Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * The User Datagram Protocol (UDP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
113aa838 | 11 | * Alan Cox, <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
12 | * Hirokazu Takahashi, <taka@valinux.co.jp> |
13 | * | |
14 | * Fixes: | |
15 | * Alan Cox : verify_area() calls | |
16 | * Alan Cox : stopped close while in use off icmp | |
17 | * messages. Not a fix but a botch that | |
18 | * for udp at least is 'valid'. | |
19 | * Alan Cox : Fixed icmp handling properly | |
20 | * Alan Cox : Correct error for oversized datagrams | |
e905a9ed YH |
21 | * Alan Cox : Tidied select() semantics. |
22 | * Alan Cox : udp_err() fixed properly, also now | |
1da177e4 LT |
23 | * select and read wake correctly on errors |
24 | * Alan Cox : udp_send verify_area moved to avoid mem leak | |
25 | * Alan Cox : UDP can count its memory | |
26 | * Alan Cox : send to an unknown connection causes | |
27 | * an ECONNREFUSED off the icmp, but | |
28 | * does NOT close. | |
29 | * Alan Cox : Switched to new sk_buff handlers. No more backlog! | |
30 | * Alan Cox : Using generic datagram code. Even smaller and the PEEK | |
31 | * bug no longer crashes it. | |
32 | * Fred Van Kempen : Net2e support for sk->broadcast. | |
33 | * Alan Cox : Uses skb_free_datagram | |
34 | * Alan Cox : Added get/set sockopt support. | |
35 | * Alan Cox : Broadcasting without option set returns EACCES. | |
36 | * Alan Cox : No wakeup calls. Instead we now use the callbacks. | |
37 | * Alan Cox : Use ip_tos and ip_ttl | |
38 | * Alan Cox : SNMP Mibs | |
39 | * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. | |
40 | * Matt Dillon : UDP length checks. | |
41 | * Alan Cox : Smarter af_inet used properly. | |
42 | * Alan Cox : Use new kernel side addressing. | |
43 | * Alan Cox : Incorrect return on truncated datagram receive. | |
44 | * Arnt Gulbrandsen : New udp_send and stuff | |
45 | * Alan Cox : Cache last socket | |
46 | * Alan Cox : Route cache | |
47 | * Jon Peatfield : Minor efficiency fix to sendto(). | |
48 | * Mike Shaver : RFC1122 checks. | |
49 | * Alan Cox : Nonblocking error fix. | |
50 | * Willy Konynenberg : Transparent proxying support. | |
51 | * Mike McLagan : Routing by source | |
52 | * David S. Miller : New socket lookup architecture. | |
53 | * Last socket cache retained as it | |
54 | * does have a high hit rate. | |
55 | * Olaf Kirch : Don't linearise iovec on sendmsg. | |
56 | * Andi Kleen : Some cleanups, cache destination entry | |
e905a9ed | 57 | * for connect. |
1da177e4 LT |
58 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. |
59 | * Melvin Smith : Check msg_name not msg_namelen in sendto(), | |
60 | * return ENOTCONN for unconnected sockets (POSIX) | |
61 | * Janos Farkas : don't deliver multi/broadcasts to a different | |
62 | * bound-to-device socket | |
63 | * Hirokazu Takahashi : HW checksumming for outgoing UDP | |
64 | * datagrams. | |
65 | * Hirokazu Takahashi : sendfile() on UDP works now. | |
66 | * Arnaldo C. Melo : convert /proc/net/udp to seq_file | |
67 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
68 | * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind | |
69 | * a single port at the same time. | |
70 | * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support | |
342f0234 | 71 | * James Chapman : Add L2TP encapsulation type. |
1da177e4 LT |
72 | * |
73 | * | |
74 | * This program is free software; you can redistribute it and/or | |
75 | * modify it under the terms of the GNU General Public License | |
76 | * as published by the Free Software Foundation; either version | |
77 | * 2 of the License, or (at your option) any later version. | |
78 | */ | |
e905a9ed | 79 | |
1da177e4 LT |
80 | #include <asm/system.h> |
81 | #include <asm/uaccess.h> | |
82 | #include <asm/ioctls.h> | |
95766fff | 83 | #include <linux/bootmem.h> |
8203efb3 ED |
84 | #include <linux/highmem.h> |
85 | #include <linux/swap.h> | |
1da177e4 LT |
86 | #include <linux/types.h> |
87 | #include <linux/fcntl.h> | |
88 | #include <linux/module.h> | |
89 | #include <linux/socket.h> | |
90 | #include <linux/sockios.h> | |
14c85021 | 91 | #include <linux/igmp.h> |
1da177e4 LT |
92 | #include <linux/in.h> |
93 | #include <linux/errno.h> | |
94 | #include <linux/timer.h> | |
95 | #include <linux/mm.h> | |
1da177e4 | 96 | #include <linux/inet.h> |
1da177e4 | 97 | #include <linux/netdevice.h> |
c752f073 | 98 | #include <net/tcp_states.h> |
1da177e4 LT |
99 | #include <linux/skbuff.h> |
100 | #include <linux/proc_fs.h> | |
101 | #include <linux/seq_file.h> | |
457c4cbc | 102 | #include <net/net_namespace.h> |
1da177e4 LT |
103 | #include <net/icmp.h> |
104 | #include <net/route.h> | |
1da177e4 LT |
105 | #include <net/checksum.h> |
106 | #include <net/xfrm.h> | |
ba4e58ec | 107 | #include "udp_impl.h" |
1da177e4 | 108 | |
f86dcc5a | 109 | struct udp_table udp_table __read_mostly; |
645ca708 | 110 | EXPORT_SYMBOL(udp_table); |
1da177e4 | 111 | |
95766fff | 112 | int sysctl_udp_mem[3] __read_mostly; |
95766fff | 113 | EXPORT_SYMBOL(sysctl_udp_mem); |
c482c568 ED |
114 | |
115 | int sysctl_udp_rmem_min __read_mostly; | |
95766fff | 116 | EXPORT_SYMBOL(sysctl_udp_rmem_min); |
c482c568 ED |
117 | |
118 | int sysctl_udp_wmem_min __read_mostly; | |
95766fff HA |
119 | EXPORT_SYMBOL(sysctl_udp_wmem_min); |
120 | ||
121 | atomic_t udp_memory_allocated; | |
122 | EXPORT_SYMBOL(udp_memory_allocated); | |
123 | ||
f86dcc5a ED |
124 | #define MAX_UDP_PORTS 65536 |
125 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 126 | |
f24d43c0 | 127 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 128 | const struct udp_hslot *hslot, |
98322f22 | 129 | unsigned long *bitmap, |
f24d43c0 ED |
130 | struct sock *sk, |
131 | int (*saddr_comp)(const struct sock *sk1, | |
f86dcc5a ED |
132 | const struct sock *sk2), |
133 | unsigned int log) | |
1da177e4 | 134 | { |
f24d43c0 | 135 | struct sock *sk2; |
88ab1932 | 136 | struct hlist_nulls_node *node; |
25030a7f | 137 | |
88ab1932 | 138 | sk_nulls_for_each(sk2, node, &hslot->head) |
f24d43c0 ED |
139 | if (net_eq(sock_net(sk2), net) && |
140 | sk2 != sk && | |
98322f22 | 141 | (bitmap || sk2->sk_hash == num) && |
f24d43c0 ED |
142 | (!sk2->sk_reuse || !sk->sk_reuse) && |
143 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if | |
144 | || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
98322f22 ED |
145 | (*saddr_comp)(sk, sk2)) { |
146 | if (bitmap) | |
f86dcc5a | 147 | __set_bit(sk2->sk_hash >> log, bitmap); |
98322f22 ED |
148 | else |
149 | return 1; | |
150 | } | |
25030a7f GR |
151 | return 0; |
152 | } | |
153 | ||
154 | /** | |
6ba5a3c5 | 155 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
156 | * |
157 | * @sk: socket struct in question | |
158 | * @snum: port number to look up | |
df2bc459 | 159 | * @saddr_comp: AF-dependent comparison of bound local IP addresses |
25030a7f | 160 | */ |
6ba5a3c5 | 161 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
df2bc459 | 162 | int (*saddr_comp)(const struct sock *sk1, |
c482c568 | 163 | const struct sock *sk2)) |
25030a7f | 164 | { |
645ca708 ED |
165 | struct udp_hslot *hslot; |
166 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
25030a7f | 167 | int error = 1; |
3b1e0a65 | 168 | struct net *net = sock_net(sk); |
1da177e4 | 169 | |
32c1da70 | 170 | if (!snum) { |
9088c560 ED |
171 | int low, high, remaining; |
172 | unsigned rand; | |
98322f22 ED |
173 | unsigned short first, last; |
174 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 175 | |
227b60f5 | 176 | inet_get_local_port_range(&low, &high); |
a25de534 | 177 | remaining = (high - low) + 1; |
227b60f5 | 178 | |
9088c560 | 179 | rand = net_random(); |
98322f22 ED |
180 | first = (((u64)rand * remaining) >> 32) + low; |
181 | /* | |
182 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
183 | */ | |
f86dcc5a ED |
184 | rand = (rand | 1) * (udptable->mask + 1); |
185 | for (last = first + udptable->mask + 1; | |
186 | first != last; | |
187 | first++) { | |
188 | hslot = udp_hashslot(udptable, net, first); | |
98322f22 | 189 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 190 | spin_lock_bh(&hslot->lock); |
98322f22 | 191 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
f86dcc5a | 192 | saddr_comp, udptable->log); |
98322f22 ED |
193 | |
194 | snum = first; | |
195 | /* | |
196 | * Iterate on all possible values of snum for this hash. | |
197 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
198 | * give us randomization and full range coverage. | |
199 | */ | |
9088c560 | 200 | do { |
98322f22 | 201 | if (low <= snum && snum <= high && |
f86dcc5a | 202 | !test_bit(snum >> udptable->log, bitmap)) |
98322f22 ED |
203 | goto found; |
204 | snum += rand; | |
205 | } while (snum != first); | |
206 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 207 | } |
98322f22 | 208 | goto fail; |
645ca708 | 209 | } else { |
f86dcc5a | 210 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 211 | spin_lock_bh(&hslot->lock); |
f86dcc5a ED |
212 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, |
213 | saddr_comp, 0)) | |
645ca708 ED |
214 | goto fail_unlock; |
215 | } | |
98322f22 | 216 | found: |
25030a7f | 217 | inet_sk(sk)->num = snum; |
df2bc459 | 218 | sk->sk_hash = snum; |
1da177e4 | 219 | if (sk_unhashed(sk)) { |
88ab1932 | 220 | sk_nulls_add_node_rcu(sk, &hslot->head); |
c29a0bc4 | 221 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
1da177e4 | 222 | } |
25030a7f | 223 | error = 0; |
645ca708 ED |
224 | fail_unlock: |
225 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 226 | fail: |
25030a7f GR |
227 | return error; |
228 | } | |
c482c568 | 229 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 230 | |
499923c7 | 231 | static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) |
db8dac20 DM |
232 | { |
233 | struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); | |
234 | ||
c482c568 ED |
235 | return (!ipv6_only_sock(sk2) && |
236 | (!inet1->rcv_saddr || !inet2->rcv_saddr || | |
237 | inet1->rcv_saddr == inet2->rcv_saddr)); | |
db8dac20 DM |
238 | } |
239 | ||
6ba5a3c5 | 240 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 241 | { |
6ba5a3c5 | 242 | return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); |
db8dac20 DM |
243 | } |
244 | ||
645ca708 ED |
245 | static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, |
246 | unsigned short hnum, | |
247 | __be16 sport, __be32 daddr, __be16 dport, int dif) | |
248 | { | |
249 | int score = -1; | |
250 | ||
251 | if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && | |
252 | !ipv6_only_sock(sk)) { | |
253 | struct inet_sock *inet = inet_sk(sk); | |
254 | ||
255 | score = (sk->sk_family == PF_INET ? 1 : 0); | |
256 | if (inet->rcv_saddr) { | |
257 | if (inet->rcv_saddr != daddr) | |
258 | return -1; | |
259 | score += 2; | |
260 | } | |
261 | if (inet->daddr) { | |
262 | if (inet->daddr != saddr) | |
263 | return -1; | |
264 | score += 2; | |
265 | } | |
266 | if (inet->dport) { | |
267 | if (inet->dport != sport) | |
268 | return -1; | |
269 | score += 2; | |
270 | } | |
271 | if (sk->sk_bound_dev_if) { | |
272 | if (sk->sk_bound_dev_if != dif) | |
273 | return -1; | |
274 | score += 2; | |
275 | } | |
276 | } | |
277 | return score; | |
278 | } | |
279 | ||
db8dac20 DM |
280 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
281 | * harder than this. -DaveM | |
282 | */ | |
283 | static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, | |
284 | __be16 sport, __be32 daddr, __be16 dport, | |
645ca708 | 285 | int dif, struct udp_table *udptable) |
db8dac20 | 286 | { |
271b72c7 | 287 | struct sock *sk, *result; |
88ab1932 | 288 | struct hlist_nulls_node *node; |
db8dac20 | 289 | unsigned short hnum = ntohs(dport); |
f86dcc5a | 290 | unsigned int hash = udp_hashfn(net, hnum, udptable->mask); |
645ca708 | 291 | struct udp_hslot *hslot = &udptable->hash[hash]; |
271b72c7 | 292 | int score, badness; |
645ca708 | 293 | |
271b72c7 ED |
294 | rcu_read_lock(); |
295 | begin: | |
296 | result = NULL; | |
297 | badness = -1; | |
88ab1932 | 298 | sk_nulls_for_each_rcu(sk, node, &hslot->head) { |
645ca708 ED |
299 | score = compute_score(sk, net, saddr, hnum, sport, |
300 | daddr, dport, dif); | |
301 | if (score > badness) { | |
302 | result = sk; | |
303 | badness = score; | |
db8dac20 DM |
304 | } |
305 | } | |
88ab1932 ED |
306 | /* |
307 | * if the nulls value we got at the end of this lookup is | |
308 | * not the expected one, we must restart lookup. | |
309 | * We probably met an item that was moved to another chain. | |
310 | */ | |
311 | if (get_nulls_value(node) != hash) | |
312 | goto begin; | |
313 | ||
271b72c7 ED |
314 | if (result) { |
315 | if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) | |
316 | result = NULL; | |
317 | else if (unlikely(compute_score(result, net, saddr, hnum, sport, | |
318 | daddr, dport, dif) < badness)) { | |
319 | sock_put(result); | |
320 | goto begin; | |
321 | } | |
322 | } | |
323 | rcu_read_unlock(); | |
db8dac20 DM |
324 | return result; |
325 | } | |
326 | ||
607c4aaf KK |
327 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
328 | __be16 sport, __be16 dport, | |
645ca708 | 329 | struct udp_table *udptable) |
607c4aaf | 330 | { |
23542618 | 331 | struct sock *sk; |
607c4aaf KK |
332 | const struct iphdr *iph = ip_hdr(skb); |
333 | ||
23542618 KK |
334 | if (unlikely(sk = skb_steal_sock(skb))) |
335 | return sk; | |
336 | else | |
adf30907 | 337 | return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, |
23542618 KK |
338 | iph->daddr, dport, inet_iif(skb), |
339 | udptable); | |
607c4aaf KK |
340 | } |
341 | ||
bcd41303 KK |
342 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
343 | __be32 daddr, __be16 dport, int dif) | |
344 | { | |
645ca708 | 345 | return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); |
bcd41303 KK |
346 | } |
347 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
348 | ||
920a4611 | 349 | static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, |
db8dac20 DM |
350 | __be16 loc_port, __be32 loc_addr, |
351 | __be16 rmt_port, __be32 rmt_addr, | |
352 | int dif) | |
353 | { | |
88ab1932 | 354 | struct hlist_nulls_node *node; |
db8dac20 DM |
355 | struct sock *s = sk; |
356 | unsigned short hnum = ntohs(loc_port); | |
357 | ||
88ab1932 | 358 | sk_nulls_for_each_from(s, node) { |
db8dac20 DM |
359 | struct inet_sock *inet = inet_sk(s); |
360 | ||
920a4611 ED |
361 | if (!net_eq(sock_net(s), net) || |
362 | s->sk_hash != hnum || | |
db8dac20 DM |
363 | (inet->daddr && inet->daddr != rmt_addr) || |
364 | (inet->dport != rmt_port && inet->dport) || | |
365 | (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || | |
366 | ipv6_only_sock(s) || | |
367 | (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) | |
368 | continue; | |
369 | if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) | |
370 | continue; | |
371 | goto found; | |
372 | } | |
373 | s = NULL; | |
374 | found: | |
375 | return s; | |
376 | } | |
377 | ||
378 | /* | |
379 | * This routine is called by the ICMP module when it gets some | |
380 | * sort of error condition. If err < 0 then the socket should | |
381 | * be closed and the error returned to the user. If err > 0 | |
382 | * it's just the icmp type << 8 | icmp code. | |
383 | * Header points to the ip header of the error packet. We move | |
384 | * on past this. Then (as it used to claim before adjustment) | |
385 | * header points to the first 8 bytes of the udp header. We need | |
386 | * to find the appropriate port. | |
387 | */ | |
388 | ||
645ca708 | 389 | void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
390 | { |
391 | struct inet_sock *inet; | |
c482c568 ED |
392 | struct iphdr *iph = (struct iphdr *)skb->data; |
393 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); | |
db8dac20 DM |
394 | const int type = icmp_hdr(skb)->type; |
395 | const int code = icmp_hdr(skb)->code; | |
396 | struct sock *sk; | |
397 | int harderr; | |
398 | int err; | |
fd54d716 | 399 | struct net *net = dev_net(skb->dev); |
db8dac20 | 400 | |
fd54d716 | 401 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
db8dac20 DM |
402 | iph->saddr, uh->source, skb->dev->ifindex, udptable); |
403 | if (sk == NULL) { | |
dcfc23ca | 404 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
db8dac20 DM |
405 | return; /* No socket for error */ |
406 | } | |
407 | ||
408 | err = 0; | |
409 | harderr = 0; | |
410 | inet = inet_sk(sk); | |
411 | ||
412 | switch (type) { | |
413 | default: | |
414 | case ICMP_TIME_EXCEEDED: | |
415 | err = EHOSTUNREACH; | |
416 | break; | |
417 | case ICMP_SOURCE_QUENCH: | |
418 | goto out; | |
419 | case ICMP_PARAMETERPROB: | |
420 | err = EPROTO; | |
421 | harderr = 1; | |
422 | break; | |
423 | case ICMP_DEST_UNREACH: | |
424 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
425 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { | |
426 | err = EMSGSIZE; | |
427 | harderr = 1; | |
428 | break; | |
429 | } | |
430 | goto out; | |
431 | } | |
432 | err = EHOSTUNREACH; | |
433 | if (code <= NR_ICMP_UNREACH) { | |
434 | harderr = icmp_err_convert[code].fatal; | |
435 | err = icmp_err_convert[code].errno; | |
436 | } | |
437 | break; | |
438 | } | |
439 | ||
440 | /* | |
441 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
442 | * 4.1.3.3. | |
443 | */ | |
444 | if (!inet->recverr) { | |
445 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
446 | goto out; | |
447 | } else { | |
c482c568 | 448 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
db8dac20 DM |
449 | } |
450 | sk->sk_err = err; | |
451 | sk->sk_error_report(sk); | |
452 | out: | |
453 | sock_put(sk); | |
454 | } | |
455 | ||
456 | void udp_err(struct sk_buff *skb, u32 info) | |
457 | { | |
645ca708 | 458 | __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
459 | } |
460 | ||
461 | /* | |
462 | * Throw away all pending data and cancel the corking. Socket is locked. | |
463 | */ | |
36d926b9 | 464 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
465 | { |
466 | struct udp_sock *up = udp_sk(sk); | |
467 | ||
468 | if (up->pending) { | |
469 | up->len = 0; | |
470 | up->pending = 0; | |
471 | ip_flush_pending_frames(sk); | |
472 | } | |
473 | } | |
36d926b9 | 474 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
475 | |
476 | /** | |
477 | * udp4_hwcsum_outgoing - handle outgoing HW checksumming | |
478 | * @sk: socket we are sending on | |
479 | * @skb: sk_buff containing the filled-in UDP header | |
480 | * (checksum field must be zeroed out) | |
481 | */ | |
482 | static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, | |
c482c568 | 483 | __be32 src, __be32 dst, int len) |
db8dac20 DM |
484 | { |
485 | unsigned int offset; | |
486 | struct udphdr *uh = udp_hdr(skb); | |
487 | __wsum csum = 0; | |
488 | ||
489 | if (skb_queue_len(&sk->sk_write_queue) == 1) { | |
490 | /* | |
491 | * Only one fragment on the socket. | |
492 | */ | |
493 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
494 | skb->csum_offset = offsetof(struct udphdr, check); | |
495 | uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); | |
496 | } else { | |
497 | /* | |
498 | * HW-checksum won't work as there are two or more | |
499 | * fragments on the socket so that all csums of sk_buffs | |
500 | * should be together | |
501 | */ | |
502 | offset = skb_transport_offset(skb); | |
503 | skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
504 | ||
505 | skb->ip_summed = CHECKSUM_NONE; | |
506 | ||
507 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
508 | csum = csum_add(csum, skb->csum); | |
509 | } | |
510 | ||
511 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); | |
512 | if (uh->check == 0) | |
513 | uh->check = CSUM_MANGLED_0; | |
514 | } | |
515 | } | |
516 | ||
517 | /* | |
518 | * Push out all pending data as one UDP datagram. Socket is locked. | |
519 | */ | |
520 | static int udp_push_pending_frames(struct sock *sk) | |
521 | { | |
522 | struct udp_sock *up = udp_sk(sk); | |
523 | struct inet_sock *inet = inet_sk(sk); | |
524 | struct flowi *fl = &inet->cork.fl; | |
525 | struct sk_buff *skb; | |
526 | struct udphdr *uh; | |
527 | int err = 0; | |
528 | int is_udplite = IS_UDPLITE(sk); | |
529 | __wsum csum = 0; | |
530 | ||
531 | /* Grab the skbuff where UDP header space exists. */ | |
532 | if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) | |
533 | goto out; | |
534 | ||
535 | /* | |
536 | * Create a UDP header | |
537 | */ | |
538 | uh = udp_hdr(skb); | |
539 | uh->source = fl->fl_ip_sport; | |
540 | uh->dest = fl->fl_ip_dport; | |
541 | uh->len = htons(up->len); | |
542 | uh->check = 0; | |
543 | ||
544 | if (is_udplite) /* UDP-Lite */ | |
545 | csum = udplite_csum_outgoing(sk, skb); | |
546 | ||
547 | else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ | |
548 | ||
549 | skb->ip_summed = CHECKSUM_NONE; | |
550 | goto send; | |
551 | ||
552 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
553 | ||
c482c568 | 554 | udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len); |
db8dac20 DM |
555 | goto send; |
556 | ||
557 | } else /* `normal' UDP */ | |
558 | csum = udp_csum_outgoing(sk, skb); | |
559 | ||
560 | /* add protocol-dependent pseudo-header */ | |
561 | uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, | |
c482c568 | 562 | sk->sk_protocol, csum); |
db8dac20 DM |
563 | if (uh->check == 0) |
564 | uh->check = CSUM_MANGLED_0; | |
565 | ||
566 | send: | |
567 | err = ip_push_pending_frames(sk); | |
6ce9e7b5 ED |
568 | if (err) { |
569 | if (err == -ENOBUFS && !inet->recverr) { | |
570 | UDP_INC_STATS_USER(sock_net(sk), | |
571 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
572 | err = 0; | |
573 | } | |
574 | } else | |
575 | UDP_INC_STATS_USER(sock_net(sk), | |
576 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
db8dac20 DM |
577 | out: |
578 | up->len = 0; | |
579 | up->pending = 0; | |
db8dac20 DM |
580 | return err; |
581 | } | |
582 | ||
583 | int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, | |
584 | size_t len) | |
585 | { | |
586 | struct inet_sock *inet = inet_sk(sk); | |
587 | struct udp_sock *up = udp_sk(sk); | |
588 | int ulen = len; | |
589 | struct ipcm_cookie ipc; | |
590 | struct rtable *rt = NULL; | |
591 | int free = 0; | |
592 | int connected = 0; | |
593 | __be32 daddr, faddr, saddr; | |
594 | __be16 dport; | |
595 | u8 tos; | |
596 | int err, is_udplite = IS_UDPLITE(sk); | |
597 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
598 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
599 | ||
600 | if (len > 0xFFFF) | |
601 | return -EMSGSIZE; | |
602 | ||
603 | /* | |
604 | * Check the flags. | |
605 | */ | |
606 | ||
c482c568 | 607 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
608 | return -EOPNOTSUPP; |
609 | ||
610 | ipc.opt = NULL; | |
51f31cab | 611 | ipc.shtx.flags = 0; |
db8dac20 DM |
612 | |
613 | if (up->pending) { | |
614 | /* | |
615 | * There are pending frames. | |
616 | * The socket lock must be held while it's corked. | |
617 | */ | |
618 | lock_sock(sk); | |
619 | if (likely(up->pending)) { | |
620 | if (unlikely(up->pending != AF_INET)) { | |
621 | release_sock(sk); | |
622 | return -EINVAL; | |
623 | } | |
624 | goto do_append_data; | |
625 | } | |
626 | release_sock(sk); | |
627 | } | |
628 | ulen += sizeof(struct udphdr); | |
629 | ||
630 | /* | |
631 | * Get and verify the address. | |
632 | */ | |
633 | if (msg->msg_name) { | |
c482c568 | 634 | struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; |
db8dac20 DM |
635 | if (msg->msg_namelen < sizeof(*usin)) |
636 | return -EINVAL; | |
637 | if (usin->sin_family != AF_INET) { | |
638 | if (usin->sin_family != AF_UNSPEC) | |
639 | return -EAFNOSUPPORT; | |
640 | } | |
641 | ||
642 | daddr = usin->sin_addr.s_addr; | |
643 | dport = usin->sin_port; | |
644 | if (dport == 0) | |
645 | return -EINVAL; | |
646 | } else { | |
647 | if (sk->sk_state != TCP_ESTABLISHED) | |
648 | return -EDESTADDRREQ; | |
649 | daddr = inet->daddr; | |
650 | dport = inet->dport; | |
651 | /* Open fast path for connected socket. | |
652 | Route will not be used, if at least one option is set. | |
653 | */ | |
654 | connected = 1; | |
655 | } | |
656 | ipc.addr = inet->saddr; | |
657 | ||
658 | ipc.oif = sk->sk_bound_dev_if; | |
51f31cab PO |
659 | err = sock_tx_timestamp(msg, sk, &ipc.shtx); |
660 | if (err) | |
661 | return err; | |
db8dac20 | 662 | if (msg->msg_controllen) { |
3b1e0a65 | 663 | err = ip_cmsg_send(sock_net(sk), msg, &ipc); |
db8dac20 DM |
664 | if (err) |
665 | return err; | |
666 | if (ipc.opt) | |
667 | free = 1; | |
668 | connected = 0; | |
669 | } | |
670 | if (!ipc.opt) | |
671 | ipc.opt = inet->opt; | |
672 | ||
673 | saddr = ipc.addr; | |
674 | ipc.addr = faddr = daddr; | |
675 | ||
676 | if (ipc.opt && ipc.opt->srr) { | |
677 | if (!daddr) | |
678 | return -EINVAL; | |
679 | faddr = ipc.opt->faddr; | |
680 | connected = 0; | |
681 | } | |
682 | tos = RT_TOS(inet->tos); | |
683 | if (sock_flag(sk, SOCK_LOCALROUTE) || | |
684 | (msg->msg_flags & MSG_DONTROUTE) || | |
685 | (ipc.opt && ipc.opt->is_strictroute)) { | |
686 | tos |= RTO_ONLINK; | |
687 | connected = 0; | |
688 | } | |
689 | ||
690 | if (ipv4_is_multicast(daddr)) { | |
691 | if (!ipc.oif) | |
692 | ipc.oif = inet->mc_index; | |
693 | if (!saddr) | |
694 | saddr = inet->mc_addr; | |
695 | connected = 0; | |
696 | } | |
697 | ||
698 | if (connected) | |
c482c568 | 699 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 DM |
700 | |
701 | if (rt == NULL) { | |
702 | struct flowi fl = { .oif = ipc.oif, | |
914a9ab3 | 703 | .mark = sk->sk_mark, |
db8dac20 DM |
704 | .nl_u = { .ip4_u = |
705 | { .daddr = faddr, | |
706 | .saddr = saddr, | |
707 | .tos = tos } }, | |
708 | .proto = sk->sk_protocol, | |
a134f85c | 709 | .flags = inet_sk_flowi_flags(sk), |
db8dac20 DM |
710 | .uli_u = { .ports = |
711 | { .sport = inet->sport, | |
712 | .dport = dport } } }; | |
84a3aa00 PE |
713 | struct net *net = sock_net(sk); |
714 | ||
db8dac20 | 715 | security_sk_classify_flow(sk, &fl); |
84a3aa00 | 716 | err = ip_route_output_flow(net, &rt, &fl, sk, 1); |
db8dac20 DM |
717 | if (err) { |
718 | if (err == -ENETUNREACH) | |
7c73a6fa | 719 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
720 | goto out; |
721 | } | |
722 | ||
723 | err = -EACCES; | |
724 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
725 | !sock_flag(sk, SOCK_BROADCAST)) | |
726 | goto out; | |
727 | if (connected) | |
728 | sk_dst_set(sk, dst_clone(&rt->u.dst)); | |
729 | } | |
730 | ||
731 | if (msg->msg_flags&MSG_CONFIRM) | |
732 | goto do_confirm; | |
733 | back_from_confirm: | |
734 | ||
735 | saddr = rt->rt_src; | |
736 | if (!ipc.addr) | |
737 | daddr = ipc.addr = rt->rt_dst; | |
738 | ||
739 | lock_sock(sk); | |
740 | if (unlikely(up->pending)) { | |
741 | /* The socket is already corked while preparing it. */ | |
742 | /* ... which is an evident application bug. --ANK */ | |
743 | release_sock(sk); | |
744 | ||
745 | LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); | |
746 | err = -EINVAL; | |
747 | goto out; | |
748 | } | |
749 | /* | |
750 | * Now cork the socket to pend data. | |
751 | */ | |
752 | inet->cork.fl.fl4_dst = daddr; | |
753 | inet->cork.fl.fl_ip_dport = dport; | |
754 | inet->cork.fl.fl4_src = saddr; | |
755 | inet->cork.fl.fl_ip_sport = inet->sport; | |
756 | up->pending = AF_INET; | |
757 | ||
758 | do_append_data: | |
759 | up->len += ulen; | |
760 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; | |
761 | err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, | |
2e77d89b | 762 | sizeof(struct udphdr), &ipc, &rt, |
db8dac20 DM |
763 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); |
764 | if (err) | |
765 | udp_flush_pending_frames(sk); | |
766 | else if (!corkreq) | |
767 | err = udp_push_pending_frames(sk); | |
768 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
769 | up->pending = 0; | |
770 | release_sock(sk); | |
771 | ||
772 | out: | |
773 | ip_rt_put(rt); | |
774 | if (free) | |
775 | kfree(ipc.opt); | |
776 | if (!err) | |
777 | return len; | |
778 | /* | |
779 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
780 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
781 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
782 | * things). We could add another new stat but at least for now that | |
783 | * seems like overkill. | |
784 | */ | |
785 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
629ca23c PE |
786 | UDP_INC_STATS_USER(sock_net(sk), |
787 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
788 | } |
789 | return err; | |
790 | ||
791 | do_confirm: | |
792 | dst_confirm(&rt->u.dst); | |
793 | if (!(msg->msg_flags&MSG_PROBE) || len) | |
794 | goto back_from_confirm; | |
795 | err = 0; | |
796 | goto out; | |
797 | } | |
c482c568 | 798 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
799 | |
800 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
801 | size_t size, int flags) | |
802 | { | |
803 | struct udp_sock *up = udp_sk(sk); | |
804 | int ret; | |
805 | ||
806 | if (!up->pending) { | |
807 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
808 | ||
809 | /* Call udp_sendmsg to specify destination address which | |
810 | * sendpage interface can't pass. | |
811 | * This will succeed only when the socket is connected. | |
812 | */ | |
813 | ret = udp_sendmsg(NULL, sk, &msg, 0); | |
814 | if (ret < 0) | |
815 | return ret; | |
816 | } | |
817 | ||
818 | lock_sock(sk); | |
819 | ||
820 | if (unlikely(!up->pending)) { | |
821 | release_sock(sk); | |
822 | ||
823 | LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); | |
824 | return -EINVAL; | |
825 | } | |
826 | ||
827 | ret = ip_append_page(sk, page, offset, size, flags); | |
828 | if (ret == -EOPNOTSUPP) { | |
829 | release_sock(sk); | |
830 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
831 | size, flags); | |
832 | } | |
833 | if (ret < 0) { | |
834 | udp_flush_pending_frames(sk); | |
835 | goto out; | |
836 | } | |
837 | ||
838 | up->len += size; | |
839 | if (!(up->corkflag || (flags&MSG_MORE))) | |
840 | ret = udp_push_pending_frames(sk); | |
841 | if (!ret) | |
842 | ret = size; | |
843 | out: | |
844 | release_sock(sk); | |
845 | return ret; | |
846 | } | |
847 | ||
1da177e4 LT |
848 | /* |
849 | * IOCTL requests applicable to the UDP protocol | |
850 | */ | |
e905a9ed | 851 | |
1da177e4 LT |
852 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
853 | { | |
6516c655 SH |
854 | switch (cmd) { |
855 | case SIOCOUTQ: | |
1da177e4 | 856 | { |
31e6d363 ED |
857 | int amount = sk_wmem_alloc_get(sk); |
858 | ||
6516c655 SH |
859 | return put_user(amount, (int __user *)arg); |
860 | } | |
1da177e4 | 861 | |
6516c655 SH |
862 | case SIOCINQ: |
863 | { | |
864 | struct sk_buff *skb; | |
865 | unsigned long amount; | |
866 | ||
867 | amount = 0; | |
868 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
869 | skb = skb_peek(&sk->sk_receive_queue); | |
870 | if (skb != NULL) { | |
871 | /* | |
872 | * We will only return the amount | |
873 | * of this packet since that is all | |
874 | * that will be read. | |
875 | */ | |
876 | amount = skb->len - sizeof(struct udphdr); | |
1da177e4 | 877 | } |
6516c655 SH |
878 | spin_unlock_bh(&sk->sk_receive_queue.lock); |
879 | return put_user(amount, (int __user *)arg); | |
880 | } | |
1da177e4 | 881 | |
6516c655 SH |
882 | default: |
883 | return -ENOIOCTLCMD; | |
1da177e4 | 884 | } |
6516c655 SH |
885 | |
886 | return 0; | |
1da177e4 | 887 | } |
c482c568 | 888 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 889 | |
db8dac20 DM |
890 | /* |
891 | * This should be easy, if there is something there we | |
892 | * return it, otherwise we block. | |
893 | */ | |
894 | ||
895 | int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, | |
896 | size_t len, int noblock, int flags, int *addr_len) | |
897 | { | |
898 | struct inet_sock *inet = inet_sk(sk); | |
899 | struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; | |
900 | struct sk_buff *skb; | |
901 | unsigned int ulen, copied; | |
902 | int peeked; | |
903 | int err; | |
904 | int is_udplite = IS_UDPLITE(sk); | |
905 | ||
906 | /* | |
907 | * Check any passed addresses | |
908 | */ | |
909 | if (addr_len) | |
c482c568 | 910 | *addr_len = sizeof(*sin); |
db8dac20 DM |
911 | |
912 | if (flags & MSG_ERRQUEUE) | |
913 | return ip_recv_error(sk, msg, len); | |
914 | ||
915 | try_again: | |
916 | skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
917 | &peeked, &err); | |
918 | if (!skb) | |
919 | goto out; | |
920 | ||
921 | ulen = skb->len - sizeof(struct udphdr); | |
922 | copied = len; | |
923 | if (copied > ulen) | |
924 | copied = ulen; | |
925 | else if (copied < ulen) | |
926 | msg->msg_flags |= MSG_TRUNC; | |
927 | ||
928 | /* | |
929 | * If checksum is needed at all, try to do it while copying the | |
930 | * data. If the data is truncated, or if we only want a partial | |
931 | * coverage checksum (UDP-Lite), do it before the copy. | |
932 | */ | |
933 | ||
934 | if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { | |
935 | if (udp_lib_checksum_complete(skb)) | |
936 | goto csum_copy_err; | |
937 | } | |
938 | ||
939 | if (skb_csum_unnecessary(skb)) | |
940 | err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), | |
c482c568 | 941 | msg->msg_iov, copied); |
db8dac20 | 942 | else { |
c482c568 ED |
943 | err = skb_copy_and_csum_datagram_iovec(skb, |
944 | sizeof(struct udphdr), | |
945 | msg->msg_iov); | |
db8dac20 DM |
946 | |
947 | if (err == -EINVAL) | |
948 | goto csum_copy_err; | |
949 | } | |
950 | ||
951 | if (err) | |
952 | goto out_free; | |
953 | ||
954 | if (!peeked) | |
629ca23c PE |
955 | UDP_INC_STATS_USER(sock_net(sk), |
956 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 DM |
957 | |
958 | sock_recv_timestamp(msg, sk, skb); | |
959 | ||
960 | /* Copy the address. */ | |
c482c568 | 961 | if (sin) { |
db8dac20 DM |
962 | sin->sin_family = AF_INET; |
963 | sin->sin_port = udp_hdr(skb)->source; | |
964 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
965 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
966 | } | |
967 | if (inet->cmsg_flags) | |
968 | ip_cmsg_recv(msg, skb); | |
969 | ||
970 | err = copied; | |
971 | if (flags & MSG_TRUNC) | |
972 | err = ulen; | |
973 | ||
974 | out_free: | |
975 | lock_sock(sk); | |
976 | skb_free_datagram(sk, skb); | |
977 | release_sock(sk); | |
978 | out: | |
979 | return err; | |
980 | ||
981 | csum_copy_err: | |
982 | lock_sock(sk); | |
983 | if (!skb_kill_datagram(sk, skb, flags)) | |
629ca23c | 984 | UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
db8dac20 DM |
985 | release_sock(sk); |
986 | ||
987 | if (noblock) | |
988 | return -EAGAIN; | |
989 | goto try_again; | |
990 | } | |
991 | ||
992 | ||
1da177e4 LT |
993 | int udp_disconnect(struct sock *sk, int flags) |
994 | { | |
995 | struct inet_sock *inet = inet_sk(sk); | |
996 | /* | |
997 | * 1003.1g - break association. | |
998 | */ | |
e905a9ed | 999 | |
1da177e4 LT |
1000 | sk->sk_state = TCP_CLOSE; |
1001 | inet->daddr = 0; | |
1002 | inet->dport = 0; | |
1003 | sk->sk_bound_dev_if = 0; | |
1004 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
1005 | inet_reset_saddr(sk); | |
1006 | ||
1007 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1008 | sk->sk_prot->unhash(sk); | |
1009 | inet->sport = 0; | |
1010 | } | |
1011 | sk_dst_reset(sk); | |
1012 | return 0; | |
1013 | } | |
c482c568 | 1014 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 1015 | |
645ca708 ED |
1016 | void udp_lib_unhash(struct sock *sk) |
1017 | { | |
723b4610 ED |
1018 | if (sk_hashed(sk)) { |
1019 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
f86dcc5a ED |
1020 | struct udp_hslot *hslot = udp_hashslot(udptable, sock_net(sk), |
1021 | sk->sk_hash); | |
645ca708 | 1022 | |
723b4610 ED |
1023 | spin_lock_bh(&hslot->lock); |
1024 | if (sk_nulls_del_node_init_rcu(sk)) { | |
1025 | inet_sk(sk)->num = 0; | |
1026 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
1027 | } | |
1028 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 1029 | } |
645ca708 ED |
1030 | } |
1031 | EXPORT_SYMBOL(udp_lib_unhash); | |
1032 | ||
93821778 HX |
1033 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
1034 | { | |
1035 | int is_udplite = IS_UDPLITE(sk); | |
1036 | int rc; | |
1037 | ||
1038 | if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) { | |
1039 | /* Note that an ENOMEM error is charged twice */ | |
e408b8dc | 1040 | if (rc == -ENOMEM) { |
93821778 HX |
1041 | UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
1042 | is_udplite); | |
e408b8dc ED |
1043 | atomic_inc(&sk->sk_drops); |
1044 | } | |
93821778 HX |
1045 | goto drop; |
1046 | } | |
1047 | ||
1048 | return 0; | |
1049 | ||
1050 | drop: | |
1051 | UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
1052 | kfree_skb(skb); | |
1053 | return -1; | |
1054 | } | |
1055 | ||
db8dac20 DM |
1056 | /* returns: |
1057 | * -1: error | |
1058 | * 0: success | |
1059 | * >0: "udp encap" protocol resubmission | |
1060 | * | |
1061 | * Note that in the success and error cases, the skb is assumed to | |
1062 | * have either been requeued or freed. | |
1063 | */ | |
c482c568 | 1064 | int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 DM |
1065 | { |
1066 | struct udp_sock *up = udp_sk(sk); | |
1067 | int rc; | |
1068 | int is_udplite = IS_UDPLITE(sk); | |
1069 | ||
1070 | /* | |
1071 | * Charge it to the socket, dropping if the queue is full. | |
1072 | */ | |
1073 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1074 | goto drop; | |
1075 | nf_reset(skb); | |
1076 | ||
1077 | if (up->encap_type) { | |
1078 | /* | |
1079 | * This is an encapsulation socket so pass the skb to | |
1080 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
1081 | * fall through and pass this up the UDP socket. | |
1082 | * up->encap_rcv() returns the following value: | |
1083 | * =0 if skb was successfully passed to the encap | |
1084 | * handler or was discarded by it. | |
1085 | * >0 if skb should be passed on to UDP. | |
1086 | * <0 if skb should be resubmitted as proto -N | |
1087 | */ | |
1088 | ||
1089 | /* if we're overly short, let UDP handle it */ | |
1090 | if (skb->len > sizeof(struct udphdr) && | |
1091 | up->encap_rcv != NULL) { | |
1092 | int ret; | |
1093 | ||
1094 | ret = (*up->encap_rcv)(sk, skb); | |
1095 | if (ret <= 0) { | |
0283328e PE |
1096 | UDP_INC_STATS_BH(sock_net(sk), |
1097 | UDP_MIB_INDATAGRAMS, | |
db8dac20 DM |
1098 | is_udplite); |
1099 | return -ret; | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | /* FALLTHROUGH -- it's a UDP Packet */ | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * UDP-Lite specific tests, ignored on UDP sockets | |
1108 | */ | |
1109 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
1110 | ||
1111 | /* | |
1112 | * MIB statistics other than incrementing the error count are | |
1113 | * disabled for the following two types of errors: these depend | |
1114 | * on the application settings, not on the functioning of the | |
1115 | * protocol stack as such. | |
1116 | * | |
1117 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
1118 | * way ... to ... at least let the receiving application block | |
1119 | * delivery of packets with coverage values less than a value | |
1120 | * provided by the application." | |
1121 | */ | |
1122 | if (up->pcrlen == 0) { /* full coverage was set */ | |
1123 | LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " | |
1124 | "%d while full coverage %d requested\n", | |
1125 | UDP_SKB_CB(skb)->cscov, skb->len); | |
1126 | goto drop; | |
1127 | } | |
1128 | /* The next case involves violating the min. coverage requested | |
1129 | * by the receiver. This is subtle: if receiver wants x and x is | |
1130 | * greater than the buffersize/MTU then receiver will complain | |
1131 | * that it wants x while sender emits packets of smaller size y. | |
1132 | * Therefore the above ...()->partial_cov statement is essential. | |
1133 | */ | |
1134 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
1135 | LIMIT_NETDEBUG(KERN_WARNING | |
1136 | "UDPLITE: coverage %d too small, need min %d\n", | |
1137 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
1138 | goto drop; | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | if (sk->sk_filter) { | |
1143 | if (udp_lib_checksum_complete(skb)) | |
1144 | goto drop; | |
1145 | } | |
1146 | ||
93821778 | 1147 | rc = 0; |
db8dac20 | 1148 | |
93821778 HX |
1149 | bh_lock_sock(sk); |
1150 | if (!sock_owned_by_user(sk)) | |
1151 | rc = __udp_queue_rcv_skb(sk, skb); | |
1152 | else | |
1153 | sk_add_backlog(sk, skb); | |
1154 | bh_unlock_sock(sk); | |
1155 | ||
1156 | return rc; | |
db8dac20 DM |
1157 | |
1158 | drop: | |
0283328e | 1159 | UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
db8dac20 DM |
1160 | kfree_skb(skb); |
1161 | return -1; | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * Multicasts and broadcasts go to each listener. | |
1166 | * | |
1167 | * Note: called only from the BH handler context, | |
1168 | * so we don't need to lock the hashes. | |
1169 | */ | |
e3163493 | 1170 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
1171 | struct udphdr *uh, |
1172 | __be32 saddr, __be32 daddr, | |
645ca708 | 1173 | struct udp_table *udptable) |
db8dac20 DM |
1174 | { |
1175 | struct sock *sk; | |
f86dcc5a | 1176 | struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); |
db8dac20 DM |
1177 | int dif; |
1178 | ||
645ca708 | 1179 | spin_lock(&hslot->lock); |
88ab1932 | 1180 | sk = sk_nulls_head(&hslot->head); |
db8dac20 | 1181 | dif = skb->dev->ifindex; |
920a4611 | 1182 | sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); |
db8dac20 DM |
1183 | if (sk) { |
1184 | struct sock *sknext = NULL; | |
1185 | ||
1186 | do { | |
1187 | struct sk_buff *skb1 = skb; | |
1188 | ||
88ab1932 | 1189 | sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, |
920a4611 ED |
1190 | daddr, uh->source, saddr, |
1191 | dif); | |
db8dac20 DM |
1192 | if (sknext) |
1193 | skb1 = skb_clone(skb, GFP_ATOMIC); | |
1194 | ||
1195 | if (skb1) { | |
93821778 | 1196 | int ret = udp_queue_rcv_skb(sk, skb1); |
db8dac20 DM |
1197 | if (ret > 0) |
1198 | /* we should probably re-process instead | |
1199 | * of dropping packets here. */ | |
1200 | kfree_skb(skb1); | |
1201 | } | |
1202 | sk = sknext; | |
1203 | } while (sknext); | |
1204 | } else | |
ead2ceb0 | 1205 | consume_skb(skb); |
645ca708 | 1206 | spin_unlock(&hslot->lock); |
db8dac20 DM |
1207 | return 0; |
1208 | } | |
1209 | ||
1210 | /* Initialize UDP checksum. If exited with zero value (success), | |
1211 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
1212 | * Otherwise, csum completion requires chacksumming packet body, | |
1213 | * including udp header and folding it to skb->csum. | |
1214 | */ | |
1215 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
1216 | int proto) | |
1217 | { | |
1218 | const struct iphdr *iph; | |
1219 | int err; | |
1220 | ||
1221 | UDP_SKB_CB(skb)->partial_cov = 0; | |
1222 | UDP_SKB_CB(skb)->cscov = skb->len; | |
1223 | ||
1224 | if (proto == IPPROTO_UDPLITE) { | |
1225 | err = udplite_checksum_init(skb, uh); | |
1226 | if (err) | |
1227 | return err; | |
1228 | } | |
1229 | ||
1230 | iph = ip_hdr(skb); | |
1231 | if (uh->check == 0) { | |
1232 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1233 | } else if (skb->ip_summed == CHECKSUM_COMPLETE) { | |
c482c568 | 1234 | if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, |
db8dac20 DM |
1235 | proto, skb->csum)) |
1236 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1237 | } | |
1238 | if (!skb_csum_unnecessary(skb)) | |
1239 | skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, | |
1240 | skb->len, proto, 0); | |
1241 | /* Probably, we should checksum udp header (it should be in cache | |
1242 | * in any case) and data in tiny packets (< rx copybreak). | |
1243 | */ | |
1244 | ||
1245 | return 0; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * All we need to do is get the socket, and then do a checksum. | |
1250 | */ | |
1251 | ||
645ca708 | 1252 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
1253 | int proto) |
1254 | { | |
1255 | struct sock *sk; | |
7b5e56f9 | 1256 | struct udphdr *uh; |
db8dac20 | 1257 | unsigned short ulen; |
adf30907 | 1258 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 1259 | __be32 saddr, daddr; |
0283328e | 1260 | struct net *net = dev_net(skb->dev); |
db8dac20 DM |
1261 | |
1262 | /* | |
1263 | * Validate the packet. | |
1264 | */ | |
1265 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
1266 | goto drop; /* No space for header. */ | |
1267 | ||
7b5e56f9 | 1268 | uh = udp_hdr(skb); |
db8dac20 DM |
1269 | ulen = ntohs(uh->len); |
1270 | if (ulen > skb->len) | |
1271 | goto short_packet; | |
1272 | ||
1273 | if (proto == IPPROTO_UDP) { | |
1274 | /* UDP validates ulen. */ | |
1275 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
1276 | goto short_packet; | |
1277 | uh = udp_hdr(skb); | |
1278 | } | |
1279 | ||
1280 | if (udp4_csum_init(skb, uh, proto)) | |
1281 | goto csum_error; | |
1282 | ||
2783ef23 JDB |
1283 | saddr = ip_hdr(skb)->saddr; |
1284 | daddr = ip_hdr(skb)->daddr; | |
1285 | ||
db8dac20 | 1286 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
e3163493 PE |
1287 | return __udp4_lib_mcast_deliver(net, skb, uh, |
1288 | saddr, daddr, udptable); | |
db8dac20 | 1289 | |
607c4aaf | 1290 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); |
db8dac20 DM |
1291 | |
1292 | if (sk != NULL) { | |
93821778 | 1293 | int ret = udp_queue_rcv_skb(sk, skb); |
db8dac20 DM |
1294 | sock_put(sk); |
1295 | ||
1296 | /* a return value > 0 means to resubmit the input, but | |
1297 | * it wants the return to be -protocol, or 0 | |
1298 | */ | |
1299 | if (ret > 0) | |
1300 | return -ret; | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1305 | goto drop; | |
1306 | nf_reset(skb); | |
1307 | ||
1308 | /* No socket. Drop packet silently, if checksum is wrong */ | |
1309 | if (udp_lib_checksum_complete(skb)) | |
1310 | goto csum_error; | |
1311 | ||
0283328e | 1312 | UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
1313 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
1314 | ||
1315 | /* | |
1316 | * Hmm. We got an UDP packet to a port to which we | |
1317 | * don't wanna listen. Ignore it. | |
1318 | */ | |
1319 | kfree_skb(skb); | |
1320 | return 0; | |
1321 | ||
1322 | short_packet: | |
673d57e7 | 1323 | LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
db8dac20 | 1324 | proto == IPPROTO_UDPLITE ? "-Lite" : "", |
673d57e7 | 1325 | &saddr, |
db8dac20 DM |
1326 | ntohs(uh->source), |
1327 | ulen, | |
1328 | skb->len, | |
673d57e7 | 1329 | &daddr, |
db8dac20 DM |
1330 | ntohs(uh->dest)); |
1331 | goto drop; | |
1332 | ||
1333 | csum_error: | |
1334 | /* | |
1335 | * RFC1122: OK. Discards the bad packet silently (as far as | |
1336 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
1337 | */ | |
673d57e7 | 1338 | LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
db8dac20 | 1339 | proto == IPPROTO_UDPLITE ? "-Lite" : "", |
673d57e7 | 1340 | &saddr, |
db8dac20 | 1341 | ntohs(uh->source), |
673d57e7 | 1342 | &daddr, |
db8dac20 DM |
1343 | ntohs(uh->dest), |
1344 | ulen); | |
1345 | drop: | |
0283328e | 1346 | UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
1347 | kfree_skb(skb); |
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | int udp_rcv(struct sk_buff *skb) | |
1352 | { | |
645ca708 | 1353 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
1354 | } |
1355 | ||
7d06b2e0 | 1356 | void udp_destroy_sock(struct sock *sk) |
db8dac20 DM |
1357 | { |
1358 | lock_sock(sk); | |
1359 | udp_flush_pending_frames(sk); | |
1360 | release_sock(sk); | |
db8dac20 DM |
1361 | } |
1362 | ||
1da177e4 LT |
1363 | /* |
1364 | * Socket option code for UDP | |
1365 | */ | |
4c0a6cb0 | 1366 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 1367 | char __user *optval, unsigned int optlen, |
4c0a6cb0 | 1368 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
1369 | { |
1370 | struct udp_sock *up = udp_sk(sk); | |
1371 | int val; | |
1372 | int err = 0; | |
b2bf1e26 | 1373 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 1374 | |
c482c568 | 1375 | if (optlen < sizeof(int)) |
1da177e4 LT |
1376 | return -EINVAL; |
1377 | ||
1378 | if (get_user(val, (int __user *)optval)) | |
1379 | return -EFAULT; | |
1380 | ||
6516c655 | 1381 | switch (optname) { |
1da177e4 LT |
1382 | case UDP_CORK: |
1383 | if (val != 0) { | |
1384 | up->corkflag = 1; | |
1385 | } else { | |
1386 | up->corkflag = 0; | |
1387 | lock_sock(sk); | |
4c0a6cb0 | 1388 | (*push_pending_frames)(sk); |
1da177e4 LT |
1389 | release_sock(sk); |
1390 | } | |
1391 | break; | |
e905a9ed | 1392 | |
1da177e4 LT |
1393 | case UDP_ENCAP: |
1394 | switch (val) { | |
1395 | case 0: | |
1396 | case UDP_ENCAP_ESPINUDP: | |
1397 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
067b207b JC |
1398 | up->encap_rcv = xfrm4_udp_encap_rcv; |
1399 | /* FALLTHROUGH */ | |
342f0234 | 1400 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 LT |
1401 | up->encap_type = val; |
1402 | break; | |
1403 | default: | |
1404 | err = -ENOPROTOOPT; | |
1405 | break; | |
1406 | } | |
1407 | break; | |
1408 | ||
ba4e58ec GR |
1409 | /* |
1410 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
1411 | */ | |
1412 | /* The sender sets actual checksum coverage length via this option. | |
1413 | * The case coverage > packet length is handled by send module. */ | |
1414 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 1415 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
1416 | return -ENOPROTOOPT; |
1417 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
1418 | val = 8; | |
47112e25 GR |
1419 | else if (val > USHORT_MAX) |
1420 | val = USHORT_MAX; | |
ba4e58ec GR |
1421 | up->pcslen = val; |
1422 | up->pcflag |= UDPLITE_SEND_CC; | |
1423 | break; | |
1424 | ||
e905a9ed YH |
1425 | /* The receiver specifies a minimum checksum coverage value. To make |
1426 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
1427 | * used, this again means full checksum coverage. */ |
1428 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 1429 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
1430 | return -ENOPROTOOPT; |
1431 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
1432 | val = 8; | |
47112e25 GR |
1433 | else if (val > USHORT_MAX) |
1434 | val = USHORT_MAX; | |
ba4e58ec GR |
1435 | up->pcrlen = val; |
1436 | up->pcflag |= UDPLITE_RECV_CC; | |
1437 | break; | |
1438 | ||
1da177e4 LT |
1439 | default: |
1440 | err = -ENOPROTOOPT; | |
1441 | break; | |
6516c655 | 1442 | } |
1da177e4 LT |
1443 | |
1444 | return err; | |
1445 | } | |
c482c568 | 1446 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 1447 | |
db8dac20 | 1448 | int udp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 1449 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
1450 | { |
1451 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1452 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1453 | udp_push_pending_frames); | |
1454 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
1455 | } | |
1456 | ||
1457 | #ifdef CONFIG_COMPAT | |
1458 | int compat_udp_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 1459 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
1460 | { |
1461 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1462 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1463 | udp_push_pending_frames); | |
1464 | return compat_ip_setsockopt(sk, level, optname, optval, optlen); | |
1465 | } | |
1466 | #endif | |
1467 | ||
4c0a6cb0 GR |
1468 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
1469 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
1470 | { |
1471 | struct udp_sock *up = udp_sk(sk); | |
1472 | int val, len; | |
1473 | ||
c482c568 | 1474 | if (get_user(len, optlen)) |
1da177e4 LT |
1475 | return -EFAULT; |
1476 | ||
1477 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 1478 | |
6516c655 | 1479 | if (len < 0) |
1da177e4 LT |
1480 | return -EINVAL; |
1481 | ||
6516c655 | 1482 | switch (optname) { |
1da177e4 LT |
1483 | case UDP_CORK: |
1484 | val = up->corkflag; | |
1485 | break; | |
1486 | ||
1487 | case UDP_ENCAP: | |
1488 | val = up->encap_type; | |
1489 | break; | |
1490 | ||
ba4e58ec GR |
1491 | /* The following two cannot be changed on UDP sockets, the return is |
1492 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
1493 | case UDPLITE_SEND_CSCOV: | |
1494 | val = up->pcslen; | |
1495 | break; | |
1496 | ||
1497 | case UDPLITE_RECV_CSCOV: | |
1498 | val = up->pcrlen; | |
1499 | break; | |
1500 | ||
1da177e4 LT |
1501 | default: |
1502 | return -ENOPROTOOPT; | |
6516c655 | 1503 | } |
1da177e4 | 1504 | |
6516c655 | 1505 | if (put_user(len, optlen)) |
e905a9ed | 1506 | return -EFAULT; |
c482c568 | 1507 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 1508 | return -EFAULT; |
e905a9ed | 1509 | return 0; |
1da177e4 | 1510 | } |
c482c568 | 1511 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 1512 | |
db8dac20 DM |
1513 | int udp_getsockopt(struct sock *sk, int level, int optname, |
1514 | char __user *optval, int __user *optlen) | |
1515 | { | |
1516 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1517 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1518 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
1519 | } | |
1520 | ||
1521 | #ifdef CONFIG_COMPAT | |
1522 | int compat_udp_getsockopt(struct sock *sk, int level, int optname, | |
1523 | char __user *optval, int __user *optlen) | |
1524 | { | |
1525 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1526 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1527 | return compat_ip_getsockopt(sk, level, optname, optval, optlen); | |
1528 | } | |
1529 | #endif | |
1da177e4 LT |
1530 | /** |
1531 | * udp_poll - wait for a UDP event. | |
1532 | * @file - file struct | |
1533 | * @sock - socket | |
1534 | * @wait - poll table | |
1535 | * | |
e905a9ed | 1536 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
1537 | * blocking sockets. If application is using a blocking fd |
1538 | * and a packet with checksum error is in the queue; | |
1539 | * then it could get return from select indicating data available | |
1540 | * but then block when reading it. Add special case code | |
1541 | * to work around these arguably broken applications. | |
1542 | */ | |
1543 | unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
1544 | { | |
1545 | unsigned int mask = datagram_poll(file, sock, wait); | |
1546 | struct sock *sk = sock->sk; | |
ba4e58ec GR |
1547 | int is_lite = IS_UDPLITE(sk); |
1548 | ||
1da177e4 | 1549 | /* Check for false positives due to checksum errors */ |
c482c568 ED |
1550 | if ((mask & POLLRDNORM) && |
1551 | !(file->f_flags & O_NONBLOCK) && | |
1552 | !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1da177e4 LT |
1553 | struct sk_buff_head *rcvq = &sk->sk_receive_queue; |
1554 | struct sk_buff *skb; | |
1555 | ||
208d8984 | 1556 | spin_lock_bh(&rcvq->lock); |
759e5d00 HX |
1557 | while ((skb = skb_peek(rcvq)) != NULL && |
1558 | udp_lib_checksum_complete(skb)) { | |
0283328e PE |
1559 | UDP_INC_STATS_BH(sock_net(sk), |
1560 | UDP_MIB_INERRORS, is_lite); | |
759e5d00 HX |
1561 | __skb_unlink(skb, rcvq); |
1562 | kfree_skb(skb); | |
1da177e4 | 1563 | } |
208d8984 | 1564 | spin_unlock_bh(&rcvq->lock); |
1da177e4 LT |
1565 | |
1566 | /* nothing to see, move along */ | |
1567 | if (skb == NULL) | |
1568 | mask &= ~(POLLIN | POLLRDNORM); | |
1569 | } | |
1570 | ||
1571 | return mask; | |
e905a9ed | 1572 | |
1da177e4 | 1573 | } |
c482c568 | 1574 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 1575 | |
db8dac20 DM |
1576 | struct proto udp_prot = { |
1577 | .name = "UDP", | |
1578 | .owner = THIS_MODULE, | |
1579 | .close = udp_lib_close, | |
1580 | .connect = ip4_datagram_connect, | |
1581 | .disconnect = udp_disconnect, | |
1582 | .ioctl = udp_ioctl, | |
1583 | .destroy = udp_destroy_sock, | |
1584 | .setsockopt = udp_setsockopt, | |
1585 | .getsockopt = udp_getsockopt, | |
1586 | .sendmsg = udp_sendmsg, | |
1587 | .recvmsg = udp_recvmsg, | |
1588 | .sendpage = udp_sendpage, | |
93821778 | 1589 | .backlog_rcv = __udp_queue_rcv_skb, |
db8dac20 DM |
1590 | .hash = udp_lib_hash, |
1591 | .unhash = udp_lib_unhash, | |
1592 | .get_port = udp_v4_get_port, | |
1593 | .memory_allocated = &udp_memory_allocated, | |
1594 | .sysctl_mem = sysctl_udp_mem, | |
1595 | .sysctl_wmem = &sysctl_udp_wmem_min, | |
1596 | .sysctl_rmem = &sysctl_udp_rmem_min, | |
1597 | .obj_size = sizeof(struct udp_sock), | |
271b72c7 | 1598 | .slab_flags = SLAB_DESTROY_BY_RCU, |
645ca708 | 1599 | .h.udp_table = &udp_table, |
db8dac20 DM |
1600 | #ifdef CONFIG_COMPAT |
1601 | .compat_setsockopt = compat_udp_setsockopt, | |
1602 | .compat_getsockopt = compat_udp_getsockopt, | |
1603 | #endif | |
db8dac20 | 1604 | }; |
c482c568 | 1605 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
1606 | |
1607 | /* ------------------------------------------------------------------------ */ | |
1608 | #ifdef CONFIG_PROC_FS | |
1609 | ||
645ca708 | 1610 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
1611 | { |
1612 | struct sock *sk; | |
1613 | struct udp_iter_state *state = seq->private; | |
6f191efe | 1614 | struct net *net = seq_file_net(seq); |
1da177e4 | 1615 | |
f86dcc5a ED |
1616 | for (state->bucket = start; state->bucket <= state->udp_table->mask; |
1617 | ++state->bucket) { | |
88ab1932 | 1618 | struct hlist_nulls_node *node; |
645ca708 | 1619 | struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; |
f86dcc5a ED |
1620 | |
1621 | if (hlist_nulls_empty(&hslot->head)) | |
1622 | continue; | |
1623 | ||
645ca708 | 1624 | spin_lock_bh(&hslot->lock); |
88ab1932 | 1625 | sk_nulls_for_each(sk, node, &hslot->head) { |
878628fb | 1626 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 1627 | continue; |
1da177e4 LT |
1628 | if (sk->sk_family == state->family) |
1629 | goto found; | |
1630 | } | |
645ca708 | 1631 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
1632 | } |
1633 | sk = NULL; | |
1634 | found: | |
1635 | return sk; | |
1636 | } | |
1637 | ||
1638 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
1639 | { | |
1640 | struct udp_iter_state *state = seq->private; | |
6f191efe | 1641 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1642 | |
1643 | do { | |
88ab1932 | 1644 | sk = sk_nulls_next(sk); |
878628fb | 1645 | } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); |
1da177e4 | 1646 | |
645ca708 | 1647 | if (!sk) { |
f86dcc5a | 1648 | if (state->bucket <= state->udp_table->mask) |
30842f29 | 1649 | spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); |
645ca708 | 1650 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
1651 | } |
1652 | return sk; | |
1653 | } | |
1654 | ||
1655 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
1656 | { | |
645ca708 | 1657 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
1658 | |
1659 | if (sk) | |
6516c655 | 1660 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
1661 | --pos; |
1662 | return pos ? NULL : sk; | |
1663 | } | |
1664 | ||
1665 | static void *udp_seq_start(struct seq_file *seq, loff_t *pos) | |
1666 | { | |
30842f29 | 1667 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 1668 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 1669 | |
b50660f1 | 1670 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 LT |
1671 | } |
1672 | ||
1673 | static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1674 | { | |
1675 | struct sock *sk; | |
1676 | ||
b50660f1 | 1677 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
1678 | sk = udp_get_idx(seq, 0); |
1679 | else | |
1680 | sk = udp_get_next(seq, v); | |
1681 | ||
1682 | ++*pos; | |
1683 | return sk; | |
1684 | } | |
1685 | ||
1686 | static void udp_seq_stop(struct seq_file *seq, void *v) | |
1687 | { | |
645ca708 ED |
1688 | struct udp_iter_state *state = seq->private; |
1689 | ||
f86dcc5a | 1690 | if (state->bucket <= state->udp_table->mask) |
645ca708 | 1691 | spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); |
1da177e4 LT |
1692 | } |
1693 | ||
1694 | static int udp_seq_open(struct inode *inode, struct file *file) | |
1695 | { | |
1696 | struct udp_seq_afinfo *afinfo = PDE(inode)->data; | |
a2be75c1 DL |
1697 | struct udp_iter_state *s; |
1698 | int err; | |
a91275ef | 1699 | |
a2be75c1 DL |
1700 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
1701 | sizeof(struct udp_iter_state)); | |
1702 | if (err < 0) | |
1703 | return err; | |
a91275ef | 1704 | |
a2be75c1 | 1705 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 1706 | s->family = afinfo->family; |
645ca708 | 1707 | s->udp_table = afinfo->udp_table; |
a2be75c1 | 1708 | return err; |
a91275ef DL |
1709 | } |
1710 | ||
1da177e4 | 1711 | /* ------------------------------------------------------------------------ */ |
0c96d8c5 | 1712 | int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) |
1da177e4 LT |
1713 | { |
1714 | struct proc_dir_entry *p; | |
1715 | int rc = 0; | |
1716 | ||
3ba9441b DL |
1717 | afinfo->seq_fops.open = udp_seq_open; |
1718 | afinfo->seq_fops.read = seq_read; | |
1719 | afinfo->seq_fops.llseek = seq_lseek; | |
1720 | afinfo->seq_fops.release = seq_release_net; | |
1da177e4 | 1721 | |
dda61925 DL |
1722 | afinfo->seq_ops.start = udp_seq_start; |
1723 | afinfo->seq_ops.next = udp_seq_next; | |
1724 | afinfo->seq_ops.stop = udp_seq_stop; | |
1725 | ||
84841c3c DL |
1726 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
1727 | &afinfo->seq_fops, afinfo); | |
1728 | if (!p) | |
1da177e4 LT |
1729 | rc = -ENOMEM; |
1730 | return rc; | |
1731 | } | |
c482c568 | 1732 | EXPORT_SYMBOL(udp_proc_register); |
1da177e4 | 1733 | |
0c96d8c5 | 1734 | void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) |
1da177e4 | 1735 | { |
0c96d8c5 | 1736 | proc_net_remove(net, afinfo->name); |
1da177e4 | 1737 | } |
c482c568 | 1738 | EXPORT_SYMBOL(udp_proc_unregister); |
db8dac20 DM |
1739 | |
1740 | /* ------------------------------------------------------------------------ */ | |
5e659e4c PE |
1741 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
1742 | int bucket, int *len) | |
db8dac20 DM |
1743 | { |
1744 | struct inet_sock *inet = inet_sk(sp); | |
1745 | __be32 dest = inet->daddr; | |
1746 | __be32 src = inet->rcv_saddr; | |
1747 | __u16 destp = ntohs(inet->dport); | |
1748 | __u16 srcp = ntohs(inet->sport); | |
1749 | ||
f86dcc5a | 1750 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
cb61cb9b | 1751 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", |
db8dac20 | 1752 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 ED |
1753 | sk_wmem_alloc_get(sp), |
1754 | sk_rmem_alloc_get(sp), | |
db8dac20 | 1755 | 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), |
cb61cb9b ED |
1756 | atomic_read(&sp->sk_refcnt), sp, |
1757 | atomic_read(&sp->sk_drops), len); | |
db8dac20 DM |
1758 | } |
1759 | ||
1760 | int udp4_seq_show(struct seq_file *seq, void *v) | |
1761 | { | |
1762 | if (v == SEQ_START_TOKEN) | |
1763 | seq_printf(seq, "%-127s\n", | |
1764 | " sl local_address rem_address st tx_queue " | |
1765 | "rx_queue tr tm->when retrnsmt uid timeout " | |
cb61cb9b | 1766 | "inode ref pointer drops"); |
db8dac20 | 1767 | else { |
db8dac20 | 1768 | struct udp_iter_state *state = seq->private; |
5e659e4c | 1769 | int len; |
db8dac20 | 1770 | |
5e659e4c | 1771 | udp4_format_sock(v, seq, state->bucket, &len); |
c482c568 | 1772 | seq_printf(seq, "%*s\n", 127 - len, ""); |
db8dac20 DM |
1773 | } |
1774 | return 0; | |
1775 | } | |
1776 | ||
1777 | /* ------------------------------------------------------------------------ */ | |
db8dac20 | 1778 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 DM |
1779 | .name = "udp", |
1780 | .family = AF_INET, | |
645ca708 | 1781 | .udp_table = &udp_table, |
4ad96d39 DL |
1782 | .seq_fops = { |
1783 | .owner = THIS_MODULE, | |
1784 | }, | |
dda61925 DL |
1785 | .seq_ops = { |
1786 | .show = udp4_seq_show, | |
1787 | }, | |
db8dac20 DM |
1788 | }; |
1789 | ||
15439feb PE |
1790 | static int udp4_proc_init_net(struct net *net) |
1791 | { | |
1792 | return udp_proc_register(net, &udp4_seq_afinfo); | |
1793 | } | |
1794 | ||
1795 | static void udp4_proc_exit_net(struct net *net) | |
1796 | { | |
1797 | udp_proc_unregister(net, &udp4_seq_afinfo); | |
1798 | } | |
1799 | ||
1800 | static struct pernet_operations udp4_net_ops = { | |
1801 | .init = udp4_proc_init_net, | |
1802 | .exit = udp4_proc_exit_net, | |
1803 | }; | |
1804 | ||
db8dac20 DM |
1805 | int __init udp4_proc_init(void) |
1806 | { | |
15439feb | 1807 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
1808 | } |
1809 | ||
1810 | void udp4_proc_exit(void) | |
1811 | { | |
15439feb | 1812 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 1813 | } |
1da177e4 LT |
1814 | #endif /* CONFIG_PROC_FS */ |
1815 | ||
f86dcc5a ED |
1816 | static __initdata unsigned long uhash_entries; |
1817 | static int __init set_uhash_entries(char *str) | |
645ca708 | 1818 | { |
f86dcc5a ED |
1819 | if (!str) |
1820 | return 0; | |
1821 | uhash_entries = simple_strtoul(str, &str, 0); | |
1822 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) | |
1823 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
1824 | return 1; | |
1825 | } | |
1826 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 1827 | |
f86dcc5a ED |
1828 | void __init udp_table_init(struct udp_table *table, const char *name) |
1829 | { | |
1830 | unsigned int i; | |
1831 | ||
1832 | if (!CONFIG_BASE_SMALL) | |
1833 | table->hash = alloc_large_system_hash(name, | |
1834 | sizeof(struct udp_hslot), | |
1835 | uhash_entries, | |
1836 | 21, /* one slot per 2 MB */ | |
1837 | 0, | |
1838 | &table->log, | |
1839 | &table->mask, | |
1840 | 64 * 1024); | |
1841 | /* | |
1842 | * Make sure hash table has the minimum size | |
1843 | */ | |
1844 | if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { | |
1845 | table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * | |
1846 | sizeof(struct udp_hslot), GFP_KERNEL); | |
1847 | if (!table->hash) | |
1848 | panic(name); | |
1849 | table->log = ilog2(UDP_HTABLE_SIZE_MIN); | |
1850 | table->mask = UDP_HTABLE_SIZE_MIN - 1; | |
1851 | } | |
1852 | for (i = 0; i <= table->mask; i++) { | |
88ab1932 | 1853 | INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); |
645ca708 ED |
1854 | spin_lock_init(&table->hash[i].lock); |
1855 | } | |
1856 | } | |
1857 | ||
95766fff HA |
1858 | void __init udp_init(void) |
1859 | { | |
8203efb3 | 1860 | unsigned long nr_pages, limit; |
95766fff | 1861 | |
f86dcc5a | 1862 | udp_table_init(&udp_table, "UDP"); |
95766fff HA |
1863 | /* Set the pressure threshold up by the same strategy of TCP. It is a |
1864 | * fraction of global memory that is up to 1/2 at 256 MB, decreasing | |
1865 | * toward zero with the amount of memory, with a floor of 128 pages. | |
1866 | */ | |
8203efb3 ED |
1867 | nr_pages = totalram_pages - totalhigh_pages; |
1868 | limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); | |
1869 | limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); | |
95766fff HA |
1870 | limit = max(limit, 128UL); |
1871 | sysctl_udp_mem[0] = limit / 4 * 3; | |
1872 | sysctl_udp_mem[1] = limit; | |
1873 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
1874 | ||
1875 | sysctl_udp_rmem_min = SK_MEM_QUANTUM; | |
1876 | sysctl_udp_wmem_min = SK_MEM_QUANTUM; | |
1877 | } | |
1878 | ||
d7ca4cc0 SS |
1879 | int udp4_ufo_send_check(struct sk_buff *skb) |
1880 | { | |
1881 | const struct iphdr *iph; | |
1882 | struct udphdr *uh; | |
1883 | ||
1884 | if (!pskb_may_pull(skb, sizeof(*uh))) | |
1885 | return -EINVAL; | |
1886 | ||
1887 | iph = ip_hdr(skb); | |
1888 | uh = udp_hdr(skb); | |
1889 | ||
1890 | uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, | |
1891 | IPPROTO_UDP, 0); | |
1892 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
1893 | skb->csum_offset = offsetof(struct udphdr, check); | |
1894 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1895 | return 0; | |
1896 | } | |
1897 | ||
1898 | struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features) | |
1899 | { | |
1900 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
1901 | unsigned int mss; | |
1902 | int offset; | |
1903 | __wsum csum; | |
1904 | ||
1905 | mss = skb_shinfo(skb)->gso_size; | |
1906 | if (unlikely(skb->len <= mss)) | |
1907 | goto out; | |
1908 | ||
1909 | if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { | |
1910 | /* Packet is from an untrusted source, reset gso_segs. */ | |
1911 | int type = skb_shinfo(skb)->gso_type; | |
1912 | ||
1913 | if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || | |
1914 | !(type & (SKB_GSO_UDP)))) | |
1915 | goto out; | |
1916 | ||
1917 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); | |
1918 | ||
1919 | segs = NULL; | |
1920 | goto out; | |
1921 | } | |
1922 | ||
1923 | /* Do software UFO. Complete and fill in the UDP checksum as HW cannot | |
1924 | * do checksum of UDP packets sent as multiple IP fragments. | |
1925 | */ | |
1926 | offset = skb->csum_start - skb_headroom(skb); | |
c482c568 | 1927 | csum = skb_checksum(skb, offset, skb->len - offset, 0); |
d7ca4cc0 SS |
1928 | offset += skb->csum_offset; |
1929 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); | |
1930 | skb->ip_summed = CHECKSUM_NONE; | |
1931 | ||
1932 | /* Fragment the skb. IP headers of the fragments are updated in | |
1933 | * inet_gso_segment() | |
1934 | */ | |
1935 | segs = skb_segment(skb, features); | |
1936 | out: | |
1937 | return segs; | |
1938 | } | |
1939 |