a5 FF |
2096 | |
2097 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
00db4124 | 2098 | if (sk) { |
a5b50476 ET |
2099 | int ret; |
2100 | ||
224d019c | 2101 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) |
2abb7cdc TH |
2102 | skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, |
2103 | inet_compute_pseudo); | |
2104 | ||
a5b50476 | 2105 | ret = udp_queue_rcv_skb(sk, skb); |
db8dac20 DM |
2106 | |
2107 | /* a return value > 0 means to resubmit the input, but | |
2108 | * it wants the return to be -protocol, or 0 | |
2109 | */ | |
2110 | if (ret > 0) | |
2111 | return -ret; | |
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2116 | goto drop; | |
2117 | nf_reset(skb); | |
2118 | ||
2119 | /* No socket. Drop packet silently, if checksum is wrong */ | |
2120 | if (udp_lib_checksum_complete(skb)) | |
2121 | goto csum_error; | |
2122 | ||
02c22347 | 2123 | __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2124 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
2125 | ||
2126 | /* | |
2127 | * Hmm. We got an UDP packet to a port to which we | |
2128 | * don't wanna listen. Ignore it. | |
2129 | */ | |
2130 | kfree_skb(skb); | |
2131 | return 0; | |
2132 | ||
2133 | short_packet: | |
ba7a46f1 JP |
2134 | net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
2135 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2136 | &saddr, ntohs(uh->source), | |
2137 | ulen, skb->len, | |
2138 | &daddr, ntohs(uh->dest)); | |
db8dac20 DM |
2139 | goto drop; |
2140 | ||
2141 | csum_error: | |
2142 | /* | |
2143 | * RFC1122: OK. Discards the bad packet silently (as far as | |
2144 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
2145 | */ | |
ba7a46f1 JP |
2146 | net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
2147 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2148 | &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), | |
2149 | ulen); | |
02c22347 | 2150 | __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 | 2151 | drop: |
02c22347 | 2152 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2153 | kfree_skb(skb); |
2154 | return 0; | |
2155 | } | |
2156 | ||
421b3885 SB |
2157 | /* We can only early demux multicast if there is a single matching socket. |
2158 | * If more than one socket found returns NULL | |
2159 | */ | |
2160 | static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, | |
2161 | __be16 loc_port, __be32 loc_addr, | |
2162 | __be16 rmt_port, __be32 rmt_addr, | |
2163 | int dif) | |
2164 | { | |
2165 | struct sock *sk, *result; | |
421b3885 | 2166 | unsigned short hnum = ntohs(loc_port); |
ca065d0c | 2167 | unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); |
421b3885 SB |
2168 | struct udp_hslot *hslot = &udp_table.hash[slot]; |
2169 | ||
63c6f81c ED |
2170 | /* Do not bother scanning a too big list */ |
2171 | if (hslot->count > 10) | |
2172 | return NULL; | |
2173 | ||
421b3885 | 2174 | result = NULL; |
ca065d0c ED |
2175 | sk_for_each_rcu(sk, &hslot->head) { |
2176 | if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, | |
2177 | rmt_port, rmt_addr, dif, hnum)) { | |
2178 | if (result) | |
2179 | return NULL; | |
421b3885 | 2180 | result = sk; |
421b3885 SB |
2181 | } |
2182 | } | |
ca065d0c | 2183 | |
421b3885 SB |
2184 | return result; |
2185 | } | |
2186 | ||
2187 | /* For unicast we should only early demux connected sockets or we can | |
2188 | * break forwarding setups. The chains here can be long so only check | |
2189 | * if the first socket is an exact match and if not move on. | |
2190 | */ | |
2191 | static struct sock *__udp4_lib_demux_lookup(struct net *net, | |
2192 | __be16 loc_port, __be32 loc_addr, | |
2193 | __be16 rmt_port, __be32 rmt_addr, | |
2194 | int dif) | |
2195 | { | |
421b3885 SB |
2196 | unsigned short hnum = ntohs(loc_port); |
2197 | unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); | |
2198 | unsigned int slot2 = hash2 & udp_table.mask; | |
2199 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
c7228317 | 2200 | INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); |
421b3885 | 2201 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); |
ca065d0c | 2202 | struct sock *sk; |
421b3885 | 2203 | |
ca065d0c ED |
2204 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
2205 | if (INET_MATCH(sk, net, acookie, rmt_addr, | |
2206 | loc_addr, ports, dif)) | |
2207 | return sk; | |
421b3885 SB |
2208 | /* Only check first socket in chain */ |
2209 | break; | |
2210 | } | |
ca065d0c | 2211 | return NULL; |
421b3885 SB |
2212 | } |
2213 | ||
2214 | void udp_v4_early_demux(struct sk_buff *skb) | |
2215 | { | |
610438b7 ED |
2216 | struct net *net = dev_net(skb->dev); |
2217 | const struct iphdr *iph; | |
2218 | const struct udphdr *uh; | |
ca065d0c | 2219 | struct sock *sk = NULL; |
421b3885 | 2220 | struct dst_entry *dst; |
421b3885 | 2221 | int dif = skb->dev->ifindex; |
6e540309 | 2222 | int ours; |
421b3885 SB |
2223 | |
2224 | /* validate the packet */ | |
2225 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) | |
2226 | return; | |
2227 | ||
610438b7 ED |
2228 | iph = ip_hdr(skb); |
2229 | uh = udp_hdr(skb); | |
2230 | ||
421b3885 | 2231 | if (skb->pkt_type == PACKET_BROADCAST || |
6e540309 SB |
2232 | skb->pkt_type == PACKET_MULTICAST) { |
2233 | struct in_device *in_dev = __in_dev_get_rcu(skb->dev); | |
2234 | ||
2235 | if (!in_dev) | |
2236 | return; | |
2237 | ||
ad0ea198 PA |
2238 | /* we are supposed to accept bcast packets */ |
2239 | if (skb->pkt_type == PACKET_MULTICAST) { | |
2240 | ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, | |
2241 | iph->protocol); | |
2242 | if (!ours) | |
2243 | return; | |
2244 | } | |
2245 | ||
421b3885 SB |
2246 | sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, |
2247 | uh->source, iph->saddr, dif); | |
6e540309 | 2248 | } else if (skb->pkt_type == PACKET_HOST) { |
421b3885 SB |
2249 | sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, |
2250 | uh->source, iph->saddr, dif); | |
6e540309 | 2251 | } |
421b3885 | 2252 | |
41c6d650 | 2253 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) |
421b3885 SB |
2254 | return; |
2255 | ||
2256 | skb->sk = sk; | |
82eabd9e | 2257 | skb->destructor = sock_efree; |
10e2eb87 | 2258 | dst = READ_ONCE(sk->sk_rx_dst); |
421b3885 SB |
2259 | |
2260 | if (dst) | |
2261 | dst = dst_check(dst, 0); | |
10e2eb87 | 2262 | if (dst) { |
d24406c8 WW |
2263 | /* set noref for now. |
2264 | * any place which wants to hold dst has to call | |
2265 | * dst_hold_safe() | |
2266 | */ | |
2267 | skb_dst_set_noref(skb, dst); | |
10e2eb87 | 2268 | } |
421b3885 SB |
2269 | } |
2270 | ||
db8dac20 DM |
2271 | int udp_rcv(struct sk_buff *skb) |
2272 | { | |
645ca708 | 2273 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
2274 | } |
2275 | ||
7d06b2e0 | 2276 | void udp_destroy_sock(struct sock *sk) |
db8dac20 | 2277 | { |
44046a59 | 2278 | struct udp_sock *up = udp_sk(sk); |
8a74ad60 | 2279 | bool slow = lock_sock_fast(sk); |
db8dac20 | 2280 | udp_flush_pending_frames(sk); |
8a74ad60 | 2281 | unlock_sock_fast(sk, slow); |
44046a59 TP |
2282 | if (static_key_false(&udp_encap_needed) && up->encap_type) { |
2283 | void (*encap_destroy)(struct sock *sk); | |
2284 | encap_destroy = ACCESS_ONCE(up->encap_destroy); | |
2285 | if (encap_destroy) | |
2286 | encap_destroy(sk); | |
2287 | } | |
db8dac20 DM |
2288 | } |
2289 | ||
1da177e4 LT |
2290 | /* |
2291 | * Socket option code for UDP | |
2292 | */ | |
4c0a6cb0 | 2293 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2294 | char __user *optval, unsigned int optlen, |
4c0a6cb0 | 2295 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
2296 | { |
2297 | struct udp_sock *up = udp_sk(sk); | |
1c19448c | 2298 | int val, valbool; |
1da177e4 | 2299 | int err = 0; |
b2bf1e26 | 2300 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 2301 | |
c482c568 | 2302 | if (optlen < sizeof(int)) |
1da177e4 LT |
2303 | return -EINVAL; |
2304 | ||
2305 | if (get_user(val, (int __user *)optval)) | |
2306 | return -EFAULT; | |
2307 | ||
1c19448c TH |
2308 | valbool = val ? 1 : 0; |
2309 | ||
6516c655 | 2310 | switch (optname) { |
1da177e4 LT |
2311 | case UDP_CORK: |
2312 | if (val != 0) { | |
2313 | up->corkflag = 1; | |
2314 | } else { | |
2315 | up->corkflag = 0; | |
2316 | lock_sock(sk); | |
4243cdc2 | 2317 | push_pending_frames(sk); |
1da177e4 LT |
2318 | release_sock(sk); |
2319 | } | |
2320 | break; | |
e905a9ed | 2321 | |
1da177e4 LT |
2322 | case UDP_ENCAP: |
2323 | switch (val) { | |
2324 | case 0: | |
2325 | case UDP_ENCAP_ESPINUDP: | |
2326 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
067b207b JC |
2327 | up->encap_rcv = xfrm4_udp_encap_rcv; |
2328 | /* FALLTHROUGH */ | |
342f0234 | 2329 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 | 2330 | up->encap_type = val; |
447167bf | 2331 | udp_encap_enable(); |
1da177e4 LT |
2332 | break; |
2333 | default: | |
2334 | err = -ENOPROTOOPT; | |
2335 | break; | |
2336 | } | |
2337 | break; | |
2338 | ||
1c19448c TH |
2339 | case UDP_NO_CHECK6_TX: |
2340 | up->no_check6_tx = valbool; | |
2341 | break; | |
2342 | ||
2343 | case UDP_NO_CHECK6_RX: | |
2344 | up->no_check6_rx = valbool; | |
2345 | break; | |
2346 | ||
ba4e58ec GR |
2347 | /* |
2348 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
2349 | */ | |
2350 | /* The sender sets actual checksum coverage length via this option. | |
2351 | * The case coverage > packet length is handled by send module. */ | |
2352 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 2353 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2354 | return -ENOPROTOOPT; |
2355 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
2356 | val = 8; | |
4be929be AD |
2357 | else if (val > USHRT_MAX) |
2358 | val = USHRT_MAX; | |
ba4e58ec GR |
2359 | up->pcslen = val; |
2360 | up->pcflag |= UDPLITE_SEND_CC; | |
2361 | break; | |
2362 | ||
e905a9ed YH |
2363 | /* The receiver specifies a minimum checksum coverage value. To make |
2364 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
2365 | * used, this again means full checksum coverage. */ |
2366 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 2367 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2368 | return -ENOPROTOOPT; |
2369 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
2370 | val = 8; | |
4be929be AD |
2371 | else if (val > USHRT_MAX) |
2372 | val = USHRT_MAX; | |
ba4e58ec GR |
2373 | up->pcrlen = val; |
2374 | up->pcflag |= UDPLITE_RECV_CC; | |
2375 | break; | |
2376 | ||
1da177e4 LT |
2377 | default: |
2378 | err = -ENOPROTOOPT; | |
2379 | break; | |
6516c655 | 2380 | } |
1da177e4 LT |
2381 | |
2382 | return err; | |
2383 | } | |
c482c568 | 2384 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 2385 | |
db8dac20 | 2386 | int udp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2387 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2388 | { |
2389 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2390 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2391 | udp_push_pending_frames); | |
2392 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
2393 | } | |
2394 | ||
2395 | #ifdef CONFIG_COMPAT | |
2396 | int compat_udp_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 2397 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2398 | { |
2399 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2400 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2401 | udp_push_pending_frames); | |
2402 | return compat_ip_setsockopt(sk, level, optname, optval, optlen); | |
2403 | } | |
2404 | #endif | |
2405 | ||
4c0a6cb0 GR |
2406 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
2407 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
2408 | { |
2409 | struct udp_sock *up = udp_sk(sk); | |
2410 | int val, len; | |
2411 | ||
c482c568 | 2412 | if (get_user(len, optlen)) |
1da177e4 LT |
2413 | return -EFAULT; |
2414 | ||
2415 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 2416 | |
6516c655 | 2417 | if (len < 0) |
1da177e4 LT |
2418 | return -EINVAL; |
2419 | ||
6516c655 | 2420 | switch (optname) { |
1da177e4 LT |
2421 | case UDP_CORK: |
2422 | val = up->corkflag; | |
2423 | break; | |
2424 | ||
2425 | case UDP_ENCAP: | |
2426 | val = up->encap_type; | |
2427 | break; | |
2428 | ||
1c19448c TH |
2429 | case UDP_NO_CHECK6_TX: |
2430 | val = up->no_check6_tx; | |
2431 | break; | |
2432 | ||
2433 | case UDP_NO_CHECK6_RX: | |
2434 | val = up->no_check6_rx; | |
2435 | break; | |
2436 | ||
ba4e58ec GR |
2437 | /* The following two cannot be changed on UDP sockets, the return is |
2438 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
2439 | case UDPLITE_SEND_CSCOV: | |
2440 | val = up->pcslen; | |
2441 | break; | |
2442 | ||
2443 | case UDPLITE_RECV_CSCOV: | |
2444 | val = up->pcrlen; | |
2445 | break; | |
2446 | ||
1da177e4 LT |
2447 | default: |
2448 | return -ENOPROTOOPT; | |
6516c655 | 2449 | } |
1da177e4 | 2450 | |
6516c655 | 2451 | if (put_user(len, optlen)) |
e905a9ed | 2452 | return -EFAULT; |
c482c568 | 2453 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 2454 | return -EFAULT; |
e905a9ed | 2455 | return 0; |
1da177e4 | 2456 | } |
c482c568 | 2457 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 2458 | |
db8dac20 DM |
2459 | int udp_getsockopt(struct sock *sk, int level, int optname, |
2460 | char __user *optval, int __user *optlen) | |
2461 | { | |
2462 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2463 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2464 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
2465 | } | |
2466 | ||
2467 | #ifdef CONFIG_COMPAT | |
2468 | int compat_udp_getsockopt(struct sock *sk, int level, int optname, | |
2469 | char __user *optval, int __user *optlen) | |
2470 | { | |
2471 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2472 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2473 | return compat_ip_getsockopt(sk, level, optname, optval, optlen); | |
2474 | } | |
2475 | #endif | |
1da177e4 LT |
2476 | /** |
2477 | * udp_poll - wait for a UDP event. | |
2478 | * @file - file struct | |
2479 | * @sock - socket | |
2480 | * @wait - poll table | |
2481 | * | |
e905a9ed | 2482 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
2483 | * blocking sockets. If application is using a blocking fd |
2484 | * and a packet with checksum error is in the queue; | |
2485 | * then it could get return from select indicating data available | |
2486 | * but then block when reading it. Add special case code | |
2487 | * to work around these arguably broken applications. | |
2488 | */ | |
2489 | unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
2490 | { | |
2491 | unsigned int mask = datagram_poll(file, sock, wait); | |
2492 | struct sock *sk = sock->sk; | |
ba4e58ec | 2493 | |
2276f58a PA |
2494 | if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) |
2495 | mask |= POLLIN | POLLRDNORM; | |
2496 | ||
c3f1dbaf DM |
2497 | sock_rps_record_flow(sk); |
2498 | ||
1da177e4 | 2499 | /* Check for false positives due to checksum errors */ |
85584672 | 2500 | if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && |
e83c6744 | 2501 | !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) |
85584672 | 2502 | mask &= ~(POLLIN | POLLRDNORM); |
1da177e4 LT |
2503 | |
2504 | return mask; | |
e905a9ed | 2505 | |
1da177e4 | 2506 | } |
c482c568 | 2507 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 2508 | |
5d77dca8 DA |
2509 | int udp_abort(struct sock *sk, int err) |
2510 | { | |
2511 | lock_sock(sk); | |
2512 | ||
2513 | sk->sk_err = err; | |
2514 | sk->sk_error_report(sk); | |
286c72de | 2515 | __udp_disconnect(sk, 0); |
5d77dca8 DA |
2516 | |
2517 | release_sock(sk); | |
2518 | ||
2519 | return 0; | |
2520 | } | |
2521 | EXPORT_SYMBOL_GPL(udp_abort); | |
2522 | ||
db8dac20 DM |
2523 | struct proto udp_prot = { |
2524 | .name = "UDP", | |
2525 | .owner = THIS_MODULE, | |
2526 | .close = udp_lib_close, | |
2527 | .connect = ip4_datagram_connect, | |
2528 | .disconnect = udp_disconnect, | |
2529 | .ioctl = udp_ioctl, | |
850cbadd | 2530 | .init = udp_init_sock, |
db8dac20 DM |
2531 | .destroy = udp_destroy_sock, |
2532 | .setsockopt = udp_setsockopt, | |
2533 | .getsockopt = udp_getsockopt, | |
2534 | .sendmsg = udp_sendmsg, | |
2535 | .recvmsg = udp_recvmsg, | |
2536 | .sendpage = udp_sendpage, | |
8141ed9f | 2537 | .release_cb = ip4_datagram_release_cb, |
db8dac20 DM |
2538 | .hash = udp_lib_hash, |
2539 | .unhash = udp_lib_unhash, | |
719f8358 | 2540 | .rehash = udp_v4_rehash, |
db8dac20 DM |
2541 | .get_port = udp_v4_get_port, |
2542 | .memory_allocated = &udp_memory_allocated, | |
2543 | .sysctl_mem = sysctl_udp_mem, | |
2544 | .sysctl_wmem = &sysctl_udp_wmem_min, | |
2545 | .sysctl_rmem = &sysctl_udp_rmem_min, | |
2546 | .obj_size = sizeof(struct udp_sock), | |
645ca708 | 2547 | .h.udp_table = &udp_table, |
db8dac20 DM |
2548 | #ifdef CONFIG_COMPAT |
2549 | .compat_setsockopt = compat_udp_setsockopt, | |
2550 | .compat_getsockopt = compat_udp_getsockopt, | |
2551 | #endif | |
5d77dca8 | 2552 | .diag_destroy = udp_abort, |
db8dac20 | 2553 | }; |
c482c568 | 2554 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
2555 | |
2556 | /* ------------------------------------------------------------------------ */ | |
2557 | #ifdef CONFIG_PROC_FS | |
2558 | ||
645ca708 | 2559 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
2560 | { |
2561 | struct sock *sk; | |
2562 | struct udp_iter_state *state = seq->private; | |
6f191efe | 2563 | struct net *net = seq_file_net(seq); |
1da177e4 | 2564 | |
f86dcc5a ED |
2565 | for (state->bucket = start; state->bucket <= state->udp_table->mask; |
2566 | ++state->bucket) { | |
645ca708 | 2567 | struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; |
f86dcc5a | 2568 | |
ca065d0c | 2569 | if (hlist_empty(&hslot->head)) |
f86dcc5a ED |
2570 | continue; |
2571 | ||
645ca708 | 2572 | spin_lock_bh(&hslot->lock); |
ca065d0c | 2573 | sk_for_each(sk, &hslot->head) { |
878628fb | 2574 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 2575 | continue; |
1da177e4 LT |
2576 | if (sk->sk_family == state->family) |
2577 | goto found; | |
2578 | } | |
645ca708 | 2579 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
2580 | } |
2581 | sk = NULL; | |
2582 | found: | |
2583 | return sk; | |
2584 | } | |
2585 | ||
2586 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
2587 | { | |
2588 | struct udp_iter_state *state = seq->private; | |
6f191efe | 2589 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2590 | |
2591 | do { | |
ca065d0c | 2592 | sk = sk_next(sk); |
878628fb | 2593 | } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); |
1da177e4 | 2594 | |
645ca708 | 2595 | if (!sk) { |
f86dcc5a | 2596 | if (state->bucket <= state->udp_table->mask) |
30842f29 | 2597 | spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); |
645ca708 | 2598 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
2599 | } |
2600 | return sk; | |
2601 | } | |
2602 | ||
2603 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
2604 | { | |
645ca708 | 2605 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
2606 | |
2607 | if (sk) | |
6516c655 | 2608 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
2609 | --pos; |
2610 | return pos ? NULL : sk; | |
2611 | } | |
2612 | ||
2613 | static void *udp_seq_start(struct seq_file *seq, loff_t *pos) | |
2614 | { | |
30842f29 | 2615 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 2616 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 2617 | |
b50660f1 | 2618 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 LT |
2619 | } |
2620 | ||
2621 | static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2622 | { | |
2623 | struct sock *sk; | |
2624 | ||
b50660f1 | 2625 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
2626 | sk = udp_get_idx(seq, 0); |
2627 | else | |
2628 | sk = udp_get_next(seq, v); | |
2629 | ||
2630 | ++*pos; | |
2631 | return sk; | |
2632 | } | |
2633 | ||
2634 | static void udp_seq_stop(struct seq_file *seq, void *v) | |
2635 | { | |
645ca708 ED |
2636 | struct udp_iter_state *state = seq->private; |
2637 | ||
f86dcc5a | 2638 | if (state->bucket <= state->udp_table->mask) |
645ca708 | 2639 | spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); |
1da177e4 LT |
2640 | } |
2641 | ||
73cb88ec | 2642 | int udp_seq_open(struct inode *inode, struct file *file) |
1da177e4 | 2643 | { |
d9dda78b | 2644 | struct udp_seq_afinfo *afinfo = PDE_DATA(inode); |
a2be75c1 DL |
2645 | struct udp_iter_state *s; |
2646 | int err; | |
a91275ef | 2647 | |
a2be75c1 DL |
2648 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2649 | sizeof(struct udp_iter_state)); | |
2650 | if (err < 0) | |
2651 | return err; | |
a91275ef | 2652 | |
a2be75c1 | 2653 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2654 | s->family = afinfo->family; |
645ca708 | 2655 | s->udp_table = afinfo->udp_table; |
a2be75c1 | 2656 | return err; |
a91275ef | 2657 | } |
73cb88ec | 2658 | EXPORT_SYMBOL(udp_seq_open); |
a91275ef | 2659 | |
1da177e4 | 2660 | /* ------------------------------------------------------------------------ */ |
0c96d8c5 | 2661 | int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) |
1da177e4 LT |
2662 | { |
2663 | struct proc_dir_entry *p; | |
2664 | int rc = 0; | |
2665 | ||
dda61925 DL |
2666 | afinfo->seq_ops.start = udp_seq_start; |
2667 | afinfo->seq_ops.next = udp_seq_next; | |
2668 | afinfo->seq_ops.stop = udp_seq_stop; | |
2669 | ||
84841c3c | 2670 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
73cb88ec | 2671 | afinfo->seq_fops, afinfo); |
84841c3c | 2672 | if (!p) |
1da177e4 LT |
2673 | rc = -ENOMEM; |
2674 | return rc; | |
2675 | } | |
c482c568 | 2676 | EXPORT_SYMBOL(udp_proc_register); |
1da177e4 | 2677 | |
0c96d8c5 | 2678 | void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) |
1da177e4 | 2679 | { |
ece31ffd | 2680 | remove_proc_entry(afinfo->name, net->proc_net); |
1da177e4 | 2681 | } |
c482c568 | 2682 | EXPORT_SYMBOL(udp_proc_unregister); |
db8dac20 DM |
2683 | |
2684 | /* ------------------------------------------------------------------------ */ | |
5e659e4c | 2685 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
652586df | 2686 | int bucket) |
db8dac20 DM |
2687 | { |
2688 | struct inet_sock *inet = inet_sk(sp); | |
c720c7e8 ED |
2689 | __be32 dest = inet->inet_daddr; |
2690 | __be32 src = inet->inet_rcv_saddr; | |
2691 | __u16 destp = ntohs(inet->inet_dport); | |
2692 | __u16 srcp = ntohs(inet->inet_sport); | |
db8dac20 | 2693 | |
f86dcc5a | 2694 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
652586df | 2695 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", |
db8dac20 | 2696 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 ED |
2697 | sk_wmem_alloc_get(sp), |
2698 | sk_rmem_alloc_get(sp), | |
a7cb5a49 EB |
2699 | 0, 0L, 0, |
2700 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), | |
2701 | 0, sock_i_ino(sp), | |
41c6d650 | 2702 | refcount_read(&sp->sk_refcnt), sp, |
652586df | 2703 | atomic_read(&sp->sk_drops)); |
db8dac20 DM |
2704 | } |
2705 | ||
2706 | int udp4_seq_show(struct seq_file *seq, void *v) | |
2707 | { | |
652586df | 2708 | seq_setwidth(seq, 127); |
db8dac20 | 2709 | if (v == SEQ_START_TOKEN) |
652586df | 2710 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
db8dac20 | 2711 | "rx_queue tr tm->when retrnsmt uid timeout " |
cb61cb9b | 2712 | "inode ref pointer drops"); |
db8dac20 | 2713 | else { |
db8dac20 DM |
2714 | struct udp_iter_state *state = seq->private; |
2715 | ||
652586df | 2716 | udp4_format_sock(v, seq, state->bucket); |
db8dac20 | 2717 | } |
652586df | 2718 | seq_pad(seq, '\n'); |
db8dac20 DM |
2719 | return 0; |
2720 | } | |
2721 | ||
73cb88ec AV |
2722 | static const struct file_operations udp_afinfo_seq_fops = { |
2723 | .owner = THIS_MODULE, | |
2724 | .open = udp_seq_open, | |
2725 | .read = seq_read, | |
2726 | .llseek = seq_lseek, | |
2727 | .release = seq_release_net | |
2728 | }; | |
2729 | ||
db8dac20 | 2730 | /* ------------------------------------------------------------------------ */ |
db8dac20 | 2731 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 DM |
2732 | .name = "udp", |
2733 | .family = AF_INET, | |
645ca708 | 2734 | .udp_table = &udp_table, |
73cb88ec | 2735 | .seq_fops = &udp_afinfo_seq_fops, |
dda61925 DL |
2736 | .seq_ops = { |
2737 | .show = udp4_seq_show, | |
2738 | }, | |
db8dac20 DM |
2739 | }; |
2740 | ||
2c8c1e72 | 2741 | static int __net_init udp4_proc_init_net(struct net *net) |
15439feb PE |
2742 | { |
2743 | return udp_proc_register(net, &udp4_seq_afinfo); | |
2744 | } | |
2745 | ||
2c8c1e72 | 2746 | static void __net_exit udp4_proc_exit_net(struct net *net) |
15439feb PE |
2747 | { |
2748 | udp_proc_unregister(net, &udp4_seq_afinfo); | |
2749 | } | |
2750 | ||
2751 | static struct pernet_operations udp4_net_ops = { | |
2752 | .init = udp4_proc_init_net, | |
2753 | .exit = udp4_proc_exit_net, | |
2754 | }; | |
2755 | ||
db8dac20 DM |
2756 | int __init udp4_proc_init(void) |
2757 | { | |
15439feb | 2758 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
2759 | } |
2760 | ||
2761 | void udp4_proc_exit(void) | |
2762 | { | |
15439feb | 2763 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 2764 | } |
1da177e4 LT |
2765 | #endif /* CONFIG_PROC_FS */ |
2766 | ||
f86dcc5a ED |
2767 | static __initdata unsigned long uhash_entries; |
2768 | static int __init set_uhash_entries(char *str) | |
645ca708 | 2769 | { |
413c27d8 EZ |
2770 | ssize_t ret; |
2771 | ||
f86dcc5a ED |
2772 | if (!str) |
2773 | return 0; | |
413c27d8 EZ |
2774 | |
2775 | ret = kstrtoul(str, 0, &uhash_entries); | |
2776 | if (ret) | |
2777 | return 0; | |
2778 | ||
f86dcc5a ED |
2779 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) |
2780 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
2781 | return 1; | |
2782 | } | |
2783 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 2784 | |
f86dcc5a ED |
2785 | void __init udp_table_init(struct udp_table *table, const char *name) |
2786 | { | |
2787 | unsigned int i; | |
2788 | ||
31fe62b9 TB |
2789 | table->hash = alloc_large_system_hash(name, |
2790 | 2 * sizeof(struct udp_hslot), | |
2791 | uhash_entries, | |
2792 | 21, /* one slot per 2 MB */ | |
2793 | 0, | |
2794 | &table->log, | |
2795 | &table->mask, | |
2796 | UDP_HTABLE_SIZE_MIN, | |
2797 | 64 * 1024); | |
2798 | ||
512615b6 | 2799 | table->hash2 = table->hash + (table->mask + 1); |
f86dcc5a | 2800 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2801 | INIT_HLIST_HEAD(&table->hash[i].head); |
fdcc8aa9 | 2802 | table->hash[i].count = 0; |
645ca708 ED |
2803 | spin_lock_init(&table->hash[i].lock); |
2804 | } | |
512615b6 | 2805 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2806 | INIT_HLIST_HEAD(&table->hash2[i].head); |
512615b6 ED |
2807 | table->hash2[i].count = 0; |
2808 | spin_lock_init(&table->hash2[i].lock); | |
2809 | } | |
645ca708 ED |
2810 | } |
2811 | ||
723b8e46 TH |
2812 | u32 udp_flow_hashrnd(void) |
2813 | { | |
2814 | static u32 hashrnd __read_mostly; | |
2815 | ||
2816 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
2817 | ||
2818 | return hashrnd; | |
2819 | } | |
2820 | EXPORT_SYMBOL(udp_flow_hashrnd); | |
2821 | ||
95766fff HA |
2822 | void __init udp_init(void) |
2823 | { | |
f03d78db | 2824 | unsigned long limit; |
4b272750 | 2825 | unsigned int i; |
95766fff | 2826 | |
f86dcc5a | 2827 | udp_table_init(&udp_table, "UDP"); |
f03d78db | 2828 | limit = nr_free_buffer_pages() / 8; |
95766fff HA |
2829 | limit = max(limit, 128UL); |
2830 | sysctl_udp_mem[0] = limit / 4 * 3; | |
2831 | sysctl_udp_mem[1] = limit; | |
2832 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
2833 | ||
2834 | sysctl_udp_rmem_min = SK_MEM_QUANTUM; | |
2835 | sysctl_udp_wmem_min = SK_MEM_QUANTUM; | |
4b272750 ED |
2836 | |
2837 | /* 16 spinlocks per cpu */ | |
2838 | udp_busylocks_log = ilog2(nr_cpu_ids) + 4; | |
2839 | udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, | |
2840 | GFP_KERNEL); | |
2841 | if (!udp_busylocks) | |
2842 | panic("UDP: failed to alloc udp_busylocks\n"); | |
2843 | for (i = 0; i < (1U << udp_busylocks_log); i++) | |
2844 | spin_lock_init(udp_busylocks + i); | |
95766fff | 2845 | } |