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 | 124 | |
8d987e5c | 125 | atomic_long_t udp_memory_allocated; |
95766fff HA |
126 | EXPORT_SYMBOL(udp_memory_allocated); |
127 | ||
f86dcc5a ED |
128 | #define MAX_UDP_PORTS 65536 |
129 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 130 | |
63a6fff3 RS |
131 | /* IPCB reference means this can not be used from early demux */ |
132 | static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) | |
133 | { | |
134 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
135 | if (!net->ipv4.sysctl_udp_l3mdev_accept && | |
136 | skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) | |
137 | return true; | |
138 | #endif | |
139 | return false; | |
140 | } | |
141 | ||
f24d43c0 | 142 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 143 | const struct udp_hslot *hslot, |
98322f22 | 144 | unsigned long *bitmap, |
fe38d2a1 | 145 | struct sock *sk, unsigned int log) |
1da177e4 | 146 | { |
f24d43c0 | 147 | struct sock *sk2; |
ba418fa3 | 148 | kuid_t uid = sock_i_uid(sk); |
25030a7f | 149 | |
ca065d0c | 150 | sk_for_each(sk2, &hslot->head) { |
9d4fb27d JP |
151 | if (net_eq(sock_net(sk2), net) && |
152 | sk2 != sk && | |
d4cada4a | 153 | (bitmap || udp_sk(sk2)->udp_port_hash == num) && |
9d4fb27d JP |
154 | (!sk2->sk_reuse || !sk->sk_reuse) && |
155 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
156 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 157 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
158 | if (sk2->sk_reuseport && sk->sk_reuseport && |
159 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
160 | uid_eq(uid, sock_i_uid(sk2))) { | |
161 | if (!bitmap) | |
162 | return 0; | |
163 | } else { | |
164 | if (!bitmap) | |
165 | return 1; | |
166 | __set_bit(udp_sk(sk2)->udp_port_hash >> log, | |
167 | bitmap); | |
168 | } | |
98322f22 | 169 | } |
4243cdc2 | 170 | } |
25030a7f GR |
171 | return 0; |
172 | } | |
173 | ||
30fff923 ED |
174 | /* |
175 | * Note: we still hold spinlock of primary hash chain, so no other writer | |
176 | * can insert/delete a socket with local_port == num | |
177 | */ | |
178 | static int udp_lib_lport_inuse2(struct net *net, __u16 num, | |
4243cdc2 | 179 | struct udp_hslot *hslot2, |
fe38d2a1 | 180 | struct sock *sk) |
30fff923 ED |
181 | { |
182 | struct sock *sk2; | |
ba418fa3 | 183 | kuid_t uid = sock_i_uid(sk); |
30fff923 ED |
184 | int res = 0; |
185 | ||
186 | spin_lock(&hslot2->lock); | |
ca065d0c | 187 | udp_portaddr_for_each_entry(sk2, &hslot2->head) { |
9d4fb27d JP |
188 | if (net_eq(sock_net(sk2), net) && |
189 | sk2 != sk && | |
190 | (udp_sk(sk2)->udp_port_hash == num) && | |
191 | (!sk2->sk_reuse || !sk->sk_reuse) && | |
192 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
193 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 194 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
195 | if (sk2->sk_reuseport && sk->sk_reuseport && |
196 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
197 | uid_eq(uid, sock_i_uid(sk2))) { | |
198 | res = 0; | |
199 | } else { | |
200 | res = 1; | |
201 | } | |
30fff923 ED |
202 | break; |
203 | } | |
4243cdc2 | 204 | } |
30fff923 ED |
205 | spin_unlock(&hslot2->lock); |
206 | return res; | |
207 | } | |
208 | ||
fe38d2a1 | 209 | static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) |
e32ea7e7 CG |
210 | { |
211 | struct net *net = sock_net(sk); | |
e32ea7e7 CG |
212 | kuid_t uid = sock_i_uid(sk); |
213 | struct sock *sk2; | |
214 | ||
ca065d0c | 215 | sk_for_each(sk2, &hslot->head) { |
e32ea7e7 CG |
216 | if (net_eq(sock_net(sk2), net) && |
217 | sk2 != sk && | |
218 | sk2->sk_family == sk->sk_family && | |
219 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
220 | (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && | |
221 | (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
222 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && | |
fe38d2a1 | 223 | inet_rcv_saddr_equal(sk, sk2, false)) { |
2dbb9b9e MKL |
224 | return reuseport_add_sock(sk, sk2, |
225 | inet_rcv_saddr_any(sk)); | |
e32ea7e7 CG |
226 | } |
227 | } | |
228 | ||
2dbb9b9e | 229 | return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); |
e32ea7e7 CG |
230 | } |
231 | ||
25030a7f | 232 | /** |
6ba5a3c5 | 233 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
234 | * |
235 | * @sk: socket struct in question | |
236 | * @snum: port number to look up | |
25985edc | 237 | * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, |
30fff923 | 238 | * with NULL address |
25030a7f | 239 | */ |
6ba5a3c5 | 240 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
30fff923 | 241 | unsigned int hash2_nulladdr) |
25030a7f | 242 | { |
512615b6 | 243 | struct udp_hslot *hslot, *hslot2; |
645ca708 | 244 | struct udp_table *udptable = sk->sk_prot->h.udp_table; |
25030a7f | 245 | int error = 1; |
3b1e0a65 | 246 | struct net *net = sock_net(sk); |
1da177e4 | 247 | |
32c1da70 | 248 | if (!snum) { |
9088c560 | 249 | int low, high, remaining; |
95c96174 | 250 | unsigned int rand; |
98322f22 ED |
251 | unsigned short first, last; |
252 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 253 | |
0bbf87d8 | 254 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 255 | remaining = (high - low) + 1; |
227b60f5 | 256 | |
63862b5b | 257 | rand = prandom_u32(); |
8fc54f68 | 258 | first = reciprocal_scale(rand, remaining) + low; |
98322f22 ED |
259 | /* |
260 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
261 | */ | |
f86dcc5a | 262 | rand = (rand | 1) * (udptable->mask + 1); |
5781b235 ED |
263 | last = first + udptable->mask + 1; |
264 | do { | |
f86dcc5a | 265 | hslot = udp_hashslot(udptable, net, first); |
98322f22 | 266 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 267 | spin_lock_bh(&hslot->lock); |
98322f22 | 268 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
fe38d2a1 | 269 | udptable->log); |
98322f22 ED |
270 | |
271 | snum = first; | |
272 | /* | |
273 | * Iterate on all possible values of snum for this hash. | |
274 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
275 | * give us randomization and full range coverage. | |
276 | */ | |
9088c560 | 277 | do { |
98322f22 | 278 | if (low <= snum && snum <= high && |
e3826f1e | 279 | !test_bit(snum >> udptable->log, bitmap) && |
122ff243 | 280 | !inet_is_local_reserved_port(net, snum)) |
98322f22 ED |
281 | goto found; |
282 | snum += rand; | |
283 | } while (snum != first); | |
284 | spin_unlock_bh(&hslot->lock); | |
df560056 | 285 | cond_resched(); |
5781b235 | 286 | } while (++first != last); |
98322f22 | 287 | goto fail; |
645ca708 | 288 | } else { |
f86dcc5a | 289 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 290 | spin_lock_bh(&hslot->lock); |
30fff923 ED |
291 | if (hslot->count > 10) { |
292 | int exist; | |
293 | unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; | |
294 | ||
295 | slot2 &= udptable->mask; | |
296 | hash2_nulladdr &= udptable->mask; | |
297 | ||
298 | hslot2 = udp_hashslot2(udptable, slot2); | |
299 | if (hslot->count < hslot2->count) | |
300 | goto scan_primary_hash; | |
301 | ||
fe38d2a1 | 302 | exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); |
30fff923 ED |
303 | if (!exist && (hash2_nulladdr != slot2)) { |
304 | hslot2 = udp_hashslot2(udptable, hash2_nulladdr); | |
305 | exist = udp_lib_lport_inuse2(net, snum, hslot2, | |
fe38d2a1 | 306 | sk); |
30fff923 ED |
307 | } |
308 | if (exist) | |
309 | goto fail_unlock; | |
310 | else | |
311 | goto found; | |
312 | } | |
313 | scan_primary_hash: | |
fe38d2a1 | 314 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) |
645ca708 ED |
315 | goto fail_unlock; |
316 | } | |
98322f22 | 317 | found: |
c720c7e8 | 318 | inet_sk(sk)->inet_num = snum; |
d4cada4a ED |
319 | udp_sk(sk)->udp_port_hash = snum; |
320 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
1da177e4 | 321 | if (sk_unhashed(sk)) { |
e32ea7e7 | 322 | if (sk->sk_reuseport && |
fe38d2a1 | 323 | udp_reuseport_add_sock(sk, hslot)) { |
e32ea7e7 CG |
324 | inet_sk(sk)->inet_num = 0; |
325 | udp_sk(sk)->udp_port_hash = 0; | |
326 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
327 | goto fail_unlock; | |
328 | } | |
329 | ||
ca065d0c | 330 | sk_add_node_rcu(sk, &hslot->head); |
fdcc8aa9 | 331 | hslot->count++; |
c29a0bc4 | 332 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
512615b6 ED |
333 | |
334 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
335 | spin_lock(&hslot2->lock); | |
d894ba18 | 336 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
1602f49b DM |
337 | sk->sk_family == AF_INET6) |
338 | hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, | |
339 | &hslot2->head); | |
d894ba18 | 340 | else |
1602f49b DM |
341 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
342 | &hslot2->head); | |
512615b6 ED |
343 | hslot2->count++; |
344 | spin_unlock(&hslot2->lock); | |
1da177e4 | 345 | } |
ca065d0c | 346 | sock_set_flag(sk, SOCK_RCU_FREE); |
25030a7f | 347 | error = 0; |
645ca708 ED |
348 | fail_unlock: |
349 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 350 | fail: |
25030a7f GR |
351 | return error; |
352 | } | |
c482c568 | 353 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 354 | |
6ba5a3c5 | 355 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 356 | { |
30fff923 | 357 | unsigned int hash2_nulladdr = |
f0b1e64c | 358 | ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); |
30fff923 | 359 | unsigned int hash2_partial = |
f0b1e64c | 360 | ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); |
30fff923 | 361 | |
d4cada4a | 362 | /* precompute partial secondary hash */ |
30fff923 | 363 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
fe38d2a1 | 364 | return udp_lib_get_port(sk, snum, hash2_nulladdr); |
db8dac20 DM |
365 | } |
366 | ||
d1e37288 SX |
367 | static int compute_score(struct sock *sk, struct net *net, |
368 | __be32 saddr, __be16 sport, | |
fb74c277 DA |
369 | __be32 daddr, unsigned short hnum, |
370 | int dif, int sdif, bool exact_dif) | |
645ca708 | 371 | { |
60c04aec JP |
372 | int score; |
373 | struct inet_sock *inet; | |
645ca708 | 374 | |
60c04aec JP |
375 | if (!net_eq(sock_net(sk), net) || |
376 | udp_sk(sk)->udp_port_hash != hnum || | |
377 | ipv6_only_sock(sk)) | |
378 | return -1; | |
645ca708 | 379 | |
60c04aec JP |
380 | score = (sk->sk_family == PF_INET) ? 2 : 1; |
381 | inet = inet_sk(sk); | |
382 | ||
383 | if (inet->inet_rcv_saddr) { | |
384 | if (inet->inet_rcv_saddr != daddr) | |
385 | return -1; | |
386 | score += 4; | |
645ca708 | 387 | } |
60c04aec JP |
388 | |
389 | if (inet->inet_daddr) { | |
390 | if (inet->inet_daddr != saddr) | |
391 | return -1; | |
392 | score += 4; | |
393 | } | |
394 | ||
395 | if (inet->inet_dport) { | |
396 | if (inet->inet_dport != sport) | |
397 | return -1; | |
398 | score += 4; | |
399 | } | |
400 | ||
63a6fff3 | 401 | if (sk->sk_bound_dev_if || exact_dif) { |
fb74c277 DA |
402 | bool dev_match = (sk->sk_bound_dev_if == dif || |
403 | sk->sk_bound_dev_if == sdif); | |
404 | ||
69678bcd | 405 | if (!dev_match) |
60c04aec | 406 | return -1; |
69678bcd | 407 | if (sk->sk_bound_dev_if) |
fb74c277 | 408 | score += 4; |
60c04aec | 409 | } |
fb74c277 | 410 | |
70da268b ED |
411 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
412 | score++; | |
645ca708 ED |
413 | return score; |
414 | } | |
415 | ||
6eada011 ED |
416 | static u32 udp_ehashfn(const struct net *net, const __be32 laddr, |
417 | const __u16 lport, const __be32 faddr, | |
418 | const __be16 fport) | |
65cd8033 | 419 | { |
1bbdceef HFS |
420 | static u32 udp_ehash_secret __read_mostly; |
421 | ||
422 | net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); | |
423 | ||
65cd8033 | 424 | return __inet_ehashfn(laddr, lport, faddr, fport, |
1bbdceef | 425 | udp_ehash_secret + net_hash_mix(net)); |
65cd8033 HFS |
426 | } |
427 | ||
d1e37288 | 428 | /* called with rcu_read_lock() */ |
5051ebd2 | 429 | static struct sock *udp4_lib_lookup2(struct net *net, |
fb74c277 DA |
430 | __be32 saddr, __be16 sport, |
431 | __be32 daddr, unsigned int hnum, | |
432 | int dif, int sdif, bool exact_dif, | |
433 | struct udp_hslot *hslot2, | |
434 | struct sk_buff *skb) | |
5051ebd2 ED |
435 | { |
436 | struct sock *sk, *result; | |
e94a62f5 | 437 | int score, badness; |
ba418fa3 | 438 | u32 hash = 0; |
5051ebd2 | 439 | |
5051ebd2 | 440 | result = NULL; |
ba418fa3 | 441 | badness = 0; |
ca065d0c | 442 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
d1e37288 | 443 | score = compute_score(sk, net, saddr, sport, |
fb74c277 | 444 | daddr, hnum, dif, sdif, exact_dif); |
5051ebd2 | 445 | if (score > badness) { |
e94a62f5 | 446 | if (sk->sk_reuseport) { |
65cd8033 HFS |
447 | hash = udp_ehashfn(net, daddr, hnum, |
448 | saddr, sport); | |
ca065d0c | 449 | result = reuseport_select_sock(sk, hash, skb, |
ed0dfffd | 450 | sizeof(struct udphdr)); |
ca065d0c ED |
451 | if (result) |
452 | return result; | |
ba418fa3 | 453 | } |
ca065d0c ED |
454 | badness = score; |
455 | result = sk; | |
5051ebd2 ED |
456 | } |
457 | } | |
5051ebd2 ED |
458 | return result; |
459 | } | |
460 | ||
db8dac20 DM |
461 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
462 | * harder than this. -DaveM | |
463 | */ | |
fce82338 | 464 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, |
fb74c277 DA |
465 | __be16 sport, __be32 daddr, __be16 dport, int dif, |
466 | int sdif, struct udp_table *udptable, struct sk_buff *skb) | |
db8dac20 | 467 | { |
271b72c7 | 468 | struct sock *sk, *result; |
db8dac20 | 469 | unsigned short hnum = ntohs(dport); |
5051ebd2 ED |
470 | unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); |
471 | struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; | |
63a6fff3 | 472 | bool exact_dif = udp_lib_exact_dif_match(net, skb); |
e94a62f5 | 473 | int score, badness; |
ba418fa3 | 474 | u32 hash = 0; |
645ca708 | 475 | |
5051ebd2 | 476 | if (hslot->count > 10) { |
f0b1e64c | 477 | hash2 = ipv4_portaddr_hash(net, daddr, hnum); |
5051ebd2 ED |
478 | slot2 = hash2 & udptable->mask; |
479 | hslot2 = &udptable->hash2[slot2]; | |
480 | if (hslot->count < hslot2->count) | |
481 | goto begin; | |
482 | ||
483 | result = udp4_lib_lookup2(net, saddr, sport, | |
fb74c277 | 484 | daddr, hnum, dif, sdif, |
63a6fff3 | 485 | exact_dif, hslot2, skb); |
5051ebd2 | 486 | if (!result) { |
d1e37288 | 487 | unsigned int old_slot2 = slot2; |
f0b1e64c | 488 | hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); |
5051ebd2 | 489 | slot2 = hash2 & udptable->mask; |
d1e37288 SX |
490 | /* avoid searching the same slot again. */ |
491 | if (unlikely(slot2 == old_slot2)) | |
492 | return result; | |
493 | ||
5051ebd2 ED |
494 | hslot2 = &udptable->hash2[slot2]; |
495 | if (hslot->count < hslot2->count) | |
496 | goto begin; | |
497 | ||
1223c67c | 498 | result = udp4_lib_lookup2(net, saddr, sport, |
fb74c277 | 499 | daddr, hnum, dif, sdif, |
63a6fff3 | 500 | exact_dif, hslot2, skb); |
5051ebd2 | 501 | } |
8217ca65 MKL |
502 | if (unlikely(IS_ERR(result))) |
503 | return NULL; | |
5051ebd2 ED |
504 | return result; |
505 | } | |
271b72c7 ED |
506 | begin: |
507 | result = NULL; | |
ba418fa3 | 508 | badness = 0; |
ca065d0c | 509 | sk_for_each_rcu(sk, &hslot->head) { |
d1e37288 | 510 | score = compute_score(sk, net, saddr, sport, |
fb74c277 | 511 | daddr, hnum, dif, sdif, exact_dif); |
645ca708 | 512 | if (score > badness) { |
e94a62f5 | 513 | if (sk->sk_reuseport) { |
65cd8033 HFS |
514 | hash = udp_ehashfn(net, daddr, hnum, |
515 | saddr, sport); | |
ca065d0c | 516 | result = reuseport_select_sock(sk, hash, skb, |
538950a1 | 517 | sizeof(struct udphdr)); |
8217ca65 MKL |
518 | if (unlikely(IS_ERR(result))) |
519 | return NULL; | |
ca065d0c ED |
520 | if (result) |
521 | return result; | |
ba418fa3 | 522 | } |
ca065d0c ED |
523 | result = sk; |
524 | badness = score; | |
db8dac20 DM |
525 | } |
526 | } | |
db8dac20 DM |
527 | return result; |
528 | } | |
fce82338 | 529 | EXPORT_SYMBOL_GPL(__udp4_lib_lookup); |
db8dac20 | 530 | |
607c4aaf KK |
531 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
532 | __be16 sport, __be16 dport, | |
645ca708 | 533 | struct udp_table *udptable) |
607c4aaf KK |
534 | { |
535 | const struct iphdr *iph = ip_hdr(skb); | |
536 | ||
ed7cbbce | 537 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, |
8afdd99a | 538 | iph->daddr, dport, inet_iif(skb), |
fb74c277 | 539 | inet_sdif(skb), udptable, skb); |
607c4aaf KK |
540 | } |
541 | ||
63058308 TH |
542 | struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, |
543 | __be16 sport, __be16 dport) | |
544 | { | |
ed7cbbce | 545 | return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); |
63058308 TH |
546 | } |
547 | EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); | |
548 | ||
ca065d0c ED |
549 | /* Must be called under rcu_read_lock(). |
550 | * Does increment socket refcount. | |
551 | */ | |
6e86000c | 552 | #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) |
bcd41303 KK |
553 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
554 | __be32 daddr, __be16 dport, int dif) | |
555 | { | |
ca065d0c ED |
556 | struct sock *sk; |
557 | ||
558 | sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, | |
fb74c277 | 559 | dif, 0, &udp_table, NULL); |
41c6d650 | 560 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) |
ca065d0c ED |
561 | sk = NULL; |
562 | return sk; | |
bcd41303 KK |
563 | } |
564 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
ca065d0c | 565 | #endif |
bcd41303 | 566 | |
421b3885 SB |
567 | static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, |
568 | __be16 loc_port, __be32 loc_addr, | |
569 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 570 | int dif, int sdif, unsigned short hnum) |
421b3885 SB |
571 | { |
572 | struct inet_sock *inet = inet_sk(sk); | |
573 | ||
574 | if (!net_eq(sock_net(sk), net) || | |
575 | udp_sk(sk)->udp_port_hash != hnum || | |
576 | (inet->inet_daddr && inet->inet_daddr != rmt_addr) || | |
577 | (inet->inet_dport != rmt_port && inet->inet_dport) || | |
578 | (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || | |
579 | ipv6_only_sock(sk) || | |
fb74c277 DA |
580 | (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && |
581 | sk->sk_bound_dev_if != sdif)) | |
421b3885 | 582 | return false; |
60d9b031 | 583 | if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) |
421b3885 SB |
584 | return false; |
585 | return true; | |
586 | } | |
587 | ||
db8dac20 DM |
588 | /* |
589 | * This routine is called by the ICMP module when it gets some | |
590 | * sort of error condition. If err < 0 then the socket should | |
591 | * be closed and the error returned to the user. If err > 0 | |
592 | * it's just the icmp type << 8 | icmp code. | |
593 | * Header points to the ip header of the error packet. We move | |
594 | * on past this. Then (as it used to claim before adjustment) | |
595 | * header points to the first 8 bytes of the udp header. We need | |
596 | * to find the appropriate port. | |
597 | */ | |
598 | ||
645ca708 | 599 | void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
600 | { |
601 | struct inet_sock *inet; | |
b71d1d42 | 602 | const struct iphdr *iph = (const struct iphdr *)skb->data; |
c482c568 | 603 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); |
db8dac20 DM |
604 | const int type = icmp_hdr(skb)->type; |
605 | const int code = icmp_hdr(skb)->code; | |
606 | struct sock *sk; | |
607 | int harderr; | |
608 | int err; | |
fd54d716 | 609 | struct net *net = dev_net(skb->dev); |
db8dac20 | 610 | |
fd54d716 | 611 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
fb74c277 DA |
612 | iph->saddr, uh->source, skb->dev->ifindex, 0, |
613 | udptable, NULL); | |
51456b29 | 614 | if (!sk) { |
5d3848bc | 615 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
db8dac20 DM |
616 | return; /* No socket for error */ |
617 | } | |
618 | ||
619 | err = 0; | |
620 | harderr = 0; | |
621 | inet = inet_sk(sk); | |
622 | ||
623 | switch (type) { | |
624 | default: | |
625 | case ICMP_TIME_EXCEEDED: | |
626 | err = EHOSTUNREACH; | |
627 | break; | |
628 | case ICMP_SOURCE_QUENCH: | |
629 | goto out; | |
630 | case ICMP_PARAMETERPROB: | |
631 | err = EPROTO; | |
632 | harderr = 1; | |
633 | break; | |
634 | case ICMP_DEST_UNREACH: | |
635 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
36393395 | 636 | ipv4_sk_update_pmtu(skb, sk, info); |
db8dac20 DM |
637 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { |
638 | err = EMSGSIZE; | |
639 | harderr = 1; | |
640 | break; | |
641 | } | |
642 | goto out; | |
643 | } | |
644 | err = EHOSTUNREACH; | |
645 | if (code <= NR_ICMP_UNREACH) { | |
646 | harderr = icmp_err_convert[code].fatal; | |
647 | err = icmp_err_convert[code].errno; | |
648 | } | |
649 | break; | |
55be7a9c DM |
650 | case ICMP_REDIRECT: |
651 | ipv4_sk_redirect(skb, sk); | |
1a462d18 | 652 | goto out; |
db8dac20 DM |
653 | } |
654 | ||
655 | /* | |
656 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
657 | * 4.1.3.3. | |
658 | */ | |
659 | if (!inet->recverr) { | |
660 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
661 | goto out; | |
b1faf566 | 662 | } else |
c482c568 | 663 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
b1faf566 | 664 | |
db8dac20 DM |
665 | sk->sk_err = err; |
666 | sk->sk_error_report(sk); | |
667 | out: | |
ca065d0c | 668 | return; |
db8dac20 DM |
669 | } |
670 | ||
671 | void udp_err(struct sk_buff *skb, u32 info) | |
672 | { | |
645ca708 | 673 | __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
674 | } |
675 | ||
676 | /* | |
677 | * Throw away all pending data and cancel the corking. Socket is locked. | |
678 | */ | |
36d926b9 | 679 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
680 | { |
681 | struct udp_sock *up = udp_sk(sk); | |
682 | ||
683 | if (up->pending) { | |
684 | up->len = 0; | |
685 | up->pending = 0; | |
686 | ip_flush_pending_frames(sk); | |
687 | } | |
688 | } | |
36d926b9 | 689 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
690 | |
691 | /** | |
f6b9664f | 692 | * udp4_hwcsum - handle outgoing HW checksumming |
db8dac20 DM |
693 | * @skb: sk_buff containing the filled-in UDP header |
694 | * (checksum field must be zeroed out) | |
f6b9664f HX |
695 | * @src: source IP address |
696 | * @dst: destination IP address | |
db8dac20 | 697 | */ |
c26bf4a5 | 698 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) |
db8dac20 | 699 | { |
db8dac20 | 700 | struct udphdr *uh = udp_hdr(skb); |
f6b9664f HX |
701 | int offset = skb_transport_offset(skb); |
702 | int len = skb->len - offset; | |
703 | int hlen = len; | |
db8dac20 DM |
704 | __wsum csum = 0; |
705 | ||
ebbe495f | 706 | if (!skb_has_frag_list(skb)) { |
db8dac20 DM |
707 | /* |
708 | * Only one fragment on the socket. | |
709 | */ | |
710 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
711 | skb->csum_offset = offsetof(struct udphdr, check); | |
f6b9664f HX |
712 | uh->check = ~csum_tcpudp_magic(src, dst, len, |
713 | IPPROTO_UDP, 0); | |
db8dac20 | 714 | } else { |
ebbe495f WC |
715 | struct sk_buff *frags; |
716 | ||
db8dac20 DM |
717 | /* |
718 | * HW-checksum won't work as there are two or more | |
719 | * fragments on the socket so that all csums of sk_buffs | |
720 | * should be together | |
721 | */ | |
ebbe495f | 722 | skb_walk_frags(skb, frags) { |
f6b9664f HX |
723 | csum = csum_add(csum, frags->csum); |
724 | hlen -= frags->len; | |
ebbe495f | 725 | } |
db8dac20 | 726 | |
f6b9664f | 727 | csum = skb_checksum(skb, offset, hlen, csum); |
db8dac20 DM |
728 | skb->ip_summed = CHECKSUM_NONE; |
729 | ||
db8dac20 DM |
730 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); |
731 | if (uh->check == 0) | |
732 | uh->check = CSUM_MANGLED_0; | |
733 | } | |
734 | } | |
c26bf4a5 | 735 | EXPORT_SYMBOL_GPL(udp4_hwcsum); |
db8dac20 | 736 | |
af5fcba7 TH |
737 | /* Function to set UDP checksum for an IPv4 UDP packet. This is intended |
738 | * for the simple case like when setting the checksum for a UDP tunnel. | |
739 | */ | |
740 | void udp_set_csum(bool nocheck, struct sk_buff *skb, | |
741 | __be32 saddr, __be32 daddr, int len) | |
742 | { | |
743 | struct udphdr *uh = udp_hdr(skb); | |
744 | ||
179bc67f | 745 | if (nocheck) { |
af5fcba7 | 746 | uh->check = 0; |
179bc67f | 747 | } else if (skb_is_gso(skb)) { |
af5fcba7 | 748 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); |
179bc67f EC |
749 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
750 | uh->check = 0; | |
751 | uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); | |
752 | if (uh->check == 0) | |
753 | uh->check = CSUM_MANGLED_0; | |
d75f1306 | 754 | } else { |
af5fcba7 TH |
755 | skb->ip_summed = CHECKSUM_PARTIAL; |
756 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
757 | skb->csum_offset = offsetof(struct udphdr, check); | |
758 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); | |
af5fcba7 TH |
759 | } |
760 | } | |
761 | EXPORT_SYMBOL(udp_set_csum); | |
762 | ||
bec1f6f6 WB |
763 | static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, |
764 | struct inet_cork *cork) | |
db8dac20 | 765 | { |
f6b9664f | 766 | struct sock *sk = skb->sk; |
db8dac20 | 767 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
768 | struct udphdr *uh; |
769 | int err = 0; | |
770 | int is_udplite = IS_UDPLITE(sk); | |
f6b9664f HX |
771 | int offset = skb_transport_offset(skb); |
772 | int len = skb->len - offset; | |
db8dac20 DM |
773 | __wsum csum = 0; |
774 | ||
db8dac20 DM |
775 | /* |
776 | * Create a UDP header | |
777 | */ | |
778 | uh = udp_hdr(skb); | |
f6b9664f | 779 | uh->source = inet->inet_sport; |
79ab0531 | 780 | uh->dest = fl4->fl4_dport; |
f6b9664f | 781 | uh->len = htons(len); |
db8dac20 DM |
782 | uh->check = 0; |
783 | ||
bec1f6f6 WB |
784 | if (cork->gso_size) { |
785 | const int hlen = skb_network_header_len(skb) + | |
786 | sizeof(struct udphdr); | |
787 | ||
788 | if (hlen + cork->gso_size > cork->fragsize) | |
789 | return -EINVAL; | |
790 | if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) | |
791 | return -EINVAL; | |
a8c744a8 WB |
792 | if (sk->sk_no_check_tx) |
793 | return -EINVAL; | |
ff06342c WB |
794 | if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || |
795 | dst_xfrm(skb_dst(skb))) | |
bec1f6f6 WB |
796 | return -EIO; |
797 | ||
798 | skb_shinfo(skb)->gso_size = cork->gso_size; | |
799 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; | |
dfec0ee2 AD |
800 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), |
801 | cork->gso_size); | |
a8c744a8 | 802 | goto csum_partial; |
bec1f6f6 WB |
803 | } |
804 | ||
db8dac20 | 805 | if (is_udplite) /* UDP-Lite */ |
f6b9664f | 806 | csum = udplite_csum(skb); |
db8dac20 | 807 | |
ab2fb7e3 | 808 | else if (sk->sk_no_check_tx) { /* UDP csum off */ |
db8dac20 DM |
809 | |
810 | skb->ip_summed = CHECKSUM_NONE; | |
811 | goto send; | |
812 | ||
813 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
a8c744a8 | 814 | csum_partial: |
db8dac20 | 815 | |
79ab0531 | 816 | udp4_hwcsum(skb, fl4->saddr, fl4->daddr); |
db8dac20 DM |
817 | goto send; |
818 | ||
f6b9664f HX |
819 | } else |
820 | csum = udp_csum(skb); | |
db8dac20 DM |
821 | |
822 | /* add protocol-dependent pseudo-header */ | |
79ab0531 | 823 | uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, |
c482c568 | 824 | sk->sk_protocol, csum); |
db8dac20 DM |
825 | if (uh->check == 0) |
826 | uh->check = CSUM_MANGLED_0; | |
827 | ||
828 | send: | |
b5ec8eea | 829 | err = ip_send_skb(sock_net(sk), skb); |
6ce9e7b5 ED |
830 | if (err) { |
831 | if (err == -ENOBUFS && !inet->recverr) { | |
6aef70a8 ED |
832 | UDP_INC_STATS(sock_net(sk), |
833 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
6ce9e7b5 ED |
834 | err = 0; |
835 | } | |
836 | } else | |
6aef70a8 ED |
837 | UDP_INC_STATS(sock_net(sk), |
838 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
f6b9664f HX |
839 | return err; |
840 | } | |
841 | ||
842 | /* | |
843 | * Push out all pending data as one UDP datagram. Socket is locked. | |
844 | */ | |
8822b64a | 845 | int udp_push_pending_frames(struct sock *sk) |
f6b9664f HX |
846 | { |
847 | struct udp_sock *up = udp_sk(sk); | |
848 | struct inet_sock *inet = inet_sk(sk); | |
b6f21b26 | 849 | struct flowi4 *fl4 = &inet->cork.fl.u.ip4; |
f6b9664f HX |
850 | struct sk_buff *skb; |
851 | int err = 0; | |
852 | ||
77968b78 | 853 | skb = ip_finish_skb(sk, fl4); |
f6b9664f HX |
854 | if (!skb) |
855 | goto out; | |
856 | ||
bec1f6f6 | 857 | err = udp_send_skb(skb, fl4, &inet->cork.base); |
f6b9664f | 858 | |
db8dac20 DM |
859 | out: |
860 | up->len = 0; | |
861 | up->pending = 0; | |
db8dac20 DM |
862 | return err; |
863 | } | |
8822b64a | 864 | EXPORT_SYMBOL(udp_push_pending_frames); |
db8dac20 | 865 | |
2e8de857 WB |
866 | static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) |
867 | { | |
868 | switch (cmsg->cmsg_type) { | |
869 | case UDP_SEGMENT: | |
870 | if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) | |
871 | return -EINVAL; | |
872 | *gso_size = *(__u16 *)CMSG_DATA(cmsg); | |
873 | return 0; | |
874 | default: | |
875 | return -EINVAL; | |
876 | } | |
877 | } | |
878 | ||
879 | int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) | |
880 | { | |
881 | struct cmsghdr *cmsg; | |
882 | bool need_ip = false; | |
883 | int err; | |
884 | ||
885 | for_each_cmsghdr(cmsg, msg) { | |
886 | if (!CMSG_OK(msg, cmsg)) | |
887 | return -EINVAL; | |
888 | ||
889 | if (cmsg->cmsg_level != SOL_UDP) { | |
890 | need_ip = true; | |
891 | continue; | |
892 | } | |
893 | ||
894 | err = __udp_cmsg_send(cmsg, gso_size); | |
895 | if (err) | |
896 | return err; | |
897 | } | |
898 | ||
899 | return need_ip; | |
900 | } | |
901 | EXPORT_SYMBOL_GPL(udp_cmsg_send); | |
902 | ||
1b784140 | 903 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
db8dac20 DM |
904 | { |
905 | struct inet_sock *inet = inet_sk(sk); | |
906 | struct udp_sock *up = udp_sk(sk); | |
1cedee13 | 907 | DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); |
e474995f | 908 | struct flowi4 fl4_stack; |
b6f21b26 | 909 | struct flowi4 *fl4; |
db8dac20 DM |
910 | int ulen = len; |
911 | struct ipcm_cookie ipc; | |
912 | struct rtable *rt = NULL; | |
913 | int free = 0; | |
914 | int connected = 0; | |
915 | __be32 daddr, faddr, saddr; | |
916 | __be16 dport; | |
917 | u8 tos; | |
918 | int err, is_udplite = IS_UDPLITE(sk); | |
919 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
920 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
903ab86d | 921 | struct sk_buff *skb; |
f6d8bd05 | 922 | struct ip_options_data opt_copy; |
db8dac20 DM |
923 | |
924 | if (len > 0xFFFF) | |
925 | return -EMSGSIZE; | |
926 | ||
927 | /* | |
928 | * Check the flags. | |
929 | */ | |
930 | ||
c482c568 | 931 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
932 | return -EOPNOTSUPP; |
933 | ||
903ab86d HX |
934 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
935 | ||
f5fca608 | 936 | fl4 = &inet->cork.fl.u.ip4; |
db8dac20 DM |
937 | if (up->pending) { |
938 | /* | |
939 | * There are pending frames. | |
940 | * The socket lock must be held while it's corked. | |
941 | */ | |
942 | lock_sock(sk); | |
943 | if (likely(up->pending)) { | |
944 | if (unlikely(up->pending != AF_INET)) { | |
945 | release_sock(sk); | |
946 | return -EINVAL; | |
947 | } | |
948 | goto do_append_data; | |
949 | } | |
950 | release_sock(sk); | |
951 | } | |
952 | ulen += sizeof(struct udphdr); | |
953 | ||
954 | /* | |
955 | * Get and verify the address. | |
956 | */ | |
1cedee13 | 957 | if (usin) { |
db8dac20 DM |
958 | if (msg->msg_namelen < sizeof(*usin)) |
959 | return -EINVAL; | |
960 | if (usin->sin_family != AF_INET) { | |
961 | if (usin->sin_family != AF_UNSPEC) | |
962 | return -EAFNOSUPPORT; | |
963 | } | |
964 | ||
965 | daddr = usin->sin_addr.s_addr; | |
966 | dport = usin->sin_port; | |
967 | if (dport == 0) | |
968 | return -EINVAL; | |
969 | } else { | |
970 | if (sk->sk_state != TCP_ESTABLISHED) | |
971 | return -EDESTADDRREQ; | |
c720c7e8 ED |
972 | daddr = inet->inet_daddr; |
973 | dport = inet->inet_dport; | |
db8dac20 DM |
974 | /* Open fast path for connected socket. |
975 | Route will not be used, if at least one option is set. | |
976 | */ | |
977 | connected = 1; | |
978 | } | |
db8dac20 | 979 | |
35178206 | 980 | ipcm_init_sk(&ipc, inet); |
bec1f6f6 | 981 | ipc.gso_size = up->gso_size; |
bf84a010 | 982 | |
db8dac20 | 983 | if (msg->msg_controllen) { |
2e8de857 WB |
984 | err = udp_cmsg_send(sk, msg, &ipc.gso_size); |
985 | if (err > 0) | |
986 | err = ip_cmsg_send(sk, msg, &ipc, | |
987 | sk->sk_family == AF_INET6); | |
988 | if (unlikely(err < 0)) { | |
91948309 | 989 | kfree(ipc.opt); |
db8dac20 | 990 | return err; |
91948309 | 991 | } |
db8dac20 DM |
992 | if (ipc.opt) |
993 | free = 1; | |
994 | connected = 0; | |
995 | } | |
f6d8bd05 ED |
996 | if (!ipc.opt) { |
997 | struct ip_options_rcu *inet_opt; | |
998 | ||
999 | rcu_read_lock(); | |
1000 | inet_opt = rcu_dereference(inet->inet_opt); | |
1001 | if (inet_opt) { | |
1002 | memcpy(&opt_copy, inet_opt, | |
1003 | sizeof(*inet_opt) + inet_opt->opt.optlen); | |
1004 | ipc.opt = &opt_copy.opt; | |
1005 | } | |
1006 | rcu_read_unlock(); | |
1007 | } | |
db8dac20 | 1008 | |
1cedee13 AI |
1009 | if (cgroup_bpf_enabled && !connected) { |
1010 | err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, | |
1011 | (struct sockaddr *)usin, &ipc.addr); | |
1012 | if (err) | |
1013 | goto out_free; | |
1014 | if (usin) { | |
1015 | if (usin->sin_port == 0) { | |
1016 | /* BPF program set invalid port. Reject it. */ | |
1017 | err = -EINVAL; | |
1018 | goto out_free; | |
1019 | } | |
1020 | daddr = usin->sin_addr.s_addr; | |
1021 | dport = usin->sin_port; | |
1022 | } | |
1023 | } | |
1024 | ||
db8dac20 DM |
1025 | saddr = ipc.addr; |
1026 | ipc.addr = faddr = daddr; | |
1027 | ||
f6d8bd05 | 1028 | if (ipc.opt && ipc.opt->opt.srr) { |
1b97013b AI |
1029 | if (!daddr) { |
1030 | err = -EINVAL; | |
1031 | goto out_free; | |
1032 | } | |
f6d8bd05 | 1033 | faddr = ipc.opt->opt.faddr; |
db8dac20 DM |
1034 | connected = 0; |
1035 | } | |
aa661581 | 1036 | tos = get_rttos(&ipc, inet); |
db8dac20 DM |
1037 | if (sock_flag(sk, SOCK_LOCALROUTE) || |
1038 | (msg->msg_flags & MSG_DONTROUTE) || | |
f6d8bd05 | 1039 | (ipc.opt && ipc.opt->opt.is_strictroute)) { |
db8dac20 DM |
1040 | tos |= RTO_ONLINK; |
1041 | connected = 0; | |
1042 | } | |
1043 | ||
1044 | if (ipv4_is_multicast(daddr)) { | |
854da991 | 1045 | if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) |
db8dac20 DM |
1046 | ipc.oif = inet->mc_index; |
1047 | if (!saddr) | |
1048 | saddr = inet->mc_addr; | |
1049 | connected = 0; | |
9515a2e0 | 1050 | } else if (!ipc.oif) { |
76e21053 | 1051 | ipc.oif = inet->uc_index; |
9515a2e0 DA |
1052 | } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { |
1053 | /* oif is set, packet is to local broadcast and | |
1054 | * and uc_index is set. oif is most likely set | |
1055 | * by sk_bound_dev_if. If uc_index != oif check if the | |
1056 | * oif is an L3 master and uc_index is an L3 slave. | |
1057 | * If so, we want to allow the send using the uc_index. | |
1058 | */ | |
1059 | if (ipc.oif != inet->uc_index && | |
1060 | ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), | |
1061 | inet->uc_index)) { | |
1062 | ipc.oif = inet->uc_index; | |
1063 | } | |
1064 | } | |
db8dac20 DM |
1065 | |
1066 | if (connected) | |
c482c568 | 1067 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 | 1068 | |
51456b29 | 1069 | if (!rt) { |
84a3aa00 | 1070 | struct net *net = sock_net(sk); |
9a24abfa | 1071 | __u8 flow_flags = inet_sk_flowi_flags(sk); |
84a3aa00 | 1072 | |
e474995f | 1073 | fl4 = &fl4_stack; |
9a24abfa | 1074 | |
e474995f | 1075 | flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, |
c0951cbc | 1076 | RT_SCOPE_UNIVERSE, sk->sk_protocol, |
9a24abfa | 1077 | flow_flags, |
e2d118a1 LC |
1078 | faddr, saddr, dport, inet->inet_sport, |
1079 | sk->sk_uid); | |
c0951cbc | 1080 | |
e474995f DM |
1081 | security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); |
1082 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe DM |
1083 | if (IS_ERR(rt)) { |
1084 | err = PTR_ERR(rt); | |
06dc94b1 | 1085 | rt = NULL; |
db8dac20 | 1086 | if (err == -ENETUNREACH) |
f1d8cba6 | 1087 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
1088 | goto out; |
1089 | } | |
1090 | ||
1091 | err = -EACCES; | |
1092 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
1093 | !sock_flag(sk, SOCK_BROADCAST)) | |
1094 | goto out; | |
1095 | if (connected) | |
d8d1f30b | 1096 | sk_dst_set(sk, dst_clone(&rt->dst)); |
db8dac20 DM |
1097 | } |
1098 | ||
1099 | if (msg->msg_flags&MSG_CONFIRM) | |
1100 | goto do_confirm; | |
1101 | back_from_confirm: | |
1102 | ||
e474995f | 1103 | saddr = fl4->saddr; |
db8dac20 | 1104 | if (!ipc.addr) |
e474995f | 1105 | daddr = ipc.addr = fl4->daddr; |
db8dac20 | 1106 | |
903ab86d HX |
1107 | /* Lockless fast path for the non-corking case. */ |
1108 | if (!corkreq) { | |
1cd7884d WB |
1109 | struct inet_cork cork; |
1110 | ||
f69e6d13 | 1111 | skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, |
903ab86d | 1112 | sizeof(struct udphdr), &ipc, &rt, |
1cd7884d | 1113 | &cork, msg->msg_flags); |
903ab86d | 1114 | err = PTR_ERR(skb); |
50c3a487 | 1115 | if (!IS_ERR_OR_NULL(skb)) |
bec1f6f6 | 1116 | err = udp_send_skb(skb, fl4, &cork); |
903ab86d HX |
1117 | goto out; |
1118 | } | |
1119 | ||
db8dac20 DM |
1120 | lock_sock(sk); |
1121 | if (unlikely(up->pending)) { | |
1122 | /* The socket is already corked while preparing it. */ | |
1123 | /* ... which is an evident application bug. --ANK */ | |
1124 | release_sock(sk); | |
1125 | ||
197df02c | 1126 | net_dbg_ratelimited("socket already corked\n"); |
db8dac20 DM |
1127 | err = -EINVAL; |
1128 | goto out; | |
1129 | } | |
1130 | /* | |
1131 | * Now cork the socket to pend data. | |
1132 | */ | |
b6f21b26 DM |
1133 | fl4 = &inet->cork.fl.u.ip4; |
1134 | fl4->daddr = daddr; | |
1135 | fl4->saddr = saddr; | |
9cce96df DM |
1136 | fl4->fl4_dport = dport; |
1137 | fl4->fl4_sport = inet->inet_sport; | |
db8dac20 DM |
1138 | up->pending = AF_INET; |
1139 | ||
1140 | do_append_data: | |
1141 | up->len += ulen; | |
f69e6d13 | 1142 | err = ip_append_data(sk, fl4, getfrag, msg, ulen, |
f5fca608 DM |
1143 | sizeof(struct udphdr), &ipc, &rt, |
1144 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); | |
db8dac20 DM |
1145 | if (err) |
1146 | udp_flush_pending_frames(sk); | |
1147 | else if (!corkreq) | |
1148 | err = udp_push_pending_frames(sk); | |
1149 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1150 | up->pending = 0; | |
1151 | release_sock(sk); | |
1152 | ||
1153 | out: | |
1154 | ip_rt_put(rt); | |
1b97013b | 1155 | out_free: |
db8dac20 DM |
1156 | if (free) |
1157 | kfree(ipc.opt); | |
1158 | if (!err) | |
1159 | return len; | |
1160 | /* | |
1161 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1162 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1163 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1164 | * things). We could add another new stat but at least for now that | |
1165 | * seems like overkill. | |
1166 | */ | |
1167 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
6aef70a8 ED |
1168 | UDP_INC_STATS(sock_net(sk), |
1169 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
1170 | } |
1171 | return err; | |
1172 | ||
1173 | do_confirm: | |
0dec879f JA |
1174 | if (msg->msg_flags & MSG_PROBE) |
1175 | dst_confirm_neigh(&rt->dst, &fl4->daddr); | |
db8dac20 DM |
1176 | if (!(msg->msg_flags&MSG_PROBE) || len) |
1177 | goto back_from_confirm; | |
1178 | err = 0; | |
1179 | goto out; | |
1180 | } | |
c482c568 | 1181 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
1182 | |
1183 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
1184 | size_t size, int flags) | |
1185 | { | |
f5fca608 | 1186 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
1187 | struct udp_sock *up = udp_sk(sk); |
1188 | int ret; | |
1189 | ||
d3f7d56a SL |
1190 | if (flags & MSG_SENDPAGE_NOTLAST) |
1191 | flags |= MSG_MORE; | |
1192 | ||
db8dac20 DM |
1193 | if (!up->pending) { |
1194 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
1195 | ||
1196 | /* Call udp_sendmsg to specify destination address which | |
1197 | * sendpage interface can't pass. | |
1198 | * This will succeed only when the socket is connected. | |
1199 | */ | |
1b784140 | 1200 | ret = udp_sendmsg(sk, &msg, 0); |
db8dac20 DM |
1201 | if (ret < 0) |
1202 | return ret; | |
1203 | } | |
1204 | ||
1205 | lock_sock(sk); | |
1206 | ||
1207 | if (unlikely(!up->pending)) { | |
1208 | release_sock(sk); | |
1209 | ||
197df02c | 1210 | net_dbg_ratelimited("cork failed\n"); |
db8dac20 DM |
1211 | return -EINVAL; |
1212 | } | |
1213 | ||
f5fca608 DM |
1214 | ret = ip_append_page(sk, &inet->cork.fl.u.ip4, |
1215 | page, offset, size, flags); | |
db8dac20 DM |
1216 | if (ret == -EOPNOTSUPP) { |
1217 | release_sock(sk); | |
1218 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
1219 | size, flags); | |
1220 | } | |
1221 | if (ret < 0) { | |
1222 | udp_flush_pending_frames(sk); | |
1223 | goto out; | |
1224 | } | |
1225 | ||
1226 | up->len += size; | |
1227 | if (!(up->corkflag || (flags&MSG_MORE))) | |
1228 | ret = udp_push_pending_frames(sk); | |
1229 | if (!ret) | |
1230 | ret = size; | |
1231 | out: | |
1232 | release_sock(sk); | |
1233 | return ret; | |
1234 | } | |
1235 | ||
dce4551c PA |
1236 | #define UDP_SKB_IS_STATELESS 0x80000000 |
1237 | ||
b65ac446 PA |
1238 | static void udp_set_dev_scratch(struct sk_buff *skb) |
1239 | { | |
dce4551c | 1240 | struct udp_dev_scratch *scratch = udp_skb_scratch(skb); |
b65ac446 PA |
1241 | |
1242 | BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); | |
dce4551c PA |
1243 | scratch->_tsize_state = skb->truesize; |
1244 | #if BITS_PER_LONG == 64 | |
b65ac446 PA |
1245 | scratch->len = skb->len; |
1246 | scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); | |
1247 | scratch->is_linear = !skb_is_nonlinear(skb); | |
dce4551c | 1248 | #endif |
3bdefdf9 PA |
1249 | /* all head states execept sp (dst, sk, nf) are always cleared by |
1250 | * udp_rcv() and we need to preserve secpath, if present, to eventually | |
1251 | * process IP_CMSG_PASSSEC at recvmsg() time | |
1252 | */ | |
1253 | if (likely(!skb_sec_path(skb))) | |
dce4551c | 1254 | scratch->_tsize_state |= UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1255 | } |
1256 | ||
1257 | static int udp_skb_truesize(struct sk_buff *skb) | |
1258 | { | |
dce4551c | 1259 | return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1260 | } |
1261 | ||
dce4551c | 1262 | static bool udp_skb_has_head_state(struct sk_buff *skb) |
b65ac446 | 1263 | { |
dce4551c | 1264 | return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); |
b65ac446 | 1265 | } |
b65ac446 | 1266 | |
7c13f97f | 1267 | /* fully reclaim rmem/fwd memory allocated for skb */ |
6dfb4367 PA |
1268 | static void udp_rmem_release(struct sock *sk, int size, int partial, |
1269 | bool rx_queue_lock_held) | |
f970bd9e | 1270 | { |
6b229cf7 | 1271 | struct udp_sock *up = udp_sk(sk); |
2276f58a | 1272 | struct sk_buff_head *sk_queue; |
f970bd9e PA |
1273 | int amt; |
1274 | ||
6b229cf7 ED |
1275 | if (likely(partial)) { |
1276 | up->forward_deficit += size; | |
1277 | size = up->forward_deficit; | |
0d4a6608 | 1278 | if (size < (sk->sk_rcvbuf >> 2)) |
6b229cf7 ED |
1279 | return; |
1280 | } else { | |
1281 | size += up->forward_deficit; | |
1282 | } | |
1283 | up->forward_deficit = 0; | |
1284 | ||
6dfb4367 PA |
1285 | /* acquire the sk_receive_queue for fwd allocated memory scheduling, |
1286 | * if the called don't held it already | |
1287 | */ | |
2276f58a | 1288 | sk_queue = &sk->sk_receive_queue; |
6dfb4367 PA |
1289 | if (!rx_queue_lock_held) |
1290 | spin_lock(&sk_queue->lock); | |
1291 | ||
2276f58a | 1292 | |
f970bd9e PA |
1293 | sk->sk_forward_alloc += size; |
1294 | amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); | |
1295 | sk->sk_forward_alloc -= amt; | |
f970bd9e PA |
1296 | |
1297 | if (amt) | |
1298 | __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); | |
02ab0d13 ED |
1299 | |
1300 | atomic_sub(size, &sk->sk_rmem_alloc); | |
2276f58a PA |
1301 | |
1302 | /* this can save us from acquiring the rx queue lock on next receive */ | |
1303 | skb_queue_splice_tail_init(sk_queue, &up->reader_queue); | |
1304 | ||
6dfb4367 PA |
1305 | if (!rx_queue_lock_held) |
1306 | spin_unlock(&sk_queue->lock); | |
f970bd9e PA |
1307 | } |
1308 | ||
2276f58a | 1309 | /* Note: called with reader_queue.lock held. |
c84d9490 ED |
1310 | * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch |
1311 | * This avoids a cache line miss while receive_queue lock is held. | |
1312 | * Look at __udp_enqueue_schedule_skb() to find where this copy is done. | |
1313 | */ | |
7c13f97f | 1314 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) |
f970bd9e | 1315 | { |
b65ac446 PA |
1316 | prefetch(&skb->data); |
1317 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); | |
f970bd9e | 1318 | } |
7c13f97f | 1319 | EXPORT_SYMBOL(udp_skb_destructor); |
f970bd9e | 1320 | |
6dfb4367 | 1321 | /* as above, but the caller held the rx queue lock, too */ |
64f5102d | 1322 | static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) |
6dfb4367 | 1323 | { |
b65ac446 PA |
1324 | prefetch(&skb->data); |
1325 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); | |
6dfb4367 PA |
1326 | } |
1327 | ||
4b272750 ED |
1328 | /* Idea of busylocks is to let producers grab an extra spinlock |
1329 | * to relieve pressure on the receive_queue spinlock shared by consumer. | |
1330 | * Under flood, this means that only one producer can be in line | |
1331 | * trying to acquire the receive_queue spinlock. | |
1332 | * These busylock can be allocated on a per cpu manner, instead of a | |
1333 | * per socket one (that would consume a cache line per socket) | |
1334 | */ | |
1335 | static int udp_busylocks_log __read_mostly; | |
1336 | static spinlock_t *udp_busylocks __read_mostly; | |
1337 | ||
1338 | static spinlock_t *busylock_acquire(void *ptr) | |
1339 | { | |
1340 | spinlock_t *busy; | |
1341 | ||
1342 | busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); | |
1343 | spin_lock(busy); | |
1344 | return busy; | |
1345 | } | |
1346 | ||
1347 | static void busylock_release(spinlock_t *busy) | |
1348 | { | |
1349 | if (busy) | |
1350 | spin_unlock(busy); | |
1351 | } | |
1352 | ||
f970bd9e PA |
1353 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) |
1354 | { | |
1355 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
1356 | int rmem, delta, amt, err = -ENOMEM; | |
4b272750 | 1357 | spinlock_t *busy = NULL; |
c8c8b127 | 1358 | int size; |
f970bd9e PA |
1359 | |
1360 | /* try to avoid the costly atomic add/sub pair when the receive | |
1361 | * queue is full; always allow at least a packet | |
1362 | */ | |
1363 | rmem = atomic_read(&sk->sk_rmem_alloc); | |
363dc73a | 1364 | if (rmem > sk->sk_rcvbuf) |
f970bd9e PA |
1365 | goto drop; |
1366 | ||
c8c8b127 ED |
1367 | /* Under mem pressure, it might be helpful to help udp_recvmsg() |
1368 | * having linear skbs : | |
1369 | * - Reduce memory overhead and thus increase receive queue capacity | |
1370 | * - Less cache line misses at copyout() time | |
1371 | * - Less work at consume_skb() (less alien page frag freeing) | |
1372 | */ | |
4b272750 | 1373 | if (rmem > (sk->sk_rcvbuf >> 1)) { |
c8c8b127 | 1374 | skb_condense(skb); |
4b272750 ED |
1375 | |
1376 | busy = busylock_acquire(sk); | |
1377 | } | |
c8c8b127 | 1378 | size = skb->truesize; |
b65ac446 | 1379 | udp_set_dev_scratch(skb); |
c8c8b127 | 1380 | |
f970bd9e PA |
1381 | /* we drop only if the receive buf is full and the receive |
1382 | * queue contains some other skb | |
1383 | */ | |
1384 | rmem = atomic_add_return(size, &sk->sk_rmem_alloc); | |
363dc73a | 1385 | if (rmem > (size + sk->sk_rcvbuf)) |
f970bd9e PA |
1386 | goto uncharge_drop; |
1387 | ||
1388 | spin_lock(&list->lock); | |
1389 | if (size >= sk->sk_forward_alloc) { | |
1390 | amt = sk_mem_pages(size); | |
1391 | delta = amt << SK_MEM_QUANTUM_SHIFT; | |
1392 | if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { | |
1393 | err = -ENOBUFS; | |
1394 | spin_unlock(&list->lock); | |
1395 | goto uncharge_drop; | |
1396 | } | |
1397 | ||
1398 | sk->sk_forward_alloc += delta; | |
1399 | } | |
1400 | ||
1401 | sk->sk_forward_alloc -= size; | |
1402 | ||
7c13f97f PA |
1403 | /* no need to setup a destructor, we will explicitly release the |
1404 | * forward allocated memory on dequeue | |
1405 | */ | |
f970bd9e PA |
1406 | sock_skb_set_dropcount(sk, skb); |
1407 | ||
1408 | __skb_queue_tail(list, skb); | |
1409 | spin_unlock(&list->lock); | |
1410 | ||
1411 | if (!sock_flag(sk, SOCK_DEAD)) | |
1412 | sk->sk_data_ready(sk); | |
1413 | ||
4b272750 | 1414 | busylock_release(busy); |
f970bd9e PA |
1415 | return 0; |
1416 | ||
1417 | uncharge_drop: | |
1418 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); | |
1419 | ||
1420 | drop: | |
1421 | atomic_inc(&sk->sk_drops); | |
4b272750 | 1422 | busylock_release(busy); |
f970bd9e PA |
1423 | return err; |
1424 | } | |
1425 | EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); | |
1426 | ||
c915fe13 | 1427 | void udp_destruct_sock(struct sock *sk) |
f970bd9e PA |
1428 | { |
1429 | /* reclaim completely the forward allocated memory */ | |
2276f58a | 1430 | struct udp_sock *up = udp_sk(sk); |
7c13f97f PA |
1431 | unsigned int total = 0; |
1432 | struct sk_buff *skb; | |
1433 | ||
2276f58a PA |
1434 | skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); |
1435 | while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { | |
7c13f97f PA |
1436 | total += skb->truesize; |
1437 | kfree_skb(skb); | |
1438 | } | |
6dfb4367 | 1439 | udp_rmem_release(sk, total, 0, true); |
7c13f97f | 1440 | |
f970bd9e PA |
1441 | inet_sock_destruct(sk); |
1442 | } | |
c915fe13 | 1443 | EXPORT_SYMBOL_GPL(udp_destruct_sock); |
f970bd9e PA |
1444 | |
1445 | int udp_init_sock(struct sock *sk) | |
1446 | { | |
2276f58a | 1447 | skb_queue_head_init(&udp_sk(sk)->reader_queue); |
f970bd9e PA |
1448 | sk->sk_destruct = udp_destruct_sock; |
1449 | return 0; | |
1450 | } | |
1451 | EXPORT_SYMBOL_GPL(udp_init_sock); | |
1452 | ||
1453 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) | |
1454 | { | |
1455 | if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { | |
1456 | bool slow = lock_sock_fast(sk); | |
1457 | ||
1458 | sk_peek_offset_bwd(sk, len); | |
1459 | unlock_sock_fast(sk, slow); | |
1460 | } | |
0a463c78 | 1461 | |
ca2c1418 PA |
1462 | if (!skb_unref(skb)) |
1463 | return; | |
1464 | ||
dce4551c PA |
1465 | /* In the more common cases we cleared the head states previously, |
1466 | * see __udp_queue_rcv_skb(). | |
0ddf3fb2 | 1467 | */ |
dce4551c | 1468 | if (unlikely(udp_skb_has_head_state(skb))) |
0ddf3fb2 | 1469 | skb_release_head_state(skb); |
ca2c1418 | 1470 | __consume_stateless_skb(skb); |
f970bd9e PA |
1471 | } |
1472 | EXPORT_SYMBOL_GPL(skb_consume_udp); | |
1473 | ||
2276f58a PA |
1474 | static struct sk_buff *__first_packet_length(struct sock *sk, |
1475 | struct sk_buff_head *rcvq, | |
1476 | int *total) | |
1477 | { | |
1478 | struct sk_buff *skb; | |
1479 | ||
9bd780f5 PA |
1480 | while ((skb = skb_peek(rcvq)) != NULL) { |
1481 | if (udp_lib_checksum_complete(skb)) { | |
1482 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1483 | IS_UDPLITE(sk)); | |
1484 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1485 | IS_UDPLITE(sk)); | |
1486 | atomic_inc(&sk->sk_drops); | |
1487 | __skb_unlink(skb, rcvq); | |
1488 | *total += skb->truesize; | |
1489 | kfree_skb(skb); | |
1490 | } else { | |
1491 | /* the csum related bits could be changed, refresh | |
1492 | * the scratch area | |
1493 | */ | |
1494 | udp_set_dev_scratch(skb); | |
1495 | break; | |
1496 | } | |
2276f58a PA |
1497 | } |
1498 | return skb; | |
1499 | } | |
1500 | ||
85584672 ED |
1501 | /** |
1502 | * first_packet_length - return length of first packet in receive queue | |
1503 | * @sk: socket | |
1504 | * | |
1505 | * Drops all bad checksum frames, until a valid one is found. | |
e83c6744 | 1506 | * Returns the length of found skb, or -1 if none is found. |
85584672 | 1507 | */ |
e83c6744 | 1508 | static int first_packet_length(struct sock *sk) |
85584672 | 1509 | { |
2276f58a PA |
1510 | struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; |
1511 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
85584672 | 1512 | struct sk_buff *skb; |
7c13f97f | 1513 | int total = 0; |
e83c6744 | 1514 | int res; |
85584672 | 1515 | |
85584672 | 1516 | spin_lock_bh(&rcvq->lock); |
2276f58a PA |
1517 | skb = __first_packet_length(sk, rcvq, &total); |
1518 | if (!skb && !skb_queue_empty(sk_queue)) { | |
1519 | spin_lock(&sk_queue->lock); | |
1520 | skb_queue_splice_tail_init(sk_queue, rcvq); | |
1521 | spin_unlock(&sk_queue->lock); | |
1522 | ||
1523 | skb = __first_packet_length(sk, rcvq, &total); | |
85584672 | 1524 | } |
e83c6744 | 1525 | res = skb ? skb->len : -1; |
7c13f97f | 1526 | if (total) |
6dfb4367 | 1527 | udp_rmem_release(sk, total, 1, false); |
85584672 | 1528 | spin_unlock_bh(&rcvq->lock); |
85584672 ED |
1529 | return res; |
1530 | } | |
1531 | ||
1da177e4 LT |
1532 | /* |
1533 | * IOCTL requests applicable to the UDP protocol | |
1534 | */ | |
e905a9ed | 1535 | |
1da177e4 LT |
1536 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
1537 | { | |
6516c655 SH |
1538 | switch (cmd) { |
1539 | case SIOCOUTQ: | |
1da177e4 | 1540 | { |
31e6d363 ED |
1541 | int amount = sk_wmem_alloc_get(sk); |
1542 | ||
6516c655 SH |
1543 | return put_user(amount, (int __user *)arg); |
1544 | } | |
1da177e4 | 1545 | |
6516c655 SH |
1546 | case SIOCINQ: |
1547 | { | |
e83c6744 | 1548 | int amount = max_t(int, 0, first_packet_length(sk)); |
6516c655 | 1549 | |
6516c655 SH |
1550 | return put_user(amount, (int __user *)arg); |
1551 | } | |
1da177e4 | 1552 | |
6516c655 SH |
1553 | default: |
1554 | return -ENOIOCTLCMD; | |
1da177e4 | 1555 | } |
6516c655 SH |
1556 | |
1557 | return 0; | |
1da177e4 | 1558 | } |
c482c568 | 1559 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 1560 | |
2276f58a PA |
1561 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, |
1562 | int noblock, int *peeked, int *off, int *err) | |
1563 | { | |
1564 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
1565 | struct sk_buff_head *queue; | |
1566 | struct sk_buff *last; | |
1567 | long timeo; | |
1568 | int error; | |
1569 | ||
1570 | queue = &udp_sk(sk)->reader_queue; | |
1571 | flags |= noblock ? MSG_DONTWAIT : 0; | |
1572 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1573 | do { | |
1574 | struct sk_buff *skb; | |
1575 | ||
1576 | error = sock_error(sk); | |
1577 | if (error) | |
1578 | break; | |
1579 | ||
1580 | error = -EAGAIN; | |
1581 | *peeked = 0; | |
1582 | do { | |
2276f58a PA |
1583 | spin_lock_bh(&queue->lock); |
1584 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
1585 | udp_skb_destructor, | |
de321ed3 | 1586 | peeked, off, err, |
2276f58a PA |
1587 | &last); |
1588 | if (skb) { | |
1589 | spin_unlock_bh(&queue->lock); | |
2276f58a PA |
1590 | return skb; |
1591 | } | |
1592 | ||
1593 | if (skb_queue_empty(sk_queue)) { | |
1594 | spin_unlock_bh(&queue->lock); | |
1595 | goto busy_check; | |
1596 | } | |
1597 | ||
6dfb4367 PA |
1598 | /* refill the reader queue and walk it again |
1599 | * keep both queues locked to avoid re-acquiring | |
1600 | * the sk_receive_queue lock if fwd memory scheduling | |
1601 | * is needed. | |
1602 | */ | |
2276f58a PA |
1603 | spin_lock(&sk_queue->lock); |
1604 | skb_queue_splice_tail_init(sk_queue, queue); | |
2276f58a PA |
1605 | |
1606 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
6dfb4367 | 1607 | udp_skb_dtor_locked, |
de321ed3 | 1608 | peeked, off, err, |
2276f58a | 1609 | &last); |
6dfb4367 | 1610 | spin_unlock(&sk_queue->lock); |
2276f58a | 1611 | spin_unlock_bh(&queue->lock); |
de321ed3 | 1612 | if (skb) |
2276f58a | 1613 | return skb; |
2276f58a PA |
1614 | |
1615 | busy_check: | |
1616 | if (!sk_can_busy_loop(sk)) | |
1617 | break; | |
1618 | ||
1619 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
1620 | } while (!skb_queue_empty(sk_queue)); | |
1621 | ||
1622 | /* sk_queue is empty, reader_queue may contain peeked packets */ | |
1623 | } while (timeo && | |
1624 | !__skb_wait_for_more_packets(sk, &error, &timeo, | |
1625 | (struct sk_buff *)sk_queue)); | |
1626 | ||
1627 | *err = error; | |
1628 | return NULL; | |
1629 | } | |
7e823644 | 1630 | EXPORT_SYMBOL(__skb_recv_udp); |
2276f58a | 1631 | |
db8dac20 DM |
1632 | /* |
1633 | * This should be easy, if there is something there we | |
1634 | * return it, otherwise we block. | |
1635 | */ | |
1636 | ||
1b784140 YX |
1637 | int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, |
1638 | int flags, int *addr_len) | |
db8dac20 DM |
1639 | { |
1640 | struct inet_sock *inet = inet_sk(sk); | |
342dfc30 | 1641 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
db8dac20 | 1642 | struct sk_buff *skb; |
59c2cdae | 1643 | unsigned int ulen, copied; |
627d2d6b | 1644 | int peeked, peeking, off; |
db8dac20 DM |
1645 | int err; |
1646 | int is_udplite = IS_UDPLITE(sk); | |
197c949e | 1647 | bool checksum_valid = false; |
db8dac20 | 1648 | |
db8dac20 | 1649 | if (flags & MSG_ERRQUEUE) |
85fbaa75 | 1650 | return ip_recv_error(sk, msg, len, addr_len); |
db8dac20 DM |
1651 | |
1652 | try_again: | |
a0917e0b MD |
1653 | peeking = flags & MSG_PEEK; |
1654 | off = sk_peek_offset(sk, flags); | |
7c13f97f | 1655 | skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); |
db8dac20 | 1656 | if (!skb) |
627d2d6b | 1657 | return err; |
db8dac20 | 1658 | |
b65ac446 | 1659 | ulen = udp_skb_len(skb); |
59c2cdae | 1660 | copied = len; |
627d2d6b | 1661 | if (copied > ulen - off) |
1662 | copied = ulen - off; | |
59c2cdae | 1663 | else if (copied < ulen) |
db8dac20 DM |
1664 | msg->msg_flags |= MSG_TRUNC; |
1665 | ||
1666 | /* | |
1667 | * If checksum is needed at all, try to do it while copying the | |
1668 | * data. If the data is truncated, or if we only want a partial | |
1669 | * coverage checksum (UDP-Lite), do it before the copy. | |
1670 | */ | |
1671 | ||
d21dbdfe ED |
1672 | if (copied < ulen || peeking || |
1673 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
b65ac446 PA |
1674 | checksum_valid = udp_skb_csum_unnecessary(skb) || |
1675 | !__udp_lib_checksum_complete(skb); | |
197c949e | 1676 | if (!checksum_valid) |
db8dac20 DM |
1677 | goto csum_copy_err; |
1678 | } | |
1679 | ||
b65ac446 PA |
1680 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { |
1681 | if (udp_skb_is_linear(skb)) | |
1682 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
1683 | else | |
1684 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
1685 | } else { | |
627d2d6b | 1686 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); |
db8dac20 DM |
1687 | |
1688 | if (err == -EINVAL) | |
1689 | goto csum_copy_err; | |
1690 | } | |
1691 | ||
22911fc5 | 1692 | if (unlikely(err)) { |
979402b1 ED |
1693 | if (!peeked) { |
1694 | atomic_inc(&sk->sk_drops); | |
6aef70a8 ED |
1695 | UDP_INC_STATS(sock_net(sk), |
1696 | UDP_MIB_INERRORS, is_udplite); | |
979402b1 | 1697 | } |
850cbadd | 1698 | kfree_skb(skb); |
627d2d6b | 1699 | return err; |
22911fc5 | 1700 | } |
db8dac20 DM |
1701 | |
1702 | if (!peeked) | |
6aef70a8 ED |
1703 | UDP_INC_STATS(sock_net(sk), |
1704 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 | 1705 | |
3b885787 | 1706 | sock_recv_ts_and_drops(msg, sk, skb); |
db8dac20 DM |
1707 | |
1708 | /* Copy the address. */ | |
c482c568 | 1709 | if (sin) { |
db8dac20 DM |
1710 | sin->sin_family = AF_INET; |
1711 | sin->sin_port = udp_hdr(skb)->source; | |
1712 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
1713 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
bceaa902 | 1714 | *addr_len = sizeof(*sin); |
db8dac20 DM |
1715 | } |
1716 | if (inet->cmsg_flags) | |
ad959036 | 1717 | ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); |
db8dac20 | 1718 | |
59c2cdae | 1719 | err = copied; |
db8dac20 DM |
1720 | if (flags & MSG_TRUNC) |
1721 | err = ulen; | |
1722 | ||
850cbadd | 1723 | skb_consume_udp(sk, skb, peeking ? -err : err); |
db8dac20 DM |
1724 | return err; |
1725 | ||
1726 | csum_copy_err: | |
2276f58a PA |
1727 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, |
1728 | udp_skb_destructor)) { | |
6aef70a8 ED |
1729 | UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
1730 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
6a5dc9e5 | 1731 | } |
850cbadd | 1732 | kfree_skb(skb); |
db8dac20 | 1733 | |
beb39db5 ED |
1734 | /* starting over for a new packet, but check if we need to yield */ |
1735 | cond_resched(); | |
9cfaa8de | 1736 | msg->msg_flags &= ~MSG_TRUNC; |
db8dac20 DM |
1737 | goto try_again; |
1738 | } | |
1739 | ||
d74bad4e AI |
1740 | int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
1741 | { | |
1742 | /* This check is replicated from __ip4_datagram_connect() and | |
1743 | * intended to prevent BPF program called below from accessing bytes | |
1744 | * that are out of the bound specified by user in addr_len. | |
1745 | */ | |
1746 | if (addr_len < sizeof(struct sockaddr_in)) | |
1747 | return -EINVAL; | |
1748 | ||
1749 | return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); | |
1750 | } | |
1751 | EXPORT_SYMBOL(udp_pre_connect); | |
1752 | ||
286c72de | 1753 | int __udp_disconnect(struct sock *sk, int flags) |
1da177e4 LT |
1754 | { |
1755 | struct inet_sock *inet = inet_sk(sk); | |
1756 | /* | |
1757 | * 1003.1g - break association. | |
1758 | */ | |
e905a9ed | 1759 | |
1da177e4 | 1760 | sk->sk_state = TCP_CLOSE; |
c720c7e8 ED |
1761 | inet->inet_daddr = 0; |
1762 | inet->inet_dport = 0; | |
bdeab991 | 1763 | sock_rps_reset_rxhash(sk); |
1da177e4 LT |
1764 | sk->sk_bound_dev_if = 0; |
1765 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
1766 | inet_reset_saddr(sk); | |
1767 | ||
1768 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1769 | sk->sk_prot->unhash(sk); | |
c720c7e8 | 1770 | inet->inet_sport = 0; |
1da177e4 LT |
1771 | } |
1772 | sk_dst_reset(sk); | |
1773 | return 0; | |
1774 | } | |
286c72de ED |
1775 | EXPORT_SYMBOL(__udp_disconnect); |
1776 | ||
1777 | int udp_disconnect(struct sock *sk, int flags) | |
1778 | { | |
1779 | lock_sock(sk); | |
1780 | __udp_disconnect(sk, flags); | |
1781 | release_sock(sk); | |
1782 | return 0; | |
1783 | } | |
c482c568 | 1784 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 1785 | |
645ca708 ED |
1786 | void udp_lib_unhash(struct sock *sk) |
1787 | { | |
723b4610 ED |
1788 | if (sk_hashed(sk)) { |
1789 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
512615b6 ED |
1790 | struct udp_hslot *hslot, *hslot2; |
1791 | ||
1792 | hslot = udp_hashslot(udptable, sock_net(sk), | |
1793 | udp_sk(sk)->udp_port_hash); | |
1794 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
645ca708 | 1795 | |
723b4610 | 1796 | spin_lock_bh(&hslot->lock); |
e32ea7e7 CG |
1797 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1798 | reuseport_detach_sock(sk); | |
ca065d0c | 1799 | if (sk_del_node_init_rcu(sk)) { |
fdcc8aa9 | 1800 | hslot->count--; |
c720c7e8 | 1801 | inet_sk(sk)->inet_num = 0; |
723b4610 | 1802 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
512615b6 ED |
1803 | |
1804 | spin_lock(&hslot2->lock); | |
ca065d0c | 1805 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
512615b6 ED |
1806 | hslot2->count--; |
1807 | spin_unlock(&hslot2->lock); | |
723b4610 ED |
1808 | } |
1809 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 1810 | } |
645ca708 ED |
1811 | } |
1812 | EXPORT_SYMBOL(udp_lib_unhash); | |
1813 | ||
719f8358 ED |
1814 | /* |
1815 | * inet_rcv_saddr was changed, we must rehash secondary hash | |
1816 | */ | |
1817 | void udp_lib_rehash(struct sock *sk, u16 newhash) | |
1818 | { | |
1819 | if (sk_hashed(sk)) { | |
1820 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
1821 | struct udp_hslot *hslot, *hslot2, *nhslot2; | |
1822 | ||
1823 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
1824 | nhslot2 = udp_hashslot2(udptable, newhash); | |
1825 | udp_sk(sk)->udp_portaddr_hash = newhash; | |
e32ea7e7 CG |
1826 | |
1827 | if (hslot2 != nhslot2 || | |
1828 | rcu_access_pointer(sk->sk_reuseport_cb)) { | |
719f8358 ED |
1829 | hslot = udp_hashslot(udptable, sock_net(sk), |
1830 | udp_sk(sk)->udp_port_hash); | |
1831 | /* we must lock primary chain too */ | |
1832 | spin_lock_bh(&hslot->lock); | |
e32ea7e7 CG |
1833 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1834 | reuseport_detach_sock(sk); | |
1835 | ||
1836 | if (hslot2 != nhslot2) { | |
1837 | spin_lock(&hslot2->lock); | |
ca065d0c | 1838 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
e32ea7e7 CG |
1839 | hslot2->count--; |
1840 | spin_unlock(&hslot2->lock); | |
1841 | ||
1842 | spin_lock(&nhslot2->lock); | |
ca065d0c | 1843 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
e32ea7e7 CG |
1844 | &nhslot2->head); |
1845 | nhslot2->count++; | |
1846 | spin_unlock(&nhslot2->lock); | |
1847 | } | |
719f8358 ED |
1848 | |
1849 | spin_unlock_bh(&hslot->lock); | |
1850 | } | |
1851 | } | |
1852 | } | |
1853 | EXPORT_SYMBOL(udp_lib_rehash); | |
1854 | ||
1855 | static void udp_v4_rehash(struct sock *sk) | |
1856 | { | |
f0b1e64c | 1857 | u16 new_hash = ipv4_portaddr_hash(sock_net(sk), |
719f8358 ED |
1858 | inet_sk(sk)->inet_rcv_saddr, |
1859 | inet_sk(sk)->inet_num); | |
1860 | udp_lib_rehash(sk, new_hash); | |
1861 | } | |
1862 | ||
a3f96c47 | 1863 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
93821778 | 1864 | { |
fec5e652 | 1865 | int rc; |
766e9037 | 1866 | |
005ec974 | 1867 | if (inet_sk(sk)->inet_daddr) { |
bdeab991 | 1868 | sock_rps_save_rxhash(sk, skb); |
005ec974 | 1869 | sk_mark_napi_id(sk, skb); |
2c8c56e1 | 1870 | sk_incoming_cpu_update(sk); |
e68b6e50 ED |
1871 | } else { |
1872 | sk_mark_napi_id_once(sk, skb); | |
005ec974 | 1873 | } |
fec5e652 | 1874 | |
850cbadd | 1875 | rc = __udp_enqueue_schedule_skb(sk, skb); |
766e9037 ED |
1876 | if (rc < 0) { |
1877 | int is_udplite = IS_UDPLITE(sk); | |
93821778 | 1878 | |
93821778 | 1879 | /* Note that an ENOMEM error is charged twice */ |
766e9037 | 1880 | if (rc == -ENOMEM) |
e61da9e2 | 1881 | UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
02c22347 | 1882 | is_udplite); |
e61da9e2 | 1883 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
766e9037 | 1884 | kfree_skb(skb); |
296f7ea7 | 1885 | trace_udp_fail_queue_rcv_skb(rc, sk); |
766e9037 | 1886 | return -1; |
93821778 HX |
1887 | } |
1888 | ||
1889 | return 0; | |
93821778 HX |
1890 | } |
1891 | ||
f2e9de21 | 1892 | DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); |
447167bf ED |
1893 | void udp_encap_enable(void) |
1894 | { | |
88ab3108 | 1895 | static_branch_enable(&udp_encap_needed_key); |
447167bf ED |
1896 | } |
1897 | EXPORT_SYMBOL(udp_encap_enable); | |
1898 | ||
db8dac20 DM |
1899 | /* returns: |
1900 | * -1: error | |
1901 | * 0: success | |
1902 | * >0: "udp encap" protocol resubmission | |
1903 | * | |
1904 | * Note that in the success and error cases, the skb is assumed to | |
1905 | * have either been requeued or freed. | |
1906 | */ | |
a3f96c47 | 1907 | static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 DM |
1908 | { |
1909 | struct udp_sock *up = udp_sk(sk); | |
db8dac20 DM |
1910 | int is_udplite = IS_UDPLITE(sk); |
1911 | ||
1912 | /* | |
1913 | * Charge it to the socket, dropping if the queue is full. | |
1914 | */ | |
1915 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1916 | goto drop; | |
1917 | nf_reset(skb); | |
1918 | ||
88ab3108 | 1919 | if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { |
0ad92ad0 ED |
1920 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
1921 | ||
db8dac20 DM |
1922 | /* |
1923 | * This is an encapsulation socket so pass the skb to | |
1924 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
1925 | * fall through and pass this up the UDP socket. | |
1926 | * up->encap_rcv() returns the following value: | |
1927 | * =0 if skb was successfully passed to the encap | |
1928 | * handler or was discarded by it. | |
1929 | * >0 if skb should be passed on to UDP. | |
1930 | * <0 if skb should be resubmitted as proto -N | |
1931 | */ | |
1932 | ||
1933 | /* if we're overly short, let UDP handle it */ | |
6aa7de05 | 1934 | encap_rcv = READ_ONCE(up->encap_rcv); |
e5aed006 | 1935 | if (encap_rcv) { |
db8dac20 DM |
1936 | int ret; |
1937 | ||
0a80966b TH |
1938 | /* Verify checksum before giving to encap */ |
1939 | if (udp_lib_checksum_complete(skb)) | |
1940 | goto csum_error; | |
1941 | ||
0ad92ad0 | 1942 | ret = encap_rcv(sk, skb); |
db8dac20 | 1943 | if (ret <= 0) { |
02c22347 ED |
1944 | __UDP_INC_STATS(sock_net(sk), |
1945 | UDP_MIB_INDATAGRAMS, | |
1946 | is_udplite); | |
db8dac20 DM |
1947 | return -ret; |
1948 | } | |
1949 | } | |
1950 | ||
1951 | /* FALLTHROUGH -- it's a UDP Packet */ | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * UDP-Lite specific tests, ignored on UDP sockets | |
1956 | */ | |
1957 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
1958 | ||
1959 | /* | |
1960 | * MIB statistics other than incrementing the error count are | |
1961 | * disabled for the following two types of errors: these depend | |
1962 | * on the application settings, not on the functioning of the | |
1963 | * protocol stack as such. | |
1964 | * | |
1965 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
1966 | * way ... to ... at least let the receiving application block | |
1967 | * delivery of packets with coverage values less than a value | |
1968 | * provided by the application." | |
1969 | */ | |
1970 | if (up->pcrlen == 0) { /* full coverage was set */ | |
ba7a46f1 JP |
1971 | net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", |
1972 | UDP_SKB_CB(skb)->cscov, skb->len); | |
db8dac20 DM |
1973 | goto drop; |
1974 | } | |
1975 | /* The next case involves violating the min. coverage requested | |
1976 | * by the receiver. This is subtle: if receiver wants x and x is | |
1977 | * greater than the buffersize/MTU then receiver will complain | |
1978 | * that it wants x while sender emits packets of smaller size y. | |
1979 | * Therefore the above ...()->partial_cov statement is essential. | |
1980 | */ | |
1981 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
ba7a46f1 JP |
1982 | net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", |
1983 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
db8dac20 DM |
1984 | goto drop; |
1985 | } | |
1986 | } | |
1987 | ||
dd99e425 | 1988 | prefetch(&sk->sk_rmem_alloc); |
ce25d66a ED |
1989 | if (rcu_access_pointer(sk->sk_filter) && |
1990 | udp_lib_checksum_complete(skb)) | |
e6afc8ac | 1991 | goto csum_error; |
ce25d66a | 1992 | |
ba66bbe5 | 1993 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) |
a6127697 | 1994 | goto drop; |
db8dac20 | 1995 | |
e6afc8ac | 1996 | udp_csum_pull_header(skb); |
db8dac20 | 1997 | |
fbf8866d | 1998 | ipv4_pktinfo_prepare(sk, skb); |
850cbadd | 1999 | return __udp_queue_rcv_skb(sk, skb); |
db8dac20 | 2000 | |
6a5dc9e5 | 2001 | csum_error: |
02c22347 | 2002 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
db8dac20 | 2003 | drop: |
02c22347 | 2004 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
8edf19c2 | 2005 | atomic_inc(&sk->sk_drops); |
db8dac20 DM |
2006 | kfree_skb(skb); |
2007 | return -1; | |
2008 | } | |
2009 | ||
97502231 | 2010 | /* For TCP sockets, sk_rx_dst is protected by socket lock |
e47eb5df | 2011 | * For UDP, we use xchg() to guard against concurrent changes. |
97502231 | 2012 | */ |
64f0f5d1 | 2013 | bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) |
421b3885 | 2014 | { |
97502231 ED |
2015 | struct dst_entry *old; |
2016 | ||
d24406c8 WW |
2017 | if (dst_hold_safe(dst)) { |
2018 | old = xchg(&sk->sk_rx_dst, dst); | |
2019 | dst_release(old); | |
64f0f5d1 | 2020 | return old != dst; |
d24406c8 | 2021 | } |
64f0f5d1 | 2022 | return false; |
421b3885 | 2023 | } |
c9f2c1ae | 2024 | EXPORT_SYMBOL(udp_sk_rx_dst_set); |
421b3885 | 2025 | |
db8dac20 DM |
2026 | /* |
2027 | * Multicasts and broadcasts go to each listener. | |
2028 | * | |
1240d137 | 2029 | * Note: called only from the BH handler context. |
db8dac20 | 2030 | */ |
e3163493 | 2031 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
2032 | struct udphdr *uh, |
2033 | __be32 saddr, __be32 daddr, | |
36cbb245 RJ |
2034 | struct udp_table *udptable, |
2035 | int proto) | |
db8dac20 | 2036 | { |
ca065d0c | 2037 | struct sock *sk, *first = NULL; |
5cf3d461 DH |
2038 | unsigned short hnum = ntohs(uh->dest); |
2039 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
2dc41cff | 2040 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); |
ca065d0c ED |
2041 | unsigned int offset = offsetof(typeof(*sk), sk_node); |
2042 | int dif = skb->dev->ifindex; | |
fb74c277 | 2043 | int sdif = inet_sdif(skb); |
ca065d0c ED |
2044 | struct hlist_node *node; |
2045 | struct sk_buff *nskb; | |
2dc41cff DH |
2046 | |
2047 | if (use_hash2) { | |
f0b1e64c | 2048 | hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & |
73e2d5e3 | 2049 | udptable->mask; |
f0b1e64c | 2050 | hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; |
2dc41cff | 2051 | start_lookup: |
73e2d5e3 | 2052 | hslot = &udptable->hash2[hash2]; |
2dc41cff DH |
2053 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); |
2054 | } | |
db8dac20 | 2055 | |
ca065d0c ED |
2056 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { |
2057 | if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, | |
fb74c277 | 2058 | uh->source, saddr, dif, sdif, hnum)) |
ca065d0c ED |
2059 | continue; |
2060 | ||
2061 | if (!first) { | |
2062 | first = sk; | |
2063 | continue; | |
1240d137 | 2064 | } |
ca065d0c | 2065 | nskb = skb_clone(skb, GFP_ATOMIC); |
1240d137 | 2066 | |
ca065d0c ED |
2067 | if (unlikely(!nskb)) { |
2068 | atomic_inc(&sk->sk_drops); | |
02c22347 ED |
2069 | __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, |
2070 | IS_UDPLITE(sk)); | |
2071 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, | |
2072 | IS_UDPLITE(sk)); | |
ca065d0c ED |
2073 | continue; |
2074 | } | |
2075 | if (udp_queue_rcv_skb(sk, nskb) > 0) | |
2076 | consume_skb(nskb); | |
2077 | } | |
1240d137 | 2078 | |
2dc41cff DH |
2079 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ |
2080 | if (use_hash2 && hash2 != hash2_any) { | |
2081 | hash2 = hash2_any; | |
2082 | goto start_lookup; | |
2083 | } | |
2084 | ||
ca065d0c ED |
2085 | if (first) { |
2086 | if (udp_queue_rcv_skb(first, skb) > 0) | |
2087 | consume_skb(skb); | |
1240d137 | 2088 | } else { |
ca065d0c | 2089 | kfree_skb(skb); |
02c22347 ED |
2090 | __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, |
2091 | proto == IPPROTO_UDPLITE); | |
1240d137 | 2092 | } |
db8dac20 DM |
2093 | return 0; |
2094 | } | |
2095 | ||
2096 | /* Initialize UDP checksum. If exited with zero value (success), | |
2097 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
2098 | * Otherwise, csum completion requires chacksumming packet body, | |
2099 | * including udp header and folding it to skb->csum. | |
2100 | */ | |
2101 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
2102 | int proto) | |
2103 | { | |
db8dac20 DM |
2104 | int err; |
2105 | ||
2106 | UDP_SKB_CB(skb)->partial_cov = 0; | |
2107 | UDP_SKB_CB(skb)->cscov = skb->len; | |
2108 | ||
2109 | if (proto == IPPROTO_UDPLITE) { | |
2110 | err = udplite_checksum_init(skb, uh); | |
2111 | if (err) | |
2112 | return err; | |
15f35d49 AK |
2113 | |
2114 | if (UDP_SKB_CB(skb)->partial_cov) { | |
2115 | skb->csum = inet_compute_pseudo(skb, proto); | |
2116 | return 0; | |
2117 | } | |
db8dac20 DM |
2118 | } |
2119 | ||
b46d9f62 HFS |
2120 | /* Note, we are only interested in != 0 or == 0, thus the |
2121 | * force to int. | |
2122 | */ | |
db4f1be3 ST |
2123 | err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, |
2124 | inet_compute_pseudo); | |
2125 | if (err) | |
2126 | return err; | |
2127 | ||
2128 | if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { | |
2129 | /* If SW calculated the value, we know it's bad */ | |
2130 | if (skb->csum_complete_sw) | |
2131 | return 1; | |
2132 | ||
2133 | /* HW says the value is bad. Let's validate that. | |
2134 | * skb->csum is no longer the full packet checksum, | |
2135 | * so don't treat it as such. | |
2136 | */ | |
2137 | skb_checksum_complete_unset(skb); | |
2138 | } | |
2139 | ||
2140 | return 0; | |
db8dac20 DM |
2141 | } |
2142 | ||
2b5a9217 PA |
2143 | /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and |
2144 | * return code conversion for ip layer consumption | |
2145 | */ | |
2146 | static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, | |
2147 | struct udphdr *uh) | |
2148 | { | |
2149 | int ret; | |
2150 | ||
2151 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) | |
2152 | skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, | |
2153 | inet_compute_pseudo); | |
2154 | ||
2155 | ret = udp_queue_rcv_skb(sk, skb); | |
2156 | ||
2157 | /* a return value > 0 means to resubmit the input, but | |
2158 | * it wants the return to be -protocol, or 0 | |
2159 | */ | |
2160 | if (ret > 0) | |
2161 | return -ret; | |
2162 | return 0; | |
2163 | } | |
2164 | ||
db8dac20 DM |
2165 | /* |
2166 | * All we need to do is get the socket, and then do a checksum. | |
2167 | */ | |
2168 | ||
645ca708 | 2169 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
2170 | int proto) |
2171 | { | |
2172 | struct sock *sk; | |
7b5e56f9 | 2173 | struct udphdr *uh; |
db8dac20 | 2174 | unsigned short ulen; |
adf30907 | 2175 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 2176 | __be32 saddr, daddr; |
0283328e | 2177 | struct net *net = dev_net(skb->dev); |
db8dac20 DM |
2178 | |
2179 | /* | |
2180 | * Validate the packet. | |
2181 | */ | |
2182 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
2183 | goto drop; /* No space for header. */ | |
2184 | ||
7b5e56f9 | 2185 | uh = udp_hdr(skb); |
db8dac20 | 2186 | ulen = ntohs(uh->len); |
ccc2d97c BM |
2187 | saddr = ip_hdr(skb)->saddr; |
2188 | daddr = ip_hdr(skb)->daddr; | |
2189 | ||
db8dac20 DM |
2190 | if (ulen > skb->len) |
2191 | goto short_packet; | |
2192 | ||
2193 | if (proto == IPPROTO_UDP) { | |
2194 | /* UDP validates ulen. */ | |
2195 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
2196 | goto short_packet; | |
2197 | uh = udp_hdr(skb); | |
2198 | } | |
2199 | ||
2200 | if (udp4_csum_init(skb, uh, proto)) | |
2201 | goto csum_error; | |
2202 | ||
8afdd99a ED |
2203 | sk = skb_steal_sock(skb); |
2204 | if (sk) { | |
97502231 | 2205 | struct dst_entry *dst = skb_dst(skb); |
421b3885 | 2206 | int ret; |
421b3885 | 2207 | |
97502231 ED |
2208 | if (unlikely(sk->sk_rx_dst != dst)) |
2209 | udp_sk_rx_dst_set(sk, dst); | |
db8dac20 | 2210 | |
2b5a9217 | 2211 | ret = udp_unicast_rcv_skb(sk, skb, uh); |
8afdd99a | 2212 | sock_put(sk); |
2b5a9217 | 2213 | return ret; |
421b3885 | 2214 | } |
db8dac20 | 2215 | |
c18450a5 FF |
2216 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
2217 | return __udp4_lib_mcast_deliver(net, skb, uh, | |
36cbb245 | 2218 | saddr, daddr, udptable, proto); |
c18450a5 FF |
2219 | |
2220 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
2b5a9217 PA |
2221 | if (sk) |
2222 | return udp_unicast_rcv_skb(sk, skb, uh); | |
db8dac20 DM |
2223 | |
2224 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2225 | goto drop; | |
2226 | nf_reset(skb); | |
2227 | ||
2228 | /* No socket. Drop packet silently, if checksum is wrong */ | |
2229 | if (udp_lib_checksum_complete(skb)) | |
2230 | goto csum_error; | |
2231 | ||
02c22347 | 2232 | __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2233 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
2234 | ||
2235 | /* | |
2236 | * Hmm. We got an UDP packet to a port to which we | |
2237 | * don't wanna listen. Ignore it. | |
2238 | */ | |
2239 | kfree_skb(skb); | |
2240 | return 0; | |
2241 | ||
2242 | short_packet: | |
ba7a46f1 JP |
2243 | net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
2244 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2245 | &saddr, ntohs(uh->source), | |
2246 | ulen, skb->len, | |
2247 | &daddr, ntohs(uh->dest)); | |
db8dac20 DM |
2248 | goto drop; |
2249 | ||
2250 | csum_error: | |
2251 | /* | |
2252 | * RFC1122: OK. Discards the bad packet silently (as far as | |
2253 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
2254 | */ | |
ba7a46f1 JP |
2255 | net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
2256 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2257 | &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), | |
2258 | ulen); | |
02c22347 | 2259 | __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 | 2260 | drop: |
02c22347 | 2261 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2262 | kfree_skb(skb); |
2263 | return 0; | |
2264 | } | |
2265 | ||
421b3885 SB |
2266 | /* We can only early demux multicast if there is a single matching socket. |
2267 | * If more than one socket found returns NULL | |
2268 | */ | |
2269 | static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, | |
2270 | __be16 loc_port, __be32 loc_addr, | |
2271 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 2272 | int dif, int sdif) |
421b3885 SB |
2273 | { |
2274 | struct sock *sk, *result; | |
421b3885 | 2275 | unsigned short hnum = ntohs(loc_port); |
ca065d0c | 2276 | unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); |
421b3885 SB |
2277 | struct udp_hslot *hslot = &udp_table.hash[slot]; |
2278 | ||
63c6f81c ED |
2279 | /* Do not bother scanning a too big list */ |
2280 | if (hslot->count > 10) | |
2281 | return NULL; | |
2282 | ||
421b3885 | 2283 | result = NULL; |
ca065d0c ED |
2284 | sk_for_each_rcu(sk, &hslot->head) { |
2285 | if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, | |
fb74c277 | 2286 | rmt_port, rmt_addr, dif, sdif, hnum)) { |
ca065d0c ED |
2287 | if (result) |
2288 | return NULL; | |
421b3885 | 2289 | result = sk; |
421b3885 SB |
2290 | } |
2291 | } | |
ca065d0c | 2292 | |
421b3885 SB |
2293 | return result; |
2294 | } | |
2295 | ||
2296 | /* For unicast we should only early demux connected sockets or we can | |
2297 | * break forwarding setups. The chains here can be long so only check | |
2298 | * if the first socket is an exact match and if not move on. | |
2299 | */ | |
2300 | static struct sock *__udp4_lib_demux_lookup(struct net *net, | |
2301 | __be16 loc_port, __be32 loc_addr, | |
2302 | __be16 rmt_port, __be32 rmt_addr, | |
3fa6f616 | 2303 | int dif, int sdif) |
421b3885 | 2304 | { |
421b3885 | 2305 | unsigned short hnum = ntohs(loc_port); |
f0b1e64c | 2306 | unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); |
421b3885 SB |
2307 | unsigned int slot2 = hash2 & udp_table.mask; |
2308 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
c7228317 | 2309 | INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); |
421b3885 | 2310 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); |
ca065d0c | 2311 | struct sock *sk; |
421b3885 | 2312 | |
ca065d0c ED |
2313 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
2314 | if (INET_MATCH(sk, net, acookie, rmt_addr, | |
3fa6f616 | 2315 | loc_addr, ports, dif, sdif)) |
ca065d0c | 2316 | return sk; |
421b3885 SB |
2317 | /* Only check first socket in chain */ |
2318 | break; | |
2319 | } | |
ca065d0c | 2320 | return NULL; |
421b3885 SB |
2321 | } |
2322 | ||
7487449c | 2323 | int udp_v4_early_demux(struct sk_buff *skb) |
421b3885 | 2324 | { |
610438b7 | 2325 | struct net *net = dev_net(skb->dev); |
bc044e8d | 2326 | struct in_device *in_dev = NULL; |
610438b7 ED |
2327 | const struct iphdr *iph; |
2328 | const struct udphdr *uh; | |
ca065d0c | 2329 | struct sock *sk = NULL; |
421b3885 | 2330 | struct dst_entry *dst; |
421b3885 | 2331 | int dif = skb->dev->ifindex; |
fb74c277 | 2332 | int sdif = inet_sdif(skb); |
6e540309 | 2333 | int ours; |
421b3885 SB |
2334 | |
2335 | /* validate the packet */ | |
2336 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) | |
7487449c | 2337 | return 0; |
421b3885 | 2338 | |
610438b7 ED |
2339 | iph = ip_hdr(skb); |
2340 | uh = udp_hdr(skb); | |
2341 | ||
996b44fc | 2342 | if (skb->pkt_type == PACKET_MULTICAST) { |
bc044e8d | 2343 | in_dev = __in_dev_get_rcu(skb->dev); |
6e540309 SB |
2344 | |
2345 | if (!in_dev) | |
7487449c | 2346 | return 0; |
6e540309 | 2347 | |
996b44fc PA |
2348 | ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, |
2349 | iph->protocol); | |
2350 | if (!ours) | |
2351 | return 0; | |
ad0ea198 | 2352 | |
421b3885 | 2353 | sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, |
fb74c277 DA |
2354 | uh->source, iph->saddr, |
2355 | dif, sdif); | |
6e540309 | 2356 | } else if (skb->pkt_type == PACKET_HOST) { |
421b3885 | 2357 | sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, |
3fa6f616 | 2358 | uh->source, iph->saddr, dif, sdif); |
6e540309 | 2359 | } |
421b3885 | 2360 | |
41c6d650 | 2361 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) |
7487449c | 2362 | return 0; |
421b3885 SB |
2363 | |
2364 | skb->sk = sk; | |
82eabd9e | 2365 | skb->destructor = sock_efree; |
10e2eb87 | 2366 | dst = READ_ONCE(sk->sk_rx_dst); |
421b3885 SB |
2367 | |
2368 | if (dst) | |
2369 | dst = dst_check(dst, 0); | |
10e2eb87 | 2370 | if (dst) { |
bc044e8d PA |
2371 | u32 itag = 0; |
2372 | ||
d24406c8 WW |
2373 | /* set noref for now. |
2374 | * any place which wants to hold dst has to call | |
2375 | * dst_hold_safe() | |
2376 | */ | |
2377 | skb_dst_set_noref(skb, dst); | |
bc044e8d PA |
2378 | |
2379 | /* for unconnected multicast sockets we need to validate | |
2380 | * the source on each packet | |
2381 | */ | |
2382 | if (!inet_sk(sk)->inet_daddr && in_dev) | |
2383 | return ip_mc_validate_source(skb, iph->daddr, | |
2384 | iph->saddr, iph->tos, | |
2385 | skb->dev, in_dev, &itag); | |
10e2eb87 | 2386 | } |
7487449c | 2387 | return 0; |
421b3885 SB |
2388 | } |
2389 | ||
db8dac20 DM |
2390 | int udp_rcv(struct sk_buff *skb) |
2391 | { | |
645ca708 | 2392 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
2393 | } |
2394 | ||
7d06b2e0 | 2395 | void udp_destroy_sock(struct sock *sk) |
db8dac20 | 2396 | { |
44046a59 | 2397 | struct udp_sock *up = udp_sk(sk); |
8a74ad60 | 2398 | bool slow = lock_sock_fast(sk); |
db8dac20 | 2399 | udp_flush_pending_frames(sk); |
8a74ad60 | 2400 | unlock_sock_fast(sk, slow); |
88ab3108 | 2401 | if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { |
44046a59 | 2402 | void (*encap_destroy)(struct sock *sk); |
6aa7de05 | 2403 | encap_destroy = READ_ONCE(up->encap_destroy); |
44046a59 TP |
2404 | if (encap_destroy) |
2405 | encap_destroy(sk); | |
2406 | } | |
db8dac20 DM |
2407 | } |
2408 | ||
1da177e4 LT |
2409 | /* |
2410 | * Socket option code for UDP | |
2411 | */ | |
4c0a6cb0 | 2412 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2413 | char __user *optval, unsigned int optlen, |
4c0a6cb0 | 2414 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
2415 | { |
2416 | struct udp_sock *up = udp_sk(sk); | |
1c19448c | 2417 | int val, valbool; |
1da177e4 | 2418 | int err = 0; |
b2bf1e26 | 2419 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 2420 | |
c482c568 | 2421 | if (optlen < sizeof(int)) |
1da177e4 LT |
2422 | return -EINVAL; |
2423 | ||
2424 | if (get_user(val, (int __user *)optval)) | |
2425 | return -EFAULT; | |
2426 | ||
1c19448c TH |
2427 | valbool = val ? 1 : 0; |
2428 | ||
6516c655 | 2429 | switch (optname) { |
1da177e4 LT |
2430 | case UDP_CORK: |
2431 | if (val != 0) { | |
2432 | up->corkflag = 1; | |
2433 | } else { | |
2434 | up->corkflag = 0; | |
2435 | lock_sock(sk); | |
4243cdc2 | 2436 | push_pending_frames(sk); |
1da177e4 LT |
2437 | release_sock(sk); |
2438 | } | |
2439 | break; | |
e905a9ed | 2440 | |
1da177e4 LT |
2441 | case UDP_ENCAP: |
2442 | switch (val) { | |
2443 | case 0: | |
2444 | case UDP_ENCAP_ESPINUDP: | |
2445 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
067b207b JC |
2446 | up->encap_rcv = xfrm4_udp_encap_rcv; |
2447 | /* FALLTHROUGH */ | |
342f0234 | 2448 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 | 2449 | up->encap_type = val; |
447167bf | 2450 | udp_encap_enable(); |
1da177e4 LT |
2451 | break; |
2452 | default: | |
2453 | err = -ENOPROTOOPT; | |
2454 | break; | |
2455 | } | |
2456 | break; | |
2457 | ||
1c19448c TH |
2458 | case UDP_NO_CHECK6_TX: |
2459 | up->no_check6_tx = valbool; | |
2460 | break; | |
2461 | ||
2462 | case UDP_NO_CHECK6_RX: | |
2463 | up->no_check6_rx = valbool; | |
2464 | break; | |
2465 | ||
bec1f6f6 WB |
2466 | case UDP_SEGMENT: |
2467 | if (val < 0 || val > USHRT_MAX) | |
2468 | return -EINVAL; | |
2469 | up->gso_size = val; | |
2470 | break; | |
2471 | ||
ba4e58ec GR |
2472 | /* |
2473 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
2474 | */ | |
2475 | /* The sender sets actual checksum coverage length via this option. | |
2476 | * The case coverage > packet length is handled by send module. */ | |
2477 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 2478 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2479 | return -ENOPROTOOPT; |
2480 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
2481 | val = 8; | |
4be929be AD |
2482 | else if (val > USHRT_MAX) |
2483 | val = USHRT_MAX; | |
ba4e58ec GR |
2484 | up->pcslen = val; |
2485 | up->pcflag |= UDPLITE_SEND_CC; | |
2486 | break; | |
2487 | ||
e905a9ed YH |
2488 | /* The receiver specifies a minimum checksum coverage value. To make |
2489 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
2490 | * used, this again means full checksum coverage. */ |
2491 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 2492 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2493 | return -ENOPROTOOPT; |
2494 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
2495 | val = 8; | |
4be929be AD |
2496 | else if (val > USHRT_MAX) |
2497 | val = USHRT_MAX; | |
ba4e58ec GR |
2498 | up->pcrlen = val; |
2499 | up->pcflag |= UDPLITE_RECV_CC; | |
2500 | break; | |
2501 | ||
1da177e4 LT |
2502 | default: |
2503 | err = -ENOPROTOOPT; | |
2504 | break; | |
6516c655 | 2505 | } |
1da177e4 LT |
2506 | |
2507 | return err; | |
2508 | } | |
c482c568 | 2509 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 2510 | |
db8dac20 | 2511 | int udp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2512 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2513 | { |
2514 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2515 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2516 | udp_push_pending_frames); | |
2517 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
2518 | } | |
2519 | ||
2520 | #ifdef CONFIG_COMPAT | |
2521 | int compat_udp_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 2522 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2523 | { |
2524 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2525 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2526 | udp_push_pending_frames); | |
2527 | return compat_ip_setsockopt(sk, level, optname, optval, optlen); | |
2528 | } | |
2529 | #endif | |
2530 | ||
4c0a6cb0 GR |
2531 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
2532 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
2533 | { |
2534 | struct udp_sock *up = udp_sk(sk); | |
2535 | int val, len; | |
2536 | ||
c482c568 | 2537 | if (get_user(len, optlen)) |
1da177e4 LT |
2538 | return -EFAULT; |
2539 | ||
2540 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 2541 | |
6516c655 | 2542 | if (len < 0) |
1da177e4 LT |
2543 | return -EINVAL; |
2544 | ||
6516c655 | 2545 | switch (optname) { |
1da177e4 LT |
2546 | case UDP_CORK: |
2547 | val = up->corkflag; | |
2548 | break; | |
2549 | ||
2550 | case UDP_ENCAP: | |
2551 | val = up->encap_type; | |
2552 | break; | |
2553 | ||
1c19448c TH |
2554 | case UDP_NO_CHECK6_TX: |
2555 | val = up->no_check6_tx; | |
2556 | break; | |
2557 | ||
2558 | case UDP_NO_CHECK6_RX: | |
2559 | val = up->no_check6_rx; | |
2560 | break; | |
2561 | ||
bec1f6f6 WB |
2562 | case UDP_SEGMENT: |
2563 | val = up->gso_size; | |
2564 | break; | |
2565 | ||
ba4e58ec GR |
2566 | /* The following two cannot be changed on UDP sockets, the return is |
2567 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
2568 | case UDPLITE_SEND_CSCOV: | |
2569 | val = up->pcslen; | |
2570 | break; | |
2571 | ||
2572 | case UDPLITE_RECV_CSCOV: | |
2573 | val = up->pcrlen; | |
2574 | break; | |
2575 | ||
1da177e4 LT |
2576 | default: |
2577 | return -ENOPROTOOPT; | |
6516c655 | 2578 | } |
1da177e4 | 2579 | |
6516c655 | 2580 | if (put_user(len, optlen)) |
e905a9ed | 2581 | return -EFAULT; |
c482c568 | 2582 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 2583 | return -EFAULT; |
e905a9ed | 2584 | return 0; |
1da177e4 | 2585 | } |
c482c568 | 2586 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 2587 | |
db8dac20 DM |
2588 | int udp_getsockopt(struct sock *sk, int level, int optname, |
2589 | char __user *optval, int __user *optlen) | |
2590 | { | |
2591 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2592 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2593 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
2594 | } | |
2595 | ||
2596 | #ifdef CONFIG_COMPAT | |
2597 | int compat_udp_getsockopt(struct sock *sk, int level, int optname, | |
2598 | char __user *optval, int __user *optlen) | |
2599 | { | |
2600 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2601 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2602 | return compat_ip_getsockopt(sk, level, optname, optval, optlen); | |
2603 | } | |
2604 | #endif | |
1da177e4 LT |
2605 | /** |
2606 | * udp_poll - wait for a UDP event. | |
2607 | * @file - file struct | |
2608 | * @sock - socket | |
a11e1d43 | 2609 | * @wait - poll table |
1da177e4 | 2610 | * |
e905a9ed | 2611 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
2612 | * blocking sockets. If application is using a blocking fd |
2613 | * and a packet with checksum error is in the queue; | |
2614 | * then it could get return from select indicating data available | |
2615 | * but then block when reading it. Add special case code | |
2616 | * to work around these arguably broken applications. | |
2617 | */ | |
a11e1d43 | 2618 | __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 2619 | { |
a11e1d43 | 2620 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 | 2621 | struct sock *sk = sock->sk; |
ba4e58ec | 2622 | |
2276f58a | 2623 | if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) |
a9a08845 | 2624 | mask |= EPOLLIN | EPOLLRDNORM; |
2276f58a | 2625 | |
1da177e4 | 2626 | /* Check for false positives due to checksum errors */ |
a11e1d43 | 2627 | if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && |
e83c6744 | 2628 | !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) |
a9a08845 | 2629 | mask &= ~(EPOLLIN | EPOLLRDNORM); |
1da177e4 LT |
2630 | |
2631 | return mask; | |
e905a9ed | 2632 | |
1da177e4 | 2633 | } |
a11e1d43 | 2634 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 2635 | |
5d77dca8 DA |
2636 | int udp_abort(struct sock *sk, int err) |
2637 | { | |
2638 | lock_sock(sk); | |
2639 | ||
2640 | sk->sk_err = err; | |
2641 | sk->sk_error_report(sk); | |
286c72de | 2642 | __udp_disconnect(sk, 0); |
5d77dca8 DA |
2643 | |
2644 | release_sock(sk); | |
2645 | ||
2646 | return 0; | |
2647 | } | |
2648 | EXPORT_SYMBOL_GPL(udp_abort); | |
2649 | ||
db8dac20 | 2650 | struct proto udp_prot = { |
1e802951 TZ |
2651 | .name = "UDP", |
2652 | .owner = THIS_MODULE, | |
2653 | .close = udp_lib_close, | |
d74bad4e | 2654 | .pre_connect = udp_pre_connect, |
1e802951 TZ |
2655 | .connect = ip4_datagram_connect, |
2656 | .disconnect = udp_disconnect, | |
2657 | .ioctl = udp_ioctl, | |
2658 | .init = udp_init_sock, | |
2659 | .destroy = udp_destroy_sock, | |
2660 | .setsockopt = udp_setsockopt, | |
2661 | .getsockopt = udp_getsockopt, | |
2662 | .sendmsg = udp_sendmsg, | |
2663 | .recvmsg = udp_recvmsg, | |
2664 | .sendpage = udp_sendpage, | |
2665 | .release_cb = ip4_datagram_release_cb, | |
2666 | .hash = udp_lib_hash, | |
2667 | .unhash = udp_lib_unhash, | |
2668 | .rehash = udp_v4_rehash, | |
2669 | .get_port = udp_v4_get_port, | |
2670 | .memory_allocated = &udp_memory_allocated, | |
2671 | .sysctl_mem = sysctl_udp_mem, | |
2672 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), | |
2673 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), | |
2674 | .obj_size = sizeof(struct udp_sock), | |
2675 | .h.udp_table = &udp_table, | |
db8dac20 | 2676 | #ifdef CONFIG_COMPAT |
1e802951 TZ |
2677 | .compat_setsockopt = compat_udp_setsockopt, |
2678 | .compat_getsockopt = compat_udp_getsockopt, | |
db8dac20 | 2679 | #endif |
1e802951 | 2680 | .diag_destroy = udp_abort, |
db8dac20 | 2681 | }; |
c482c568 | 2682 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
2683 | |
2684 | /* ------------------------------------------------------------------------ */ | |
2685 | #ifdef CONFIG_PROC_FS | |
2686 | ||
645ca708 | 2687 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
2688 | { |
2689 | struct sock *sk; | |
a3d2599b | 2690 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2691 | struct udp_iter_state *state = seq->private; |
6f191efe | 2692 | struct net *net = seq_file_net(seq); |
1da177e4 | 2693 | |
a3d2599b | 2694 | for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; |
f86dcc5a | 2695 | ++state->bucket) { |
a3d2599b | 2696 | struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; |
f86dcc5a | 2697 | |
ca065d0c | 2698 | if (hlist_empty(&hslot->head)) |
f86dcc5a ED |
2699 | continue; |
2700 | ||
645ca708 | 2701 | spin_lock_bh(&hslot->lock); |
ca065d0c | 2702 | sk_for_each(sk, &hslot->head) { |
878628fb | 2703 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 2704 | continue; |
a3d2599b | 2705 | if (sk->sk_family == afinfo->family) |
1da177e4 LT |
2706 | goto found; |
2707 | } | |
645ca708 | 2708 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
2709 | } |
2710 | sk = NULL; | |
2711 | found: | |
2712 | return sk; | |
2713 | } | |
2714 | ||
2715 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
2716 | { | |
a3d2599b | 2717 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2718 | struct udp_iter_state *state = seq->private; |
6f191efe | 2719 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2720 | |
2721 | do { | |
ca065d0c | 2722 | sk = sk_next(sk); |
a3d2599b | 2723 | } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); |
1da177e4 | 2724 | |
645ca708 | 2725 | if (!sk) { |
a3d2599b CH |
2726 | if (state->bucket <= afinfo->udp_table->mask) |
2727 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
645ca708 | 2728 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
2729 | } |
2730 | return sk; | |
2731 | } | |
2732 | ||
2733 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
2734 | { | |
645ca708 | 2735 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
2736 | |
2737 | if (sk) | |
6516c655 | 2738 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
2739 | --pos; |
2740 | return pos ? NULL : sk; | |
2741 | } | |
2742 | ||
a3d2599b | 2743 | void *udp_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 2744 | { |
30842f29 | 2745 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 2746 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 2747 | |
b50660f1 | 2748 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 | 2749 | } |
a3d2599b | 2750 | EXPORT_SYMBOL(udp_seq_start); |
1da177e4 | 2751 | |
a3d2599b | 2752 | void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1da177e4 LT |
2753 | { |
2754 | struct sock *sk; | |
2755 | ||
b50660f1 | 2756 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
2757 | sk = udp_get_idx(seq, 0); |
2758 | else | |
2759 | sk = udp_get_next(seq, v); | |
2760 | ||
2761 | ++*pos; | |
2762 | return sk; | |
2763 | } | |
a3d2599b | 2764 | EXPORT_SYMBOL(udp_seq_next); |
1da177e4 | 2765 | |
a3d2599b | 2766 | void udp_seq_stop(struct seq_file *seq, void *v) |
1da177e4 | 2767 | { |
a3d2599b | 2768 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
645ca708 ED |
2769 | struct udp_iter_state *state = seq->private; |
2770 | ||
a3d2599b CH |
2771 | if (state->bucket <= afinfo->udp_table->mask) |
2772 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
1da177e4 | 2773 | } |
a3d2599b | 2774 | EXPORT_SYMBOL(udp_seq_stop); |
db8dac20 DM |
2775 | |
2776 | /* ------------------------------------------------------------------------ */ | |
5e659e4c | 2777 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
652586df | 2778 | int bucket) |
db8dac20 DM |
2779 | { |
2780 | struct inet_sock *inet = inet_sk(sp); | |
c720c7e8 ED |
2781 | __be32 dest = inet->inet_daddr; |
2782 | __be32 src = inet->inet_rcv_saddr; | |
2783 | __u16 destp = ntohs(inet->inet_dport); | |
2784 | __u16 srcp = ntohs(inet->inet_sport); | |
db8dac20 | 2785 | |
f86dcc5a | 2786 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
652586df | 2787 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", |
db8dac20 | 2788 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 | 2789 | sk_wmem_alloc_get(sp), |
6c206b20 | 2790 | udp_rqueue_get(sp), |
a7cb5a49 EB |
2791 | 0, 0L, 0, |
2792 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), | |
2793 | 0, sock_i_ino(sp), | |
41c6d650 | 2794 | refcount_read(&sp->sk_refcnt), sp, |
652586df | 2795 | atomic_read(&sp->sk_drops)); |
db8dac20 DM |
2796 | } |
2797 | ||
2798 | int udp4_seq_show(struct seq_file *seq, void *v) | |
2799 | { | |
652586df | 2800 | seq_setwidth(seq, 127); |
db8dac20 | 2801 | if (v == SEQ_START_TOKEN) |
652586df | 2802 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
db8dac20 | 2803 | "rx_queue tr tm->when retrnsmt uid timeout " |
cb61cb9b | 2804 | "inode ref pointer drops"); |
db8dac20 | 2805 | else { |
db8dac20 DM |
2806 | struct udp_iter_state *state = seq->private; |
2807 | ||
652586df | 2808 | udp4_format_sock(v, seq, state->bucket); |
db8dac20 | 2809 | } |
652586df | 2810 | seq_pad(seq, '\n'); |
db8dac20 DM |
2811 | return 0; |
2812 | } | |
2813 | ||
c3506372 | 2814 | const struct seq_operations udp_seq_ops = { |
a3d2599b CH |
2815 | .start = udp_seq_start, |
2816 | .next = udp_seq_next, | |
2817 | .stop = udp_seq_stop, | |
2818 | .show = udp4_seq_show, | |
2819 | }; | |
c3506372 | 2820 | EXPORT_SYMBOL(udp_seq_ops); |
73cb88ec | 2821 | |
db8dac20 | 2822 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 | 2823 | .family = AF_INET, |
645ca708 | 2824 | .udp_table = &udp_table, |
db8dac20 DM |
2825 | }; |
2826 | ||
2c8c1e72 | 2827 | static int __net_init udp4_proc_init_net(struct net *net) |
15439feb | 2828 | { |
c3506372 CH |
2829 | if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, |
2830 | sizeof(struct udp_iter_state), &udp4_seq_afinfo)) | |
a3d2599b CH |
2831 | return -ENOMEM; |
2832 | return 0; | |
15439feb PE |
2833 | } |
2834 | ||
2c8c1e72 | 2835 | static void __net_exit udp4_proc_exit_net(struct net *net) |
15439feb | 2836 | { |
a3d2599b | 2837 | remove_proc_entry("udp", net->proc_net); |
15439feb PE |
2838 | } |
2839 | ||
2840 | static struct pernet_operations udp4_net_ops = { | |
2841 | .init = udp4_proc_init_net, | |
2842 | .exit = udp4_proc_exit_net, | |
2843 | }; | |
2844 | ||
db8dac20 DM |
2845 | int __init udp4_proc_init(void) |
2846 | { | |
15439feb | 2847 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
2848 | } |
2849 | ||
2850 | void udp4_proc_exit(void) | |
2851 | { | |
15439feb | 2852 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 2853 | } |
1da177e4 LT |
2854 | #endif /* CONFIG_PROC_FS */ |
2855 | ||
f86dcc5a ED |
2856 | static __initdata unsigned long uhash_entries; |
2857 | static int __init set_uhash_entries(char *str) | |
645ca708 | 2858 | { |
413c27d8 EZ |
2859 | ssize_t ret; |
2860 | ||
f86dcc5a ED |
2861 | if (!str) |
2862 | return 0; | |
413c27d8 EZ |
2863 | |
2864 | ret = kstrtoul(str, 0, &uhash_entries); | |
2865 | if (ret) | |
2866 | return 0; | |
2867 | ||
f86dcc5a ED |
2868 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) |
2869 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
2870 | return 1; | |
2871 | } | |
2872 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 2873 | |
f86dcc5a ED |
2874 | void __init udp_table_init(struct udp_table *table, const char *name) |
2875 | { | |
2876 | unsigned int i; | |
2877 | ||
31fe62b9 TB |
2878 | table->hash = alloc_large_system_hash(name, |
2879 | 2 * sizeof(struct udp_hslot), | |
2880 | uhash_entries, | |
2881 | 21, /* one slot per 2 MB */ | |
2882 | 0, | |
2883 | &table->log, | |
2884 | &table->mask, | |
2885 | UDP_HTABLE_SIZE_MIN, | |
2886 | 64 * 1024); | |
2887 | ||
512615b6 | 2888 | table->hash2 = table->hash + (table->mask + 1); |
f86dcc5a | 2889 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2890 | INIT_HLIST_HEAD(&table->hash[i].head); |
fdcc8aa9 | 2891 | table->hash[i].count = 0; |
645ca708 ED |
2892 | spin_lock_init(&table->hash[i].lock); |
2893 | } | |
512615b6 | 2894 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2895 | INIT_HLIST_HEAD(&table->hash2[i].head); |
512615b6 ED |
2896 | table->hash2[i].count = 0; |
2897 | spin_lock_init(&table->hash2[i].lock); | |
2898 | } | |
645ca708 ED |
2899 | } |
2900 | ||
723b8e46 TH |
2901 | u32 udp_flow_hashrnd(void) |
2902 | { | |
2903 | static u32 hashrnd __read_mostly; | |
2904 | ||
2905 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
2906 | ||
2907 | return hashrnd; | |
2908 | } | |
2909 | EXPORT_SYMBOL(udp_flow_hashrnd); | |
2910 | ||
1e802951 TZ |
2911 | static void __udp_sysctl_init(struct net *net) |
2912 | { | |
2913 | net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; | |
2914 | net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; | |
2915 | ||
2916 | #ifdef CONFIG_NET_L3_MASTER_DEV | |
2917 | net->ipv4.sysctl_udp_l3mdev_accept = 0; | |
2918 | #endif | |
2919 | } | |
2920 | ||
2921 | static int __net_init udp_sysctl_init(struct net *net) | |
2922 | { | |
2923 | __udp_sysctl_init(net); | |
2924 | return 0; | |
2925 | } | |
2926 | ||
2927 | static struct pernet_operations __net_initdata udp_sysctl_ops = { | |
fc18999e | 2928 | .init = udp_sysctl_init, |
1e802951 TZ |
2929 | }; |
2930 | ||
95766fff HA |
2931 | void __init udp_init(void) |
2932 | { | |
f03d78db | 2933 | unsigned long limit; |
4b272750 | 2934 | unsigned int i; |
95766fff | 2935 | |
f86dcc5a | 2936 | udp_table_init(&udp_table, "UDP"); |
f03d78db | 2937 | limit = nr_free_buffer_pages() / 8; |
95766fff HA |
2938 | limit = max(limit, 128UL); |
2939 | sysctl_udp_mem[0] = limit / 4 * 3; | |
2940 | sysctl_udp_mem[1] = limit; | |
2941 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
2942 | ||
1e802951 | 2943 | __udp_sysctl_init(&init_net); |
4b272750 ED |
2944 | |
2945 | /* 16 spinlocks per cpu */ | |
2946 | udp_busylocks_log = ilog2(nr_cpu_ids) + 4; | |
2947 | udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, | |
2948 | GFP_KERNEL); | |
2949 | if (!udp_busylocks) | |
2950 | panic("UDP: failed to alloc udp_busylocks\n"); | |
2951 | for (i = 0; i < (1U << udp_busylocks_log); i++) | |
2952 | spin_lock_init(udp_busylocks + i); | |
1e802951 TZ |
2953 | |
2954 | if (register_pernet_subsys(&udp_sysctl_ops)) | |
2955 | panic("UDP: failed to init sysctl parameters.\n"); | |
95766fff | 2956 | } |