Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * The User Datagram Protocol (UDP). | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
113aa838 | 12 | * Alan Cox, <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
13 | * Hirokazu Takahashi, <taka@valinux.co.jp> |
14 | * | |
15 | * Fixes: | |
16 | * Alan Cox : verify_area() calls | |
17 | * Alan Cox : stopped close while in use off icmp | |
18 | * messages. Not a fix but a botch that | |
19 | * for udp at least is 'valid'. | |
20 | * Alan Cox : Fixed icmp handling properly | |
21 | * Alan Cox : Correct error for oversized datagrams | |
e905a9ed YH |
22 | * Alan Cox : Tidied select() semantics. |
23 | * Alan Cox : udp_err() fixed properly, also now | |
1da177e4 LT |
24 | * select and read wake correctly on errors |
25 | * Alan Cox : udp_send verify_area moved to avoid mem leak | |
26 | * Alan Cox : UDP can count its memory | |
27 | * Alan Cox : send to an unknown connection causes | |
28 | * an ECONNREFUSED off the icmp, but | |
29 | * does NOT close. | |
30 | * Alan Cox : Switched to new sk_buff handlers. No more backlog! | |
31 | * Alan Cox : Using generic datagram code. Even smaller and the PEEK | |
32 | * bug no longer crashes it. | |
33 | * Fred Van Kempen : Net2e support for sk->broadcast. | |
34 | * Alan Cox : Uses skb_free_datagram | |
35 | * Alan Cox : Added get/set sockopt support. | |
36 | * Alan Cox : Broadcasting without option set returns EACCES. | |
37 | * Alan Cox : No wakeup calls. Instead we now use the callbacks. | |
38 | * Alan Cox : Use ip_tos and ip_ttl | |
39 | * Alan Cox : SNMP Mibs | |
40 | * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. | |
41 | * Matt Dillon : UDP length checks. | |
42 | * Alan Cox : Smarter af_inet used properly. | |
43 | * Alan Cox : Use new kernel side addressing. | |
44 | * Alan Cox : Incorrect return on truncated datagram receive. | |
45 | * Arnt Gulbrandsen : New udp_send and stuff | |
46 | * Alan Cox : Cache last socket | |
47 | * Alan Cox : Route cache | |
48 | * Jon Peatfield : Minor efficiency fix to sendto(). | |
49 | * Mike Shaver : RFC1122 checks. | |
50 | * Alan Cox : Nonblocking error fix. | |
51 | * Willy Konynenberg : Transparent proxying support. | |
52 | * Mike McLagan : Routing by source | |
53 | * David S. Miller : New socket lookup architecture. | |
54 | * Last socket cache retained as it | |
55 | * does have a high hit rate. | |
56 | * Olaf Kirch : Don't linearise iovec on sendmsg. | |
57 | * Andi Kleen : Some cleanups, cache destination entry | |
e905a9ed | 58 | * for connect. |
1da177e4 LT |
59 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. |
60 | * Melvin Smith : Check msg_name not msg_namelen in sendto(), | |
61 | * return ENOTCONN for unconnected sockets (POSIX) | |
62 | * Janos Farkas : don't deliver multi/broadcasts to a different | |
63 | * bound-to-device socket | |
64 | * Hirokazu Takahashi : HW checksumming for outgoing UDP | |
65 | * datagrams. | |
66 | * Hirokazu Takahashi : sendfile() on UDP works now. | |
67 | * Arnaldo C. Melo : convert /proc/net/udp to seq_file | |
68 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
69 | * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind | |
70 | * a single port at the same time. | |
71 | * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support | |
342f0234 | 72 | * James Chapman : Add L2TP encapsulation type. |
1da177e4 | 73 | */ |
e905a9ed | 74 | |
afd46503 JP |
75 | #define pr_fmt(fmt) "UDP: " fmt |
76 | ||
aef2feda | 77 | #include <linux/bpf-cgroup.h> |
7c0f6ba6 | 78 | #include <linux/uaccess.h> |
1da177e4 | 79 | #include <asm/ioctls.h> |
57c8a661 | 80 | #include <linux/memblock.h> |
8203efb3 | 81 | #include <linux/highmem.h> |
1da177e4 LT |
82 | #include <linux/types.h> |
83 | #include <linux/fcntl.h> | |
84 | #include <linux/module.h> | |
85 | #include <linux/socket.h> | |
86 | #include <linux/sockios.h> | |
14c85021 | 87 | #include <linux/igmp.h> |
6e540309 | 88 | #include <linux/inetdevice.h> |
1da177e4 LT |
89 | #include <linux/in.h> |
90 | #include <linux/errno.h> | |
91 | #include <linux/timer.h> | |
92 | #include <linux/mm.h> | |
1da177e4 | 93 | #include <linux/inet.h> |
1da177e4 | 94 | #include <linux/netdevice.h> |
5a0e3ad6 | 95 | #include <linux/slab.h> |
c752f073 | 96 | #include <net/tcp_states.h> |
1da177e4 LT |
97 | #include <linux/skbuff.h> |
98 | #include <linux/proc_fs.h> | |
99 | #include <linux/seq_file.h> | |
457c4cbc | 100 | #include <net/net_namespace.h> |
1da177e4 | 101 | #include <net/icmp.h> |
421b3885 | 102 | #include <net/inet_hashtables.h> |
e7cc0824 | 103 | #include <net/ip_tunnels.h> |
1da177e4 | 104 | #include <net/route.h> |
1da177e4 LT |
105 | #include <net/checksum.h> |
106 | #include <net/xfrm.h> | |
296f7ea7 | 107 | #include <trace/events/udp.h> |
447167bf | 108 | #include <linux/static_key.h> |
951cf368 | 109 | #include <linux/btf_ids.h> |
22911fc5 | 110 | #include <trace/events/skb.h> |
076bb0c8 | 111 | #include <net/busy_poll.h> |
ba4e58ec | 112 | #include "udp_impl.h" |
e32ea7e7 | 113 | #include <net/sock_reuseport.h> |
217375a0 | 114 | #include <net/addrconf.h> |
60fb9567 | 115 | #include <net/udp_tunnel.h> |
0146dca7 SD |
116 | #if IS_ENABLED(CONFIG_IPV6) |
117 | #include <net/ipv6_stubs.h> | |
118 | #endif | |
1da177e4 | 119 | |
f86dcc5a | 120 | struct udp_table udp_table __read_mostly; |
645ca708 | 121 | EXPORT_SYMBOL(udp_table); |
1da177e4 | 122 | |
8d987e5c | 123 | long sysctl_udp_mem[3] __read_mostly; |
95766fff | 124 | EXPORT_SYMBOL(sysctl_udp_mem); |
c482c568 | 125 | |
91b6d325 | 126 | atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; |
95766fff | 127 | EXPORT_SYMBOL(udp_memory_allocated); |
0defbb0a ED |
128 | DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); |
129 | EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); | |
95766fff | 130 | |
f86dcc5a ED |
131 | #define MAX_UDP_PORTS 65536 |
132 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 133 | |
f24d43c0 | 134 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 135 | const struct udp_hslot *hslot, |
98322f22 | 136 | unsigned long *bitmap, |
fe38d2a1 | 137 | struct sock *sk, unsigned int log) |
1da177e4 | 138 | { |
f24d43c0 | 139 | struct sock *sk2; |
ba418fa3 | 140 | kuid_t uid = sock_i_uid(sk); |
25030a7f | 141 | |
ca065d0c | 142 | sk_for_each(sk2, &hslot->head) { |
9d4fb27d JP |
143 | if (net_eq(sock_net(sk2), net) && |
144 | sk2 != sk && | |
d4cada4a | 145 | (bitmap || udp_sk(sk2)->udp_port_hash == num) && |
9d4fb27d JP |
146 | (!sk2->sk_reuse || !sk->sk_reuse) && |
147 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
148 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 149 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
150 | if (sk2->sk_reuseport && sk->sk_reuseport && |
151 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
152 | uid_eq(uid, sock_i_uid(sk2))) { | |
153 | if (!bitmap) | |
154 | return 0; | |
155 | } else { | |
156 | if (!bitmap) | |
157 | return 1; | |
158 | __set_bit(udp_sk(sk2)->udp_port_hash >> log, | |
159 | bitmap); | |
160 | } | |
98322f22 | 161 | } |
4243cdc2 | 162 | } |
25030a7f GR |
163 | return 0; |
164 | } | |
165 | ||
30fff923 ED |
166 | /* |
167 | * Note: we still hold spinlock of primary hash chain, so no other writer | |
168 | * can insert/delete a socket with local_port == num | |
169 | */ | |
170 | static int udp_lib_lport_inuse2(struct net *net, __u16 num, | |
4243cdc2 | 171 | struct udp_hslot *hslot2, |
fe38d2a1 | 172 | struct sock *sk) |
30fff923 ED |
173 | { |
174 | struct sock *sk2; | |
ba418fa3 | 175 | kuid_t uid = sock_i_uid(sk); |
30fff923 ED |
176 | int res = 0; |
177 | ||
178 | spin_lock(&hslot2->lock); | |
ca065d0c | 179 | udp_portaddr_for_each_entry(sk2, &hslot2->head) { |
9d4fb27d JP |
180 | if (net_eq(sock_net(sk2), net) && |
181 | sk2 != sk && | |
182 | (udp_sk(sk2)->udp_port_hash == num) && | |
183 | (!sk2->sk_reuse || !sk->sk_reuse) && | |
184 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
185 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 186 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
187 | if (sk2->sk_reuseport && sk->sk_reuseport && |
188 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
189 | uid_eq(uid, sock_i_uid(sk2))) { | |
190 | res = 0; | |
191 | } else { | |
192 | res = 1; | |
193 | } | |
30fff923 ED |
194 | break; |
195 | } | |
4243cdc2 | 196 | } |
30fff923 ED |
197 | spin_unlock(&hslot2->lock); |
198 | return res; | |
199 | } | |
200 | ||
fe38d2a1 | 201 | static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) |
e32ea7e7 CG |
202 | { |
203 | struct net *net = sock_net(sk); | |
e32ea7e7 CG |
204 | kuid_t uid = sock_i_uid(sk); |
205 | struct sock *sk2; | |
206 | ||
ca065d0c | 207 | sk_for_each(sk2, &hslot->head) { |
e32ea7e7 CG |
208 | if (net_eq(sock_net(sk2), net) && |
209 | sk2 != sk && | |
210 | sk2->sk_family == sk->sk_family && | |
211 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
212 | (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && | |
213 | (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
214 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && | |
fe38d2a1 | 215 | inet_rcv_saddr_equal(sk, sk2, false)) { |
2dbb9b9e MKL |
216 | return reuseport_add_sock(sk, sk2, |
217 | inet_rcv_saddr_any(sk)); | |
e32ea7e7 CG |
218 | } |
219 | } | |
220 | ||
2dbb9b9e | 221 | return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); |
e32ea7e7 CG |
222 | } |
223 | ||
25030a7f | 224 | /** |
6ba5a3c5 | 225 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
226 | * |
227 | * @sk: socket struct in question | |
228 | * @snum: port number to look up | |
25985edc | 229 | * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, |
30fff923 | 230 | * with NULL address |
25030a7f | 231 | */ |
6ba5a3c5 | 232 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
30fff923 | 233 | unsigned int hash2_nulladdr) |
25030a7f | 234 | { |
512615b6 | 235 | struct udp_hslot *hslot, *hslot2; |
645ca708 | 236 | struct udp_table *udptable = sk->sk_prot->h.udp_table; |
25030a7f | 237 | int error = 1; |
3b1e0a65 | 238 | struct net *net = sock_net(sk); |
1da177e4 | 239 | |
32c1da70 | 240 | if (!snum) { |
9088c560 | 241 | int low, high, remaining; |
95c96174 | 242 | unsigned int rand; |
98322f22 ED |
243 | unsigned short first, last; |
244 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 245 | |
0bbf87d8 | 246 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 247 | remaining = (high - low) + 1; |
227b60f5 | 248 | |
63862b5b | 249 | rand = prandom_u32(); |
8fc54f68 | 250 | first = reciprocal_scale(rand, remaining) + low; |
98322f22 ED |
251 | /* |
252 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
253 | */ | |
f86dcc5a | 254 | rand = (rand | 1) * (udptable->mask + 1); |
5781b235 ED |
255 | last = first + udptable->mask + 1; |
256 | do { | |
f86dcc5a | 257 | hslot = udp_hashslot(udptable, net, first); |
98322f22 | 258 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 259 | spin_lock_bh(&hslot->lock); |
98322f22 | 260 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
fe38d2a1 | 261 | udptable->log); |
98322f22 ED |
262 | |
263 | snum = first; | |
264 | /* | |
265 | * Iterate on all possible values of snum for this hash. | |
266 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
267 | * give us randomization and full range coverage. | |
268 | */ | |
9088c560 | 269 | do { |
98322f22 | 270 | if (low <= snum && snum <= high && |
e3826f1e | 271 | !test_bit(snum >> udptable->log, bitmap) && |
122ff243 | 272 | !inet_is_local_reserved_port(net, snum)) |
98322f22 ED |
273 | goto found; |
274 | snum += rand; | |
275 | } while (snum != first); | |
276 | spin_unlock_bh(&hslot->lock); | |
df560056 | 277 | cond_resched(); |
5781b235 | 278 | } while (++first != last); |
98322f22 | 279 | goto fail; |
645ca708 | 280 | } else { |
f86dcc5a | 281 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 282 | spin_lock_bh(&hslot->lock); |
30fff923 ED |
283 | if (hslot->count > 10) { |
284 | int exist; | |
285 | unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; | |
286 | ||
287 | slot2 &= udptable->mask; | |
288 | hash2_nulladdr &= udptable->mask; | |
289 | ||
290 | hslot2 = udp_hashslot2(udptable, slot2); | |
291 | if (hslot->count < hslot2->count) | |
292 | goto scan_primary_hash; | |
293 | ||
fe38d2a1 | 294 | exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); |
30fff923 ED |
295 | if (!exist && (hash2_nulladdr != slot2)) { |
296 | hslot2 = udp_hashslot2(udptable, hash2_nulladdr); | |
297 | exist = udp_lib_lport_inuse2(net, snum, hslot2, | |
fe38d2a1 | 298 | sk); |
30fff923 ED |
299 | } |
300 | if (exist) | |
301 | goto fail_unlock; | |
302 | else | |
303 | goto found; | |
304 | } | |
305 | scan_primary_hash: | |
fe38d2a1 | 306 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) |
645ca708 ED |
307 | goto fail_unlock; |
308 | } | |
98322f22 | 309 | found: |
c720c7e8 | 310 | inet_sk(sk)->inet_num = snum; |
d4cada4a ED |
311 | udp_sk(sk)->udp_port_hash = snum; |
312 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
1da177e4 | 313 | if (sk_unhashed(sk)) { |
e32ea7e7 | 314 | if (sk->sk_reuseport && |
fe38d2a1 | 315 | udp_reuseport_add_sock(sk, hslot)) { |
e32ea7e7 CG |
316 | inet_sk(sk)->inet_num = 0; |
317 | udp_sk(sk)->udp_port_hash = 0; | |
318 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
319 | goto fail_unlock; | |
320 | } | |
321 | ||
ca065d0c | 322 | sk_add_node_rcu(sk, &hslot->head); |
fdcc8aa9 | 323 | hslot->count++; |
c29a0bc4 | 324 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
512615b6 ED |
325 | |
326 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
327 | spin_lock(&hslot2->lock); | |
d894ba18 | 328 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
1602f49b DM |
329 | sk->sk_family == AF_INET6) |
330 | hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, | |
331 | &hslot2->head); | |
d894ba18 | 332 | else |
1602f49b DM |
333 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
334 | &hslot2->head); | |
512615b6 ED |
335 | hslot2->count++; |
336 | spin_unlock(&hslot2->lock); | |
1da177e4 | 337 | } |
ca065d0c | 338 | sock_set_flag(sk, SOCK_RCU_FREE); |
25030a7f | 339 | error = 0; |
645ca708 ED |
340 | fail_unlock: |
341 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 342 | fail: |
25030a7f GR |
343 | return error; |
344 | } | |
c482c568 | 345 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 346 | |
6ba5a3c5 | 347 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 348 | { |
30fff923 | 349 | unsigned int hash2_nulladdr = |
f0b1e64c | 350 | ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); |
30fff923 | 351 | unsigned int hash2_partial = |
f0b1e64c | 352 | ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); |
30fff923 | 353 | |
d4cada4a | 354 | /* precompute partial secondary hash */ |
30fff923 | 355 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
fe38d2a1 | 356 | return udp_lib_get_port(sk, snum, hash2_nulladdr); |
db8dac20 DM |
357 | } |
358 | ||
d1e37288 SX |
359 | static int compute_score(struct sock *sk, struct net *net, |
360 | __be32 saddr, __be16 sport, | |
fb74c277 | 361 | __be32 daddr, unsigned short hnum, |
73545373 | 362 | int dif, int sdif) |
645ca708 | 363 | { |
60c04aec JP |
364 | int score; |
365 | struct inet_sock *inet; | |
6da5b0f0 | 366 | bool dev_match; |
645ca708 | 367 | |
60c04aec JP |
368 | if (!net_eq(sock_net(sk), net) || |
369 | udp_sk(sk)->udp_port_hash != hnum || | |
370 | ipv6_only_sock(sk)) | |
371 | return -1; | |
645ca708 | 372 | |
4cdeeee9 PO |
373 | if (sk->sk_rcv_saddr != daddr) |
374 | return -1; | |
60c04aec | 375 | |
4cdeeee9 | 376 | score = (sk->sk_family == PF_INET) ? 2 : 1; |
60c04aec | 377 | |
4cdeeee9 | 378 | inet = inet_sk(sk); |
60c04aec JP |
379 | if (inet->inet_daddr) { |
380 | if (inet->inet_daddr != saddr) | |
381 | return -1; | |
382 | score += 4; | |
383 | } | |
384 | ||
385 | if (inet->inet_dport) { | |
386 | if (inet->inet_dport != sport) | |
387 | return -1; | |
388 | score += 4; | |
389 | } | |
390 | ||
6da5b0f0 MM |
391 | dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, |
392 | dif, sdif); | |
393 | if (!dev_match) | |
394 | return -1; | |
8d6c414c MM |
395 | if (sk->sk_bound_dev_if) |
396 | score += 4; | |
fb74c277 | 397 | |
7170a977 | 398 | if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) |
70da268b | 399 | score++; |
645ca708 ED |
400 | return score; |
401 | } | |
402 | ||
6eada011 ED |
403 | static u32 udp_ehashfn(const struct net *net, const __be32 laddr, |
404 | const __u16 lport, const __be32 faddr, | |
405 | const __be16 fport) | |
65cd8033 | 406 | { |
1bbdceef HFS |
407 | static u32 udp_ehash_secret __read_mostly; |
408 | ||
409 | net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); | |
410 | ||
65cd8033 | 411 | return __inet_ehashfn(laddr, lport, faddr, fport, |
1bbdceef | 412 | udp_ehash_secret + net_hash_mix(net)); |
65cd8033 HFS |
413 | } |
414 | ||
a57066b1 DM |
415 | static struct sock *lookup_reuseport(struct net *net, struct sock *sk, |
416 | struct sk_buff *skb, | |
417 | __be32 saddr, __be16 sport, | |
418 | __be32 daddr, unsigned short hnum) | |
7629c73a JS |
419 | { |
420 | struct sock *reuse_sk = NULL; | |
421 | u32 hash; | |
422 | ||
423 | if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { | |
424 | hash = udp_ehashfn(net, daddr, hnum, saddr, sport); | |
425 | reuse_sk = reuseport_select_sock(sk, hash, skb, | |
426 | sizeof(struct udphdr)); | |
7629c73a JS |
427 | } |
428 | return reuse_sk; | |
429 | } | |
430 | ||
d1e37288 | 431 | /* called with rcu_read_lock() */ |
5051ebd2 | 432 | static struct sock *udp4_lib_lookup2(struct net *net, |
fb74c277 DA |
433 | __be32 saddr, __be16 sport, |
434 | __be32 daddr, unsigned int hnum, | |
73545373 | 435 | int dif, int sdif, |
fb74c277 DA |
436 | struct udp_hslot *hslot2, |
437 | struct sk_buff *skb) | |
5051ebd2 ED |
438 | { |
439 | struct sock *sk, *result; | |
e94a62f5 | 440 | int score, badness; |
5051ebd2 | 441 | |
5051ebd2 | 442 | result = NULL; |
ba418fa3 | 443 | badness = 0; |
ca065d0c | 444 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
d1e37288 | 445 | score = compute_score(sk, net, saddr, sport, |
73545373 | 446 | daddr, hnum, dif, sdif); |
5051ebd2 | 447 | if (score > badness) { |
7629c73a JS |
448 | result = lookup_reuseport(net, sk, skb, |
449 | saddr, sport, daddr, hnum); | |
a57066b1 DM |
450 | /* Fall back to scoring if group has connections */ |
451 | if (result && !reuseport_has_conns(sk, false)) | |
7629c73a JS |
452 | return result; |
453 | ||
a57066b1 | 454 | result = result ? : sk; |
ca065d0c | 455 | badness = score; |
5051ebd2 ED |
456 | } |
457 | } | |
5051ebd2 ED |
458 | return result; |
459 | } | |
460 | ||
a57066b1 DM |
461 | static struct sock *udp4_lookup_run_bpf(struct net *net, |
462 | struct udp_table *udptable, | |
463 | struct sk_buff *skb, | |
464 | __be32 saddr, __be16 sport, | |
f8931565 | 465 | __be32 daddr, u16 hnum, const int dif) |
72f7e944 JS |
466 | { |
467 | struct sock *sk, *reuse_sk; | |
468 | bool no_reuseport; | |
469 | ||
470 | if (udptable != &udp_table) | |
471 | return NULL; /* only UDP is supported */ | |
472 | ||
f8931565 MP |
473 | no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, |
474 | daddr, hnum, dif, &sk); | |
72f7e944 JS |
475 | if (no_reuseport || IS_ERR_OR_NULL(sk)) |
476 | return sk; | |
477 | ||
478 | reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); | |
c64c9c28 | 479 | if (reuse_sk) |
72f7e944 JS |
480 | sk = reuse_sk; |
481 | return sk; | |
482 | } | |
483 | ||
db8dac20 DM |
484 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
485 | * harder than this. -DaveM | |
486 | */ | |
fce82338 | 487 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, |
fb74c277 DA |
488 | __be16 sport, __be32 daddr, __be16 dport, int dif, |
489 | int sdif, struct udp_table *udptable, struct sk_buff *skb) | |
db8dac20 | 490 | { |
db8dac20 | 491 | unsigned short hnum = ntohs(dport); |
4cdeeee9 PO |
492 | unsigned int hash2, slot2; |
493 | struct udp_hslot *hslot2; | |
72f7e944 | 494 | struct sock *result, *sk; |
645ca708 | 495 | |
4cdeeee9 PO |
496 | hash2 = ipv4_portaddr_hash(net, daddr, hnum); |
497 | slot2 = hash2 & udptable->mask; | |
498 | hslot2 = &udptable->hash2[slot2]; | |
499 | ||
72f7e944 | 500 | /* Lookup connected or non-wildcard socket */ |
4cdeeee9 PO |
501 | result = udp4_lib_lookup2(net, saddr, sport, |
502 | daddr, hnum, dif, sdif, | |
73545373 | 503 | hslot2, skb); |
72f7e944 JS |
504 | if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) |
505 | goto done; | |
506 | ||
507 | /* Lookup redirect from BPF */ | |
508 | if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { | |
509 | sk = udp4_lookup_run_bpf(net, udptable, skb, | |
f8931565 | 510 | saddr, sport, daddr, hnum, dif); |
72f7e944 JS |
511 | if (sk) { |
512 | result = sk; | |
513 | goto done; | |
514 | } | |
db8dac20 | 515 | } |
72f7e944 JS |
516 | |
517 | /* Got non-wildcard socket or error on first lookup */ | |
518 | if (result) | |
519 | goto done; | |
520 | ||
521 | /* Lookup wildcard sockets */ | |
522 | hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); | |
523 | slot2 = hash2 & udptable->mask; | |
524 | hslot2 = &udptable->hash2[slot2]; | |
525 | ||
526 | result = udp4_lib_lookup2(net, saddr, sport, | |
527 | htonl(INADDR_ANY), hnum, dif, sdif, | |
528 | hslot2, skb); | |
529 | done: | |
88e235b8 | 530 | if (IS_ERR(result)) |
4cdeeee9 | 531 | return NULL; |
db8dac20 DM |
532 | return result; |
533 | } | |
fce82338 | 534 | EXPORT_SYMBOL_GPL(__udp4_lib_lookup); |
db8dac20 | 535 | |
607c4aaf KK |
536 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
537 | __be16 sport, __be16 dport, | |
645ca708 | 538 | struct udp_table *udptable) |
607c4aaf KK |
539 | { |
540 | const struct iphdr *iph = ip_hdr(skb); | |
541 | ||
ed7cbbce | 542 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, |
8afdd99a | 543 | iph->daddr, dport, inet_iif(skb), |
fb74c277 | 544 | inet_sdif(skb), udptable, skb); |
607c4aaf KK |
545 | } |
546 | ||
7b58e63e | 547 | struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, |
63058308 TH |
548 | __be16 sport, __be16 dport) |
549 | { | |
257a525f MKL |
550 | const struct iphdr *iph = ip_hdr(skb); |
551 | ||
552 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, | |
553 | iph->daddr, dport, inet_iif(skb), | |
554 | inet_sdif(skb), &udp_table, NULL); | |
63058308 | 555 | } |
63058308 | 556 | |
ca065d0c ED |
557 | /* Must be called under rcu_read_lock(). |
558 | * Does increment socket refcount. | |
559 | */ | |
6e86000c | 560 | #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) |
bcd41303 KK |
561 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
562 | __be32 daddr, __be16 dport, int dif) | |
563 | { | |
ca065d0c ED |
564 | struct sock *sk; |
565 | ||
566 | sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, | |
fb74c277 | 567 | dif, 0, &udp_table, NULL); |
41c6d650 | 568 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) |
ca065d0c ED |
569 | sk = NULL; |
570 | return sk; | |
bcd41303 KK |
571 | } |
572 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
ca065d0c | 573 | #endif |
bcd41303 | 574 | |
421b3885 SB |
575 | static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, |
576 | __be16 loc_port, __be32 loc_addr, | |
577 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 578 | int dif, int sdif, unsigned short hnum) |
421b3885 SB |
579 | { |
580 | struct inet_sock *inet = inet_sk(sk); | |
581 | ||
582 | if (!net_eq(sock_net(sk), net) || | |
583 | udp_sk(sk)->udp_port_hash != hnum || | |
584 | (inet->inet_daddr && inet->inet_daddr != rmt_addr) || | |
585 | (inet->inet_dport != rmt_port && inet->inet_dport) || | |
586 | (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || | |
587 | ipv6_only_sock(sk) || | |
82ba25c6 | 588 | !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) |
421b3885 | 589 | return false; |
60d9b031 | 590 | if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) |
421b3885 SB |
591 | return false; |
592 | return true; | |
593 | } | |
594 | ||
a36e185e SB |
595 | DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); |
596 | void udp_encap_enable(void) | |
597 | { | |
9c480601 | 598 | static_branch_inc(&udp_encap_needed_key); |
a36e185e SB |
599 | } |
600 | EXPORT_SYMBOL(udp_encap_enable); | |
601 | ||
a4a600dd XL |
602 | void udp_encap_disable(void) |
603 | { | |
604 | static_branch_dec(&udp_encap_needed_key); | |
605 | } | |
606 | EXPORT_SYMBOL(udp_encap_disable); | |
607 | ||
e7cc0824 SB |
608 | /* Handler for tunnels with arbitrary destination ports: no socket lookup, go |
609 | * through error handlers in encapsulations looking for a match. | |
610 | */ | |
611 | static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) | |
612 | { | |
613 | int i; | |
614 | ||
615 | for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { | |
616 | int (*handler)(struct sk_buff *skb, u32 info); | |
92b95364 | 617 | const struct ip_tunnel_encap_ops *encap; |
e7cc0824 | 618 | |
92b95364 PA |
619 | encap = rcu_dereference(iptun_encaps[i]); |
620 | if (!encap) | |
e7cc0824 | 621 | continue; |
92b95364 | 622 | handler = encap->err_handler; |
e7cc0824 SB |
623 | if (handler && !handler(skb, info)) |
624 | return 0; | |
625 | } | |
626 | ||
627 | return -ENOENT; | |
628 | } | |
629 | ||
a36e185e SB |
630 | /* Try to match ICMP errors to UDP tunnels by looking up a socket without |
631 | * reversing source and destination port: this will match tunnels that force the | |
632 | * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that | |
633 | * lwtunnels might actually break this assumption by being configured with | |
634 | * different destination ports on endpoints, in this case we won't be able to | |
635 | * trace ICMP messages back to them. | |
636 | * | |
e7cc0824 SB |
637 | * If this doesn't match any socket, probe tunnels with arbitrary destination |
638 | * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port | |
639 | * we've sent packets to won't necessarily match the local destination port. | |
640 | * | |
a36e185e SB |
641 | * Then ask the tunnel implementation to match the error against a valid |
642 | * association. | |
643 | * | |
e7cc0824 SB |
644 | * Return an error if we can't find a match, the socket if we need further |
645 | * processing, zero otherwise. | |
a36e185e SB |
646 | */ |
647 | static struct sock *__udp4_lib_err_encap(struct net *net, | |
648 | const struct iphdr *iph, | |
649 | struct udphdr *uh, | |
650 | struct udp_table *udptable, | |
9bfce73c | 651 | struct sock *sk, |
e7cc0824 | 652 | struct sk_buff *skb, u32 info) |
a36e185e | 653 | { |
9bfce73c | 654 | int (*lookup)(struct sock *sk, struct sk_buff *skb); |
a36e185e | 655 | int network_offset, transport_offset; |
9bfce73c | 656 | struct udp_sock *up; |
a36e185e | 657 | |
a36e185e SB |
658 | network_offset = skb_network_offset(skb); |
659 | transport_offset = skb_transport_offset(skb); | |
660 | ||
661 | /* Network header needs to point to the outer IPv4 header inside ICMP */ | |
662 | skb_reset_network_header(skb); | |
663 | ||
664 | /* Transport header needs to point to the UDP header */ | |
665 | skb_set_transport_header(skb, iph->ihl << 2); | |
666 | ||
9bfce73c VF |
667 | if (sk) { |
668 | up = udp_sk(sk); | |
669 | ||
670 | lookup = READ_ONCE(up->encap_err_lookup); | |
671 | if (lookup && lookup(sk, skb)) | |
672 | sk = NULL; | |
673 | ||
674 | goto out; | |
675 | } | |
676 | ||
e7cc0824 SB |
677 | sk = __udp4_lib_lookup(net, iph->daddr, uh->source, |
678 | iph->saddr, uh->dest, skb->dev->ifindex, 0, | |
679 | udptable, NULL); | |
680 | if (sk) { | |
9bfce73c | 681 | up = udp_sk(sk); |
e7cc0824 SB |
682 | |
683 | lookup = READ_ONCE(up->encap_err_lookup); | |
684 | if (!lookup || lookup(sk, skb)) | |
685 | sk = NULL; | |
686 | } | |
687 | ||
9bfce73c | 688 | out: |
e7cc0824 SB |
689 | if (!sk) |
690 | sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); | |
a36e185e SB |
691 | |
692 | skb_set_transport_header(skb, transport_offset); | |
693 | skb_set_network_header(skb, network_offset); | |
694 | ||
695 | return sk; | |
696 | } | |
697 | ||
db8dac20 DM |
698 | /* |
699 | * This routine is called by the ICMP module when it gets some | |
700 | * sort of error condition. If err < 0 then the socket should | |
701 | * be closed and the error returned to the user. If err > 0 | |
702 | * it's just the icmp type << 8 | icmp code. | |
703 | * Header points to the ip header of the error packet. We move | |
704 | * on past this. Then (as it used to claim before adjustment) | |
705 | * header points to the first 8 bytes of the udp header. We need | |
706 | * to find the appropriate port. | |
707 | */ | |
708 | ||
32bbd879 | 709 | int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
710 | { |
711 | struct inet_sock *inet; | |
b71d1d42 | 712 | const struct iphdr *iph = (const struct iphdr *)skb->data; |
c482c568 | 713 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); |
db8dac20 DM |
714 | const int type = icmp_hdr(skb)->type; |
715 | const int code = icmp_hdr(skb)->code; | |
a36e185e | 716 | bool tunnel = false; |
db8dac20 DM |
717 | struct sock *sk; |
718 | int harderr; | |
719 | int err; | |
fd54d716 | 720 | struct net *net = dev_net(skb->dev); |
db8dac20 | 721 | |
fd54d716 | 722 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
f64bf6b8 MM |
723 | iph->saddr, uh->source, skb->dev->ifindex, |
724 | inet_sdif(skb), udptable, NULL); | |
9bfce73c | 725 | |
d26796ae | 726 | if (!sk || udp_sk(sk)->encap_type) { |
a36e185e | 727 | /* No socket for error: try tunnels before discarding */ |
e7cc0824 | 728 | if (static_branch_unlikely(&udp_encap_needed_key)) { |
9bfce73c | 729 | sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, |
e7cc0824 SB |
730 | info); |
731 | if (!sk) | |
732 | return 0; | |
9bfce73c VF |
733 | } else |
734 | sk = ERR_PTR(-ENOENT); | |
a36e185e | 735 | |
e7cc0824 | 736 | if (IS_ERR(sk)) { |
a36e185e | 737 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
e7cc0824 | 738 | return PTR_ERR(sk); |
a36e185e | 739 | } |
e7cc0824 | 740 | |
a36e185e | 741 | tunnel = true; |
db8dac20 DM |
742 | } |
743 | ||
744 | err = 0; | |
745 | harderr = 0; | |
746 | inet = inet_sk(sk); | |
747 | ||
748 | switch (type) { | |
749 | default: | |
750 | case ICMP_TIME_EXCEEDED: | |
751 | err = EHOSTUNREACH; | |
752 | break; | |
753 | case ICMP_SOURCE_QUENCH: | |
754 | goto out; | |
755 | case ICMP_PARAMETERPROB: | |
756 | err = EPROTO; | |
757 | harderr = 1; | |
758 | break; | |
759 | case ICMP_DEST_UNREACH: | |
760 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
36393395 | 761 | ipv4_sk_update_pmtu(skb, sk, info); |
db8dac20 DM |
762 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { |
763 | err = EMSGSIZE; | |
764 | harderr = 1; | |
765 | break; | |
766 | } | |
767 | goto out; | |
768 | } | |
769 | err = EHOSTUNREACH; | |
770 | if (code <= NR_ICMP_UNREACH) { | |
771 | harderr = icmp_err_convert[code].fatal; | |
772 | err = icmp_err_convert[code].errno; | |
773 | } | |
774 | break; | |
55be7a9c DM |
775 | case ICMP_REDIRECT: |
776 | ipv4_sk_redirect(skb, sk); | |
1a462d18 | 777 | goto out; |
db8dac20 DM |
778 | } |
779 | ||
780 | /* | |
781 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
782 | * 4.1.3.3. | |
783 | */ | |
a36e185e SB |
784 | if (tunnel) { |
785 | /* ...not for tunnels though: we don't have a sending socket */ | |
786 | goto out; | |
787 | } | |
db8dac20 DM |
788 | if (!inet->recverr) { |
789 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
790 | goto out; | |
b1faf566 | 791 | } else |
c482c568 | 792 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
b1faf566 | 793 | |
db8dac20 | 794 | sk->sk_err = err; |
e3ae2365 | 795 | sk_error_report(sk); |
db8dac20 | 796 | out: |
32bbd879 | 797 | return 0; |
db8dac20 DM |
798 | } |
799 | ||
32bbd879 | 800 | int udp_err(struct sk_buff *skb, u32 info) |
db8dac20 | 801 | { |
32bbd879 | 802 | return __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
803 | } |
804 | ||
805 | /* | |
806 | * Throw away all pending data and cancel the corking. Socket is locked. | |
807 | */ | |
36d926b9 | 808 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
809 | { |
810 | struct udp_sock *up = udp_sk(sk); | |
811 | ||
812 | if (up->pending) { | |
813 | up->len = 0; | |
814 | up->pending = 0; | |
815 | ip_flush_pending_frames(sk); | |
816 | } | |
817 | } | |
36d926b9 | 818 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
819 | |
820 | /** | |
f6b9664f | 821 | * udp4_hwcsum - handle outgoing HW checksumming |
db8dac20 DM |
822 | * @skb: sk_buff containing the filled-in UDP header |
823 | * (checksum field must be zeroed out) | |
f6b9664f HX |
824 | * @src: source IP address |
825 | * @dst: destination IP address | |
db8dac20 | 826 | */ |
c26bf4a5 | 827 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) |
db8dac20 | 828 | { |
db8dac20 | 829 | struct udphdr *uh = udp_hdr(skb); |
f6b9664f HX |
830 | int offset = skb_transport_offset(skb); |
831 | int len = skb->len - offset; | |
832 | int hlen = len; | |
db8dac20 DM |
833 | __wsum csum = 0; |
834 | ||
ebbe495f | 835 | if (!skb_has_frag_list(skb)) { |
db8dac20 DM |
836 | /* |
837 | * Only one fragment on the socket. | |
838 | */ | |
839 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
840 | skb->csum_offset = offsetof(struct udphdr, check); | |
f6b9664f HX |
841 | uh->check = ~csum_tcpudp_magic(src, dst, len, |
842 | IPPROTO_UDP, 0); | |
db8dac20 | 843 | } else { |
ebbe495f WC |
844 | struct sk_buff *frags; |
845 | ||
db8dac20 DM |
846 | /* |
847 | * HW-checksum won't work as there are two or more | |
848 | * fragments on the socket so that all csums of sk_buffs | |
849 | * should be together | |
850 | */ | |
ebbe495f | 851 | skb_walk_frags(skb, frags) { |
f6b9664f HX |
852 | csum = csum_add(csum, frags->csum); |
853 | hlen -= frags->len; | |
ebbe495f | 854 | } |
db8dac20 | 855 | |
f6b9664f | 856 | csum = skb_checksum(skb, offset, hlen, csum); |
db8dac20 DM |
857 | skb->ip_summed = CHECKSUM_NONE; |
858 | ||
db8dac20 DM |
859 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); |
860 | if (uh->check == 0) | |
861 | uh->check = CSUM_MANGLED_0; | |
862 | } | |
863 | } | |
c26bf4a5 | 864 | EXPORT_SYMBOL_GPL(udp4_hwcsum); |
db8dac20 | 865 | |
af5fcba7 TH |
866 | /* Function to set UDP checksum for an IPv4 UDP packet. This is intended |
867 | * for the simple case like when setting the checksum for a UDP tunnel. | |
868 | */ | |
869 | void udp_set_csum(bool nocheck, struct sk_buff *skb, | |
870 | __be32 saddr, __be32 daddr, int len) | |
871 | { | |
872 | struct udphdr *uh = udp_hdr(skb); | |
873 | ||
179bc67f | 874 | if (nocheck) { |
af5fcba7 | 875 | uh->check = 0; |
179bc67f | 876 | } else if (skb_is_gso(skb)) { |
af5fcba7 | 877 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); |
179bc67f EC |
878 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
879 | uh->check = 0; | |
880 | uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); | |
881 | if (uh->check == 0) | |
882 | uh->check = CSUM_MANGLED_0; | |
d75f1306 | 883 | } else { |
af5fcba7 TH |
884 | skb->ip_summed = CHECKSUM_PARTIAL; |
885 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
886 | skb->csum_offset = offsetof(struct udphdr, check); | |
887 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); | |
af5fcba7 TH |
888 | } |
889 | } | |
890 | EXPORT_SYMBOL(udp_set_csum); | |
891 | ||
bec1f6f6 WB |
892 | static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, |
893 | struct inet_cork *cork) | |
db8dac20 | 894 | { |
f6b9664f | 895 | struct sock *sk = skb->sk; |
db8dac20 | 896 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 | 897 | struct udphdr *uh; |
cffb8f61 | 898 | int err; |
db8dac20 | 899 | int is_udplite = IS_UDPLITE(sk); |
f6b9664f HX |
900 | int offset = skb_transport_offset(skb); |
901 | int len = skb->len - offset; | |
4094871d | 902 | int datalen = len - sizeof(*uh); |
db8dac20 DM |
903 | __wsum csum = 0; |
904 | ||
db8dac20 DM |
905 | /* |
906 | * Create a UDP header | |
907 | */ | |
908 | uh = udp_hdr(skb); | |
f6b9664f | 909 | uh->source = inet->inet_sport; |
79ab0531 | 910 | uh->dest = fl4->fl4_dport; |
f6b9664f | 911 | uh->len = htons(len); |
db8dac20 DM |
912 | uh->check = 0; |
913 | ||
bec1f6f6 WB |
914 | if (cork->gso_size) { |
915 | const int hlen = skb_network_header_len(skb) + | |
916 | sizeof(struct udphdr); | |
917 | ||
0f149c9f WB |
918 | if (hlen + cork->gso_size > cork->fragsize) { |
919 | kfree_skb(skb); | |
bec1f6f6 | 920 | return -EINVAL; |
0f149c9f | 921 | } |
158390e4 | 922 | if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { |
0f149c9f | 923 | kfree_skb(skb); |
bec1f6f6 | 924 | return -EINVAL; |
0f149c9f WB |
925 | } |
926 | if (sk->sk_no_check_tx) { | |
927 | kfree_skb(skb); | |
a8c744a8 | 928 | return -EINVAL; |
0f149c9f | 929 | } |
ff06342c | 930 | if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || |
0f149c9f WB |
931 | dst_xfrm(skb_dst(skb))) { |
932 | kfree_skb(skb); | |
bec1f6f6 | 933 | return -EIO; |
0f149c9f | 934 | } |
bec1f6f6 | 935 | |
4094871d JH |
936 | if (datalen > cork->gso_size) { |
937 | skb_shinfo(skb)->gso_size = cork->gso_size; | |
938 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; | |
939 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, | |
940 | cork->gso_size); | |
941 | } | |
a8c744a8 | 942 | goto csum_partial; |
bec1f6f6 WB |
943 | } |
944 | ||
db8dac20 | 945 | if (is_udplite) /* UDP-Lite */ |
f6b9664f | 946 | csum = udplite_csum(skb); |
db8dac20 | 947 | |
ab2fb7e3 | 948 | else if (sk->sk_no_check_tx) { /* UDP csum off */ |
db8dac20 DM |
949 | |
950 | skb->ip_summed = CHECKSUM_NONE; | |
951 | goto send; | |
952 | ||
953 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
a8c744a8 | 954 | csum_partial: |
db8dac20 | 955 | |
79ab0531 | 956 | udp4_hwcsum(skb, fl4->saddr, fl4->daddr); |
db8dac20 DM |
957 | goto send; |
958 | ||
f6b9664f HX |
959 | } else |
960 | csum = udp_csum(skb); | |
db8dac20 DM |
961 | |
962 | /* add protocol-dependent pseudo-header */ | |
79ab0531 | 963 | uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, |
c482c568 | 964 | sk->sk_protocol, csum); |
db8dac20 DM |
965 | if (uh->check == 0) |
966 | uh->check = CSUM_MANGLED_0; | |
967 | ||
968 | send: | |
b5ec8eea | 969 | err = ip_send_skb(sock_net(sk), skb); |
6ce9e7b5 ED |
970 | if (err) { |
971 | if (err == -ENOBUFS && !inet->recverr) { | |
6aef70a8 ED |
972 | UDP_INC_STATS(sock_net(sk), |
973 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
6ce9e7b5 ED |
974 | err = 0; |
975 | } | |
976 | } else | |
6aef70a8 ED |
977 | UDP_INC_STATS(sock_net(sk), |
978 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
f6b9664f HX |
979 | return err; |
980 | } | |
981 | ||
982 | /* | |
983 | * Push out all pending data as one UDP datagram. Socket is locked. | |
984 | */ | |
8822b64a | 985 | int udp_push_pending_frames(struct sock *sk) |
f6b9664f HX |
986 | { |
987 | struct udp_sock *up = udp_sk(sk); | |
988 | struct inet_sock *inet = inet_sk(sk); | |
b6f21b26 | 989 | struct flowi4 *fl4 = &inet->cork.fl.u.ip4; |
f6b9664f HX |
990 | struct sk_buff *skb; |
991 | int err = 0; | |
992 | ||
77968b78 | 993 | skb = ip_finish_skb(sk, fl4); |
f6b9664f HX |
994 | if (!skb) |
995 | goto out; | |
996 | ||
bec1f6f6 | 997 | err = udp_send_skb(skb, fl4, &inet->cork.base); |
f6b9664f | 998 | |
db8dac20 DM |
999 | out: |
1000 | up->len = 0; | |
1001 | up->pending = 0; | |
db8dac20 DM |
1002 | return err; |
1003 | } | |
8822b64a | 1004 | EXPORT_SYMBOL(udp_push_pending_frames); |
db8dac20 | 1005 | |
2e8de857 WB |
1006 | static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) |
1007 | { | |
1008 | switch (cmsg->cmsg_type) { | |
1009 | case UDP_SEGMENT: | |
1010 | if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) | |
1011 | return -EINVAL; | |
1012 | *gso_size = *(__u16 *)CMSG_DATA(cmsg); | |
1013 | return 0; | |
1014 | default: | |
1015 | return -EINVAL; | |
1016 | } | |
1017 | } | |
1018 | ||
1019 | int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) | |
1020 | { | |
1021 | struct cmsghdr *cmsg; | |
1022 | bool need_ip = false; | |
1023 | int err; | |
1024 | ||
1025 | for_each_cmsghdr(cmsg, msg) { | |
1026 | if (!CMSG_OK(msg, cmsg)) | |
1027 | return -EINVAL; | |
1028 | ||
1029 | if (cmsg->cmsg_level != SOL_UDP) { | |
1030 | need_ip = true; | |
1031 | continue; | |
1032 | } | |
1033 | ||
1034 | err = __udp_cmsg_send(cmsg, gso_size); | |
1035 | if (err) | |
1036 | return err; | |
1037 | } | |
1038 | ||
1039 | return need_ip; | |
1040 | } | |
1041 | EXPORT_SYMBOL_GPL(udp_cmsg_send); | |
1042 | ||
1b784140 | 1043 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
db8dac20 DM |
1044 | { |
1045 | struct inet_sock *inet = inet_sk(sk); | |
1046 | struct udp_sock *up = udp_sk(sk); | |
1cedee13 | 1047 | DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); |
e474995f | 1048 | struct flowi4 fl4_stack; |
b6f21b26 | 1049 | struct flowi4 *fl4; |
db8dac20 DM |
1050 | int ulen = len; |
1051 | struct ipcm_cookie ipc; | |
1052 | struct rtable *rt = NULL; | |
1053 | int free = 0; | |
1054 | int connected = 0; | |
1055 | __be32 daddr, faddr, saddr; | |
1056 | __be16 dport; | |
1057 | u8 tos; | |
1058 | int err, is_udplite = IS_UDPLITE(sk); | |
a9f59707 | 1059 | int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; |
db8dac20 | 1060 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); |
903ab86d | 1061 | struct sk_buff *skb; |
f6d8bd05 | 1062 | struct ip_options_data opt_copy; |
db8dac20 DM |
1063 | |
1064 | if (len > 0xFFFF) | |
1065 | return -EMSGSIZE; | |
1066 | ||
1067 | /* | |
1068 | * Check the flags. | |
1069 | */ | |
1070 | ||
c482c568 | 1071 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
1072 | return -EOPNOTSUPP; |
1073 | ||
903ab86d HX |
1074 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
1075 | ||
f5fca608 | 1076 | fl4 = &inet->cork.fl.u.ip4; |
db8dac20 DM |
1077 | if (up->pending) { |
1078 | /* | |
1079 | * There are pending frames. | |
1080 | * The socket lock must be held while it's corked. | |
1081 | */ | |
1082 | lock_sock(sk); | |
1083 | if (likely(up->pending)) { | |
1084 | if (unlikely(up->pending != AF_INET)) { | |
1085 | release_sock(sk); | |
1086 | return -EINVAL; | |
1087 | } | |
1088 | goto do_append_data; | |
1089 | } | |
1090 | release_sock(sk); | |
1091 | } | |
1092 | ulen += sizeof(struct udphdr); | |
1093 | ||
1094 | /* | |
1095 | * Get and verify the address. | |
1096 | */ | |
1cedee13 | 1097 | if (usin) { |
db8dac20 DM |
1098 | if (msg->msg_namelen < sizeof(*usin)) |
1099 | return -EINVAL; | |
1100 | if (usin->sin_family != AF_INET) { | |
1101 | if (usin->sin_family != AF_UNSPEC) | |
1102 | return -EAFNOSUPPORT; | |
1103 | } | |
1104 | ||
1105 | daddr = usin->sin_addr.s_addr; | |
1106 | dport = usin->sin_port; | |
1107 | if (dport == 0) | |
1108 | return -EINVAL; | |
1109 | } else { | |
1110 | if (sk->sk_state != TCP_ESTABLISHED) | |
1111 | return -EDESTADDRREQ; | |
c720c7e8 ED |
1112 | daddr = inet->inet_daddr; |
1113 | dport = inet->inet_dport; | |
db8dac20 DM |
1114 | /* Open fast path for connected socket. |
1115 | Route will not be used, if at least one option is set. | |
1116 | */ | |
1117 | connected = 1; | |
1118 | } | |
db8dac20 | 1119 | |
35178206 | 1120 | ipcm_init_sk(&ipc, inet); |
18a419ba | 1121 | ipc.gso_size = READ_ONCE(up->gso_size); |
bf84a010 | 1122 | |
db8dac20 | 1123 | if (msg->msg_controllen) { |
2e8de857 WB |
1124 | err = udp_cmsg_send(sk, msg, &ipc.gso_size); |
1125 | if (err > 0) | |
1126 | err = ip_cmsg_send(sk, msg, &ipc, | |
1127 | sk->sk_family == AF_INET6); | |
1128 | if (unlikely(err < 0)) { | |
91948309 | 1129 | kfree(ipc.opt); |
db8dac20 | 1130 | return err; |
91948309 | 1131 | } |
db8dac20 DM |
1132 | if (ipc.opt) |
1133 | free = 1; | |
1134 | connected = 0; | |
1135 | } | |
f6d8bd05 ED |
1136 | if (!ipc.opt) { |
1137 | struct ip_options_rcu *inet_opt; | |
1138 | ||
1139 | rcu_read_lock(); | |
1140 | inet_opt = rcu_dereference(inet->inet_opt); | |
1141 | if (inet_opt) { | |
1142 | memcpy(&opt_copy, inet_opt, | |
1143 | sizeof(*inet_opt) + inet_opt->opt.optlen); | |
1144 | ipc.opt = &opt_copy.opt; | |
1145 | } | |
1146 | rcu_read_unlock(); | |
1147 | } | |
db8dac20 | 1148 | |
6fc88c35 | 1149 | if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { |
1cedee13 AI |
1150 | err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, |
1151 | (struct sockaddr *)usin, &ipc.addr); | |
1152 | if (err) | |
1153 | goto out_free; | |
1154 | if (usin) { | |
1155 | if (usin->sin_port == 0) { | |
1156 | /* BPF program set invalid port. Reject it. */ | |
1157 | err = -EINVAL; | |
1158 | goto out_free; | |
1159 | } | |
1160 | daddr = usin->sin_addr.s_addr; | |
1161 | dport = usin->sin_port; | |
1162 | } | |
1163 | } | |
1164 | ||
db8dac20 DM |
1165 | saddr = ipc.addr; |
1166 | ipc.addr = faddr = daddr; | |
1167 | ||
f6d8bd05 | 1168 | if (ipc.opt && ipc.opt->opt.srr) { |
1b97013b AI |
1169 | if (!daddr) { |
1170 | err = -EINVAL; | |
1171 | goto out_free; | |
1172 | } | |
f6d8bd05 | 1173 | faddr = ipc.opt->opt.faddr; |
db8dac20 DM |
1174 | connected = 0; |
1175 | } | |
aa661581 | 1176 | tos = get_rttos(&ipc, inet); |
db8dac20 DM |
1177 | if (sock_flag(sk, SOCK_LOCALROUTE) || |
1178 | (msg->msg_flags & MSG_DONTROUTE) || | |
f6d8bd05 | 1179 | (ipc.opt && ipc.opt->opt.is_strictroute)) { |
db8dac20 DM |
1180 | tos |= RTO_ONLINK; |
1181 | connected = 0; | |
1182 | } | |
1183 | ||
1184 | if (ipv4_is_multicast(daddr)) { | |
854da991 | 1185 | if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) |
db8dac20 DM |
1186 | ipc.oif = inet->mc_index; |
1187 | if (!saddr) | |
1188 | saddr = inet->mc_addr; | |
1189 | connected = 0; | |
9515a2e0 | 1190 | } else if (!ipc.oif) { |
76e21053 | 1191 | ipc.oif = inet->uc_index; |
9515a2e0 DA |
1192 | } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { |
1193 | /* oif is set, packet is to local broadcast and | |
2bdcc73c | 1194 | * uc_index is set. oif is most likely set |
9515a2e0 DA |
1195 | * by sk_bound_dev_if. If uc_index != oif check if the |
1196 | * oif is an L3 master and uc_index is an L3 slave. | |
1197 | * If so, we want to allow the send using the uc_index. | |
1198 | */ | |
1199 | if (ipc.oif != inet->uc_index && | |
1200 | ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), | |
1201 | inet->uc_index)) { | |
1202 | ipc.oif = inet->uc_index; | |
1203 | } | |
1204 | } | |
db8dac20 DM |
1205 | |
1206 | if (connected) | |
c482c568 | 1207 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 | 1208 | |
51456b29 | 1209 | if (!rt) { |
84a3aa00 | 1210 | struct net *net = sock_net(sk); |
9a24abfa | 1211 | __u8 flow_flags = inet_sk_flowi_flags(sk); |
84a3aa00 | 1212 | |
e474995f | 1213 | fl4 = &fl4_stack; |
9a24abfa | 1214 | |
c6af0c22 | 1215 | flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, |
c0951cbc | 1216 | RT_SCOPE_UNIVERSE, sk->sk_protocol, |
9a24abfa | 1217 | flow_flags, |
e2d118a1 LC |
1218 | faddr, saddr, dport, inet->inet_sport, |
1219 | sk->sk_uid); | |
c0951cbc | 1220 | |
3df98d79 | 1221 | security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); |
e474995f | 1222 | rt = ip_route_output_flow(net, fl4, sk); |
b23dd4fe DM |
1223 | if (IS_ERR(rt)) { |
1224 | err = PTR_ERR(rt); | |
06dc94b1 | 1225 | rt = NULL; |
db8dac20 | 1226 | if (err == -ENETUNREACH) |
f1d8cba6 | 1227 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
1228 | goto out; |
1229 | } | |
1230 | ||
1231 | err = -EACCES; | |
1232 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
1233 | !sock_flag(sk, SOCK_BROADCAST)) | |
1234 | goto out; | |
1235 | if (connected) | |
d8d1f30b | 1236 | sk_dst_set(sk, dst_clone(&rt->dst)); |
db8dac20 DM |
1237 | } |
1238 | ||
1239 | if (msg->msg_flags&MSG_CONFIRM) | |
1240 | goto do_confirm; | |
1241 | back_from_confirm: | |
1242 | ||
e474995f | 1243 | saddr = fl4->saddr; |
db8dac20 | 1244 | if (!ipc.addr) |
e474995f | 1245 | daddr = ipc.addr = fl4->daddr; |
db8dac20 | 1246 | |
903ab86d HX |
1247 | /* Lockless fast path for the non-corking case. */ |
1248 | if (!corkreq) { | |
1cd7884d WB |
1249 | struct inet_cork cork; |
1250 | ||
f69e6d13 | 1251 | skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, |
903ab86d | 1252 | sizeof(struct udphdr), &ipc, &rt, |
1cd7884d | 1253 | &cork, msg->msg_flags); |
903ab86d | 1254 | err = PTR_ERR(skb); |
50c3a487 | 1255 | if (!IS_ERR_OR_NULL(skb)) |
bec1f6f6 | 1256 | err = udp_send_skb(skb, fl4, &cork); |
903ab86d HX |
1257 | goto out; |
1258 | } | |
1259 | ||
db8dac20 DM |
1260 | lock_sock(sk); |
1261 | if (unlikely(up->pending)) { | |
1262 | /* The socket is already corked while preparing it. */ | |
1263 | /* ... which is an evident application bug. --ANK */ | |
1264 | release_sock(sk); | |
1265 | ||
197df02c | 1266 | net_dbg_ratelimited("socket already corked\n"); |
db8dac20 DM |
1267 | err = -EINVAL; |
1268 | goto out; | |
1269 | } | |
1270 | /* | |
1271 | * Now cork the socket to pend data. | |
1272 | */ | |
b6f21b26 DM |
1273 | fl4 = &inet->cork.fl.u.ip4; |
1274 | fl4->daddr = daddr; | |
1275 | fl4->saddr = saddr; | |
9cce96df DM |
1276 | fl4->fl4_dport = dport; |
1277 | fl4->fl4_sport = inet->inet_sport; | |
db8dac20 DM |
1278 | up->pending = AF_INET; |
1279 | ||
1280 | do_append_data: | |
1281 | up->len += ulen; | |
f69e6d13 | 1282 | err = ip_append_data(sk, fl4, getfrag, msg, ulen, |
f5fca608 DM |
1283 | sizeof(struct udphdr), &ipc, &rt, |
1284 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); | |
db8dac20 DM |
1285 | if (err) |
1286 | udp_flush_pending_frames(sk); | |
1287 | else if (!corkreq) | |
1288 | err = udp_push_pending_frames(sk); | |
1289 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1290 | up->pending = 0; | |
1291 | release_sock(sk); | |
1292 | ||
1293 | out: | |
1294 | ip_rt_put(rt); | |
1b97013b | 1295 | out_free: |
db8dac20 DM |
1296 | if (free) |
1297 | kfree(ipc.opt); | |
1298 | if (!err) | |
1299 | return len; | |
1300 | /* | |
1301 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1302 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1303 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1304 | * things). We could add another new stat but at least for now that | |
1305 | * seems like overkill. | |
1306 | */ | |
1307 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
6aef70a8 ED |
1308 | UDP_INC_STATS(sock_net(sk), |
1309 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
1310 | } |
1311 | return err; | |
1312 | ||
1313 | do_confirm: | |
0dec879f JA |
1314 | if (msg->msg_flags & MSG_PROBE) |
1315 | dst_confirm_neigh(&rt->dst, &fl4->daddr); | |
db8dac20 DM |
1316 | if (!(msg->msg_flags&MSG_PROBE) || len) |
1317 | goto back_from_confirm; | |
1318 | err = 0; | |
1319 | goto out; | |
1320 | } | |
c482c568 | 1321 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
1322 | |
1323 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
1324 | size_t size, int flags) | |
1325 | { | |
f5fca608 | 1326 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
1327 | struct udp_sock *up = udp_sk(sk); |
1328 | int ret; | |
1329 | ||
d3f7d56a SL |
1330 | if (flags & MSG_SENDPAGE_NOTLAST) |
1331 | flags |= MSG_MORE; | |
1332 | ||
db8dac20 DM |
1333 | if (!up->pending) { |
1334 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
1335 | ||
1336 | /* Call udp_sendmsg to specify destination address which | |
1337 | * sendpage interface can't pass. | |
1338 | * This will succeed only when the socket is connected. | |
1339 | */ | |
1b784140 | 1340 | ret = udp_sendmsg(sk, &msg, 0); |
db8dac20 DM |
1341 | if (ret < 0) |
1342 | return ret; | |
1343 | } | |
1344 | ||
1345 | lock_sock(sk); | |
1346 | ||
1347 | if (unlikely(!up->pending)) { | |
1348 | release_sock(sk); | |
1349 | ||
197df02c | 1350 | net_dbg_ratelimited("cork failed\n"); |
db8dac20 DM |
1351 | return -EINVAL; |
1352 | } | |
1353 | ||
f5fca608 DM |
1354 | ret = ip_append_page(sk, &inet->cork.fl.u.ip4, |
1355 | page, offset, size, flags); | |
db8dac20 DM |
1356 | if (ret == -EOPNOTSUPP) { |
1357 | release_sock(sk); | |
1358 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
1359 | size, flags); | |
1360 | } | |
1361 | if (ret < 0) { | |
1362 | udp_flush_pending_frames(sk); | |
1363 | goto out; | |
1364 | } | |
1365 | ||
1366 | up->len += size; | |
a9f59707 | 1367 | if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) |
db8dac20 DM |
1368 | ret = udp_push_pending_frames(sk); |
1369 | if (!ret) | |
1370 | ret = size; | |
1371 | out: | |
1372 | release_sock(sk); | |
1373 | return ret; | |
1374 | } | |
1375 | ||
dce4551c PA |
1376 | #define UDP_SKB_IS_STATELESS 0x80000000 |
1377 | ||
677bf08c FW |
1378 | /* all head states (dst, sk, nf conntrack) except skb extensions are |
1379 | * cleared by udp_rcv(). | |
1380 | * | |
1381 | * We need to preserve secpath, if present, to eventually process | |
1382 | * IP_CMSG_PASSSEC at recvmsg() time. | |
1383 | * | |
1384 | * Other extensions can be cleared. | |
1385 | */ | |
1386 | static bool udp_try_make_stateless(struct sk_buff *skb) | |
1387 | { | |
1388 | if (!skb_has_extensions(skb)) | |
1389 | return true; | |
1390 | ||
1391 | if (!secpath_exists(skb)) { | |
1392 | skb_ext_reset(skb); | |
1393 | return true; | |
1394 | } | |
1395 | ||
1396 | return false; | |
1397 | } | |
1398 | ||
b65ac446 PA |
1399 | static void udp_set_dev_scratch(struct sk_buff *skb) |
1400 | { | |
dce4551c | 1401 | struct udp_dev_scratch *scratch = udp_skb_scratch(skb); |
b65ac446 PA |
1402 | |
1403 | BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); | |
dce4551c PA |
1404 | scratch->_tsize_state = skb->truesize; |
1405 | #if BITS_PER_LONG == 64 | |
b65ac446 PA |
1406 | scratch->len = skb->len; |
1407 | scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); | |
1408 | scratch->is_linear = !skb_is_nonlinear(skb); | |
dce4551c | 1409 | #endif |
677bf08c | 1410 | if (udp_try_make_stateless(skb)) |
dce4551c | 1411 | scratch->_tsize_state |= UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1412 | } |
1413 | ||
a793183c ED |
1414 | static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) |
1415 | { | |
1416 | /* We come here after udp_lib_checksum_complete() returned 0. | |
1417 | * This means that __skb_checksum_complete() might have | |
1418 | * set skb->csum_valid to 1. | |
1419 | * On 64bit platforms, we can set csum_unnecessary | |
1420 | * to true, but only if the skb is not shared. | |
1421 | */ | |
1422 | #if BITS_PER_LONG == 64 | |
1423 | if (!skb_shared(skb)) | |
1424 | udp_skb_scratch(skb)->csum_unnecessary = true; | |
1425 | #endif | |
1426 | } | |
1427 | ||
b65ac446 PA |
1428 | static int udp_skb_truesize(struct sk_buff *skb) |
1429 | { | |
dce4551c | 1430 | return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1431 | } |
1432 | ||
dce4551c | 1433 | static bool udp_skb_has_head_state(struct sk_buff *skb) |
b65ac446 | 1434 | { |
dce4551c | 1435 | return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); |
b65ac446 | 1436 | } |
b65ac446 | 1437 | |
7c13f97f | 1438 | /* fully reclaim rmem/fwd memory allocated for skb */ |
6dfb4367 PA |
1439 | static void udp_rmem_release(struct sock *sk, int size, int partial, |
1440 | bool rx_queue_lock_held) | |
f970bd9e | 1441 | { |
6b229cf7 | 1442 | struct udp_sock *up = udp_sk(sk); |
2276f58a | 1443 | struct sk_buff_head *sk_queue; |
f970bd9e PA |
1444 | int amt; |
1445 | ||
6b229cf7 ED |
1446 | if (likely(partial)) { |
1447 | up->forward_deficit += size; | |
1448 | size = up->forward_deficit; | |
d39ca259 PA |
1449 | if (size < (sk->sk_rcvbuf >> 2) && |
1450 | !skb_queue_empty(&up->reader_queue)) | |
6b229cf7 ED |
1451 | return; |
1452 | } else { | |
1453 | size += up->forward_deficit; | |
1454 | } | |
1455 | up->forward_deficit = 0; | |
1456 | ||
6dfb4367 PA |
1457 | /* acquire the sk_receive_queue for fwd allocated memory scheduling, |
1458 | * if the called don't held it already | |
1459 | */ | |
2276f58a | 1460 | sk_queue = &sk->sk_receive_queue; |
6dfb4367 PA |
1461 | if (!rx_queue_lock_held) |
1462 | spin_lock(&sk_queue->lock); | |
1463 | ||
2276f58a | 1464 | |
f970bd9e | 1465 | sk->sk_forward_alloc += size; |
100fdd1f | 1466 | amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); |
f970bd9e | 1467 | sk->sk_forward_alloc -= amt; |
f970bd9e PA |
1468 | |
1469 | if (amt) | |
100fdd1f | 1470 | __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); |
02ab0d13 ED |
1471 | |
1472 | atomic_sub(size, &sk->sk_rmem_alloc); | |
2276f58a PA |
1473 | |
1474 | /* this can save us from acquiring the rx queue lock on next receive */ | |
1475 | skb_queue_splice_tail_init(sk_queue, &up->reader_queue); | |
1476 | ||
6dfb4367 PA |
1477 | if (!rx_queue_lock_held) |
1478 | spin_unlock(&sk_queue->lock); | |
f970bd9e PA |
1479 | } |
1480 | ||
2276f58a | 1481 | /* Note: called with reader_queue.lock held. |
c84d9490 ED |
1482 | * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch |
1483 | * This avoids a cache line miss while receive_queue lock is held. | |
1484 | * Look at __udp_enqueue_schedule_skb() to find where this copy is done. | |
1485 | */ | |
7c13f97f | 1486 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) |
f970bd9e | 1487 | { |
b65ac446 PA |
1488 | prefetch(&skb->data); |
1489 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); | |
f970bd9e | 1490 | } |
7c13f97f | 1491 | EXPORT_SYMBOL(udp_skb_destructor); |
f970bd9e | 1492 | |
6dfb4367 | 1493 | /* as above, but the caller held the rx queue lock, too */ |
64f5102d | 1494 | static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) |
6dfb4367 | 1495 | { |
b65ac446 PA |
1496 | prefetch(&skb->data); |
1497 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); | |
6dfb4367 PA |
1498 | } |
1499 | ||
4b272750 ED |
1500 | /* Idea of busylocks is to let producers grab an extra spinlock |
1501 | * to relieve pressure on the receive_queue spinlock shared by consumer. | |
1502 | * Under flood, this means that only one producer can be in line | |
1503 | * trying to acquire the receive_queue spinlock. | |
1504 | * These busylock can be allocated on a per cpu manner, instead of a | |
1505 | * per socket one (that would consume a cache line per socket) | |
1506 | */ | |
1507 | static int udp_busylocks_log __read_mostly; | |
1508 | static spinlock_t *udp_busylocks __read_mostly; | |
1509 | ||
1510 | static spinlock_t *busylock_acquire(void *ptr) | |
1511 | { | |
1512 | spinlock_t *busy; | |
1513 | ||
1514 | busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); | |
1515 | spin_lock(busy); | |
1516 | return busy; | |
1517 | } | |
1518 | ||
1519 | static void busylock_release(spinlock_t *busy) | |
1520 | { | |
1521 | if (busy) | |
1522 | spin_unlock(busy); | |
1523 | } | |
1524 | ||
f970bd9e PA |
1525 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) |
1526 | { | |
1527 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
1528 | int rmem, delta, amt, err = -ENOMEM; | |
4b272750 | 1529 | spinlock_t *busy = NULL; |
c8c8b127 | 1530 | int size; |
f970bd9e PA |
1531 | |
1532 | /* try to avoid the costly atomic add/sub pair when the receive | |
1533 | * queue is full; always allow at least a packet | |
1534 | */ | |
1535 | rmem = atomic_read(&sk->sk_rmem_alloc); | |
363dc73a | 1536 | if (rmem > sk->sk_rcvbuf) |
f970bd9e PA |
1537 | goto drop; |
1538 | ||
c8c8b127 ED |
1539 | /* Under mem pressure, it might be helpful to help udp_recvmsg() |
1540 | * having linear skbs : | |
1541 | * - Reduce memory overhead and thus increase receive queue capacity | |
1542 | * - Less cache line misses at copyout() time | |
1543 | * - Less work at consume_skb() (less alien page frag freeing) | |
1544 | */ | |
4b272750 | 1545 | if (rmem > (sk->sk_rcvbuf >> 1)) { |
c8c8b127 | 1546 | skb_condense(skb); |
4b272750 ED |
1547 | |
1548 | busy = busylock_acquire(sk); | |
1549 | } | |
c8c8b127 | 1550 | size = skb->truesize; |
b65ac446 | 1551 | udp_set_dev_scratch(skb); |
c8c8b127 | 1552 | |
f970bd9e PA |
1553 | /* we drop only if the receive buf is full and the receive |
1554 | * queue contains some other skb | |
1555 | */ | |
1556 | rmem = atomic_add_return(size, &sk->sk_rmem_alloc); | |
feed8a4f | 1557 | if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) |
f970bd9e PA |
1558 | goto uncharge_drop; |
1559 | ||
1560 | spin_lock(&list->lock); | |
1561 | if (size >= sk->sk_forward_alloc) { | |
1562 | amt = sk_mem_pages(size); | |
100fdd1f | 1563 | delta = amt << PAGE_SHIFT; |
f970bd9e PA |
1564 | if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { |
1565 | err = -ENOBUFS; | |
1566 | spin_unlock(&list->lock); | |
1567 | goto uncharge_drop; | |
1568 | } | |
1569 | ||
1570 | sk->sk_forward_alloc += delta; | |
1571 | } | |
1572 | ||
1573 | sk->sk_forward_alloc -= size; | |
1574 | ||
7c13f97f PA |
1575 | /* no need to setup a destructor, we will explicitly release the |
1576 | * forward allocated memory on dequeue | |
1577 | */ | |
f970bd9e PA |
1578 | sock_skb_set_dropcount(sk, skb); |
1579 | ||
1580 | __skb_queue_tail(list, skb); | |
1581 | spin_unlock(&list->lock); | |
1582 | ||
1583 | if (!sock_flag(sk, SOCK_DEAD)) | |
1584 | sk->sk_data_ready(sk); | |
1585 | ||
4b272750 | 1586 | busylock_release(busy); |
f970bd9e PA |
1587 | return 0; |
1588 | ||
1589 | uncharge_drop: | |
1590 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); | |
1591 | ||
1592 | drop: | |
1593 | atomic_inc(&sk->sk_drops); | |
4b272750 | 1594 | busylock_release(busy); |
f970bd9e PA |
1595 | return err; |
1596 | } | |
1597 | EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); | |
1598 | ||
c915fe13 | 1599 | void udp_destruct_sock(struct sock *sk) |
f970bd9e PA |
1600 | { |
1601 | /* reclaim completely the forward allocated memory */ | |
2276f58a | 1602 | struct udp_sock *up = udp_sk(sk); |
7c13f97f PA |
1603 | unsigned int total = 0; |
1604 | struct sk_buff *skb; | |
1605 | ||
2276f58a PA |
1606 | skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); |
1607 | while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { | |
7c13f97f PA |
1608 | total += skb->truesize; |
1609 | kfree_skb(skb); | |
1610 | } | |
6dfb4367 | 1611 | udp_rmem_release(sk, total, 0, true); |
7c13f97f | 1612 | |
f970bd9e PA |
1613 | inet_sock_destruct(sk); |
1614 | } | |
c915fe13 | 1615 | EXPORT_SYMBOL_GPL(udp_destruct_sock); |
f970bd9e PA |
1616 | |
1617 | int udp_init_sock(struct sock *sk) | |
1618 | { | |
2276f58a | 1619 | skb_queue_head_init(&udp_sk(sk)->reader_queue); |
f970bd9e PA |
1620 | sk->sk_destruct = udp_destruct_sock; |
1621 | return 0; | |
1622 | } | |
1623 | EXPORT_SYMBOL_GPL(udp_init_sock); | |
1624 | ||
1625 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) | |
1626 | { | |
1627 | if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { | |
1628 | bool slow = lock_sock_fast(sk); | |
1629 | ||
1630 | sk_peek_offset_bwd(sk, len); | |
1631 | unlock_sock_fast(sk, slow); | |
1632 | } | |
0a463c78 | 1633 | |
ca2c1418 PA |
1634 | if (!skb_unref(skb)) |
1635 | return; | |
1636 | ||
dce4551c PA |
1637 | /* In the more common cases we cleared the head states previously, |
1638 | * see __udp_queue_rcv_skb(). | |
0ddf3fb2 | 1639 | */ |
dce4551c | 1640 | if (unlikely(udp_skb_has_head_state(skb))) |
0ddf3fb2 | 1641 | skb_release_head_state(skb); |
ca2c1418 | 1642 | __consume_stateless_skb(skb); |
f970bd9e PA |
1643 | } |
1644 | EXPORT_SYMBOL_GPL(skb_consume_udp); | |
1645 | ||
2276f58a PA |
1646 | static struct sk_buff *__first_packet_length(struct sock *sk, |
1647 | struct sk_buff_head *rcvq, | |
1648 | int *total) | |
1649 | { | |
1650 | struct sk_buff *skb; | |
1651 | ||
9bd780f5 PA |
1652 | while ((skb = skb_peek(rcvq)) != NULL) { |
1653 | if (udp_lib_checksum_complete(skb)) { | |
1654 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1655 | IS_UDPLITE(sk)); | |
1656 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1657 | IS_UDPLITE(sk)); | |
1658 | atomic_inc(&sk->sk_drops); | |
1659 | __skb_unlink(skb, rcvq); | |
1660 | *total += skb->truesize; | |
1661 | kfree_skb(skb); | |
1662 | } else { | |
a793183c | 1663 | udp_skb_csum_unnecessary_set(skb); |
9bd780f5 PA |
1664 | break; |
1665 | } | |
2276f58a PA |
1666 | } |
1667 | return skb; | |
1668 | } | |
1669 | ||
85584672 ED |
1670 | /** |
1671 | * first_packet_length - return length of first packet in receive queue | |
1672 | * @sk: socket | |
1673 | * | |
1674 | * Drops all bad checksum frames, until a valid one is found. | |
e83c6744 | 1675 | * Returns the length of found skb, or -1 if none is found. |
85584672 | 1676 | */ |
e83c6744 | 1677 | static int first_packet_length(struct sock *sk) |
85584672 | 1678 | { |
2276f58a PA |
1679 | struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; |
1680 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
85584672 | 1681 | struct sk_buff *skb; |
7c13f97f | 1682 | int total = 0; |
e83c6744 | 1683 | int res; |
85584672 | 1684 | |
85584672 | 1685 | spin_lock_bh(&rcvq->lock); |
2276f58a | 1686 | skb = __first_packet_length(sk, rcvq, &total); |
137a0dbe | 1687 | if (!skb && !skb_queue_empty_lockless(sk_queue)) { |
2276f58a PA |
1688 | spin_lock(&sk_queue->lock); |
1689 | skb_queue_splice_tail_init(sk_queue, rcvq); | |
1690 | spin_unlock(&sk_queue->lock); | |
1691 | ||
1692 | skb = __first_packet_length(sk, rcvq, &total); | |
85584672 | 1693 | } |
e83c6744 | 1694 | res = skb ? skb->len : -1; |
7c13f97f | 1695 | if (total) |
6dfb4367 | 1696 | udp_rmem_release(sk, total, 1, false); |
85584672 | 1697 | spin_unlock_bh(&rcvq->lock); |
85584672 ED |
1698 | return res; |
1699 | } | |
1700 | ||
1da177e4 LT |
1701 | /* |
1702 | * IOCTL requests applicable to the UDP protocol | |
1703 | */ | |
e905a9ed | 1704 | |
1da177e4 LT |
1705 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
1706 | { | |
6516c655 SH |
1707 | switch (cmd) { |
1708 | case SIOCOUTQ: | |
1da177e4 | 1709 | { |
31e6d363 ED |
1710 | int amount = sk_wmem_alloc_get(sk); |
1711 | ||
6516c655 SH |
1712 | return put_user(amount, (int __user *)arg); |
1713 | } | |
1da177e4 | 1714 | |
6516c655 SH |
1715 | case SIOCINQ: |
1716 | { | |
e83c6744 | 1717 | int amount = max_t(int, 0, first_packet_length(sk)); |
6516c655 | 1718 | |
6516c655 SH |
1719 | return put_user(amount, (int __user *)arg); |
1720 | } | |
1da177e4 | 1721 | |
6516c655 SH |
1722 | default: |
1723 | return -ENOIOCTLCMD; | |
1da177e4 | 1724 | } |
6516c655 SH |
1725 | |
1726 | return 0; | |
1da177e4 | 1727 | } |
c482c568 | 1728 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 1729 | |
2276f58a | 1730 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, |
ec095263 | 1731 | int *off, int *err) |
2276f58a PA |
1732 | { |
1733 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
1734 | struct sk_buff_head *queue; | |
1735 | struct sk_buff *last; | |
1736 | long timeo; | |
1737 | int error; | |
1738 | ||
1739 | queue = &udp_sk(sk)->reader_queue; | |
2276f58a PA |
1740 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1741 | do { | |
1742 | struct sk_buff *skb; | |
1743 | ||
1744 | error = sock_error(sk); | |
1745 | if (error) | |
1746 | break; | |
1747 | ||
1748 | error = -EAGAIN; | |
2276f58a | 1749 | do { |
2276f58a | 1750 | spin_lock_bh(&queue->lock); |
e427cad6 PA |
1751 | skb = __skb_try_recv_from_queue(sk, queue, flags, off, |
1752 | err, &last); | |
2276f58a | 1753 | if (skb) { |
e427cad6 PA |
1754 | if (!(flags & MSG_PEEK)) |
1755 | udp_skb_destructor(sk, skb); | |
2276f58a | 1756 | spin_unlock_bh(&queue->lock); |
2276f58a PA |
1757 | return skb; |
1758 | } | |
1759 | ||
137a0dbe | 1760 | if (skb_queue_empty_lockless(sk_queue)) { |
2276f58a PA |
1761 | spin_unlock_bh(&queue->lock); |
1762 | goto busy_check; | |
1763 | } | |
1764 | ||
6dfb4367 PA |
1765 | /* refill the reader queue and walk it again |
1766 | * keep both queues locked to avoid re-acquiring | |
1767 | * the sk_receive_queue lock if fwd memory scheduling | |
1768 | * is needed. | |
1769 | */ | |
2276f58a PA |
1770 | spin_lock(&sk_queue->lock); |
1771 | skb_queue_splice_tail_init(sk_queue, queue); | |
2276f58a | 1772 | |
e427cad6 PA |
1773 | skb = __skb_try_recv_from_queue(sk, queue, flags, off, |
1774 | err, &last); | |
1775 | if (skb && !(flags & MSG_PEEK)) | |
1776 | udp_skb_dtor_locked(sk, skb); | |
6dfb4367 | 1777 | spin_unlock(&sk_queue->lock); |
2276f58a | 1778 | spin_unlock_bh(&queue->lock); |
de321ed3 | 1779 | if (skb) |
2276f58a | 1780 | return skb; |
2276f58a PA |
1781 | |
1782 | busy_check: | |
1783 | if (!sk_can_busy_loop(sk)) | |
1784 | break; | |
1785 | ||
1786 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
137a0dbe | 1787 | } while (!skb_queue_empty_lockless(sk_queue)); |
2276f58a PA |
1788 | |
1789 | /* sk_queue is empty, reader_queue may contain peeked packets */ | |
1790 | } while (timeo && | |
b50b0580 SD |
1791 | !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, |
1792 | &error, &timeo, | |
2276f58a PA |
1793 | (struct sk_buff *)sk_queue)); |
1794 | ||
1795 | *err = error; | |
1796 | return NULL; | |
1797 | } | |
7e823644 | 1798 | EXPORT_SYMBOL(__skb_recv_udp); |
2276f58a | 1799 | |
d7f57118 CW |
1800 | int udp_read_sock(struct sock *sk, read_descriptor_t *desc, |
1801 | sk_read_actor_t recv_actor) | |
1802 | { | |
1803 | int copied = 0; | |
1804 | ||
1805 | while (1) { | |
1806 | struct sk_buff *skb; | |
1807 | int err, used; | |
1808 | ||
ec095263 | 1809 | skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); |
d7f57118 CW |
1810 | if (!skb) |
1811 | return err; | |
099f896f CW |
1812 | |
1813 | if (udp_lib_checksum_complete(skb)) { | |
1814 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1815 | IS_UDPLITE(sk)); | |
1816 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1817 | IS_UDPLITE(sk)); | |
1818 | atomic_inc(&sk->sk_drops); | |
1819 | kfree_skb(skb); | |
1820 | continue; | |
1821 | } | |
1822 | ||
d7f57118 CW |
1823 | used = recv_actor(desc, skb, 0, skb->len); |
1824 | if (used <= 0) { | |
1825 | if (!copied) | |
1826 | copied = used; | |
e00a5c33 | 1827 | kfree_skb(skb); |
d7f57118 CW |
1828 | break; |
1829 | } else if (used <= skb->len) { | |
1830 | copied += used; | |
1831 | } | |
1832 | ||
e00a5c33 | 1833 | kfree_skb(skb); |
d7f57118 CW |
1834 | if (!desc->count) |
1835 | break; | |
1836 | } | |
1837 | ||
1838 | return copied; | |
1839 | } | |
1840 | EXPORT_SYMBOL(udp_read_sock); | |
1841 | ||
db8dac20 DM |
1842 | /* |
1843 | * This should be easy, if there is something there we | |
1844 | * return it, otherwise we block. | |
1845 | */ | |
1846 | ||
ec095263 OH |
1847 | int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, |
1848 | int *addr_len) | |
db8dac20 DM |
1849 | { |
1850 | struct inet_sock *inet = inet_sk(sk); | |
342dfc30 | 1851 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
db8dac20 | 1852 | struct sk_buff *skb; |
59c2cdae | 1853 | unsigned int ulen, copied; |
fd69c399 | 1854 | int off, err, peeking = flags & MSG_PEEK; |
db8dac20 | 1855 | int is_udplite = IS_UDPLITE(sk); |
197c949e | 1856 | bool checksum_valid = false; |
db8dac20 | 1857 | |
db8dac20 | 1858 | if (flags & MSG_ERRQUEUE) |
85fbaa75 | 1859 | return ip_recv_error(sk, msg, len, addr_len); |
db8dac20 DM |
1860 | |
1861 | try_again: | |
a0917e0b | 1862 | off = sk_peek_offset(sk, flags); |
ec095263 | 1863 | skb = __skb_recv_udp(sk, flags, &off, &err); |
db8dac20 | 1864 | if (!skb) |
627d2d6b | 1865 | return err; |
db8dac20 | 1866 | |
b65ac446 | 1867 | ulen = udp_skb_len(skb); |
59c2cdae | 1868 | copied = len; |
627d2d6b | 1869 | if (copied > ulen - off) |
1870 | copied = ulen - off; | |
59c2cdae | 1871 | else if (copied < ulen) |
db8dac20 DM |
1872 | msg->msg_flags |= MSG_TRUNC; |
1873 | ||
1874 | /* | |
1875 | * If checksum is needed at all, try to do it while copying the | |
1876 | * data. If the data is truncated, or if we only want a partial | |
1877 | * coverage checksum (UDP-Lite), do it before the copy. | |
1878 | */ | |
1879 | ||
d21dbdfe ED |
1880 | if (copied < ulen || peeking || |
1881 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
b65ac446 PA |
1882 | checksum_valid = udp_skb_csum_unnecessary(skb) || |
1883 | !__udp_lib_checksum_complete(skb); | |
197c949e | 1884 | if (!checksum_valid) |
db8dac20 DM |
1885 | goto csum_copy_err; |
1886 | } | |
1887 | ||
b65ac446 PA |
1888 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { |
1889 | if (udp_skb_is_linear(skb)) | |
1890 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
1891 | else | |
1892 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
1893 | } else { | |
627d2d6b | 1894 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); |
db8dac20 DM |
1895 | |
1896 | if (err == -EINVAL) | |
1897 | goto csum_copy_err; | |
1898 | } | |
1899 | ||
22911fc5 | 1900 | if (unlikely(err)) { |
fd69c399 | 1901 | if (!peeking) { |
979402b1 | 1902 | atomic_inc(&sk->sk_drops); |
6aef70a8 ED |
1903 | UDP_INC_STATS(sock_net(sk), |
1904 | UDP_MIB_INERRORS, is_udplite); | |
979402b1 | 1905 | } |
850cbadd | 1906 | kfree_skb(skb); |
627d2d6b | 1907 | return err; |
22911fc5 | 1908 | } |
db8dac20 | 1909 | |
fd69c399 | 1910 | if (!peeking) |
6aef70a8 ED |
1911 | UDP_INC_STATS(sock_net(sk), |
1912 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 | 1913 | |
6fd1d51c | 1914 | sock_recv_cmsgs(msg, sk, skb); |
db8dac20 DM |
1915 | |
1916 | /* Copy the address. */ | |
c482c568 | 1917 | if (sin) { |
db8dac20 DM |
1918 | sin->sin_family = AF_INET; |
1919 | sin->sin_port = udp_hdr(skb)->source; | |
1920 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
1921 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
bceaa902 | 1922 | *addr_len = sizeof(*sin); |
983695fa | 1923 | |
a9ed15da SF |
1924 | BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, |
1925 | (struct sockaddr *)sin); | |
db8dac20 | 1926 | } |
bcd1665e PA |
1927 | |
1928 | if (udp_sk(sk)->gro_enabled) | |
1929 | udp_cmsg_recv(msg, sk, skb); | |
1930 | ||
db8dac20 | 1931 | if (inet->cmsg_flags) |
ad959036 | 1932 | ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); |
db8dac20 | 1933 | |
59c2cdae | 1934 | err = copied; |
db8dac20 DM |
1935 | if (flags & MSG_TRUNC) |
1936 | err = ulen; | |
1937 | ||
850cbadd | 1938 | skb_consume_udp(sk, skb, peeking ? -err : err); |
db8dac20 DM |
1939 | return err; |
1940 | ||
1941 | csum_copy_err: | |
2276f58a PA |
1942 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, |
1943 | udp_skb_destructor)) { | |
6aef70a8 ED |
1944 | UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
1945 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
6a5dc9e5 | 1946 | } |
850cbadd | 1947 | kfree_skb(skb); |
db8dac20 | 1948 | |
beb39db5 ED |
1949 | /* starting over for a new packet, but check if we need to yield */ |
1950 | cond_resched(); | |
9cfaa8de | 1951 | msg->msg_flags &= ~MSG_TRUNC; |
db8dac20 DM |
1952 | goto try_again; |
1953 | } | |
1954 | ||
d74bad4e AI |
1955 | int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
1956 | { | |
1957 | /* This check is replicated from __ip4_datagram_connect() and | |
1958 | * intended to prevent BPF program called below from accessing bytes | |
1959 | * that are out of the bound specified by user in addr_len. | |
1960 | */ | |
1961 | if (addr_len < sizeof(struct sockaddr_in)) | |
1962 | return -EINVAL; | |
1963 | ||
1964 | return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); | |
1965 | } | |
1966 | EXPORT_SYMBOL(udp_pre_connect); | |
1967 | ||
286c72de | 1968 | int __udp_disconnect(struct sock *sk, int flags) |
1da177e4 LT |
1969 | { |
1970 | struct inet_sock *inet = inet_sk(sk); | |
1971 | /* | |
1972 | * 1003.1g - break association. | |
1973 | */ | |
e905a9ed | 1974 | |
1da177e4 | 1975 | sk->sk_state = TCP_CLOSE; |
c720c7e8 ED |
1976 | inet->inet_daddr = 0; |
1977 | inet->inet_dport = 0; | |
bdeab991 | 1978 | sock_rps_reset_rxhash(sk); |
1da177e4 | 1979 | sk->sk_bound_dev_if = 0; |
303d0403 | 1980 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { |
1da177e4 | 1981 | inet_reset_saddr(sk); |
303d0403 WB |
1982 | if (sk->sk_prot->rehash && |
1983 | (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
1984 | sk->sk_prot->rehash(sk); | |
1985 | } | |
1da177e4 LT |
1986 | |
1987 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1988 | sk->sk_prot->unhash(sk); | |
c720c7e8 | 1989 | inet->inet_sport = 0; |
1da177e4 LT |
1990 | } |
1991 | sk_dst_reset(sk); | |
1992 | return 0; | |
1993 | } | |
286c72de ED |
1994 | EXPORT_SYMBOL(__udp_disconnect); |
1995 | ||
1996 | int udp_disconnect(struct sock *sk, int flags) | |
1997 | { | |
1998 | lock_sock(sk); | |
1999 | __udp_disconnect(sk, flags); | |
2000 | release_sock(sk); | |
2001 | return 0; | |
2002 | } | |
c482c568 | 2003 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 2004 | |
645ca708 ED |
2005 | void udp_lib_unhash(struct sock *sk) |
2006 | { | |
723b4610 ED |
2007 | if (sk_hashed(sk)) { |
2008 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
512615b6 ED |
2009 | struct udp_hslot *hslot, *hslot2; |
2010 | ||
2011 | hslot = udp_hashslot(udptable, sock_net(sk), | |
2012 | udp_sk(sk)->udp_port_hash); | |
2013 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
645ca708 | 2014 | |
723b4610 | 2015 | spin_lock_bh(&hslot->lock); |
e32ea7e7 CG |
2016 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
2017 | reuseport_detach_sock(sk); | |
ca065d0c | 2018 | if (sk_del_node_init_rcu(sk)) { |
fdcc8aa9 | 2019 | hslot->count--; |
c720c7e8 | 2020 | inet_sk(sk)->inet_num = 0; |
723b4610 | 2021 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
512615b6 ED |
2022 | |
2023 | spin_lock(&hslot2->lock); | |
ca065d0c | 2024 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
512615b6 ED |
2025 | hslot2->count--; |
2026 | spin_unlock(&hslot2->lock); | |
723b4610 ED |
2027 | } |
2028 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 2029 | } |
645ca708 ED |
2030 | } |
2031 | EXPORT_SYMBOL(udp_lib_unhash); | |
2032 | ||
719f8358 ED |
2033 | /* |
2034 | * inet_rcv_saddr was changed, we must rehash secondary hash | |
2035 | */ | |
2036 | void udp_lib_rehash(struct sock *sk, u16 newhash) | |
2037 | { | |
2038 | if (sk_hashed(sk)) { | |
2039 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
2040 | struct udp_hslot *hslot, *hslot2, *nhslot2; | |
2041 | ||
2042 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
2043 | nhslot2 = udp_hashslot2(udptable, newhash); | |
2044 | udp_sk(sk)->udp_portaddr_hash = newhash; | |
e32ea7e7 CG |
2045 | |
2046 | if (hslot2 != nhslot2 || | |
2047 | rcu_access_pointer(sk->sk_reuseport_cb)) { | |
719f8358 ED |
2048 | hslot = udp_hashslot(udptable, sock_net(sk), |
2049 | udp_sk(sk)->udp_port_hash); | |
2050 | /* we must lock primary chain too */ | |
2051 | spin_lock_bh(&hslot->lock); | |
e32ea7e7 CG |
2052 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
2053 | reuseport_detach_sock(sk); | |
2054 | ||
2055 | if (hslot2 != nhslot2) { | |
2056 | spin_lock(&hslot2->lock); | |
ca065d0c | 2057 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
e32ea7e7 CG |
2058 | hslot2->count--; |
2059 | spin_unlock(&hslot2->lock); | |
2060 | ||
2061 | spin_lock(&nhslot2->lock); | |
ca065d0c | 2062 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
e32ea7e7 CG |
2063 | &nhslot2->head); |
2064 | nhslot2->count++; | |
2065 | spin_unlock(&nhslot2->lock); | |
2066 | } | |
719f8358 ED |
2067 | |
2068 | spin_unlock_bh(&hslot->lock); | |
2069 | } | |
2070 | } | |
2071 | } | |
2072 | EXPORT_SYMBOL(udp_lib_rehash); | |
2073 | ||
8f6b5392 | 2074 | void udp_v4_rehash(struct sock *sk) |
719f8358 | 2075 | { |
f0b1e64c | 2076 | u16 new_hash = ipv4_portaddr_hash(sock_net(sk), |
719f8358 ED |
2077 | inet_sk(sk)->inet_rcv_saddr, |
2078 | inet_sk(sk)->inet_num); | |
2079 | udp_lib_rehash(sk, new_hash); | |
2080 | } | |
2081 | ||
a3f96c47 | 2082 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
93821778 | 2083 | { |
fec5e652 | 2084 | int rc; |
766e9037 | 2085 | |
005ec974 | 2086 | if (inet_sk(sk)->inet_daddr) { |
bdeab991 | 2087 | sock_rps_save_rxhash(sk, skb); |
005ec974 | 2088 | sk_mark_napi_id(sk, skb); |
2c8c56e1 | 2089 | sk_incoming_cpu_update(sk); |
e68b6e50 ED |
2090 | } else { |
2091 | sk_mark_napi_id_once(sk, skb); | |
005ec974 | 2092 | } |
fec5e652 | 2093 | |
850cbadd | 2094 | rc = __udp_enqueue_schedule_skb(sk, skb); |
766e9037 ED |
2095 | if (rc < 0) { |
2096 | int is_udplite = IS_UDPLITE(sk); | |
08d4c037 | 2097 | int drop_reason; |
93821778 | 2098 | |
93821778 | 2099 | /* Note that an ENOMEM error is charged twice */ |
08d4c037 | 2100 | if (rc == -ENOMEM) { |
e61da9e2 | 2101 | UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
02c22347 | 2102 | is_udplite); |
08d4c037 MD |
2103 | drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; |
2104 | } else { | |
a3ce2b10 MD |
2105 | UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, |
2106 | is_udplite); | |
08d4c037 MD |
2107 | drop_reason = SKB_DROP_REASON_PROTO_MEM; |
2108 | } | |
e61da9e2 | 2109 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
08d4c037 | 2110 | kfree_skb_reason(skb, drop_reason); |
296f7ea7 | 2111 | trace_udp_fail_queue_rcv_skb(rc, sk); |
766e9037 | 2112 | return -1; |
93821778 HX |
2113 | } |
2114 | ||
2115 | return 0; | |
93821778 HX |
2116 | } |
2117 | ||
db8dac20 DM |
2118 | /* returns: |
2119 | * -1: error | |
2120 | * 0: success | |
2121 | * >0: "udp encap" protocol resubmission | |
2122 | * | |
2123 | * Note that in the success and error cases, the skb is assumed to | |
2124 | * have either been requeued or freed. | |
2125 | */ | |
cf329aa4 | 2126 | static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 | 2127 | { |
1379a92d | 2128 | int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; |
db8dac20 | 2129 | struct udp_sock *up = udp_sk(sk); |
db8dac20 DM |
2130 | int is_udplite = IS_UDPLITE(sk); |
2131 | ||
2132 | /* | |
2133 | * Charge it to the socket, dropping if the queue is full. | |
2134 | */ | |
1379a92d MD |
2135 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { |
2136 | drop_reason = SKB_DROP_REASON_XFRM_POLICY; | |
db8dac20 | 2137 | goto drop; |
1379a92d | 2138 | } |
895b5c9f | 2139 | nf_reset_ct(skb); |
db8dac20 | 2140 | |
88ab3108 | 2141 | if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { |
0ad92ad0 ED |
2142 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
2143 | ||
db8dac20 DM |
2144 | /* |
2145 | * This is an encapsulation socket so pass the skb to | |
2146 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
2147 | * fall through and pass this up the UDP socket. | |
2148 | * up->encap_rcv() returns the following value: | |
2149 | * =0 if skb was successfully passed to the encap | |
2150 | * handler or was discarded by it. | |
2151 | * >0 if skb should be passed on to UDP. | |
2152 | * <0 if skb should be resubmitted as proto -N | |
2153 | */ | |
2154 | ||
2155 | /* if we're overly short, let UDP handle it */ | |
6aa7de05 | 2156 | encap_rcv = READ_ONCE(up->encap_rcv); |
e5aed006 | 2157 | if (encap_rcv) { |
db8dac20 DM |
2158 | int ret; |
2159 | ||
0a80966b TH |
2160 | /* Verify checksum before giving to encap */ |
2161 | if (udp_lib_checksum_complete(skb)) | |
2162 | goto csum_error; | |
2163 | ||
0ad92ad0 | 2164 | ret = encap_rcv(sk, skb); |
db8dac20 | 2165 | if (ret <= 0) { |
02c22347 ED |
2166 | __UDP_INC_STATS(sock_net(sk), |
2167 | UDP_MIB_INDATAGRAMS, | |
2168 | is_udplite); | |
db8dac20 DM |
2169 | return -ret; |
2170 | } | |
2171 | } | |
2172 | ||
2173 | /* FALLTHROUGH -- it's a UDP Packet */ | |
2174 | } | |
2175 | ||
2176 | /* | |
2177 | * UDP-Lite specific tests, ignored on UDP sockets | |
2178 | */ | |
b0a42277 | 2179 | if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { |
db8dac20 DM |
2180 | |
2181 | /* | |
2182 | * MIB statistics other than incrementing the error count are | |
2183 | * disabled for the following two types of errors: these depend | |
2184 | * on the application settings, not on the functioning of the | |
2185 | * protocol stack as such. | |
2186 | * | |
2187 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
2188 | * way ... to ... at least let the receiving application block | |
2189 | * delivery of packets with coverage values less than a value | |
2190 | * provided by the application." | |
2191 | */ | |
2192 | if (up->pcrlen == 0) { /* full coverage was set */ | |
ba7a46f1 JP |
2193 | net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", |
2194 | UDP_SKB_CB(skb)->cscov, skb->len); | |
db8dac20 DM |
2195 | goto drop; |
2196 | } | |
2197 | /* The next case involves violating the min. coverage requested | |
2198 | * by the receiver. This is subtle: if receiver wants x and x is | |
2199 | * greater than the buffersize/MTU then receiver will complain | |
2200 | * that it wants x while sender emits packets of smaller size y. | |
2201 | * Therefore the above ...()->partial_cov statement is essential. | |
2202 | */ | |
2203 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
ba7a46f1 JP |
2204 | net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", |
2205 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
db8dac20 DM |
2206 | goto drop; |
2207 | } | |
2208 | } | |
2209 | ||
dd99e425 | 2210 | prefetch(&sk->sk_rmem_alloc); |
ce25d66a ED |
2211 | if (rcu_access_pointer(sk->sk_filter) && |
2212 | udp_lib_checksum_complete(skb)) | |
e6afc8ac | 2213 | goto csum_error; |
ce25d66a | 2214 | |
1379a92d MD |
2215 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { |
2216 | drop_reason = SKB_DROP_REASON_SOCKET_FILTER; | |
a6127697 | 2217 | goto drop; |
1379a92d | 2218 | } |
db8dac20 | 2219 | |
e6afc8ac | 2220 | udp_csum_pull_header(skb); |
db8dac20 | 2221 | |
fbf8866d | 2222 | ipv4_pktinfo_prepare(sk, skb); |
850cbadd | 2223 | return __udp_queue_rcv_skb(sk, skb); |
db8dac20 | 2224 | |
6a5dc9e5 | 2225 | csum_error: |
1379a92d | 2226 | drop_reason = SKB_DROP_REASON_UDP_CSUM; |
02c22347 | 2227 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
db8dac20 | 2228 | drop: |
02c22347 | 2229 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
8edf19c2 | 2230 | atomic_inc(&sk->sk_drops); |
1379a92d | 2231 | kfree_skb_reason(skb, drop_reason); |
db8dac20 DM |
2232 | return -1; |
2233 | } | |
2234 | ||
cf329aa4 PA |
2235 | static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
2236 | { | |
2237 | struct sk_buff *next, *segs; | |
2238 | int ret; | |
2239 | ||
2240 | if (likely(!udp_unexpected_gso(sk, skb))) | |
2241 | return udp_queue_rcv_one_skb(sk, skb); | |
2242 | ||
a08e7fd9 | 2243 | BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); |
cf329aa4 PA |
2244 | __skb_push(skb, -skb_mac_offset(skb)); |
2245 | segs = udp_rcv_segment(sk, skb, true); | |
1a186c14 | 2246 | skb_list_walk_safe(segs, skb, next) { |
cf329aa4 | 2247 | __skb_pull(skb, skb_transport_offset(skb)); |
000ac44d PA |
2248 | |
2249 | udp_post_segment_fix_csum(skb); | |
cf329aa4 PA |
2250 | ret = udp_queue_rcv_one_skb(sk, skb); |
2251 | if (ret > 0) | |
10c678bd | 2252 | ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); |
cf329aa4 PA |
2253 | } |
2254 | return 0; | |
2255 | } | |
2256 | ||
97502231 | 2257 | /* For TCP sockets, sk_rx_dst is protected by socket lock |
e47eb5df | 2258 | * For UDP, we use xchg() to guard against concurrent changes. |
97502231 | 2259 | */ |
64f0f5d1 | 2260 | bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) |
421b3885 | 2261 | { |
97502231 ED |
2262 | struct dst_entry *old; |
2263 | ||
d24406c8 | 2264 | if (dst_hold_safe(dst)) { |
8f905c0e | 2265 | old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); |
d24406c8 | 2266 | dst_release(old); |
64f0f5d1 | 2267 | return old != dst; |
d24406c8 | 2268 | } |
64f0f5d1 | 2269 | return false; |
421b3885 | 2270 | } |
c9f2c1ae | 2271 | EXPORT_SYMBOL(udp_sk_rx_dst_set); |
421b3885 | 2272 | |
db8dac20 DM |
2273 | /* |
2274 | * Multicasts and broadcasts go to each listener. | |
2275 | * | |
1240d137 | 2276 | * Note: called only from the BH handler context. |
db8dac20 | 2277 | */ |
e3163493 | 2278 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
2279 | struct udphdr *uh, |
2280 | __be32 saddr, __be32 daddr, | |
36cbb245 RJ |
2281 | struct udp_table *udptable, |
2282 | int proto) | |
db8dac20 | 2283 | { |
ca065d0c | 2284 | struct sock *sk, *first = NULL; |
5cf3d461 DH |
2285 | unsigned short hnum = ntohs(uh->dest); |
2286 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
2dc41cff | 2287 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); |
ca065d0c ED |
2288 | unsigned int offset = offsetof(typeof(*sk), sk_node); |
2289 | int dif = skb->dev->ifindex; | |
fb74c277 | 2290 | int sdif = inet_sdif(skb); |
ca065d0c ED |
2291 | struct hlist_node *node; |
2292 | struct sk_buff *nskb; | |
2dc41cff DH |
2293 | |
2294 | if (use_hash2) { | |
f0b1e64c | 2295 | hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & |
73e2d5e3 | 2296 | udptable->mask; |
f0b1e64c | 2297 | hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; |
2dc41cff | 2298 | start_lookup: |
73e2d5e3 | 2299 | hslot = &udptable->hash2[hash2]; |
2dc41cff DH |
2300 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); |
2301 | } | |
db8dac20 | 2302 | |
ca065d0c ED |
2303 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { |
2304 | if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, | |
fb74c277 | 2305 | uh->source, saddr, dif, sdif, hnum)) |
ca065d0c ED |
2306 | continue; |
2307 | ||
2308 | if (!first) { | |
2309 | first = sk; | |
2310 | continue; | |
1240d137 | 2311 | } |
ca065d0c | 2312 | nskb = skb_clone(skb, GFP_ATOMIC); |
1240d137 | 2313 | |
ca065d0c ED |
2314 | if (unlikely(!nskb)) { |
2315 | atomic_inc(&sk->sk_drops); | |
02c22347 ED |
2316 | __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, |
2317 | IS_UDPLITE(sk)); | |
2318 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, | |
2319 | IS_UDPLITE(sk)); | |
ca065d0c ED |
2320 | continue; |
2321 | } | |
2322 | if (udp_queue_rcv_skb(sk, nskb) > 0) | |
2323 | consume_skb(nskb); | |
2324 | } | |
1240d137 | 2325 | |
2dc41cff DH |
2326 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ |
2327 | if (use_hash2 && hash2 != hash2_any) { | |
2328 | hash2 = hash2_any; | |
2329 | goto start_lookup; | |
2330 | } | |
2331 | ||
ca065d0c ED |
2332 | if (first) { |
2333 | if (udp_queue_rcv_skb(first, skb) > 0) | |
2334 | consume_skb(skb); | |
1240d137 | 2335 | } else { |
ca065d0c | 2336 | kfree_skb(skb); |
02c22347 ED |
2337 | __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, |
2338 | proto == IPPROTO_UDPLITE); | |
1240d137 | 2339 | } |
db8dac20 DM |
2340 | return 0; |
2341 | } | |
2342 | ||
2343 | /* Initialize UDP checksum. If exited with zero value (success), | |
2344 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
666a3d6e | 2345 | * Otherwise, csum completion requires checksumming packet body, |
db8dac20 DM |
2346 | * including udp header and folding it to skb->csum. |
2347 | */ | |
2348 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
2349 | int proto) | |
2350 | { | |
db8dac20 DM |
2351 | int err; |
2352 | ||
2353 | UDP_SKB_CB(skb)->partial_cov = 0; | |
2354 | UDP_SKB_CB(skb)->cscov = skb->len; | |
2355 | ||
2356 | if (proto == IPPROTO_UDPLITE) { | |
2357 | err = udplite_checksum_init(skb, uh); | |
2358 | if (err) | |
2359 | return err; | |
15f35d49 AK |
2360 | |
2361 | if (UDP_SKB_CB(skb)->partial_cov) { | |
2362 | skb->csum = inet_compute_pseudo(skb, proto); | |
2363 | return 0; | |
2364 | } | |
db8dac20 DM |
2365 | } |
2366 | ||
b46d9f62 HFS |
2367 | /* Note, we are only interested in != 0 or == 0, thus the |
2368 | * force to int. | |
2369 | */ | |
db4f1be3 ST |
2370 | err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, |
2371 | inet_compute_pseudo); | |
2372 | if (err) | |
2373 | return err; | |
2374 | ||
2375 | if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { | |
2376 | /* If SW calculated the value, we know it's bad */ | |
2377 | if (skb->csum_complete_sw) | |
2378 | return 1; | |
2379 | ||
2380 | /* HW says the value is bad. Let's validate that. | |
2381 | * skb->csum is no longer the full packet checksum, | |
2382 | * so don't treat it as such. | |
2383 | */ | |
2384 | skb_checksum_complete_unset(skb); | |
2385 | } | |
2386 | ||
2387 | return 0; | |
db8dac20 DM |
2388 | } |
2389 | ||
2b5a9217 PA |
2390 | /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and |
2391 | * return code conversion for ip layer consumption | |
2392 | */ | |
2393 | static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, | |
2394 | struct udphdr *uh) | |
2395 | { | |
2396 | int ret; | |
2397 | ||
2398 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) | |
e4aa33ad | 2399 | skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); |
2b5a9217 PA |
2400 | |
2401 | ret = udp_queue_rcv_skb(sk, skb); | |
2402 | ||
2403 | /* a return value > 0 means to resubmit the input, but | |
2404 | * it wants the return to be -protocol, or 0 | |
2405 | */ | |
2406 | if (ret > 0) | |
2407 | return -ret; | |
2408 | return 0; | |
2409 | } | |
2410 | ||
db8dac20 DM |
2411 | /* |
2412 | * All we need to do is get the socket, and then do a checksum. | |
2413 | */ | |
2414 | ||
645ca708 | 2415 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
2416 | int proto) |
2417 | { | |
2418 | struct sock *sk; | |
7b5e56f9 | 2419 | struct udphdr *uh; |
db8dac20 | 2420 | unsigned short ulen; |
adf30907 | 2421 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 2422 | __be32 saddr, daddr; |
0283328e | 2423 | struct net *net = dev_net(skb->dev); |
71489e21 | 2424 | bool refcounted; |
1c7fab70 MD |
2425 | int drop_reason; |
2426 | ||
2427 | drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; | |
db8dac20 DM |
2428 | |
2429 | /* | |
2430 | * Validate the packet. | |
2431 | */ | |
2432 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
2433 | goto drop; /* No space for header. */ | |
2434 | ||
7b5e56f9 | 2435 | uh = udp_hdr(skb); |
db8dac20 | 2436 | ulen = ntohs(uh->len); |
ccc2d97c BM |
2437 | saddr = ip_hdr(skb)->saddr; |
2438 | daddr = ip_hdr(skb)->daddr; | |
2439 | ||
db8dac20 DM |
2440 | if (ulen > skb->len) |
2441 | goto short_packet; | |
2442 | ||
2443 | if (proto == IPPROTO_UDP) { | |
2444 | /* UDP validates ulen. */ | |
2445 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
2446 | goto short_packet; | |
2447 | uh = udp_hdr(skb); | |
2448 | } | |
2449 | ||
2450 | if (udp4_csum_init(skb, uh, proto)) | |
2451 | goto csum_error; | |
2452 | ||
71489e21 | 2453 | sk = skb_steal_sock(skb, &refcounted); |
8afdd99a | 2454 | if (sk) { |
97502231 | 2455 | struct dst_entry *dst = skb_dst(skb); |
421b3885 | 2456 | int ret; |
421b3885 | 2457 | |
8f905c0e | 2458 | if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) |
97502231 | 2459 | udp_sk_rx_dst_set(sk, dst); |
db8dac20 | 2460 | |
2b5a9217 | 2461 | ret = udp_unicast_rcv_skb(sk, skb, uh); |
71489e21 JS |
2462 | if (refcounted) |
2463 | sock_put(sk); | |
2b5a9217 | 2464 | return ret; |
421b3885 | 2465 | } |
db8dac20 | 2466 | |
c18450a5 FF |
2467 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
2468 | return __udp4_lib_mcast_deliver(net, skb, uh, | |
36cbb245 | 2469 | saddr, daddr, udptable, proto); |
c18450a5 FF |
2470 | |
2471 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
2b5a9217 PA |
2472 | if (sk) |
2473 | return udp_unicast_rcv_skb(sk, skb, uh); | |
db8dac20 DM |
2474 | |
2475 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2476 | goto drop; | |
895b5c9f | 2477 | nf_reset_ct(skb); |
db8dac20 DM |
2478 | |
2479 | /* No socket. Drop packet silently, if checksum is wrong */ | |
2480 | if (udp_lib_checksum_complete(skb)) | |
2481 | goto csum_error; | |
2482 | ||
1c7fab70 | 2483 | drop_reason = SKB_DROP_REASON_NO_SOCKET; |
02c22347 | 2484 | __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2485 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
2486 | ||
2487 | /* | |
2488 | * Hmm. We got an UDP packet to a port to which we | |
2489 | * don't wanna listen. Ignore it. | |
2490 | */ | |
1c7fab70 | 2491 | kfree_skb_reason(skb, drop_reason); |
db8dac20 DM |
2492 | return 0; |
2493 | ||
2494 | short_packet: | |
1c7fab70 | 2495 | drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; |
ba7a46f1 JP |
2496 | net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
2497 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2498 | &saddr, ntohs(uh->source), | |
2499 | ulen, skb->len, | |
2500 | &daddr, ntohs(uh->dest)); | |
db8dac20 DM |
2501 | goto drop; |
2502 | ||
2503 | csum_error: | |
2504 | /* | |
2505 | * RFC1122: OK. Discards the bad packet silently (as far as | |
2506 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
2507 | */ | |
1c7fab70 | 2508 | drop_reason = SKB_DROP_REASON_UDP_CSUM; |
ba7a46f1 JP |
2509 | net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
2510 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2511 | &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), | |
2512 | ulen); | |
02c22347 | 2513 | __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 | 2514 | drop: |
02c22347 | 2515 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
1c7fab70 | 2516 | kfree_skb_reason(skb, drop_reason); |
db8dac20 DM |
2517 | return 0; |
2518 | } | |
2519 | ||
421b3885 SB |
2520 | /* We can only early demux multicast if there is a single matching socket. |
2521 | * If more than one socket found returns NULL | |
2522 | */ | |
2523 | static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, | |
2524 | __be16 loc_port, __be32 loc_addr, | |
2525 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 2526 | int dif, int sdif) |
421b3885 SB |
2527 | { |
2528 | struct sock *sk, *result; | |
421b3885 | 2529 | unsigned short hnum = ntohs(loc_port); |
ca065d0c | 2530 | unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); |
421b3885 SB |
2531 | struct udp_hslot *hslot = &udp_table.hash[slot]; |
2532 | ||
63c6f81c ED |
2533 | /* Do not bother scanning a too big list */ |
2534 | if (hslot->count > 10) | |
2535 | return NULL; | |
2536 | ||
421b3885 | 2537 | result = NULL; |
ca065d0c ED |
2538 | sk_for_each_rcu(sk, &hslot->head) { |
2539 | if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, | |
fb74c277 | 2540 | rmt_port, rmt_addr, dif, sdif, hnum)) { |
ca065d0c ED |
2541 | if (result) |
2542 | return NULL; | |
421b3885 | 2543 | result = sk; |
421b3885 SB |
2544 | } |
2545 | } | |
ca065d0c | 2546 | |
421b3885 SB |
2547 | return result; |
2548 | } | |
2549 | ||
2550 | /* For unicast we should only early demux connected sockets or we can | |
2551 | * break forwarding setups. The chains here can be long so only check | |
2552 | * if the first socket is an exact match and if not move on. | |
2553 | */ | |
2554 | static struct sock *__udp4_lib_demux_lookup(struct net *net, | |
2555 | __be16 loc_port, __be32 loc_addr, | |
2556 | __be16 rmt_port, __be32 rmt_addr, | |
3fa6f616 | 2557 | int dif, int sdif) |
421b3885 | 2558 | { |
421b3885 | 2559 | unsigned short hnum = ntohs(loc_port); |
f0b1e64c | 2560 | unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); |
421b3885 SB |
2561 | unsigned int slot2 = hash2 & udp_table.mask; |
2562 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
c7228317 | 2563 | INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); |
421b3885 | 2564 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); |
ca065d0c | 2565 | struct sock *sk; |
421b3885 | 2566 | |
ca065d0c | 2567 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
eda090c3 | 2568 | if (inet_match(net, sk, acookie, ports, dif, sdif)) |
ca065d0c | 2569 | return sk; |
421b3885 SB |
2570 | /* Only check first socket in chain */ |
2571 | break; | |
2572 | } | |
ca065d0c | 2573 | return NULL; |
421b3885 SB |
2574 | } |
2575 | ||
7487449c | 2576 | int udp_v4_early_demux(struct sk_buff *skb) |
421b3885 | 2577 | { |
610438b7 | 2578 | struct net *net = dev_net(skb->dev); |
bc044e8d | 2579 | struct in_device *in_dev = NULL; |
610438b7 ED |
2580 | const struct iphdr *iph; |
2581 | const struct udphdr *uh; | |
ca065d0c | 2582 | struct sock *sk = NULL; |
421b3885 | 2583 | struct dst_entry *dst; |
421b3885 | 2584 | int dif = skb->dev->ifindex; |
fb74c277 | 2585 | int sdif = inet_sdif(skb); |
6e540309 | 2586 | int ours; |
421b3885 SB |
2587 | |
2588 | /* validate the packet */ | |
2589 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) | |
7487449c | 2590 | return 0; |
421b3885 | 2591 | |
610438b7 ED |
2592 | iph = ip_hdr(skb); |
2593 | uh = udp_hdr(skb); | |
2594 | ||
996b44fc | 2595 | if (skb->pkt_type == PACKET_MULTICAST) { |
bc044e8d | 2596 | in_dev = __in_dev_get_rcu(skb->dev); |
6e540309 SB |
2597 | |
2598 | if (!in_dev) | |
7487449c | 2599 | return 0; |
6e540309 | 2600 | |
996b44fc PA |
2601 | ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, |
2602 | iph->protocol); | |
2603 | if (!ours) | |
2604 | return 0; | |
ad0ea198 | 2605 | |
421b3885 | 2606 | sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, |
fb74c277 DA |
2607 | uh->source, iph->saddr, |
2608 | dif, sdif); | |
6e540309 | 2609 | } else if (skb->pkt_type == PACKET_HOST) { |
421b3885 | 2610 | sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, |
3fa6f616 | 2611 | uh->source, iph->saddr, dif, sdif); |
6e540309 | 2612 | } |
421b3885 | 2613 | |
41c6d650 | 2614 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) |
7487449c | 2615 | return 0; |
421b3885 SB |
2616 | |
2617 | skb->sk = sk; | |
82eabd9e | 2618 | skb->destructor = sock_efree; |
8f905c0e | 2619 | dst = rcu_dereference(sk->sk_rx_dst); |
421b3885 SB |
2620 | |
2621 | if (dst) | |
2622 | dst = dst_check(dst, 0); | |
10e2eb87 | 2623 | if (dst) { |
bc044e8d PA |
2624 | u32 itag = 0; |
2625 | ||
d24406c8 WW |
2626 | /* set noref for now. |
2627 | * any place which wants to hold dst has to call | |
2628 | * dst_hold_safe() | |
2629 | */ | |
2630 | skb_dst_set_noref(skb, dst); | |
bc044e8d PA |
2631 | |
2632 | /* for unconnected multicast sockets we need to validate | |
2633 | * the source on each packet | |
2634 | */ | |
2635 | if (!inet_sk(sk)->inet_daddr && in_dev) | |
2636 | return ip_mc_validate_source(skb, iph->daddr, | |
8d2b51b0 GN |
2637 | iph->saddr, |
2638 | iph->tos & IPTOS_RT_MASK, | |
bc044e8d | 2639 | skb->dev, in_dev, &itag); |
10e2eb87 | 2640 | } |
7487449c | 2641 | return 0; |
421b3885 SB |
2642 | } |
2643 | ||
db8dac20 DM |
2644 | int udp_rcv(struct sk_buff *skb) |
2645 | { | |
645ca708 | 2646 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
2647 | } |
2648 | ||
7d06b2e0 | 2649 | void udp_destroy_sock(struct sock *sk) |
db8dac20 | 2650 | { |
44046a59 | 2651 | struct udp_sock *up = udp_sk(sk); |
8a74ad60 | 2652 | bool slow = lock_sock_fast(sk); |
a8b897c7 PA |
2653 | |
2654 | /* protects from races with udp_abort() */ | |
2655 | sock_set_flag(sk, SOCK_DEAD); | |
db8dac20 | 2656 | udp_flush_pending_frames(sk); |
8a74ad60 | 2657 | unlock_sock_fast(sk, slow); |
60fb9567 PA |
2658 | if (static_branch_unlikely(&udp_encap_needed_key)) { |
2659 | if (up->encap_type) { | |
2660 | void (*encap_destroy)(struct sock *sk); | |
2661 | encap_destroy = READ_ONCE(up->encap_destroy); | |
2662 | if (encap_destroy) | |
2663 | encap_destroy(sk); | |
2664 | } | |
2665 | if (up->encap_enabled) | |
9c480601 | 2666 | static_branch_dec(&udp_encap_needed_key); |
44046a59 | 2667 | } |
db8dac20 DM |
2668 | } |
2669 | ||
1da177e4 LT |
2670 | /* |
2671 | * Socket option code for UDP | |
2672 | */ | |
4c0a6cb0 | 2673 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
91ac1cca | 2674 | sockptr_t optval, unsigned int optlen, |
4c0a6cb0 | 2675 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
2676 | { |
2677 | struct udp_sock *up = udp_sk(sk); | |
1c19448c | 2678 | int val, valbool; |
1da177e4 | 2679 | int err = 0; |
b2bf1e26 | 2680 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 2681 | |
c482c568 | 2682 | if (optlen < sizeof(int)) |
1da177e4 LT |
2683 | return -EINVAL; |
2684 | ||
91ac1cca | 2685 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
1da177e4 LT |
2686 | return -EFAULT; |
2687 | ||
1c19448c TH |
2688 | valbool = val ? 1 : 0; |
2689 | ||
6516c655 | 2690 | switch (optname) { |
1da177e4 LT |
2691 | case UDP_CORK: |
2692 | if (val != 0) { | |
a9f59707 | 2693 | WRITE_ONCE(up->corkflag, 1); |
1da177e4 | 2694 | } else { |
a9f59707 | 2695 | WRITE_ONCE(up->corkflag, 0); |
1da177e4 | 2696 | lock_sock(sk); |
4243cdc2 | 2697 | push_pending_frames(sk); |
1da177e4 LT |
2698 | release_sock(sk); |
2699 | } | |
2700 | break; | |
e905a9ed | 2701 | |
1da177e4 LT |
2702 | case UDP_ENCAP: |
2703 | switch (val) { | |
2704 | case 0: | |
fd1ac07f | 2705 | #ifdef CONFIG_XFRM |
1da177e4 LT |
2706 | case UDP_ENCAP_ESPINUDP: |
2707 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
0146dca7 SD |
2708 | #if IS_ENABLED(CONFIG_IPV6) |
2709 | if (sk->sk_family == AF_INET6) | |
2710 | up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; | |
2711 | else | |
2712 | #endif | |
2713 | up->encap_rcv = xfrm4_udp_encap_rcv; | |
fd1ac07f | 2714 | #endif |
a8eceea8 | 2715 | fallthrough; |
342f0234 | 2716 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 | 2717 | up->encap_type = val; |
60fb9567 PA |
2718 | lock_sock(sk); |
2719 | udp_tunnel_encap_enable(sk->sk_socket); | |
2720 | release_sock(sk); | |
1da177e4 LT |
2721 | break; |
2722 | default: | |
2723 | err = -ENOPROTOOPT; | |
2724 | break; | |
2725 | } | |
2726 | break; | |
2727 | ||
1c19448c TH |
2728 | case UDP_NO_CHECK6_TX: |
2729 | up->no_check6_tx = valbool; | |
2730 | break; | |
2731 | ||
2732 | case UDP_NO_CHECK6_RX: | |
2733 | up->no_check6_rx = valbool; | |
2734 | break; | |
2735 | ||
bec1f6f6 WB |
2736 | case UDP_SEGMENT: |
2737 | if (val < 0 || val > USHRT_MAX) | |
2738 | return -EINVAL; | |
18a419ba | 2739 | WRITE_ONCE(up->gso_size, val); |
bec1f6f6 WB |
2740 | break; |
2741 | ||
e20cf8d3 PA |
2742 | case UDP_GRO: |
2743 | lock_sock(sk); | |
78352f73 PA |
2744 | |
2745 | /* when enabling GRO, accept the related GSO packet type */ | |
e20cf8d3 PA |
2746 | if (valbool) |
2747 | udp_tunnel_encap_enable(sk->sk_socket); | |
2748 | up->gro_enabled = valbool; | |
78352f73 | 2749 | up->accept_udp_l4 = valbool; |
e20cf8d3 PA |
2750 | release_sock(sk); |
2751 | break; | |
2752 | ||
ba4e58ec GR |
2753 | /* |
2754 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
2755 | */ | |
2756 | /* The sender sets actual checksum coverage length via this option. | |
2757 | * The case coverage > packet length is handled by send module. */ | |
2758 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 2759 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2760 | return -ENOPROTOOPT; |
2761 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
2762 | val = 8; | |
4be929be AD |
2763 | else if (val > USHRT_MAX) |
2764 | val = USHRT_MAX; | |
ba4e58ec GR |
2765 | up->pcslen = val; |
2766 | up->pcflag |= UDPLITE_SEND_CC; | |
2767 | break; | |
2768 | ||
e905a9ed YH |
2769 | /* The receiver specifies a minimum checksum coverage value. To make |
2770 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
2771 | * used, this again means full checksum coverage. */ |
2772 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 2773 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2774 | return -ENOPROTOOPT; |
2775 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
2776 | val = 8; | |
4be929be AD |
2777 | else if (val > USHRT_MAX) |
2778 | val = USHRT_MAX; | |
ba4e58ec GR |
2779 | up->pcrlen = val; |
2780 | up->pcflag |= UDPLITE_RECV_CC; | |
2781 | break; | |
2782 | ||
1da177e4 LT |
2783 | default: |
2784 | err = -ENOPROTOOPT; | |
2785 | break; | |
6516c655 | 2786 | } |
1da177e4 LT |
2787 | |
2788 | return err; | |
2789 | } | |
c482c568 | 2790 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 2791 | |
a7b75c5a CH |
2792 | int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, |
2793 | unsigned int optlen) | |
db8dac20 DM |
2794 | { |
2795 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
91ac1cca | 2796 | return udp_lib_setsockopt(sk, level, optname, |
a7b75c5a | 2797 | optval, optlen, |
db8dac20 DM |
2798 | udp_push_pending_frames); |
2799 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
2800 | } | |
2801 | ||
4c0a6cb0 GR |
2802 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
2803 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
2804 | { |
2805 | struct udp_sock *up = udp_sk(sk); | |
2806 | int val, len; | |
2807 | ||
c482c568 | 2808 | if (get_user(len, optlen)) |
1da177e4 LT |
2809 | return -EFAULT; |
2810 | ||
2811 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 2812 | |
6516c655 | 2813 | if (len < 0) |
1da177e4 LT |
2814 | return -EINVAL; |
2815 | ||
6516c655 | 2816 | switch (optname) { |
1da177e4 | 2817 | case UDP_CORK: |
a9f59707 | 2818 | val = READ_ONCE(up->corkflag); |
1da177e4 LT |
2819 | break; |
2820 | ||
2821 | case UDP_ENCAP: | |
2822 | val = up->encap_type; | |
2823 | break; | |
2824 | ||
1c19448c TH |
2825 | case UDP_NO_CHECK6_TX: |
2826 | val = up->no_check6_tx; | |
2827 | break; | |
2828 | ||
2829 | case UDP_NO_CHECK6_RX: | |
2830 | val = up->no_check6_rx; | |
2831 | break; | |
2832 | ||
bec1f6f6 | 2833 | case UDP_SEGMENT: |
18a419ba | 2834 | val = READ_ONCE(up->gso_size); |
bec1f6f6 WB |
2835 | break; |
2836 | ||
98184612 NM |
2837 | case UDP_GRO: |
2838 | val = up->gro_enabled; | |
2839 | break; | |
2840 | ||
ba4e58ec GR |
2841 | /* The following two cannot be changed on UDP sockets, the return is |
2842 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
2843 | case UDPLITE_SEND_CSCOV: | |
2844 | val = up->pcslen; | |
2845 | break; | |
2846 | ||
2847 | case UDPLITE_RECV_CSCOV: | |
2848 | val = up->pcrlen; | |
2849 | break; | |
2850 | ||
1da177e4 LT |
2851 | default: |
2852 | return -ENOPROTOOPT; | |
6516c655 | 2853 | } |
1da177e4 | 2854 | |
6516c655 | 2855 | if (put_user(len, optlen)) |
e905a9ed | 2856 | return -EFAULT; |
c482c568 | 2857 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 2858 | return -EFAULT; |
e905a9ed | 2859 | return 0; |
1da177e4 | 2860 | } |
c482c568 | 2861 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 2862 | |
db8dac20 DM |
2863 | int udp_getsockopt(struct sock *sk, int level, int optname, |
2864 | char __user *optval, int __user *optlen) | |
2865 | { | |
2866 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2867 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2868 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
2869 | } | |
2870 | ||
1da177e4 LT |
2871 | /** |
2872 | * udp_poll - wait for a UDP event. | |
3628e3cb AL |
2873 | * @file: - file struct |
2874 | * @sock: - socket | |
2875 | * @wait: - poll table | |
1da177e4 | 2876 | * |
e905a9ed | 2877 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
2878 | * blocking sockets. If application is using a blocking fd |
2879 | * and a packet with checksum error is in the queue; | |
2880 | * then it could get return from select indicating data available | |
2881 | * but then block when reading it. Add special case code | |
2882 | * to work around these arguably broken applications. | |
2883 | */ | |
a11e1d43 | 2884 | __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 2885 | { |
a11e1d43 | 2886 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 | 2887 | struct sock *sk = sock->sk; |
ba4e58ec | 2888 | |
3ef7cf57 | 2889 | if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) |
a9a08845 | 2890 | mask |= EPOLLIN | EPOLLRDNORM; |
2276f58a | 2891 | |
1da177e4 | 2892 | /* Check for false positives due to checksum errors */ |
a11e1d43 | 2893 | if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && |
e83c6744 | 2894 | !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) |
a9a08845 | 2895 | mask &= ~(EPOLLIN | EPOLLRDNORM); |
1da177e4 | 2896 | |
af493388 CW |
2897 | /* psock ingress_msg queue should not contain any bad checksum frames */ |
2898 | if (sk_is_readable(sk)) | |
2899 | mask |= EPOLLIN | EPOLLRDNORM; | |
1da177e4 | 2900 | return mask; |
e905a9ed | 2901 | |
1da177e4 | 2902 | } |
a11e1d43 | 2903 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 2904 | |
5d77dca8 DA |
2905 | int udp_abort(struct sock *sk, int err) |
2906 | { | |
2907 | lock_sock(sk); | |
2908 | ||
a8b897c7 PA |
2909 | /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing |
2910 | * with close() | |
2911 | */ | |
2912 | if (sock_flag(sk, SOCK_DEAD)) | |
2913 | goto out; | |
2914 | ||
5d77dca8 | 2915 | sk->sk_err = err; |
e3ae2365 | 2916 | sk_error_report(sk); |
286c72de | 2917 | __udp_disconnect(sk, 0); |
5d77dca8 | 2918 | |
a8b897c7 | 2919 | out: |
5d77dca8 DA |
2920 | release_sock(sk); |
2921 | ||
2922 | return 0; | |
2923 | } | |
2924 | EXPORT_SYMBOL_GPL(udp_abort); | |
2925 | ||
db8dac20 | 2926 | struct proto udp_prot = { |
1e802951 TZ |
2927 | .name = "UDP", |
2928 | .owner = THIS_MODULE, | |
2929 | .close = udp_lib_close, | |
d74bad4e | 2930 | .pre_connect = udp_pre_connect, |
1e802951 TZ |
2931 | .connect = ip4_datagram_connect, |
2932 | .disconnect = udp_disconnect, | |
2933 | .ioctl = udp_ioctl, | |
2934 | .init = udp_init_sock, | |
2935 | .destroy = udp_destroy_sock, | |
2936 | .setsockopt = udp_setsockopt, | |
2937 | .getsockopt = udp_getsockopt, | |
2938 | .sendmsg = udp_sendmsg, | |
2939 | .recvmsg = udp_recvmsg, | |
2940 | .sendpage = udp_sendpage, | |
2941 | .release_cb = ip4_datagram_release_cb, | |
2942 | .hash = udp_lib_hash, | |
2943 | .unhash = udp_lib_unhash, | |
2944 | .rehash = udp_v4_rehash, | |
2945 | .get_port = udp_v4_get_port, | |
91a760b2 | 2946 | .put_port = udp_lib_unhash, |
8a59f9d1 CW |
2947 | #ifdef CONFIG_BPF_SYSCALL |
2948 | .psock_update_sk_prot = udp_bpf_update_proto, | |
2949 | #endif | |
1e802951 | 2950 | .memory_allocated = &udp_memory_allocated, |
0defbb0a ED |
2951 | .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, |
2952 | ||
1e802951 TZ |
2953 | .sysctl_mem = sysctl_udp_mem, |
2954 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), | |
2955 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), | |
2956 | .obj_size = sizeof(struct udp_sock), | |
2957 | .h.udp_table = &udp_table, | |
1e802951 | 2958 | .diag_destroy = udp_abort, |
db8dac20 | 2959 | }; |
c482c568 | 2960 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
2961 | |
2962 | /* ------------------------------------------------------------------------ */ | |
2963 | #ifdef CONFIG_PROC_FS | |
2964 | ||
645ca708 | 2965 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
2966 | { |
2967 | struct sock *sk; | |
9e8ca27a | 2968 | struct udp_seq_afinfo *afinfo; |
1da177e4 | 2969 | struct udp_iter_state *state = seq->private; |
6f191efe | 2970 | struct net *net = seq_file_net(seq); |
1da177e4 | 2971 | |
9e8ca27a YS |
2972 | if (state->bpf_seq_afinfo) |
2973 | afinfo = state->bpf_seq_afinfo; | |
2974 | else | |
359745d7 | 2975 | afinfo = pde_data(file_inode(seq->file)); |
9e8ca27a | 2976 | |
a3d2599b | 2977 | for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; |
f86dcc5a | 2978 | ++state->bucket) { |
a3d2599b | 2979 | struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; |
f86dcc5a | 2980 | |
ca065d0c | 2981 | if (hlist_empty(&hslot->head)) |
f86dcc5a ED |
2982 | continue; |
2983 | ||
645ca708 | 2984 | spin_lock_bh(&hslot->lock); |
ca065d0c | 2985 | sk_for_each(sk, &hslot->head) { |
878628fb | 2986 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 2987 | continue; |
9e8ca27a YS |
2988 | if (afinfo->family == AF_UNSPEC || |
2989 | sk->sk_family == afinfo->family) | |
1da177e4 LT |
2990 | goto found; |
2991 | } | |
645ca708 | 2992 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
2993 | } |
2994 | sk = NULL; | |
2995 | found: | |
2996 | return sk; | |
2997 | } | |
2998 | ||
2999 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
3000 | { | |
9e8ca27a | 3001 | struct udp_seq_afinfo *afinfo; |
1da177e4 | 3002 | struct udp_iter_state *state = seq->private; |
6f191efe | 3003 | struct net *net = seq_file_net(seq); |
1da177e4 | 3004 | |
9e8ca27a YS |
3005 | if (state->bpf_seq_afinfo) |
3006 | afinfo = state->bpf_seq_afinfo; | |
3007 | else | |
359745d7 | 3008 | afinfo = pde_data(file_inode(seq->file)); |
9e8ca27a | 3009 | |
1da177e4 | 3010 | do { |
ca065d0c | 3011 | sk = sk_next(sk); |
9e8ca27a YS |
3012 | } while (sk && (!net_eq(sock_net(sk), net) || |
3013 | (afinfo->family != AF_UNSPEC && | |
3014 | sk->sk_family != afinfo->family))); | |
1da177e4 | 3015 | |
645ca708 | 3016 | if (!sk) { |
a3d2599b CH |
3017 | if (state->bucket <= afinfo->udp_table->mask) |
3018 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
645ca708 | 3019 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
3020 | } |
3021 | return sk; | |
3022 | } | |
3023 | ||
3024 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
3025 | { | |
645ca708 | 3026 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
3027 | |
3028 | if (sk) | |
6516c655 | 3029 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
3030 | --pos; |
3031 | return pos ? NULL : sk; | |
3032 | } | |
3033 | ||
a3d2599b | 3034 | void *udp_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 3035 | { |
30842f29 | 3036 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 3037 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 3038 | |
b50660f1 | 3039 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 | 3040 | } |
a3d2599b | 3041 | EXPORT_SYMBOL(udp_seq_start); |
1da177e4 | 3042 | |
a3d2599b | 3043 | void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1da177e4 LT |
3044 | { |
3045 | struct sock *sk; | |
3046 | ||
b50660f1 | 3047 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
3048 | sk = udp_get_idx(seq, 0); |
3049 | else | |
3050 | sk = udp_get_next(seq, v); | |
3051 | ||
3052 | ++*pos; | |
3053 | return sk; | |
3054 | } | |
a3d2599b | 3055 | EXPORT_SYMBOL(udp_seq_next); |
1da177e4 | 3056 | |
a3d2599b | 3057 | void udp_seq_stop(struct seq_file *seq, void *v) |
1da177e4 | 3058 | { |
9e8ca27a | 3059 | struct udp_seq_afinfo *afinfo; |
645ca708 ED |
3060 | struct udp_iter_state *state = seq->private; |
3061 | ||
9e8ca27a YS |
3062 | if (state->bpf_seq_afinfo) |
3063 | afinfo = state->bpf_seq_afinfo; | |
3064 | else | |
359745d7 | 3065 | afinfo = pde_data(file_inode(seq->file)); |
9e8ca27a | 3066 | |
a3d2599b CH |
3067 | if (state->bucket <= afinfo->udp_table->mask) |
3068 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
1da177e4 | 3069 | } |
a3d2599b | 3070 | EXPORT_SYMBOL(udp_seq_stop); |
db8dac20 DM |
3071 | |
3072 | /* ------------------------------------------------------------------------ */ | |
5e659e4c | 3073 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
652586df | 3074 | int bucket) |
db8dac20 DM |
3075 | { |
3076 | struct inet_sock *inet = inet_sk(sp); | |
c720c7e8 ED |
3077 | __be32 dest = inet->inet_daddr; |
3078 | __be32 src = inet->inet_rcv_saddr; | |
3079 | __u16 destp = ntohs(inet->inet_dport); | |
3080 | __u16 srcp = ntohs(inet->inet_sport); | |
db8dac20 | 3081 | |
f86dcc5a | 3082 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
ea9a0379 | 3083 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", |
db8dac20 | 3084 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 | 3085 | sk_wmem_alloc_get(sp), |
6c206b20 | 3086 | udp_rqueue_get(sp), |
a7cb5a49 EB |
3087 | 0, 0L, 0, |
3088 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), | |
3089 | 0, sock_i_ino(sp), | |
41c6d650 | 3090 | refcount_read(&sp->sk_refcnt), sp, |
652586df | 3091 | atomic_read(&sp->sk_drops)); |
db8dac20 DM |
3092 | } |
3093 | ||
3094 | int udp4_seq_show(struct seq_file *seq, void *v) | |
3095 | { | |
652586df | 3096 | seq_setwidth(seq, 127); |
db8dac20 | 3097 | if (v == SEQ_START_TOKEN) |
6c25449e | 3098 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
db8dac20 | 3099 | "rx_queue tr tm->when retrnsmt uid timeout " |
cb61cb9b | 3100 | "inode ref pointer drops"); |
db8dac20 | 3101 | else { |
db8dac20 DM |
3102 | struct udp_iter_state *state = seq->private; |
3103 | ||
652586df | 3104 | udp4_format_sock(v, seq, state->bucket); |
db8dac20 | 3105 | } |
652586df | 3106 | seq_pad(seq, '\n'); |
db8dac20 DM |
3107 | return 0; |
3108 | } | |
3109 | ||
5788b3a0 YS |
3110 | #ifdef CONFIG_BPF_SYSCALL |
3111 | struct bpf_iter__udp { | |
3112 | __bpf_md_ptr(struct bpf_iter_meta *, meta); | |
3113 | __bpf_md_ptr(struct udp_sock *, udp_sk); | |
3114 | uid_t uid __aligned(8); | |
3115 | int bucket __aligned(8); | |
3116 | }; | |
3117 | ||
3118 | static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, | |
3119 | struct udp_sock *udp_sk, uid_t uid, int bucket) | |
3120 | { | |
3121 | struct bpf_iter__udp ctx; | |
3122 | ||
3123 | meta->seq_num--; /* skip SEQ_START_TOKEN */ | |
3124 | ctx.meta = meta; | |
3125 | ctx.udp_sk = udp_sk; | |
3126 | ctx.uid = uid; | |
3127 | ctx.bucket = bucket; | |
3128 | return bpf_iter_run_prog(prog, &ctx); | |
3129 | } | |
3130 | ||
3131 | static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) | |
3132 | { | |
3133 | struct udp_iter_state *state = seq->private; | |
3134 | struct bpf_iter_meta meta; | |
3135 | struct bpf_prog *prog; | |
3136 | struct sock *sk = v; | |
3137 | uid_t uid; | |
3138 | ||
3139 | if (v == SEQ_START_TOKEN) | |
3140 | return 0; | |
3141 | ||
3142 | uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); | |
3143 | meta.seq = seq; | |
3144 | prog = bpf_iter_get_info(&meta, false); | |
3145 | return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); | |
3146 | } | |
3147 | ||
3148 | static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) | |
3149 | { | |
3150 | struct bpf_iter_meta meta; | |
3151 | struct bpf_prog *prog; | |
3152 | ||
3153 | if (!v) { | |
3154 | meta.seq = seq; | |
3155 | prog = bpf_iter_get_info(&meta, true); | |
3156 | if (prog) | |
3157 | (void)udp_prog_seq_show(prog, &meta, v, 0, 0); | |
3158 | } | |
3159 | ||
3160 | udp_seq_stop(seq, v); | |
3161 | } | |
3162 | ||
3163 | static const struct seq_operations bpf_iter_udp_seq_ops = { | |
3164 | .start = udp_seq_start, | |
3165 | .next = udp_seq_next, | |
3166 | .stop = bpf_iter_udp_seq_stop, | |
3167 | .show = bpf_iter_udp_seq_show, | |
3168 | }; | |
3169 | #endif | |
3170 | ||
c3506372 | 3171 | const struct seq_operations udp_seq_ops = { |
a3d2599b CH |
3172 | .start = udp_seq_start, |
3173 | .next = udp_seq_next, | |
3174 | .stop = udp_seq_stop, | |
3175 | .show = udp4_seq_show, | |
3176 | }; | |
c3506372 | 3177 | EXPORT_SYMBOL(udp_seq_ops); |
73cb88ec | 3178 | |
db8dac20 | 3179 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 | 3180 | .family = AF_INET, |
645ca708 | 3181 | .udp_table = &udp_table, |
db8dac20 DM |
3182 | }; |
3183 | ||
2c8c1e72 | 3184 | static int __net_init udp4_proc_init_net(struct net *net) |
15439feb | 3185 | { |
c3506372 CH |
3186 | if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, |
3187 | sizeof(struct udp_iter_state), &udp4_seq_afinfo)) | |
a3d2599b CH |
3188 | return -ENOMEM; |
3189 | return 0; | |
15439feb PE |
3190 | } |
3191 | ||
2c8c1e72 | 3192 | static void __net_exit udp4_proc_exit_net(struct net *net) |
15439feb | 3193 | { |
a3d2599b | 3194 | remove_proc_entry("udp", net->proc_net); |
15439feb PE |
3195 | } |
3196 | ||
3197 | static struct pernet_operations udp4_net_ops = { | |
3198 | .init = udp4_proc_init_net, | |
3199 | .exit = udp4_proc_exit_net, | |
3200 | }; | |
3201 | ||
db8dac20 DM |
3202 | int __init udp4_proc_init(void) |
3203 | { | |
15439feb | 3204 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
3205 | } |
3206 | ||
3207 | void udp4_proc_exit(void) | |
3208 | { | |
15439feb | 3209 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 3210 | } |
1da177e4 LT |
3211 | #endif /* CONFIG_PROC_FS */ |
3212 | ||
f86dcc5a ED |
3213 | static __initdata unsigned long uhash_entries; |
3214 | static int __init set_uhash_entries(char *str) | |
645ca708 | 3215 | { |
413c27d8 EZ |
3216 | ssize_t ret; |
3217 | ||
f86dcc5a ED |
3218 | if (!str) |
3219 | return 0; | |
413c27d8 EZ |
3220 | |
3221 | ret = kstrtoul(str, 0, &uhash_entries); | |
3222 | if (ret) | |
3223 | return 0; | |
3224 | ||
f86dcc5a ED |
3225 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) |
3226 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
3227 | return 1; | |
3228 | } | |
3229 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 3230 | |
f86dcc5a ED |
3231 | void __init udp_table_init(struct udp_table *table, const char *name) |
3232 | { | |
3233 | unsigned int i; | |
3234 | ||
31fe62b9 TB |
3235 | table->hash = alloc_large_system_hash(name, |
3236 | 2 * sizeof(struct udp_hslot), | |
3237 | uhash_entries, | |
3238 | 21, /* one slot per 2 MB */ | |
3239 | 0, | |
3240 | &table->log, | |
3241 | &table->mask, | |
3242 | UDP_HTABLE_SIZE_MIN, | |
3243 | 64 * 1024); | |
3244 | ||
512615b6 | 3245 | table->hash2 = table->hash + (table->mask + 1); |
f86dcc5a | 3246 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 3247 | INIT_HLIST_HEAD(&table->hash[i].head); |
fdcc8aa9 | 3248 | table->hash[i].count = 0; |
645ca708 ED |
3249 | spin_lock_init(&table->hash[i].lock); |
3250 | } | |
512615b6 | 3251 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 3252 | INIT_HLIST_HEAD(&table->hash2[i].head); |
512615b6 ED |
3253 | table->hash2[i].count = 0; |
3254 | spin_lock_init(&table->hash2[i].lock); | |
3255 | } | |
645ca708 ED |
3256 | } |
3257 | ||
723b8e46 TH |
3258 | u32 udp_flow_hashrnd(void) |
3259 | { | |
3260 | static u32 hashrnd __read_mostly; | |
3261 | ||
3262 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
3263 | ||
3264 | return hashrnd; | |
3265 | } | |
3266 | EXPORT_SYMBOL(udp_flow_hashrnd); | |
3267 | ||
1e802951 TZ |
3268 | static void __udp_sysctl_init(struct net *net) |
3269 | { | |
100fdd1f ED |
3270 | net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; |
3271 | net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; | |
1e802951 TZ |
3272 | |
3273 | #ifdef CONFIG_NET_L3_MASTER_DEV | |
3274 | net->ipv4.sysctl_udp_l3mdev_accept = 0; | |
3275 | #endif | |
3276 | } | |
3277 | ||
3278 | static int __net_init udp_sysctl_init(struct net *net) | |
3279 | { | |
3280 | __udp_sysctl_init(net); | |
3281 | return 0; | |
3282 | } | |
3283 | ||
3284 | static struct pernet_operations __net_initdata udp_sysctl_ops = { | |
fc18999e | 3285 | .init = udp_sysctl_init, |
1e802951 TZ |
3286 | }; |
3287 | ||
5788b3a0 YS |
3288 | #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) |
3289 | DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, | |
3290 | struct udp_sock *udp_sk, uid_t uid, int bucket) | |
3291 | ||
f9c79272 | 3292 | static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) |
5788b3a0 YS |
3293 | { |
3294 | struct udp_iter_state *st = priv_data; | |
3295 | struct udp_seq_afinfo *afinfo; | |
3296 | int ret; | |
3297 | ||
3298 | afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); | |
3299 | if (!afinfo) | |
3300 | return -ENOMEM; | |
3301 | ||
3302 | afinfo->family = AF_UNSPEC; | |
3303 | afinfo->udp_table = &udp_table; | |
3304 | st->bpf_seq_afinfo = afinfo; | |
f9c79272 | 3305 | ret = bpf_iter_init_seq_net(priv_data, aux); |
5788b3a0 YS |
3306 | if (ret) |
3307 | kfree(afinfo); | |
3308 | return ret; | |
3309 | } | |
3310 | ||
3311 | static void bpf_iter_fini_udp(void *priv_data) | |
3312 | { | |
3313 | struct udp_iter_state *st = priv_data; | |
3314 | ||
3315 | kfree(st->bpf_seq_afinfo); | |
3316 | bpf_iter_fini_seq_net(priv_data); | |
3317 | } | |
3318 | ||
14fc6bd6 | 3319 | static const struct bpf_iter_seq_info udp_seq_info = { |
5788b3a0 YS |
3320 | .seq_ops = &bpf_iter_udp_seq_ops, |
3321 | .init_seq_private = bpf_iter_init_udp, | |
3322 | .fini_seq_private = bpf_iter_fini_udp, | |
3323 | .seq_priv_size = sizeof(struct udp_iter_state), | |
14fc6bd6 YS |
3324 | }; |
3325 | ||
3326 | static struct bpf_iter_reg udp_reg_info = { | |
3327 | .target = "udp", | |
5788b3a0 YS |
3328 | .ctx_arg_info_size = 1, |
3329 | .ctx_arg_info = { | |
3330 | { offsetof(struct bpf_iter__udp, udp_sk), | |
3331 | PTR_TO_BTF_ID_OR_NULL }, | |
3332 | }, | |
14fc6bd6 | 3333 | .seq_info = &udp_seq_info, |
5788b3a0 YS |
3334 | }; |
3335 | ||
3336 | static void __init bpf_iter_register(void) | |
3337 | { | |
951cf368 | 3338 | udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; |
5788b3a0 YS |
3339 | if (bpf_iter_reg_target(&udp_reg_info)) |
3340 | pr_warn("Warning: could not register bpf iterator udp\n"); | |
3341 | } | |
3342 | #endif | |
3343 | ||
95766fff HA |
3344 | void __init udp_init(void) |
3345 | { | |
f03d78db | 3346 | unsigned long limit; |
4b272750 | 3347 | unsigned int i; |
95766fff | 3348 | |
f86dcc5a | 3349 | udp_table_init(&udp_table, "UDP"); |
f03d78db | 3350 | limit = nr_free_buffer_pages() / 8; |
95766fff HA |
3351 | limit = max(limit, 128UL); |
3352 | sysctl_udp_mem[0] = limit / 4 * 3; | |
3353 | sysctl_udp_mem[1] = limit; | |
3354 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
3355 | ||
1e802951 | 3356 | __udp_sysctl_init(&init_net); |
4b272750 ED |
3357 | |
3358 | /* 16 spinlocks per cpu */ | |
3359 | udp_busylocks_log = ilog2(nr_cpu_ids) + 4; | |
3360 | udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, | |
3361 | GFP_KERNEL); | |
3362 | if (!udp_busylocks) | |
3363 | panic("UDP: failed to alloc udp_busylocks\n"); | |
3364 | for (i = 0; i < (1U << udp_busylocks_log); i++) | |
3365 | spin_lock_init(udp_busylocks + i); | |
1e802951 TZ |
3366 | |
3367 | if (register_pernet_subsys(&udp_sysctl_ops)) | |
3368 | panic("UDP: failed to init sysctl parameters.\n"); | |
5788b3a0 YS |
3369 | |
3370 | #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) | |
3371 | bpf_iter_register(); | |
3372 | #endif | |
95766fff | 3373 | } |