Print this page
OS-4647 lx fails to mount nfs share - Transport endpoint is already connected
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/inet/udp/udp.c
+++ new/usr/src/uts/common/inet/udp/udp.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
24 24 * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
25 25 */
26 26 /* Copyright (c) 1990 Mentat Inc. */
27 27
28 28 #include <sys/sysmacros.h>
29 29 #include <sys/types.h>
30 30 #include <sys/stream.h>
31 31 #include <sys/stropts.h>
32 32 #include <sys/strlog.h>
33 33 #include <sys/strsun.h>
34 34 #define _SUN_TPI_VERSION 2
35 35 #include <sys/tihdr.h>
36 36 #include <sys/timod.h>
37 37 #include <sys/ddi.h>
38 38 #include <sys/sunddi.h>
39 39 #include <sys/strsubr.h>
40 40 #include <sys/suntpi.h>
41 41 #include <sys/xti_inet.h>
42 42 #include <sys/kmem.h>
43 43 #include <sys/cred_impl.h>
44 44 #include <sys/policy.h>
45 45 #include <sys/priv.h>
46 46 #include <sys/ucred.h>
47 47 #include <sys/zone.h>
48 48
49 49 #include <sys/socket.h>
50 50 #include <sys/socketvar.h>
51 51 #include <sys/sockio.h>
52 52 #include <sys/vtrace.h>
53 53 #include <sys/sdt.h>
54 54 #include <sys/debug.h>
55 55 #include <sys/isa_defs.h>
56 56 #include <sys/random.h>
57 57 #include <netinet/in.h>
58 58 #include <netinet/ip6.h>
59 59 #include <netinet/icmp6.h>
60 60 #include <netinet/udp.h>
61 61
62 62 #include <inet/common.h>
63 63 #include <inet/ip.h>
64 64 #include <inet/ip_impl.h>
65 65 #include <inet/ipsec_impl.h>
66 66 #include <inet/ip6.h>
67 67 #include <inet/ip_ire.h>
68 68 #include <inet/ip_if.h>
69 69 #include <inet/ip_multi.h>
70 70 #include <inet/ip_ndp.h>
71 71 #include <inet/proto_set.h>
72 72 #include <inet/mib2.h>
73 73 #include <inet/optcom.h>
74 74 #include <inet/snmpcom.h>
75 75 #include <inet/kstatcom.h>
76 76 #include <inet/ipclassifier.h>
77 77 #include <sys/squeue_impl.h>
78 78 #include <inet/ipnet.h>
79 79 #include <sys/ethernet.h>
80 80
81 81 #include <sys/tsol/label.h>
82 82 #include <sys/tsol/tnet.h>
83 83 #include <rpc/pmap_prot.h>
84 84
85 85 #include <inet/udp_impl.h>
86 86
87 87 /*
88 88 * Synchronization notes:
89 89 *
90 90 * UDP is MT and uses the usual kernel synchronization primitives. There are 2
91 91 * locks, the fanout lock (uf_lock) and conn_lock. conn_lock
92 92 * protects the contents of the udp_t. uf_lock protects the address and the
93 93 * fanout information.
94 94 * The lock order is conn_lock -> uf_lock.
95 95 *
96 96 * The fanout lock uf_lock:
97 97 * When a UDP endpoint is bound to a local port, it is inserted into
98 98 * a bind hash list. The list consists of an array of udp_fanout_t buckets.
99 99 * The size of the array is controlled by the udp_bind_fanout_size variable.
100 100 * This variable can be changed in /etc/system if the default value is
101 101 * not large enough. Each bind hash bucket is protected by a per bucket
102 102 * lock. It protects the udp_bind_hash and udp_ptpbhn fields in the udp_t
103 103 * structure and a few other fields in the udp_t. A UDP endpoint is removed
104 104 * from the bind hash list only when it is being unbound or being closed.
105 105 * The per bucket lock also protects a UDP endpoint's state changes.
106 106 *
107 107 * Plumbing notes:
108 108 * UDP is always a device driver. For compatibility with mibopen() code
109 109 * it is possible to I_PUSH "udp", but that results in pushing a passthrough
110 110 * dummy module.
111 111 *
112 112 * The above implies that we don't support any intermediate module to
113 113 * reside in between /dev/ip and udp -- in fact, we never supported such
114 114 * scenario in the past as the inter-layer communication semantics have
115 115 * always been private.
116 116 */
117 117
118 118 /* For /etc/system control */
119 119 uint_t udp_bind_fanout_size = UDP_BIND_FANOUT_SIZE;
120 120
121 121 static void udp_addr_req(queue_t *q, mblk_t *mp);
122 122 static void udp_tpi_bind(queue_t *q, mblk_t *mp);
123 123 static void udp_bind_hash_insert(udp_fanout_t *uf, udp_t *udp);
124 124 static void udp_bind_hash_remove(udp_t *udp, boolean_t caller_holds_lock);
125 125 static int udp_build_hdr_template(conn_t *, const in6_addr_t *,
126 126 const in6_addr_t *, in_port_t, uint32_t);
127 127 static void udp_capability_req(queue_t *q, mblk_t *mp);
128 128 static int udp_tpi_close(queue_t *q, int flags);
129 129 static void udp_close_free(conn_t *);
130 130 static void udp_tpi_connect(queue_t *q, mblk_t *mp);
131 131 static void udp_tpi_disconnect(queue_t *q, mblk_t *mp);
132 132 static void udp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error,
133 133 int sys_error);
134 134 static void udp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
135 135 t_scalar_t tlierr, int sys_error);
136 136 static int udp_extra_priv_ports_get(queue_t *q, mblk_t *mp, caddr_t cp,
137 137 cred_t *cr);
138 138 static int udp_extra_priv_ports_add(queue_t *q, mblk_t *mp,
139 139 char *value, caddr_t cp, cred_t *cr);
140 140 static int udp_extra_priv_ports_del(queue_t *q, mblk_t *mp,
141 141 char *value, caddr_t cp, cred_t *cr);
142 142 static void udp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
143 143 static void udp_icmp_error_ipv6(conn_t *connp, mblk_t *mp,
144 144 ip_recv_attr_t *ira);
145 145 static void udp_info_req(queue_t *q, mblk_t *mp);
146 146 static void udp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
147 147 static void udp_lrput(queue_t *, mblk_t *);
148 148 static void udp_lwput(queue_t *, mblk_t *);
149 149 static int udp_open(queue_t *q, dev_t *devp, int flag, int sflag,
150 150 cred_t *credp, boolean_t isv6);
151 151 static int udp_openv4(queue_t *q, dev_t *devp, int flag, int sflag,
152 152 cred_t *credp);
153 153 static int udp_openv6(queue_t *q, dev_t *devp, int flag, int sflag,
154 154 cred_t *credp);
155 155 static boolean_t udp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name);
156 156 int udp_opt_set(conn_t *connp, uint_t optset_context,
157 157 int level, int name, uint_t inlen,
158 158 uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
159 159 void *thisdg_attrs, cred_t *cr);
160 160 int udp_opt_get(conn_t *connp, int level, int name,
161 161 uchar_t *ptr);
162 162 static int udp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr,
163 163 pid_t pid);
164 164 static int udp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr,
165 165 pid_t pid, ip_xmit_attr_t *ixa);
166 166 static int udp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin,
167 167 sin6_t *sin6, ushort_t ipversion, cred_t *cr, pid_t,
168 168 ip_xmit_attr_t *ixa);
169 169 static mblk_t *udp_prepend_hdr(conn_t *, ip_xmit_attr_t *, const ip_pkt_t *,
170 170 const in6_addr_t *, const in6_addr_t *, in_port_t, uint32_t, mblk_t *,
171 171 int *);
172 172 static mblk_t *udp_prepend_header_template(conn_t *, ip_xmit_attr_t *,
173 173 mblk_t *, const in6_addr_t *, in_port_t, uint32_t, int *);
174 174 static void udp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err);
175 175 static void udp_ud_err_connected(conn_t *, t_scalar_t);
176 176 static void udp_tpi_unbind(queue_t *q, mblk_t *mp);
177 177 static in_port_t udp_update_next_port(udp_t *udp, in_port_t port,
178 178 boolean_t random);
179 179 static void udp_wput_other(queue_t *q, mblk_t *mp);
180 180 static void udp_wput_iocdata(queue_t *q, mblk_t *mp);
181 181 static void udp_wput_fallback(queue_t *q, mblk_t *mp);
182 182 static size_t udp_set_rcv_hiwat(udp_t *udp, size_t size);
183 183
184 184 static void *udp_stack_init(netstackid_t stackid, netstack_t *ns);
185 185 static void udp_stack_fini(netstackid_t stackid, void *arg);
186 186
187 187 /* Common routines for TPI and socket module */
188 188 static void udp_ulp_recv(conn_t *, mblk_t *, uint_t, ip_recv_attr_t *);
189 189
190 190 /* Common routine for TPI and socket module */
191 191 static conn_t *udp_do_open(cred_t *, boolean_t, int, int *);
192 192 static void udp_do_close(conn_t *);
193 193 static int udp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
194 194 boolean_t);
195 195 static int udp_do_unbind(conn_t *);
196 196
197 197 int udp_getsockname(sock_lower_handle_t,
198 198 struct sockaddr *, socklen_t *, cred_t *);
199 199 int udp_getpeername(sock_lower_handle_t,
200 200 struct sockaddr *, socklen_t *, cred_t *);
201 201 static int udp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
202 202 cred_t *, pid_t);
203 203
204 204 #pragma inline(udp_output_connected, udp_output_newdst, udp_output_lastdst)
205 205
206 206 /*
207 207 * Checks if the given destination addr/port is allowed out.
208 208 * If allowed, registers the (dest_addr/port, node_ID) mapping at Cluster.
209 209 * Called for each connect() and for sendto()/sendmsg() to a different
210 210 * destination.
211 211 * For connect(), called in udp_connect().
212 212 * For sendto()/sendmsg(), called in udp_output_newdst().
213 213 *
214 214 * This macro assumes that the cl_inet_connect2 hook is not NULL.
215 215 * Please check this before calling this macro.
216 216 *
217 217 * void
218 218 * CL_INET_UDP_CONNECT(conn_t cp, udp_t *udp, boolean_t is_outgoing,
219 219 * in6_addr_t *faddrp, in_port_t (or uint16_t) fport, int err);
220 220 */
221 221 #define CL_INET_UDP_CONNECT(cp, is_outgoing, faddrp, fport, err) { \
222 222 (err) = 0; \
223 223 /* \
224 224 * Running in cluster mode - check and register active \
225 225 * "connection" information \
226 226 */ \
227 227 if ((cp)->conn_ipversion == IPV4_VERSION) \
228 228 (err) = (*cl_inet_connect2)( \
229 229 (cp)->conn_netstack->netstack_stackid, \
230 230 IPPROTO_UDP, is_outgoing, AF_INET, \
231 231 (uint8_t *)&((cp)->conn_laddr_v4), \
232 232 (cp)->conn_lport, \
233 233 (uint8_t *)&(V4_PART_OF_V6(*faddrp)), \
234 234 (in_port_t)(fport), NULL); \
235 235 else \
236 236 (err) = (*cl_inet_connect2)( \
237 237 (cp)->conn_netstack->netstack_stackid, \
238 238 IPPROTO_UDP, is_outgoing, AF_INET6, \
239 239 (uint8_t *)&((cp)->conn_laddr_v6), \
240 240 (cp)->conn_lport, \
241 241 (uint8_t *)(faddrp), (in_port_t)(fport), NULL); \
242 242 }
243 243
244 244 static struct module_info udp_mod_info = {
245 245 UDP_MOD_ID, UDP_MOD_NAME, 1, INFPSZ, UDP_RECV_HIWATER, UDP_RECV_LOWATER
246 246 };
247 247
248 248 /*
249 249 * Entry points for UDP as a device.
250 250 * We have separate open functions for the /dev/udp and /dev/udp6 devices.
251 251 */
252 252 static struct qinit udp_rinitv4 = {
253 253 NULL, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info, NULL
254 254 };
255 255
256 256 static struct qinit udp_rinitv6 = {
257 257 NULL, NULL, udp_openv6, udp_tpi_close, NULL, &udp_mod_info, NULL
258 258 };
259 259
260 260 static struct qinit udp_winit = {
261 261 (pfi_t)udp_wput, (pfi_t)ip_wsrv, NULL, NULL, NULL, &udp_mod_info
262 262 };
263 263
264 264 /* UDP entry point during fallback */
265 265 struct qinit udp_fallback_sock_winit = {
266 266 (pfi_t)udp_wput_fallback, NULL, NULL, NULL, NULL, &udp_mod_info
267 267 };
268 268
269 269 /*
270 270 * UDP needs to handle I_LINK and I_PLINK since ifconfig
271 271 * likes to use it as a place to hang the various streams.
272 272 */
273 273 static struct qinit udp_lrinit = {
274 274 (pfi_t)udp_lrput, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info
275 275 };
276 276
277 277 static struct qinit udp_lwinit = {
278 278 (pfi_t)udp_lwput, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info
279 279 };
280 280
281 281 /* For AF_INET aka /dev/udp */
282 282 struct streamtab udpinfov4 = {
283 283 &udp_rinitv4, &udp_winit, &udp_lrinit, &udp_lwinit
284 284 };
285 285
286 286 /* For AF_INET6 aka /dev/udp6 */
287 287 struct streamtab udpinfov6 = {
288 288 &udp_rinitv6, &udp_winit, &udp_lrinit, &udp_lwinit
289 289 };
290 290
291 291 #define UDP_MAXPACKET_IPV4 (IP_MAXPACKET - UDPH_SIZE - IP_SIMPLE_HDR_LENGTH)
292 292
293 293 /* Default structure copied into T_INFO_ACK messages */
294 294 static struct T_info_ack udp_g_t_info_ack_ipv4 = {
295 295 T_INFO_ACK,
296 296 UDP_MAXPACKET_IPV4, /* TSDU_size. Excl. headers */
297 297 T_INVALID, /* ETSU_size. udp does not support expedited data. */
298 298 T_INVALID, /* CDATA_size. udp does not support connect data. */
299 299 T_INVALID, /* DDATA_size. udp does not support disconnect data. */
300 300 sizeof (sin_t), /* ADDR_size. */
301 301 0, /* OPT_size - not initialized here */
302 302 UDP_MAXPACKET_IPV4, /* TIDU_size. Excl. headers */
303 303 T_CLTS, /* SERV_type. udp supports connection-less. */
304 304 TS_UNBND, /* CURRENT_state. This is set from udp_state. */
305 305 (XPG4_1|SENDZERO) /* PROVIDER_flag */
306 306 };
307 307
308 308 #define UDP_MAXPACKET_IPV6 (IP_MAXPACKET - UDPH_SIZE - IPV6_HDR_LEN)
309 309
310 310 static struct T_info_ack udp_g_t_info_ack_ipv6 = {
311 311 T_INFO_ACK,
312 312 UDP_MAXPACKET_IPV6, /* TSDU_size. Excl. headers */
313 313 T_INVALID, /* ETSU_size. udp does not support expedited data. */
314 314 T_INVALID, /* CDATA_size. udp does not support connect data. */
315 315 T_INVALID, /* DDATA_size. udp does not support disconnect data. */
316 316 sizeof (sin6_t), /* ADDR_size. */
317 317 0, /* OPT_size - not initialized here */
318 318 UDP_MAXPACKET_IPV6, /* TIDU_size. Excl. headers */
319 319 T_CLTS, /* SERV_type. udp supports connection-less. */
320 320 TS_UNBND, /* CURRENT_state. This is set from udp_state. */
321 321 (XPG4_1|SENDZERO) /* PROVIDER_flag */
322 322 };
323 323
324 324 /*
325 325 * UDP tunables related declarations. Definitions are in udp_tunables.c
326 326 */
327 327 extern mod_prop_info_t udp_propinfo_tbl[];
328 328 extern int udp_propinfo_count;
329 329
330 330 /* Setable in /etc/system */
331 331 /* If set to 0, pick ephemeral port sequentially; otherwise randomly. */
332 332 uint32_t udp_random_anon_port = 1;
333 333
334 334 /*
335 335 * Hook functions to enable cluster networking.
336 336 * On non-clustered systems these vectors must always be NULL
337 337 */
338 338
339 339 void (*cl_inet_bind)(netstackid_t stack_id, uchar_t protocol,
340 340 sa_family_t addr_family, uint8_t *laddrp, in_port_t lport,
341 341 void *args) = NULL;
342 342 void (*cl_inet_unbind)(netstackid_t stack_id, uint8_t protocol,
343 343 sa_family_t addr_family, uint8_t *laddrp, in_port_t lport,
344 344 void *args) = NULL;
345 345
346 346 typedef union T_primitives *t_primp_t;
347 347
348 348 /*
349 349 * Return the next anonymous port in the privileged port range for
350 350 * bind checking.
351 351 *
352 352 * Trusted Extension (TX) notes: TX allows administrator to mark or
353 353 * reserve ports as Multilevel ports (MLP). MLP has special function
354 354 * on TX systems. Once a port is made MLP, it's not available as
355 355 * ordinary port. This creates "holes" in the port name space. It
356 356 * may be necessary to skip the "holes" find a suitable anon port.
357 357 */
358 358 static in_port_t
359 359 udp_get_next_priv_port(udp_t *udp)
360 360 {
361 361 static in_port_t next_priv_port = IPPORT_RESERVED - 1;
362 362 in_port_t nextport;
363 363 boolean_t restart = B_FALSE;
364 364 udp_stack_t *us = udp->udp_us;
365 365
366 366 retry:
367 367 if (next_priv_port < us->us_min_anonpriv_port ||
368 368 next_priv_port >= IPPORT_RESERVED) {
369 369 next_priv_port = IPPORT_RESERVED - 1;
370 370 if (restart)
371 371 return (0);
372 372 restart = B_TRUE;
373 373 }
374 374
375 375 if (is_system_labeled() &&
376 376 (nextport = tsol_next_port(crgetzone(udp->udp_connp->conn_cred),
377 377 next_priv_port, IPPROTO_UDP, B_FALSE)) != 0) {
378 378 next_priv_port = nextport;
379 379 goto retry;
380 380 }
381 381
382 382 return (next_priv_port--);
383 383 }
384 384
385 385 /*
386 386 * Hash list removal routine for udp_t structures.
387 387 */
388 388 static void
389 389 udp_bind_hash_remove(udp_t *udp, boolean_t caller_holds_lock)
390 390 {
391 391 udp_t *udpnext;
392 392 kmutex_t *lockp;
393 393 udp_stack_t *us = udp->udp_us;
394 394 conn_t *connp = udp->udp_connp;
395 395
396 396 if (udp->udp_ptpbhn == NULL)
397 397 return;
398 398
399 399 /*
400 400 * Extract the lock pointer in case there are concurrent
401 401 * hash_remove's for this instance.
402 402 */
403 403 ASSERT(connp->conn_lport != 0);
404 404 if (!caller_holds_lock) {
405 405 lockp = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
406 406 us->us_bind_fanout_size)].uf_lock;
407 407 ASSERT(lockp != NULL);
408 408 mutex_enter(lockp);
409 409 }
410 410 if (udp->udp_ptpbhn != NULL) {
411 411 udpnext = udp->udp_bind_hash;
412 412 if (udpnext != NULL) {
413 413 udpnext->udp_ptpbhn = udp->udp_ptpbhn;
414 414 udp->udp_bind_hash = NULL;
415 415 }
416 416 *udp->udp_ptpbhn = udpnext;
417 417 udp->udp_ptpbhn = NULL;
418 418 }
419 419 if (!caller_holds_lock) {
420 420 mutex_exit(lockp);
421 421 }
422 422 }
423 423
424 424 static void
425 425 udp_bind_hash_insert(udp_fanout_t *uf, udp_t *udp)
426 426 {
427 427 conn_t *connp = udp->udp_connp;
428 428 udp_t **udpp;
429 429 udp_t *udpnext;
430 430 conn_t *connext;
431 431
432 432 ASSERT(MUTEX_HELD(&uf->uf_lock));
433 433 ASSERT(udp->udp_ptpbhn == NULL);
434 434 udpp = &uf->uf_udp;
435 435 udpnext = udpp[0];
436 436 if (udpnext != NULL) {
437 437 /*
438 438 * If the new udp bound to the INADDR_ANY address
439 439 * and the first one in the list is not bound to
440 440 * INADDR_ANY we skip all entries until we find the
441 441 * first one bound to INADDR_ANY.
442 442 * This makes sure that applications binding to a
443 443 * specific address get preference over those binding to
444 444 * INADDR_ANY.
445 445 */
446 446 connext = udpnext->udp_connp;
447 447 if (V6_OR_V4_INADDR_ANY(connp->conn_bound_addr_v6) &&
448 448 !V6_OR_V4_INADDR_ANY(connext->conn_bound_addr_v6)) {
449 449 while ((udpnext = udpp[0]) != NULL &&
450 450 !V6_OR_V4_INADDR_ANY(connext->conn_bound_addr_v6)) {
451 451 udpp = &(udpnext->udp_bind_hash);
452 452 }
453 453 if (udpnext != NULL)
454 454 udpnext->udp_ptpbhn = &udp->udp_bind_hash;
455 455 } else {
456 456 udpnext->udp_ptpbhn = &udp->udp_bind_hash;
457 457 }
458 458 }
459 459 udp->udp_bind_hash = udpnext;
460 460 udp->udp_ptpbhn = udpp;
461 461 udpp[0] = udp;
462 462 }
463 463
464 464 /*
465 465 * This routine is called to handle each O_T_BIND_REQ/T_BIND_REQ message
466 466 * passed to udp_wput.
467 467 * It associates a port number and local address with the stream.
468 468 * It calls IP to verify the local IP address, and calls IP to insert
469 469 * the conn_t in the fanout table.
470 470 * If everything is ok it then sends the T_BIND_ACK back up.
471 471 *
472 472 * Note that UDP over IPv4 and IPv6 sockets can use the same port number
473 473 * without setting SO_REUSEADDR. This is needed so that they
474 474 * can be viewed as two independent transport protocols.
475 475 * However, anonymouns ports are allocated from the same range to avoid
476 476 * duplicating the us->us_next_port_to_try.
477 477 */
478 478 static void
479 479 udp_tpi_bind(queue_t *q, mblk_t *mp)
480 480 {
481 481 sin_t *sin;
482 482 sin6_t *sin6;
483 483 mblk_t *mp1;
484 484 struct T_bind_req *tbr;
485 485 conn_t *connp;
486 486 udp_t *udp;
487 487 int error;
488 488 struct sockaddr *sa;
489 489 cred_t *cr;
490 490
491 491 /*
492 492 * All Solaris components should pass a db_credp
493 493 * for this TPI message, hence we ASSERT.
494 494 * But in case there is some other M_PROTO that looks
495 495 * like a TPI message sent by some other kernel
496 496 * component, we check and return an error.
497 497 */
498 498 cr = msg_getcred(mp, NULL);
499 499 ASSERT(cr != NULL);
500 500 if (cr == NULL) {
501 501 udp_err_ack(q, mp, TSYSERR, EINVAL);
502 502 return;
503 503 }
504 504
505 505 connp = Q_TO_CONN(q);
506 506 udp = connp->conn_udp;
507 507 if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
508 508 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
509 509 "udp_bind: bad req, len %u",
510 510 (uint_t)(mp->b_wptr - mp->b_rptr));
511 511 udp_err_ack(q, mp, TPROTO, 0);
512 512 return;
513 513 }
514 514 if (udp->udp_state != TS_UNBND) {
515 515 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
516 516 "udp_bind: bad state, %u", udp->udp_state);
517 517 udp_err_ack(q, mp, TOUTSTATE, 0);
518 518 return;
519 519 }
520 520 /*
521 521 * Reallocate the message to make sure we have enough room for an
522 522 * address.
523 523 */
524 524 mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
525 525 if (mp1 == NULL) {
526 526 udp_err_ack(q, mp, TSYSERR, ENOMEM);
527 527 return;
528 528 }
529 529
530 530 mp = mp1;
531 531
532 532 /* Reset the message type in preparation for shipping it back. */
533 533 DB_TYPE(mp) = M_PCPROTO;
534 534
535 535 tbr = (struct T_bind_req *)mp->b_rptr;
536 536 switch (tbr->ADDR_length) {
537 537 case 0: /* Request for a generic port */
538 538 tbr->ADDR_offset = sizeof (struct T_bind_req);
539 539 if (connp->conn_family == AF_INET) {
540 540 tbr->ADDR_length = sizeof (sin_t);
541 541 sin = (sin_t *)&tbr[1];
542 542 *sin = sin_null;
543 543 sin->sin_family = AF_INET;
544 544 mp->b_wptr = (uchar_t *)&sin[1];
545 545 sa = (struct sockaddr *)sin;
546 546 } else {
547 547 ASSERT(connp->conn_family == AF_INET6);
548 548 tbr->ADDR_length = sizeof (sin6_t);
549 549 sin6 = (sin6_t *)&tbr[1];
550 550 *sin6 = sin6_null;
551 551 sin6->sin6_family = AF_INET6;
552 552 mp->b_wptr = (uchar_t *)&sin6[1];
553 553 sa = (struct sockaddr *)sin6;
554 554 }
555 555 break;
556 556
557 557 case sizeof (sin_t): /* Complete IPv4 address */
558 558 sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
559 559 sizeof (sin_t));
560 560 if (sa == NULL || !OK_32PTR((char *)sa)) {
561 561 udp_err_ack(q, mp, TSYSERR, EINVAL);
562 562 return;
563 563 }
564 564 if (connp->conn_family != AF_INET ||
565 565 sa->sa_family != AF_INET) {
566 566 udp_err_ack(q, mp, TSYSERR, EAFNOSUPPORT);
567 567 return;
568 568 }
569 569 break;
570 570
571 571 case sizeof (sin6_t): /* complete IPv6 address */
572 572 sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
573 573 sizeof (sin6_t));
574 574 if (sa == NULL || !OK_32PTR((char *)sa)) {
575 575 udp_err_ack(q, mp, TSYSERR, EINVAL);
576 576 return;
577 577 }
578 578 if (connp->conn_family != AF_INET6 ||
579 579 sa->sa_family != AF_INET6) {
580 580 udp_err_ack(q, mp, TSYSERR, EAFNOSUPPORT);
581 581 return;
582 582 }
583 583 break;
584 584
585 585 default: /* Invalid request */
586 586 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
587 587 "udp_bind: bad ADDR_length length %u", tbr->ADDR_length);
588 588 udp_err_ack(q, mp, TBADADDR, 0);
589 589 return;
590 590 }
591 591
592 592 error = udp_do_bind(connp, sa, tbr->ADDR_length, cr,
593 593 tbr->PRIM_type != O_T_BIND_REQ);
594 594
595 595 if (error != 0) {
596 596 if (error > 0) {
597 597 udp_err_ack(q, mp, TSYSERR, error);
598 598 } else {
599 599 udp_err_ack(q, mp, -error, 0);
600 600 }
601 601 } else {
602 602 tbr->PRIM_type = T_BIND_ACK;
603 603 qreply(q, mp);
604 604 }
605 605 }
606 606
607 607 /*
608 608 * This routine handles each T_CONN_REQ message passed to udp. It
609 609 * associates a default destination address with the stream.
610 610 *
611 611 * After various error checks are completed, udp_connect() lays
612 612 * the target address and port into the composite header template.
613 613 * Then we ask IP for information, including a source address if we didn't
614 614 * already have one. Finally we send up the T_OK_ACK reply message.
615 615 */
616 616 static void
617 617 udp_tpi_connect(queue_t *q, mblk_t *mp)
618 618 {
619 619 conn_t *connp = Q_TO_CONN(q);
620 620 int error;
621 621 socklen_t len;
622 622 struct sockaddr *sa;
623 623 struct T_conn_req *tcr;
624 624 cred_t *cr;
625 625 pid_t pid;
626 626 /*
627 627 * All Solaris components should pass a db_credp
628 628 * for this TPI message, hence we ASSERT.
629 629 * But in case there is some other M_PROTO that looks
630 630 * like a TPI message sent by some other kernel
631 631 * component, we check and return an error.
632 632 */
633 633 cr = msg_getcred(mp, &pid);
634 634 ASSERT(cr != NULL);
635 635 if (cr == NULL) {
636 636 udp_err_ack(q, mp, TSYSERR, EINVAL);
637 637 return;
638 638 }
639 639
640 640 tcr = (struct T_conn_req *)mp->b_rptr;
641 641
642 642 /* A bit of sanity checking */
643 643 if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_conn_req)) {
644 644 udp_err_ack(q, mp, TPROTO, 0);
645 645 return;
646 646 }
647 647
648 648 if (tcr->OPT_length != 0) {
649 649 udp_err_ack(q, mp, TBADOPT, 0);
650 650 return;
651 651 }
652 652
653 653 /*
654 654 * Determine packet type based on type of address passed in
655 655 * the request should contain an IPv4 or IPv6 address.
656 656 * Make sure that address family matches the type of
657 657 * family of the address passed down.
658 658 */
659 659 len = tcr->DEST_length;
660 660 switch (tcr->DEST_length) {
661 661 default:
662 662 udp_err_ack(q, mp, TBADADDR, 0);
663 663 return;
664 664
665 665 case sizeof (sin_t):
666 666 sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
667 667 sizeof (sin_t));
668 668 break;
669 669
670 670 case sizeof (sin6_t):
671 671 sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
672 672 sizeof (sin6_t));
673 673 break;
674 674 }
675 675
676 676 error = proto_verify_ip_addr(connp->conn_family, sa, len);
677 677 if (error != 0) {
678 678 udp_err_ack(q, mp, TSYSERR, error);
679 679 return;
680 680 }
681 681
682 682 error = udp_do_connect(connp, sa, len, cr, pid);
683 683 if (error != 0) {
684 684 if (error < 0)
685 685 udp_err_ack(q, mp, -error, 0);
686 686 else
687 687 udp_err_ack(q, mp, TSYSERR, error);
688 688 } else {
689 689 mblk_t *mp1;
690 690 /*
691 691 * We have to send a connection confirmation to
692 692 * keep TLI happy.
693 693 */
694 694 if (connp->conn_family == AF_INET) {
695 695 mp1 = mi_tpi_conn_con(NULL, (char *)sa,
696 696 sizeof (sin_t), NULL, 0);
697 697 } else {
698 698 mp1 = mi_tpi_conn_con(NULL, (char *)sa,
699 699 sizeof (sin6_t), NULL, 0);
700 700 }
701 701 if (mp1 == NULL) {
702 702 udp_err_ack(q, mp, TSYSERR, ENOMEM);
703 703 return;
704 704 }
705 705
706 706 /*
707 707 * Send ok_ack for T_CONN_REQ
708 708 */
709 709 mp = mi_tpi_ok_ack_alloc(mp);
710 710 if (mp == NULL) {
711 711 /* Unable to reuse the T_CONN_REQ for the ack. */
712 712 udp_err_ack_prim(q, mp1, T_CONN_REQ, TSYSERR, ENOMEM);
713 713 return;
714 714 }
715 715
716 716 putnext(connp->conn_rq, mp);
717 717 putnext(connp->conn_rq, mp1);
718 718 }
719 719 }
720 720
721 721 static int
722 722 udp_tpi_close(queue_t *q, int flags)
723 723 {
724 724 conn_t *connp;
725 725
726 726 if (flags & SO_FALLBACK) {
727 727 /*
728 728 * stream is being closed while in fallback
729 729 * simply free the resources that were allocated
730 730 */
731 731 inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
732 732 qprocsoff(q);
733 733 goto done;
734 734 }
735 735
736 736 connp = Q_TO_CONN(q);
737 737 udp_do_close(connp);
738 738 done:
739 739 q->q_ptr = WR(q)->q_ptr = NULL;
740 740 return (0);
741 741 }
742 742
743 743 static void
744 744 udp_close_free(conn_t *connp)
745 745 {
746 746 udp_t *udp = connp->conn_udp;
747 747
748 748 /* If there are any options associated with the stream, free them. */
749 749 if (udp->udp_recv_ipp.ipp_fields != 0)
750 750 ip_pkt_free(&udp->udp_recv_ipp);
751 751
752 752 /*
753 753 * Clear any fields which the kmem_cache constructor clears.
754 754 * Only udp_connp needs to be preserved.
755 755 * TBD: We should make this more efficient to avoid clearing
756 756 * everything.
757 757 */
758 758 ASSERT(udp->udp_connp == connp);
759 759 bzero(udp, sizeof (udp_t));
760 760 udp->udp_connp = connp;
761 761 }
762 762
763 763 static int
764 764 udp_do_disconnect(conn_t *connp)
765 765 {
766 766 udp_t *udp;
767 767 udp_fanout_t *udpf;
768 768 udp_stack_t *us;
769 769 int error;
770 770
771 771 udp = connp->conn_udp;
772 772 us = udp->udp_us;
773 773 mutex_enter(&connp->conn_lock);
774 774 if (udp->udp_state != TS_DATA_XFER) {
775 775 mutex_exit(&connp->conn_lock);
776 776 return (-TOUTSTATE);
777 777 }
778 778 udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
779 779 us->us_bind_fanout_size)];
780 780 mutex_enter(&udpf->uf_lock);
781 781 if (connp->conn_mcbc_bind)
782 782 connp->conn_saddr_v6 = ipv6_all_zeros;
783 783 else
784 784 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
785 785 connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
786 786 connp->conn_faddr_v6 = ipv6_all_zeros;
787 787 connp->conn_fport = 0;
788 788 udp->udp_state = TS_IDLE;
789 789 mutex_exit(&udpf->uf_lock);
790 790
791 791 /* Remove any remnants of mapped address binding */
792 792 if (connp->conn_family == AF_INET6)
793 793 connp->conn_ipversion = IPV6_VERSION;
794 794
795 795 connp->conn_v6lastdst = ipv6_all_zeros;
796 796 error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
797 797 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
798 798 mutex_exit(&connp->conn_lock);
799 799 if (error != 0)
800 800 return (error);
801 801
802 802 /*
803 803 * Tell IP to remove the full binding and revert
804 804 * to the local address binding.
805 805 */
806 806 return (ip_laddr_fanout_insert(connp));
807 807 }
808 808
809 809 static void
810 810 udp_tpi_disconnect(queue_t *q, mblk_t *mp)
811 811 {
812 812 conn_t *connp = Q_TO_CONN(q);
813 813 int error;
814 814
815 815 /*
816 816 * Allocate the largest primitive we need to send back
817 817 * T_error_ack is > than T_ok_ack
818 818 */
819 819 mp = reallocb(mp, sizeof (struct T_error_ack), 1);
820 820 if (mp == NULL) {
821 821 /* Unable to reuse the T_DISCON_REQ for the ack. */
822 822 udp_err_ack_prim(q, mp, T_DISCON_REQ, TSYSERR, ENOMEM);
823 823 return;
824 824 }
825 825
826 826 error = udp_do_disconnect(connp);
827 827
828 828 if (error != 0) {
829 829 if (error < 0) {
830 830 udp_err_ack(q, mp, -error, 0);
831 831 } else {
832 832 udp_err_ack(q, mp, TSYSERR, error);
833 833 }
834 834 } else {
835 835 mp = mi_tpi_ok_ack_alloc(mp);
836 836 ASSERT(mp != NULL);
837 837 qreply(q, mp);
838 838 }
839 839 }
840 840
841 841 int
842 842 udp_disconnect(conn_t *connp)
843 843 {
844 844 int error;
845 845
846 846 connp->conn_dgram_errind = B_FALSE;
847 847 error = udp_do_disconnect(connp);
848 848 if (error < 0)
849 849 error = proto_tlitosyserr(-error);
850 850
851 851 return (error);
852 852 }
853 853
854 854 /* This routine creates a T_ERROR_ACK message and passes it upstream. */
855 855 static void
856 856 udp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error, int sys_error)
857 857 {
858 858 if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
859 859 qreply(q, mp);
860 860 }
861 861
862 862 /* Shorthand to generate and send TPI error acks to our client */
863 863 static void
864 864 udp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
865 865 t_scalar_t t_error, int sys_error)
866 866 {
867 867 struct T_error_ack *teackp;
868 868
869 869 if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
870 870 M_PCPROTO, T_ERROR_ACK)) != NULL) {
871 871 teackp = (struct T_error_ack *)mp->b_rptr;
872 872 teackp->ERROR_prim = primitive;
873 873 teackp->TLI_error = t_error;
874 874 teackp->UNIX_error = sys_error;
875 875 qreply(q, mp);
876 876 }
877 877 }
878 878
879 879 /* At minimum we need 4 bytes of UDP header */
880 880 #define ICMP_MIN_UDP_HDR 4
881 881
882 882 /*
883 883 * udp_icmp_input is called as conn_recvicmp to process ICMP messages.
884 884 * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
885 885 * Assumes that IP has pulled up everything up to and including the ICMP header.
886 886 */
887 887 /* ARGSUSED2 */
888 888 static void
889 889 udp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
890 890 {
891 891 conn_t *connp = (conn_t *)arg1;
892 892 icmph_t *icmph;
893 893 ipha_t *ipha;
894 894 int iph_hdr_length;
895 895 udpha_t *udpha;
896 896 sin_t sin;
897 897 sin6_t sin6;
898 898 mblk_t *mp1;
899 899 int error = 0;
900 900 udp_t *udp = connp->conn_udp;
901 901
902 902 ipha = (ipha_t *)mp->b_rptr;
903 903
904 904 ASSERT(OK_32PTR(mp->b_rptr));
905 905
906 906 if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
907 907 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
908 908 udp_icmp_error_ipv6(connp, mp, ira);
909 909 return;
910 910 }
911 911 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
912 912
913 913 /* Skip past the outer IP and ICMP headers */
914 914 ASSERT(IPH_HDR_LENGTH(ipha) == ira->ira_ip_hdr_length);
915 915 iph_hdr_length = ira->ira_ip_hdr_length;
916 916 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
917 917 ipha = (ipha_t *)&icmph[1]; /* Inner IP header */
918 918
919 919 /* Skip past the inner IP and find the ULP header */
920 920 iph_hdr_length = IPH_HDR_LENGTH(ipha);
921 921 udpha = (udpha_t *)((char *)ipha + iph_hdr_length);
922 922
923 923 switch (icmph->icmph_type) {
924 924 case ICMP_DEST_UNREACHABLE:
925 925 switch (icmph->icmph_code) {
926 926 case ICMP_FRAGMENTATION_NEEDED: {
927 927 ipha_t *ipha;
928 928 ip_xmit_attr_t *ixa;
929 929 /*
930 930 * IP has already adjusted the path MTU.
931 931 * But we need to adjust DF for IPv4.
932 932 */
933 933 if (connp->conn_ipversion != IPV4_VERSION)
934 934 break;
935 935
936 936 ixa = conn_get_ixa(connp, B_FALSE);
937 937 if (ixa == NULL || ixa->ixa_ire == NULL) {
938 938 /*
939 939 * Some other thread holds conn_ixa. We will
940 940 * redo this on the next ICMP too big.
941 941 */
942 942 if (ixa != NULL)
943 943 ixa_refrele(ixa);
944 944 break;
945 945 }
946 946 (void) ip_get_pmtu(ixa);
947 947
948 948 mutex_enter(&connp->conn_lock);
949 949 ipha = (ipha_t *)connp->conn_ht_iphc;
950 950 if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
951 951 ipha->ipha_fragment_offset_and_flags |=
952 952 IPH_DF_HTONS;
953 953 } else {
954 954 ipha->ipha_fragment_offset_and_flags &=
955 955 ~IPH_DF_HTONS;
956 956 }
957 957 mutex_exit(&connp->conn_lock);
958 958 ixa_refrele(ixa);
959 959 break;
960 960 }
961 961 case ICMP_PORT_UNREACHABLE:
962 962 case ICMP_PROTOCOL_UNREACHABLE:
963 963 error = ECONNREFUSED;
964 964 break;
965 965 default:
966 966 /* Transient errors */
967 967 break;
968 968 }
969 969 break;
970 970 default:
971 971 /* Transient errors */
972 972 break;
973 973 }
974 974 if (error == 0) {
975 975 freemsg(mp);
976 976 return;
977 977 }
978 978
979 979 /*
980 980 * Deliver T_UDERROR_IND when the application has asked for it.
981 981 * The socket layer enables this automatically when connected.
982 982 */
983 983 if (!connp->conn_dgram_errind) {
984 984 freemsg(mp);
985 985 return;
986 986 }
987 987
988 988 switch (connp->conn_family) {
989 989 case AF_INET:
990 990 sin = sin_null;
991 991 sin.sin_family = AF_INET;
992 992 sin.sin_addr.s_addr = ipha->ipha_dst;
993 993 sin.sin_port = udpha->uha_dst_port;
994 994 if (IPCL_IS_NONSTR(connp)) {
995 995 mutex_enter(&connp->conn_lock);
996 996 if (udp->udp_state == TS_DATA_XFER) {
997 997 if (sin.sin_port == connp->conn_fport &&
998 998 sin.sin_addr.s_addr ==
999 999 connp->conn_faddr_v4) {
1000 1000 mutex_exit(&connp->conn_lock);
1001 1001 (*connp->conn_upcalls->su_set_error)
1002 1002 (connp->conn_upper_handle, error);
1003 1003 goto done;
1004 1004 }
1005 1005 } else {
1006 1006 udp->udp_delayed_error = error;
1007 1007 *((sin_t *)&udp->udp_delayed_addr) = sin;
1008 1008 }
1009 1009 mutex_exit(&connp->conn_lock);
1010 1010 } else {
1011 1011 mp1 = mi_tpi_uderror_ind((char *)&sin, sizeof (sin_t),
1012 1012 NULL, 0, error);
1013 1013 if (mp1 != NULL)
1014 1014 putnext(connp->conn_rq, mp1);
1015 1015 }
1016 1016 break;
1017 1017 case AF_INET6:
1018 1018 sin6 = sin6_null;
1019 1019 sin6.sin6_family = AF_INET6;
1020 1020 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &sin6.sin6_addr);
1021 1021 sin6.sin6_port = udpha->uha_dst_port;
1022 1022 if (IPCL_IS_NONSTR(connp)) {
1023 1023 mutex_enter(&connp->conn_lock);
1024 1024 if (udp->udp_state == TS_DATA_XFER) {
1025 1025 if (sin6.sin6_port == connp->conn_fport &&
1026 1026 IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1027 1027 &connp->conn_faddr_v6)) {
1028 1028 mutex_exit(&connp->conn_lock);
1029 1029 (*connp->conn_upcalls->su_set_error)
1030 1030 (connp->conn_upper_handle, error);
1031 1031 goto done;
1032 1032 }
1033 1033 } else {
1034 1034 udp->udp_delayed_error = error;
1035 1035 *((sin6_t *)&udp->udp_delayed_addr) = sin6;
1036 1036 }
1037 1037 mutex_exit(&connp->conn_lock);
1038 1038 } else {
1039 1039 mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1040 1040 NULL, 0, error);
1041 1041 if (mp1 != NULL)
1042 1042 putnext(connp->conn_rq, mp1);
1043 1043 }
1044 1044 break;
1045 1045 }
1046 1046 done:
1047 1047 freemsg(mp);
1048 1048 }
1049 1049
1050 1050 /*
1051 1051 * udp_icmp_error_ipv6 is called by udp_icmp_error to process ICMP for IPv6.
1052 1052 * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1053 1053 * Assumes that IP has pulled up all the extension headers as well as the
1054 1054 * ICMPv6 header.
1055 1055 */
1056 1056 static void
1057 1057 udp_icmp_error_ipv6(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira)
1058 1058 {
1059 1059 icmp6_t *icmp6;
1060 1060 ip6_t *ip6h, *outer_ip6h;
1061 1061 uint16_t iph_hdr_length;
1062 1062 uint8_t *nexthdrp;
1063 1063 udpha_t *udpha;
1064 1064 sin6_t sin6;
1065 1065 mblk_t *mp1;
1066 1066 int error = 0;
1067 1067 udp_t *udp = connp->conn_udp;
1068 1068 udp_stack_t *us = udp->udp_us;
1069 1069
1070 1070 outer_ip6h = (ip6_t *)mp->b_rptr;
1071 1071 #ifdef DEBUG
1072 1072 if (outer_ip6h->ip6_nxt != IPPROTO_ICMPV6)
1073 1073 iph_hdr_length = ip_hdr_length_v6(mp, outer_ip6h);
1074 1074 else
1075 1075 iph_hdr_length = IPV6_HDR_LEN;
1076 1076 ASSERT(iph_hdr_length == ira->ira_ip_hdr_length);
1077 1077 #endif
1078 1078 /* Skip past the outer IP and ICMP headers */
1079 1079 iph_hdr_length = ira->ira_ip_hdr_length;
1080 1080 icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1081 1081
1082 1082 /* Skip past the inner IP and find the ULP header */
1083 1083 ip6h = (ip6_t *)&icmp6[1]; /* Inner IP header */
1084 1084 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1085 1085 freemsg(mp);
1086 1086 return;
1087 1087 }
1088 1088 udpha = (udpha_t *)((char *)ip6h + iph_hdr_length);
1089 1089
1090 1090 switch (icmp6->icmp6_type) {
1091 1091 case ICMP6_DST_UNREACH:
1092 1092 switch (icmp6->icmp6_code) {
1093 1093 case ICMP6_DST_UNREACH_NOPORT:
1094 1094 error = ECONNREFUSED;
1095 1095 break;
1096 1096 case ICMP6_DST_UNREACH_ADMIN:
1097 1097 case ICMP6_DST_UNREACH_NOROUTE:
1098 1098 case ICMP6_DST_UNREACH_BEYONDSCOPE:
1099 1099 case ICMP6_DST_UNREACH_ADDR:
1100 1100 /* Transient errors */
1101 1101 break;
1102 1102 default:
1103 1103 break;
1104 1104 }
1105 1105 break;
1106 1106 case ICMP6_PACKET_TOO_BIG: {
1107 1107 struct T_unitdata_ind *tudi;
1108 1108 struct T_opthdr *toh;
1109 1109 size_t udi_size;
1110 1110 mblk_t *newmp;
1111 1111 t_scalar_t opt_length = sizeof (struct T_opthdr) +
1112 1112 sizeof (struct ip6_mtuinfo);
1113 1113 sin6_t *sin6;
1114 1114 struct ip6_mtuinfo *mtuinfo;
1115 1115
1116 1116 /*
1117 1117 * If the application has requested to receive path mtu
1118 1118 * information, send up an empty message containing an
1119 1119 * IPV6_PATHMTU ancillary data item.
1120 1120 */
1121 1121 if (!connp->conn_ipv6_recvpathmtu)
1122 1122 break;
1123 1123
1124 1124 udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t) +
1125 1125 opt_length;
1126 1126 if ((newmp = allocb(udi_size, BPRI_MED)) == NULL) {
1127 1127 UDPS_BUMP_MIB(us, udpInErrors);
1128 1128 break;
1129 1129 }
1130 1130
1131 1131 /*
1132 1132 * newmp->b_cont is left to NULL on purpose. This is an
1133 1133 * empty message containing only ancillary data.
1134 1134 */
1135 1135 newmp->b_datap->db_type = M_PROTO;
1136 1136 tudi = (struct T_unitdata_ind *)newmp->b_rptr;
1137 1137 newmp->b_wptr = (uchar_t *)tudi + udi_size;
1138 1138 tudi->PRIM_type = T_UNITDATA_IND;
1139 1139 tudi->SRC_length = sizeof (sin6_t);
1140 1140 tudi->SRC_offset = sizeof (struct T_unitdata_ind);
1141 1141 tudi->OPT_offset = tudi->SRC_offset + sizeof (sin6_t);
1142 1142 tudi->OPT_length = opt_length;
1143 1143
1144 1144 sin6 = (sin6_t *)&tudi[1];
1145 1145 bzero(sin6, sizeof (sin6_t));
1146 1146 sin6->sin6_family = AF_INET6;
1147 1147 sin6->sin6_addr = connp->conn_faddr_v6;
1148 1148
1149 1149 toh = (struct T_opthdr *)&sin6[1];
1150 1150 toh->level = IPPROTO_IPV6;
1151 1151 toh->name = IPV6_PATHMTU;
1152 1152 toh->len = opt_length;
1153 1153 toh->status = 0;
1154 1154
1155 1155 mtuinfo = (struct ip6_mtuinfo *)&toh[1];
1156 1156 bzero(mtuinfo, sizeof (struct ip6_mtuinfo));
1157 1157 mtuinfo->ip6m_addr.sin6_family = AF_INET6;
1158 1158 mtuinfo->ip6m_addr.sin6_addr = ip6h->ip6_dst;
1159 1159 mtuinfo->ip6m_mtu = icmp6->icmp6_mtu;
1160 1160 /*
1161 1161 * We've consumed everything we need from the original
1162 1162 * message. Free it, then send our empty message.
1163 1163 */
1164 1164 freemsg(mp);
1165 1165 udp_ulp_recv(connp, newmp, msgdsize(newmp), ira);
1166 1166 return;
1167 1167 }
1168 1168 case ICMP6_TIME_EXCEEDED:
1169 1169 /* Transient errors */
1170 1170 break;
1171 1171 case ICMP6_PARAM_PROB:
1172 1172 /* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1173 1173 if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1174 1174 (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1175 1175 (uchar_t *)nexthdrp) {
1176 1176 error = ECONNREFUSED;
1177 1177 break;
1178 1178 }
1179 1179 break;
1180 1180 }
1181 1181 if (error == 0) {
1182 1182 freemsg(mp);
1183 1183 return;
1184 1184 }
1185 1185
1186 1186 /*
1187 1187 * Deliver T_UDERROR_IND when the application has asked for it.
1188 1188 * The socket layer enables this automatically when connected.
1189 1189 */
1190 1190 if (!connp->conn_dgram_errind) {
1191 1191 freemsg(mp);
1192 1192 return;
1193 1193 }
1194 1194
1195 1195 sin6 = sin6_null;
1196 1196 sin6.sin6_family = AF_INET6;
1197 1197 sin6.sin6_addr = ip6h->ip6_dst;
1198 1198 sin6.sin6_port = udpha->uha_dst_port;
1199 1199 sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
1200 1200
1201 1201 if (IPCL_IS_NONSTR(connp)) {
1202 1202 mutex_enter(&connp->conn_lock);
1203 1203 if (udp->udp_state == TS_DATA_XFER) {
1204 1204 if (sin6.sin6_port == connp->conn_fport &&
1205 1205 IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1206 1206 &connp->conn_faddr_v6)) {
1207 1207 mutex_exit(&connp->conn_lock);
1208 1208 (*connp->conn_upcalls->su_set_error)
1209 1209 (connp->conn_upper_handle, error);
1210 1210 goto done;
1211 1211 }
1212 1212 } else {
1213 1213 udp->udp_delayed_error = error;
1214 1214 *((sin6_t *)&udp->udp_delayed_addr) = sin6;
1215 1215 }
1216 1216 mutex_exit(&connp->conn_lock);
1217 1217 } else {
1218 1218 mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1219 1219 NULL, 0, error);
1220 1220 if (mp1 != NULL)
1221 1221 putnext(connp->conn_rq, mp1);
1222 1222 }
1223 1223 done:
1224 1224 freemsg(mp);
1225 1225 }
1226 1226
1227 1227 /*
1228 1228 * This routine responds to T_ADDR_REQ messages. It is called by udp_wput.
1229 1229 * The local address is filled in if endpoint is bound. The remote address
1230 1230 * is filled in if remote address has been precified ("connected endpoint")
1231 1231 * (The concept of connected CLTS sockets is alien to published TPI
1232 1232 * but we support it anyway).
1233 1233 */
1234 1234 static void
1235 1235 udp_addr_req(queue_t *q, mblk_t *mp)
1236 1236 {
1237 1237 struct sockaddr *sa;
1238 1238 mblk_t *ackmp;
1239 1239 struct T_addr_ack *taa;
1240 1240 udp_t *udp = Q_TO_UDP(q);
1241 1241 conn_t *connp = udp->udp_connp;
1242 1242 uint_t addrlen;
1243 1243
1244 1244 /* Make it large enough for worst case */
1245 1245 ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
1246 1246 2 * sizeof (sin6_t), 1);
1247 1247 if (ackmp == NULL) {
1248 1248 udp_err_ack(q, mp, TSYSERR, ENOMEM);
1249 1249 return;
1250 1250 }
1251 1251 taa = (struct T_addr_ack *)ackmp->b_rptr;
1252 1252
1253 1253 bzero(taa, sizeof (struct T_addr_ack));
1254 1254 ackmp->b_wptr = (uchar_t *)&taa[1];
1255 1255
1256 1256 taa->PRIM_type = T_ADDR_ACK;
1257 1257 ackmp->b_datap->db_type = M_PCPROTO;
1258 1258
1259 1259 if (connp->conn_family == AF_INET)
1260 1260 addrlen = sizeof (sin_t);
1261 1261 else
1262 1262 addrlen = sizeof (sin6_t);
1263 1263
1264 1264 mutex_enter(&connp->conn_lock);
1265 1265 /*
1266 1266 * Note: Following code assumes 32 bit alignment of basic
1267 1267 * data structures like sin_t and struct T_addr_ack.
1268 1268 */
1269 1269 if (udp->udp_state != TS_UNBND) {
1270 1270 /*
1271 1271 * Fill in local address first
1272 1272 */
1273 1273 taa->LOCADDR_offset = sizeof (*taa);
1274 1274 taa->LOCADDR_length = addrlen;
1275 1275 sa = (struct sockaddr *)&taa[1];
1276 1276 (void) conn_getsockname(connp, sa, &addrlen);
1277 1277 ackmp->b_wptr += addrlen;
1278 1278 }
1279 1279 if (udp->udp_state == TS_DATA_XFER) {
1280 1280 /*
1281 1281 * connected, fill remote address too
1282 1282 */
1283 1283 taa->REMADDR_length = addrlen;
1284 1284 /* assumed 32-bit alignment */
1285 1285 taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
1286 1286 sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
1287 1287 (void) conn_getpeername(connp, sa, &addrlen);
1288 1288 ackmp->b_wptr += addrlen;
1289 1289 }
1290 1290 mutex_exit(&connp->conn_lock);
1291 1291 ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
1292 1292 qreply(q, ackmp);
1293 1293 }
1294 1294
1295 1295 static void
1296 1296 udp_copy_info(struct T_info_ack *tap, udp_t *udp)
1297 1297 {
1298 1298 conn_t *connp = udp->udp_connp;
1299 1299
1300 1300 if (connp->conn_family == AF_INET) {
1301 1301 *tap = udp_g_t_info_ack_ipv4;
1302 1302 } else {
1303 1303 *tap = udp_g_t_info_ack_ipv6;
1304 1304 }
1305 1305 tap->CURRENT_state = udp->udp_state;
1306 1306 tap->OPT_size = udp_max_optsize;
1307 1307 }
1308 1308
1309 1309 static void
1310 1310 udp_do_capability_ack(udp_t *udp, struct T_capability_ack *tcap,
1311 1311 t_uscalar_t cap_bits1)
1312 1312 {
1313 1313 tcap->CAP_bits1 = 0;
1314 1314
1315 1315 if (cap_bits1 & TC1_INFO) {
1316 1316 udp_copy_info(&tcap->INFO_ack, udp);
1317 1317 tcap->CAP_bits1 |= TC1_INFO;
1318 1318 }
1319 1319 }
1320 1320
1321 1321 /*
1322 1322 * This routine responds to T_CAPABILITY_REQ messages. It is called by
1323 1323 * udp_wput. Much of the T_CAPABILITY_ACK information is copied from
1324 1324 * udp_g_t_info_ack. The current state of the stream is copied from
1325 1325 * udp_state.
1326 1326 */
1327 1327 static void
1328 1328 udp_capability_req(queue_t *q, mblk_t *mp)
1329 1329 {
1330 1330 t_uscalar_t cap_bits1;
1331 1331 struct T_capability_ack *tcap;
1332 1332 udp_t *udp = Q_TO_UDP(q);
1333 1333
1334 1334 cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
1335 1335
1336 1336 mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
1337 1337 mp->b_datap->db_type, T_CAPABILITY_ACK);
1338 1338 if (!mp)
1339 1339 return;
1340 1340
1341 1341 tcap = (struct T_capability_ack *)mp->b_rptr;
1342 1342 udp_do_capability_ack(udp, tcap, cap_bits1);
1343 1343
1344 1344 qreply(q, mp);
1345 1345 }
1346 1346
1347 1347 /*
1348 1348 * This routine responds to T_INFO_REQ messages. It is called by udp_wput.
1349 1349 * Most of the T_INFO_ACK information is copied from udp_g_t_info_ack.
1350 1350 * The current state of the stream is copied from udp_state.
1351 1351 */
1352 1352 static void
1353 1353 udp_info_req(queue_t *q, mblk_t *mp)
1354 1354 {
1355 1355 udp_t *udp = Q_TO_UDP(q);
1356 1356
1357 1357 /* Create a T_INFO_ACK message. */
1358 1358 mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
1359 1359 T_INFO_ACK);
1360 1360 if (!mp)
1361 1361 return;
1362 1362 udp_copy_info((struct T_info_ack *)mp->b_rptr, udp);
1363 1363 qreply(q, mp);
1364 1364 }
1365 1365
1366 1366 /* For /dev/udp aka AF_INET open */
1367 1367 static int
1368 1368 udp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1369 1369 {
1370 1370 return (udp_open(q, devp, flag, sflag, credp, B_FALSE));
1371 1371 }
1372 1372
1373 1373 /* For /dev/udp6 aka AF_INET6 open */
1374 1374 static int
1375 1375 udp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1376 1376 {
1377 1377 return (udp_open(q, devp, flag, sflag, credp, B_TRUE));
1378 1378 }
1379 1379
1380 1380 /*
1381 1381 * This is the open routine for udp. It allocates a udp_t structure for
1382 1382 * the stream and, on the first open of the module, creates an ND table.
1383 1383 */
1384 1384 static int
1385 1385 udp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
1386 1386 boolean_t isv6)
1387 1387 {
1388 1388 udp_t *udp;
1389 1389 conn_t *connp;
1390 1390 dev_t conn_dev;
1391 1391 vmem_t *minor_arena;
1392 1392 int err;
1393 1393
1394 1394 /* If the stream is already open, return immediately. */
1395 1395 if (q->q_ptr != NULL)
1396 1396 return (0);
1397 1397
1398 1398 if (sflag == MODOPEN)
1399 1399 return (EINVAL);
1400 1400
1401 1401 if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
1402 1402 ((conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
1403 1403 minor_arena = ip_minor_arena_la;
1404 1404 } else {
1405 1405 /*
1406 1406 * Either minor numbers in the large arena were exhausted
1407 1407 * or a non socket application is doing the open.
1408 1408 * Try to allocate from the small arena.
1409 1409 */
1410 1410 if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0)
1411 1411 return (EBUSY);
1412 1412
1413 1413 minor_arena = ip_minor_arena_sa;
1414 1414 }
1415 1415
1416 1416 if (flag & SO_FALLBACK) {
1417 1417 /*
1418 1418 * Non streams socket needs a stream to fallback to
1419 1419 */
1420 1420 RD(q)->q_ptr = (void *)conn_dev;
1421 1421 WR(q)->q_qinfo = &udp_fallback_sock_winit;
1422 1422 WR(q)->q_ptr = (void *)minor_arena;
1423 1423 qprocson(q);
1424 1424 return (0);
1425 1425 }
1426 1426
1427 1427 connp = udp_do_open(credp, isv6, KM_SLEEP, &err);
1428 1428 if (connp == NULL) {
1429 1429 inet_minor_free(minor_arena, conn_dev);
1430 1430 return (err);
1431 1431 }
1432 1432 udp = connp->conn_udp;
1433 1433
1434 1434 *devp = makedevice(getemajor(*devp), (minor_t)conn_dev);
1435 1435 connp->conn_dev = conn_dev;
1436 1436 connp->conn_minor_arena = minor_arena;
1437 1437
1438 1438 /*
1439 1439 * Initialize the udp_t structure for this stream.
1440 1440 */
1441 1441 q->q_ptr = connp;
1442 1442 WR(q)->q_ptr = connp;
1443 1443 connp->conn_rq = q;
1444 1444 connp->conn_wq = WR(q);
1445 1445
1446 1446 /*
1447 1447 * Since this conn_t/udp_t is not yet visible to anybody else we don't
1448 1448 * need to lock anything.
1449 1449 */
1450 1450 ASSERT(connp->conn_proto == IPPROTO_UDP);
1451 1451 ASSERT(connp->conn_udp == udp);
1452 1452 ASSERT(udp->udp_connp == connp);
1453 1453
1454 1454 if (flag & SO_SOCKSTR) {
1455 1455 udp->udp_issocket = B_TRUE;
1456 1456 }
1457 1457
1458 1458 WR(q)->q_hiwat = connp->conn_sndbuf;
1459 1459 WR(q)->q_lowat = connp->conn_sndlowat;
1460 1460
1461 1461 qprocson(q);
1462 1462
1463 1463 /* Set the Stream head write offset and high watermark. */
1464 1464 (void) proto_set_tx_wroff(q, connp, connp->conn_wroff);
1465 1465 (void) proto_set_rx_hiwat(q, connp,
1466 1466 udp_set_rcv_hiwat(udp, connp->conn_rcvbuf));
1467 1467
1468 1468 mutex_enter(&connp->conn_lock);
1469 1469 connp->conn_state_flags &= ~CONN_INCIPIENT;
1470 1470 mutex_exit(&connp->conn_lock);
1471 1471 return (0);
1472 1472 }
1473 1473
1474 1474 /*
1475 1475 * Which UDP options OK to set through T_UNITDATA_REQ...
1476 1476 */
1477 1477 /* ARGSUSED */
1478 1478 static boolean_t
1479 1479 udp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name)
1480 1480 {
1481 1481 return (B_TRUE);
1482 1482 }
1483 1483
1484 1484 /*
1485 1485 * This routine gets default values of certain options whose default
1486 1486 * values are maintained by protcol specific code
1487 1487 */
1488 1488 int
1489 1489 udp_opt_default(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1490 1490 {
1491 1491 udp_t *udp = Q_TO_UDP(q);
1492 1492 udp_stack_t *us = udp->udp_us;
1493 1493 int *i1 = (int *)ptr;
1494 1494
1495 1495 switch (level) {
1496 1496 case IPPROTO_IP:
1497 1497 switch (name) {
1498 1498 case IP_MULTICAST_TTL:
1499 1499 *ptr = (uchar_t)IP_DEFAULT_MULTICAST_TTL;
1500 1500 return (sizeof (uchar_t));
1501 1501 case IP_MULTICAST_LOOP:
1502 1502 *ptr = (uchar_t)IP_DEFAULT_MULTICAST_LOOP;
1503 1503 return (sizeof (uchar_t));
1504 1504 }
1505 1505 break;
1506 1506 case IPPROTO_IPV6:
1507 1507 switch (name) {
1508 1508 case IPV6_MULTICAST_HOPS:
1509 1509 *i1 = IP_DEFAULT_MULTICAST_TTL;
1510 1510 return (sizeof (int));
1511 1511 case IPV6_MULTICAST_LOOP:
1512 1512 *i1 = IP_DEFAULT_MULTICAST_LOOP;
1513 1513 return (sizeof (int));
1514 1514 case IPV6_UNICAST_HOPS:
1515 1515 *i1 = us->us_ipv6_hoplimit;
1516 1516 return (sizeof (int));
1517 1517 }
1518 1518 break;
1519 1519 }
1520 1520 return (-1);
1521 1521 }
1522 1522
1523 1523 /*
1524 1524 * This routine retrieves the current status of socket options.
1525 1525 * It returns the size of the option retrieved, or -1.
1526 1526 */
1527 1527 int
1528 1528 udp_opt_get(conn_t *connp, t_scalar_t level, t_scalar_t name,
1529 1529 uchar_t *ptr)
1530 1530 {
1531 1531 int *i1 = (int *)ptr;
1532 1532 udp_t *udp = connp->conn_udp;
1533 1533 int len;
1534 1534 conn_opt_arg_t coas;
1535 1535 int retval;
1536 1536
1537 1537 coas.coa_connp = connp;
1538 1538 coas.coa_ixa = connp->conn_ixa;
1539 1539 coas.coa_ipp = &connp->conn_xmit_ipp;
1540 1540 coas.coa_ancillary = B_FALSE;
1541 1541 coas.coa_changed = 0;
1542 1542
1543 1543 /*
1544 1544 * We assume that the optcom framework has checked for the set
1545 1545 * of levels and names that are supported, hence we don't worry
1546 1546 * about rejecting based on that.
1547 1547 * First check for UDP specific handling, then pass to common routine.
1548 1548 */
1549 1549 switch (level) {
1550 1550 case IPPROTO_IP:
1551 1551 /*
1552 1552 * Only allow IPv4 option processing on IPv4 sockets.
1553 1553 */
1554 1554 if (connp->conn_family != AF_INET)
1555 1555 return (-1);
1556 1556
1557 1557 switch (name) {
1558 1558 case IP_OPTIONS:
1559 1559 case T_IP_OPTIONS:
1560 1560 mutex_enter(&connp->conn_lock);
1561 1561 if (!(udp->udp_recv_ipp.ipp_fields &
1562 1562 IPPF_IPV4_OPTIONS)) {
1563 1563 mutex_exit(&connp->conn_lock);
1564 1564 return (0);
1565 1565 }
1566 1566
1567 1567 len = udp->udp_recv_ipp.ipp_ipv4_options_len;
1568 1568 ASSERT(len != 0);
1569 1569 bcopy(udp->udp_recv_ipp.ipp_ipv4_options, ptr, len);
1570 1570 mutex_exit(&connp->conn_lock);
1571 1571 return (len);
1572 1572 }
1573 1573 break;
1574 1574 case IPPROTO_UDP:
1575 1575 switch (name) {
|
↓ open down ↓ |
1575 lines elided |
↑ open up ↑ |
1576 1576 case UDP_NAT_T_ENDPOINT:
1577 1577 mutex_enter(&connp->conn_lock);
1578 1578 *i1 = udp->udp_nat_t_endpoint;
1579 1579 mutex_exit(&connp->conn_lock);
1580 1580 return (sizeof (int));
1581 1581 case UDP_RCVHDR:
1582 1582 mutex_enter(&connp->conn_lock);
1583 1583 *i1 = udp->udp_rcvhdr ? 1 : 0;
1584 1584 mutex_exit(&connp->conn_lock);
1585 1585 return (sizeof (int));
1586 + case UDP_SND_TO_CONNECTED:
1587 + mutex_enter(&connp->conn_lock);
1588 + *i1 = udp->udp_snd_to_conn ? 1 : 0;
1589 + mutex_exit(&connp->conn_lock);
1590 + return (sizeof (int));
1586 1591 }
1587 1592 }
1588 1593 mutex_enter(&connp->conn_lock);
1589 1594 retval = conn_opt_get(&coas, level, name, ptr);
1590 1595 mutex_exit(&connp->conn_lock);
1591 1596 return (retval);
1592 1597 }
1593 1598
1594 1599 /*
1595 1600 * This routine retrieves the current status of socket options.
1596 1601 * It returns the size of the option retrieved, or -1.
1597 1602 */
1598 1603 int
1599 1604 udp_tpi_opt_get(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1600 1605 {
1601 1606 conn_t *connp = Q_TO_CONN(q);
1602 1607 int err;
1603 1608
1604 1609 err = udp_opt_get(connp, level, name, ptr);
1605 1610 return (err);
1606 1611 }
1607 1612
1608 1613 /*
1609 1614 * This routine sets socket options.
1610 1615 */
1611 1616 int
1612 1617 udp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
1613 1618 uint_t inlen, uchar_t *invalp, cred_t *cr, boolean_t checkonly)
1614 1619 {
1615 1620 conn_t *connp = coa->coa_connp;
1616 1621 ip_xmit_attr_t *ixa = coa->coa_ixa;
1617 1622 udp_t *udp = connp->conn_udp;
1618 1623 udp_stack_t *us = udp->udp_us;
1619 1624 int *i1 = (int *)invalp;
1620 1625 boolean_t onoff = (*i1 == 0) ? 0 : 1;
1621 1626 int error;
1622 1627
1623 1628 ASSERT(MUTEX_NOT_HELD(&coa->coa_connp->conn_lock));
1624 1629 /*
1625 1630 * First do UDP specific sanity checks and handle UDP specific
1626 1631 * options. Note that some IPPROTO_UDP options are handled
1627 1632 * by conn_opt_set.
1628 1633 */
1629 1634 switch (level) {
1630 1635 case SOL_SOCKET:
1631 1636 switch (name) {
1632 1637 case SO_SNDBUF:
1633 1638 if (*i1 > us->us_max_buf) {
1634 1639 return (ENOBUFS);
1635 1640 }
1636 1641 break;
1637 1642 case SO_RCVBUF:
1638 1643 if (*i1 > us->us_max_buf) {
1639 1644 return (ENOBUFS);
1640 1645 }
1641 1646 break;
1642 1647
1643 1648 case SCM_UCRED: {
1644 1649 struct ucred_s *ucr;
1645 1650 cred_t *newcr;
1646 1651 ts_label_t *tsl;
1647 1652
1648 1653 /*
1649 1654 * Only sockets that have proper privileges and are
1650 1655 * bound to MLPs will have any other value here, so
1651 1656 * this implicitly tests for privilege to set label.
1652 1657 */
1653 1658 if (connp->conn_mlp_type == mlptSingle)
1654 1659 break;
1655 1660
1656 1661 ucr = (struct ucred_s *)invalp;
1657 1662 if (inlen < sizeof (*ucr) + sizeof (bslabel_t) ||
1658 1663 ucr->uc_labeloff < sizeof (*ucr) ||
1659 1664 ucr->uc_labeloff + sizeof (bslabel_t) > inlen)
1660 1665 return (EINVAL);
1661 1666 if (!checkonly) {
1662 1667 /*
1663 1668 * Set ixa_tsl to the new label.
1664 1669 * We assume that crgetzoneid doesn't change
1665 1670 * as part of the SCM_UCRED.
1666 1671 */
1667 1672 ASSERT(cr != NULL);
1668 1673 if ((tsl = crgetlabel(cr)) == NULL)
1669 1674 return (EINVAL);
1670 1675 newcr = copycred_from_bslabel(cr, UCLABEL(ucr),
1671 1676 tsl->tsl_doi, KM_NOSLEEP);
1672 1677 if (newcr == NULL)
1673 1678 return (ENOSR);
1674 1679 ASSERT(newcr->cr_label != NULL);
1675 1680 /*
1676 1681 * Move the hold on the cr_label to ixa_tsl by
1677 1682 * setting cr_label to NULL. Then release newcr.
1678 1683 */
1679 1684 ip_xmit_attr_replace_tsl(ixa, newcr->cr_label);
1680 1685 ixa->ixa_flags |= IXAF_UCRED_TSL;
1681 1686 newcr->cr_label = NULL;
1682 1687 crfree(newcr);
1683 1688 coa->coa_changed |= COA_HEADER_CHANGED;
1684 1689 coa->coa_changed |= COA_WROFF_CHANGED;
1685 1690 }
1686 1691 /* Fully handled this option. */
1687 1692 return (0);
1688 1693 }
1689 1694 }
1690 1695 break;
1691 1696 case IPPROTO_UDP:
1692 1697 switch (name) {
1693 1698 case UDP_NAT_T_ENDPOINT:
1694 1699 if ((error = secpolicy_ip_config(cr, B_FALSE)) != 0) {
1695 1700 return (error);
1696 1701 }
1697 1702
1698 1703 /*
1699 1704 * Use conn_family instead so we can avoid ambiguitites
1700 1705 * with AF_INET6 sockets that may switch from IPv4
1701 1706 * to IPv6.
1702 1707 */
1703 1708 if (connp->conn_family != AF_INET) {
1704 1709 return (EAFNOSUPPORT);
1705 1710 }
1706 1711
1707 1712 if (!checkonly) {
1708 1713 mutex_enter(&connp->conn_lock);
1709 1714 udp->udp_nat_t_endpoint = onoff;
1710 1715 mutex_exit(&connp->conn_lock);
|
↓ open down ↓ |
115 lines elided |
↑ open up ↑ |
1711 1716 coa->coa_changed |= COA_HEADER_CHANGED;
1712 1717 coa->coa_changed |= COA_WROFF_CHANGED;
1713 1718 }
1714 1719 /* Fully handled this option. */
1715 1720 return (0);
1716 1721 case UDP_RCVHDR:
1717 1722 mutex_enter(&connp->conn_lock);
1718 1723 udp->udp_rcvhdr = onoff;
1719 1724 mutex_exit(&connp->conn_lock);
1720 1725 return (0);
1726 + case UDP_SND_TO_CONNECTED:
1727 + mutex_enter(&connp->conn_lock);
1728 + udp->udp_snd_to_conn = onoff;
1729 + mutex_exit(&connp->conn_lock);
1730 + return (0);
1721 1731 }
1722 1732 break;
1723 1733 }
1724 1734 error = conn_opt_set(coa, level, name, inlen, invalp,
1725 1735 checkonly, cr);
1726 1736 return (error);
1727 1737 }
1728 1738
1729 1739 /*
1730 1740 * This routine sets socket options.
1731 1741 */
1732 1742 int
1733 1743 udp_opt_set(conn_t *connp, uint_t optset_context, int level,
1734 1744 int name, uint_t inlen, uchar_t *invalp, uint_t *outlenp,
1735 1745 uchar_t *outvalp, void *thisdg_attrs, cred_t *cr)
1736 1746 {
1737 1747 udp_t *udp = connp->conn_udp;
1738 1748 int err;
1739 1749 conn_opt_arg_t coas, *coa;
1740 1750 boolean_t checkonly;
1741 1751 udp_stack_t *us = udp->udp_us;
1742 1752
1743 1753 switch (optset_context) {
1744 1754 case SETFN_OPTCOM_CHECKONLY:
1745 1755 checkonly = B_TRUE;
1746 1756 /*
1747 1757 * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
1748 1758 * inlen != 0 implies value supplied and
1749 1759 * we have to "pretend" to set it.
1750 1760 * inlen == 0 implies that there is no
1751 1761 * value part in T_CHECK request and just validation
1752 1762 * done elsewhere should be enough, we just return here.
1753 1763 */
1754 1764 if (inlen == 0) {
1755 1765 *outlenp = 0;
1756 1766 return (0);
1757 1767 }
1758 1768 break;
1759 1769 case SETFN_OPTCOM_NEGOTIATE:
1760 1770 checkonly = B_FALSE;
1761 1771 break;
1762 1772 case SETFN_UD_NEGOTIATE:
1763 1773 case SETFN_CONN_NEGOTIATE:
1764 1774 checkonly = B_FALSE;
1765 1775 /*
1766 1776 * Negotiating local and "association-related" options
1767 1777 * through T_UNITDATA_REQ.
1768 1778 *
1769 1779 * Following routine can filter out ones we do not
1770 1780 * want to be "set" this way.
1771 1781 */
1772 1782 if (!udp_opt_allow_udr_set(level, name)) {
1773 1783 *outlenp = 0;
1774 1784 return (EINVAL);
1775 1785 }
1776 1786 break;
1777 1787 default:
1778 1788 /*
1779 1789 * We should never get here
1780 1790 */
1781 1791 *outlenp = 0;
1782 1792 return (EINVAL);
1783 1793 }
1784 1794
1785 1795 ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
1786 1796 (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
1787 1797
1788 1798 if (thisdg_attrs != NULL) {
1789 1799 /* Options from T_UNITDATA_REQ */
1790 1800 coa = (conn_opt_arg_t *)thisdg_attrs;
1791 1801 ASSERT(coa->coa_connp == connp);
1792 1802 ASSERT(coa->coa_ixa != NULL);
1793 1803 ASSERT(coa->coa_ipp != NULL);
1794 1804 ASSERT(coa->coa_ancillary);
1795 1805 } else {
1796 1806 coa = &coas;
1797 1807 coas.coa_connp = connp;
1798 1808 /* Get a reference on conn_ixa to prevent concurrent mods */
1799 1809 coas.coa_ixa = conn_get_ixa(connp, B_TRUE);
1800 1810 if (coas.coa_ixa == NULL) {
1801 1811 *outlenp = 0;
1802 1812 return (ENOMEM);
1803 1813 }
1804 1814 coas.coa_ipp = &connp->conn_xmit_ipp;
1805 1815 coas.coa_ancillary = B_FALSE;
1806 1816 coas.coa_changed = 0;
1807 1817 }
1808 1818
1809 1819 err = udp_do_opt_set(coa, level, name, inlen, invalp,
1810 1820 cr, checkonly);
1811 1821 if (err != 0) {
1812 1822 errout:
1813 1823 if (!coa->coa_ancillary)
1814 1824 ixa_refrele(coa->coa_ixa);
1815 1825 *outlenp = 0;
1816 1826 return (err);
1817 1827 }
1818 1828 /* Handle DHCPINIT here outside of lock */
1819 1829 if (level == IPPROTO_IP && name == IP_DHCPINIT_IF) {
1820 1830 uint_t ifindex;
1821 1831 ill_t *ill;
1822 1832
1823 1833 ifindex = *(uint_t *)invalp;
1824 1834 if (ifindex == 0) {
1825 1835 ill = NULL;
1826 1836 } else {
1827 1837 ill = ill_lookup_on_ifindex(ifindex, B_FALSE,
1828 1838 coa->coa_ixa->ixa_ipst);
1829 1839 if (ill == NULL) {
1830 1840 err = ENXIO;
1831 1841 goto errout;
1832 1842 }
1833 1843
1834 1844 mutex_enter(&ill->ill_lock);
1835 1845 if (ill->ill_state_flags & ILL_CONDEMNED) {
1836 1846 mutex_exit(&ill->ill_lock);
1837 1847 ill_refrele(ill);
1838 1848 err = ENXIO;
1839 1849 goto errout;
1840 1850 }
1841 1851 if (IS_VNI(ill)) {
1842 1852 mutex_exit(&ill->ill_lock);
1843 1853 ill_refrele(ill);
1844 1854 err = EINVAL;
1845 1855 goto errout;
1846 1856 }
1847 1857 }
1848 1858 mutex_enter(&connp->conn_lock);
1849 1859
1850 1860 if (connp->conn_dhcpinit_ill != NULL) {
1851 1861 /*
1852 1862 * We've locked the conn so conn_cleanup_ill()
1853 1863 * cannot clear conn_dhcpinit_ill -- so it's
1854 1864 * safe to access the ill.
1855 1865 */
1856 1866 ill_t *oill = connp->conn_dhcpinit_ill;
1857 1867
1858 1868 ASSERT(oill->ill_dhcpinit != 0);
1859 1869 atomic_dec_32(&oill->ill_dhcpinit);
1860 1870 ill_set_inputfn(connp->conn_dhcpinit_ill);
1861 1871 connp->conn_dhcpinit_ill = NULL;
1862 1872 }
1863 1873
1864 1874 if (ill != NULL) {
1865 1875 connp->conn_dhcpinit_ill = ill;
1866 1876 atomic_inc_32(&ill->ill_dhcpinit);
1867 1877 ill_set_inputfn(ill);
1868 1878 mutex_exit(&connp->conn_lock);
1869 1879 mutex_exit(&ill->ill_lock);
1870 1880 ill_refrele(ill);
1871 1881 } else {
1872 1882 mutex_exit(&connp->conn_lock);
1873 1883 }
1874 1884 }
1875 1885
1876 1886 /*
1877 1887 * Common case of OK return with outval same as inval.
1878 1888 */
1879 1889 if (invalp != outvalp) {
1880 1890 /* don't trust bcopy for identical src/dst */
1881 1891 (void) bcopy(invalp, outvalp, inlen);
1882 1892 }
1883 1893 *outlenp = inlen;
1884 1894
1885 1895 /*
1886 1896 * If this was not ancillary data, then we rebuild the headers,
1887 1897 * update the IRE/NCE, and IPsec as needed.
1888 1898 * Since the label depends on the destination we go through
1889 1899 * ip_set_destination first.
1890 1900 */
1891 1901 if (coa->coa_ancillary) {
1892 1902 return (0);
1893 1903 }
1894 1904
1895 1905 if (coa->coa_changed & COA_ROUTE_CHANGED) {
1896 1906 in6_addr_t saddr, faddr, nexthop;
1897 1907 in_port_t fport;
1898 1908
1899 1909 /*
1900 1910 * We clear lastdst to make sure we pick up the change
1901 1911 * next time sending.
1902 1912 * If we are connected we re-cache the information.
1903 1913 * We ignore errors to preserve BSD behavior.
1904 1914 * Note that we don't redo IPsec policy lookup here
1905 1915 * since the final destination (or source) didn't change.
1906 1916 */
1907 1917 mutex_enter(&connp->conn_lock);
1908 1918 connp->conn_v6lastdst = ipv6_all_zeros;
1909 1919
1910 1920 ip_attr_nexthop(coa->coa_ipp, coa->coa_ixa,
1911 1921 &connp->conn_faddr_v6, &nexthop);
1912 1922 saddr = connp->conn_saddr_v6;
1913 1923 faddr = connp->conn_faddr_v6;
1914 1924 fport = connp->conn_fport;
1915 1925 mutex_exit(&connp->conn_lock);
1916 1926
1917 1927 if (!IN6_IS_ADDR_UNSPECIFIED(&faddr) &&
1918 1928 !IN6_IS_ADDR_V4MAPPED_ANY(&faddr)) {
1919 1929 (void) ip_attr_connect(connp, coa->coa_ixa,
1920 1930 &saddr, &faddr, &nexthop, fport, NULL, NULL,
1921 1931 IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
1922 1932 }
1923 1933 }
1924 1934
1925 1935 ixa_refrele(coa->coa_ixa);
1926 1936
1927 1937 if (coa->coa_changed & COA_HEADER_CHANGED) {
1928 1938 /*
1929 1939 * Rebuild the header template if we are connected.
1930 1940 * Otherwise clear conn_v6lastdst so we rebuild the header
1931 1941 * in the data path.
1932 1942 */
1933 1943 mutex_enter(&connp->conn_lock);
1934 1944 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
1935 1945 !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
1936 1946 err = udp_build_hdr_template(connp,
1937 1947 &connp->conn_saddr_v6, &connp->conn_faddr_v6,
1938 1948 connp->conn_fport, connp->conn_flowinfo);
1939 1949 if (err != 0) {
1940 1950 mutex_exit(&connp->conn_lock);
1941 1951 return (err);
1942 1952 }
1943 1953 } else {
1944 1954 connp->conn_v6lastdst = ipv6_all_zeros;
1945 1955 }
1946 1956 mutex_exit(&connp->conn_lock);
1947 1957 }
1948 1958 if (coa->coa_changed & COA_RCVBUF_CHANGED) {
1949 1959 (void) proto_set_rx_hiwat(connp->conn_rq, connp,
1950 1960 connp->conn_rcvbuf);
1951 1961 }
1952 1962 if ((coa->coa_changed & COA_SNDBUF_CHANGED) && !IPCL_IS_NONSTR(connp)) {
1953 1963 connp->conn_wq->q_hiwat = connp->conn_sndbuf;
1954 1964 }
1955 1965 if (coa->coa_changed & COA_WROFF_CHANGED) {
1956 1966 /* Increase wroff if needed */
1957 1967 uint_t wroff;
1958 1968
1959 1969 mutex_enter(&connp->conn_lock);
1960 1970 wroff = connp->conn_ht_iphc_allocated + us->us_wroff_extra;
1961 1971 if (udp->udp_nat_t_endpoint)
1962 1972 wroff += sizeof (uint32_t);
1963 1973 if (wroff > connp->conn_wroff) {
1964 1974 connp->conn_wroff = wroff;
1965 1975 mutex_exit(&connp->conn_lock);
1966 1976 (void) proto_set_tx_wroff(connp->conn_rq, connp, wroff);
1967 1977 } else {
1968 1978 mutex_exit(&connp->conn_lock);
1969 1979 }
1970 1980 }
1971 1981 return (err);
1972 1982 }
1973 1983
1974 1984 /* This routine sets socket options. */
1975 1985 int
1976 1986 udp_tpi_opt_set(queue_t *q, uint_t optset_context, int level, int name,
1977 1987 uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
1978 1988 void *thisdg_attrs, cred_t *cr)
1979 1989 {
1980 1990 conn_t *connp = Q_TO_CONN(q);
1981 1991 int error;
1982 1992
1983 1993 error = udp_opt_set(connp, optset_context, level, name, inlen, invalp,
1984 1994 outlenp, outvalp, thisdg_attrs, cr);
1985 1995 return (error);
1986 1996 }
1987 1997
1988 1998 /*
1989 1999 * Setup IP and UDP headers.
1990 2000 * Returns NULL on allocation failure, in which case data_mp is freed.
1991 2001 */
1992 2002 mblk_t *
1993 2003 udp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
1994 2004 const in6_addr_t *v6src, const in6_addr_t *v6dst, in_port_t dstport,
1995 2005 uint32_t flowinfo, mblk_t *data_mp, int *errorp)
1996 2006 {
1997 2007 mblk_t *mp;
1998 2008 udpha_t *udpha;
1999 2009 udp_stack_t *us = connp->conn_netstack->netstack_udp;
2000 2010 uint_t data_len;
2001 2011 uint32_t cksum;
2002 2012 udp_t *udp = connp->conn_udp;
2003 2013 boolean_t insert_spi = udp->udp_nat_t_endpoint;
2004 2014 uint_t ulp_hdr_len;
2005 2015
2006 2016 data_len = msgdsize(data_mp);
2007 2017 ulp_hdr_len = UDPH_SIZE;
2008 2018 if (insert_spi)
2009 2019 ulp_hdr_len += sizeof (uint32_t);
2010 2020
2011 2021 mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, IPPROTO_UDP, flowinfo,
2012 2022 ulp_hdr_len, data_mp, data_len, us->us_wroff_extra, &cksum, errorp);
2013 2023 if (mp == NULL) {
2014 2024 ASSERT(*errorp != 0);
2015 2025 return (NULL);
2016 2026 }
2017 2027
2018 2028 data_len += ulp_hdr_len;
2019 2029 ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
2020 2030
2021 2031 udpha = (udpha_t *)(mp->b_rptr + ixa->ixa_ip_hdr_length);
2022 2032 udpha->uha_src_port = connp->conn_lport;
2023 2033 udpha->uha_dst_port = dstport;
2024 2034 udpha->uha_checksum = 0;
2025 2035 udpha->uha_length = htons(data_len);
2026 2036
2027 2037 /*
2028 2038 * If there was a routing option/header then conn_prepend_hdr
2029 2039 * has massaged it and placed the pseudo-header checksum difference
2030 2040 * in the cksum argument.
2031 2041 *
2032 2042 * Setup header length and prepare for ULP checksum done in IP.
2033 2043 *
2034 2044 * We make it easy for IP to include our pseudo header
2035 2045 * by putting our length in uha_checksum.
2036 2046 * The IP source, destination, and length have already been set by
2037 2047 * conn_prepend_hdr.
2038 2048 */
2039 2049 cksum += data_len;
2040 2050 cksum = (cksum >> 16) + (cksum & 0xFFFF);
2041 2051 ASSERT(cksum < 0x10000);
2042 2052
2043 2053 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2044 2054 ipha_t *ipha = (ipha_t *)mp->b_rptr;
2045 2055
2046 2056 ASSERT(ntohs(ipha->ipha_length) == ixa->ixa_pktlen);
2047 2057
2048 2058 /* IP does the checksum if uha_checksum is non-zero */
2049 2059 if (us->us_do_checksum) {
2050 2060 if (cksum == 0)
2051 2061 udpha->uha_checksum = 0xffff;
2052 2062 else
2053 2063 udpha->uha_checksum = htons(cksum);
2054 2064 } else {
2055 2065 udpha->uha_checksum = 0;
2056 2066 }
2057 2067 } else {
2058 2068 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
2059 2069
2060 2070 ASSERT(ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN == ixa->ixa_pktlen);
2061 2071 if (cksum == 0)
2062 2072 udpha->uha_checksum = 0xffff;
2063 2073 else
2064 2074 udpha->uha_checksum = htons(cksum);
2065 2075 }
2066 2076
2067 2077 /* Insert all-0s SPI now. */
2068 2078 if (insert_spi)
2069 2079 *((uint32_t *)(udpha + 1)) = 0;
2070 2080
2071 2081 return (mp);
2072 2082 }
2073 2083
2074 2084 static int
2075 2085 udp_build_hdr_template(conn_t *connp, const in6_addr_t *v6src,
2076 2086 const in6_addr_t *v6dst, in_port_t dstport, uint32_t flowinfo)
2077 2087 {
2078 2088 udpha_t *udpha;
2079 2089 int error;
2080 2090
2081 2091 ASSERT(MUTEX_HELD(&connp->conn_lock));
2082 2092 /*
2083 2093 * We clear lastdst to make sure we don't use the lastdst path
2084 2094 * next time sending since we might not have set v6dst yet.
2085 2095 */
2086 2096 connp->conn_v6lastdst = ipv6_all_zeros;
2087 2097
2088 2098 error = conn_build_hdr_template(connp, UDPH_SIZE, 0, v6src, v6dst,
2089 2099 flowinfo);
2090 2100 if (error != 0)
2091 2101 return (error);
2092 2102
2093 2103 /*
2094 2104 * Any routing header/option has been massaged. The checksum difference
2095 2105 * is stored in conn_sum.
2096 2106 */
2097 2107 udpha = (udpha_t *)connp->conn_ht_ulp;
2098 2108 udpha->uha_src_port = connp->conn_lport;
2099 2109 udpha->uha_dst_port = dstport;
2100 2110 udpha->uha_checksum = 0;
2101 2111 udpha->uha_length = htons(UDPH_SIZE); /* Filled in later */
2102 2112 return (0);
2103 2113 }
2104 2114
2105 2115 static mblk_t *
2106 2116 udp_queue_fallback(udp_t *udp, mblk_t *mp)
2107 2117 {
2108 2118 ASSERT(MUTEX_HELD(&udp->udp_recv_lock));
2109 2119 if (IPCL_IS_NONSTR(udp->udp_connp)) {
2110 2120 /*
2111 2121 * fallback has started but messages have not been moved yet
2112 2122 */
2113 2123 if (udp->udp_fallback_queue_head == NULL) {
2114 2124 ASSERT(udp->udp_fallback_queue_tail == NULL);
2115 2125 udp->udp_fallback_queue_head = mp;
2116 2126 udp->udp_fallback_queue_tail = mp;
2117 2127 } else {
2118 2128 ASSERT(udp->udp_fallback_queue_tail != NULL);
2119 2129 udp->udp_fallback_queue_tail->b_next = mp;
2120 2130 udp->udp_fallback_queue_tail = mp;
2121 2131 }
2122 2132 return (NULL);
2123 2133 } else {
2124 2134 /*
2125 2135 * Fallback completed, let the caller putnext() the mblk.
2126 2136 */
2127 2137 return (mp);
2128 2138 }
2129 2139 }
2130 2140
2131 2141 /*
2132 2142 * Deliver data to ULP. In case we have a socket, and it's falling back to
2133 2143 * TPI, then we'll queue the mp for later processing.
2134 2144 */
2135 2145 static void
2136 2146 udp_ulp_recv(conn_t *connp, mblk_t *mp, uint_t len, ip_recv_attr_t *ira)
2137 2147 {
2138 2148 if (IPCL_IS_NONSTR(connp)) {
2139 2149 udp_t *udp = connp->conn_udp;
2140 2150 int error;
2141 2151
2142 2152 ASSERT(len == msgdsize(mp));
2143 2153 if ((*connp->conn_upcalls->su_recv)
2144 2154 (connp->conn_upper_handle, mp, len, 0, &error, NULL) < 0) {
2145 2155 mutex_enter(&udp->udp_recv_lock);
2146 2156 if (error == ENOSPC) {
2147 2157 /*
2148 2158 * let's confirm while holding the lock
2149 2159 */
2150 2160 if ((*connp->conn_upcalls->su_recv)
2151 2161 (connp->conn_upper_handle, NULL, 0, 0,
2152 2162 &error, NULL) < 0) {
2153 2163 ASSERT(error == ENOSPC);
2154 2164 if (error == ENOSPC) {
2155 2165 connp->conn_flow_cntrld =
2156 2166 B_TRUE;
2157 2167 }
2158 2168 }
2159 2169 mutex_exit(&udp->udp_recv_lock);
2160 2170 } else {
2161 2171 ASSERT(error == EOPNOTSUPP);
2162 2172 mp = udp_queue_fallback(udp, mp);
2163 2173 mutex_exit(&udp->udp_recv_lock);
2164 2174 if (mp != NULL)
2165 2175 putnext(connp->conn_rq, mp);
2166 2176 }
2167 2177 }
2168 2178 ASSERT(MUTEX_NOT_HELD(&udp->udp_recv_lock));
2169 2179 } else {
2170 2180 if (is_system_labeled()) {
2171 2181 ASSERT(ira->ira_cred != NULL);
2172 2182 /*
2173 2183 * Provide for protocols above UDP such as RPC
2174 2184 * NOPID leaves db_cpid unchanged.
2175 2185 */
2176 2186 mblk_setcred(mp, ira->ira_cred, NOPID);
2177 2187 }
2178 2188
2179 2189 putnext(connp->conn_rq, mp);
2180 2190 }
2181 2191 }
2182 2192
2183 2193 /*
2184 2194 * This is the inbound data path.
2185 2195 * IP has already pulled up the IP plus UDP headers and verified alignment
2186 2196 * etc.
2187 2197 */
2188 2198 /* ARGSUSED2 */
2189 2199 static void
2190 2200 udp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2191 2201 {
2192 2202 conn_t *connp = (conn_t *)arg1;
2193 2203 struct T_unitdata_ind *tudi;
2194 2204 uchar_t *rptr; /* Pointer to IP header */
2195 2205 int hdr_length; /* Length of IP+UDP headers */
2196 2206 int udi_size; /* Size of T_unitdata_ind */
2197 2207 int pkt_len;
2198 2208 udp_t *udp;
2199 2209 udpha_t *udpha;
2200 2210 ip_pkt_t ipps;
2201 2211 ip6_t *ip6h;
2202 2212 mblk_t *mp1;
2203 2213 uint32_t udp_ipv4_options_len;
2204 2214 crb_t recv_ancillary;
2205 2215 udp_stack_t *us;
2206 2216
2207 2217 ASSERT(connp->conn_flags & IPCL_UDPCONN);
2208 2218
2209 2219 udp = connp->conn_udp;
2210 2220 us = udp->udp_us;
2211 2221 rptr = mp->b_rptr;
2212 2222
2213 2223 ASSERT(DB_TYPE(mp) == M_DATA);
2214 2224 ASSERT(OK_32PTR(rptr));
2215 2225 ASSERT(ira->ira_pktlen == msgdsize(mp));
2216 2226 pkt_len = ira->ira_pktlen;
2217 2227
2218 2228 /*
2219 2229 * Get a snapshot of these and allow other threads to change
2220 2230 * them after that. We need the same recv_ancillary when determining
2221 2231 * the size as when adding the ancillary data items.
2222 2232 */
2223 2233 mutex_enter(&connp->conn_lock);
2224 2234 udp_ipv4_options_len = udp->udp_recv_ipp.ipp_ipv4_options_len;
2225 2235 recv_ancillary = connp->conn_recv_ancillary;
2226 2236 mutex_exit(&connp->conn_lock);
2227 2237
2228 2238 hdr_length = ira->ira_ip_hdr_length;
2229 2239
2230 2240 /*
2231 2241 * IP inspected the UDP header thus all of it must be in the mblk.
2232 2242 * UDP length check is performed for IPv6 packets and IPv4 packets
2233 2243 * to check if the size of the packet as specified
2234 2244 * by the UDP header is the same as the length derived from the IP
2235 2245 * header.
2236 2246 */
2237 2247 udpha = (udpha_t *)(rptr + hdr_length);
2238 2248 if (pkt_len != ntohs(udpha->uha_length) + hdr_length)
2239 2249 goto tossit;
2240 2250
2241 2251 hdr_length += UDPH_SIZE;
2242 2252 ASSERT(MBLKL(mp) >= hdr_length); /* IP did a pullup */
2243 2253
2244 2254 /* Initialize regardless of IP version */
2245 2255 ipps.ipp_fields = 0;
2246 2256
2247 2257 if (((ira->ira_flags & IRAF_IPV4_OPTIONS) ||
2248 2258 udp_ipv4_options_len > 0) &&
2249 2259 connp->conn_family == AF_INET) {
2250 2260 int err;
2251 2261
2252 2262 /*
2253 2263 * Record/update udp_recv_ipp with the lock
2254 2264 * held. Not needed for AF_INET6 sockets
2255 2265 * since they don't support a getsockopt of IP_OPTIONS.
2256 2266 */
2257 2267 mutex_enter(&connp->conn_lock);
2258 2268 err = ip_find_hdr_v4((ipha_t *)rptr, &udp->udp_recv_ipp,
2259 2269 B_TRUE);
2260 2270 if (err != 0) {
2261 2271 /* Allocation failed. Drop packet */
2262 2272 mutex_exit(&connp->conn_lock);
2263 2273 freemsg(mp);
2264 2274 UDPS_BUMP_MIB(us, udpInErrors);
2265 2275 return;
2266 2276 }
2267 2277 mutex_exit(&connp->conn_lock);
2268 2278 }
2269 2279
2270 2280 if (recv_ancillary.crb_all != 0) {
2271 2281 /*
2272 2282 * Record packet information in the ip_pkt_t
2273 2283 */
2274 2284 if (ira->ira_flags & IRAF_IS_IPV4) {
2275 2285 ASSERT(IPH_HDR_VERSION(rptr) == IPV4_VERSION);
2276 2286 ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2277 2287 ASSERT(((ipha_t *)rptr)->ipha_protocol == IPPROTO_UDP);
2278 2288 ASSERT(ira->ira_ip_hdr_length == IPH_HDR_LENGTH(rptr));
2279 2289
2280 2290 (void) ip_find_hdr_v4((ipha_t *)rptr, &ipps, B_FALSE);
2281 2291 } else {
2282 2292 uint8_t nexthdrp;
2283 2293
2284 2294 ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION);
2285 2295 /*
2286 2296 * IPv6 packets can only be received by applications
2287 2297 * that are prepared to receive IPv6 addresses.
2288 2298 * The IP fanout must ensure this.
2289 2299 */
2290 2300 ASSERT(connp->conn_family == AF_INET6);
2291 2301
2292 2302 ip6h = (ip6_t *)rptr;
2293 2303
2294 2304 /* We don't care about the length, but need the ipp */
2295 2305 hdr_length = ip_find_hdr_v6(mp, ip6h, B_TRUE, &ipps,
2296 2306 &nexthdrp);
2297 2307 ASSERT(hdr_length == ira->ira_ip_hdr_length);
2298 2308 /* Restore */
2299 2309 hdr_length = ira->ira_ip_hdr_length + UDPH_SIZE;
2300 2310 ASSERT(nexthdrp == IPPROTO_UDP);
2301 2311 }
2302 2312 }
2303 2313
2304 2314 /*
2305 2315 * This is the inbound data path. Packets are passed upstream as
2306 2316 * T_UNITDATA_IND messages.
2307 2317 */
2308 2318 if (connp->conn_family == AF_INET) {
2309 2319 sin_t *sin;
2310 2320
2311 2321 ASSERT(IPH_HDR_VERSION((ipha_t *)rptr) == IPV4_VERSION);
2312 2322
2313 2323 /*
2314 2324 * Normally only send up the source address.
2315 2325 * If any ancillary data items are wanted we add those.
2316 2326 */
2317 2327 udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin_t);
2318 2328 if (recv_ancillary.crb_all != 0) {
2319 2329 udi_size += conn_recvancillary_size(connp,
2320 2330 recv_ancillary, ira, mp, &ipps);
2321 2331 }
2322 2332
2323 2333 /* Allocate a message block for the T_UNITDATA_IND structure. */
2324 2334 mp1 = allocb(udi_size, BPRI_MED);
2325 2335 if (mp1 == NULL) {
2326 2336 freemsg(mp);
2327 2337 UDPS_BUMP_MIB(us, udpInErrors);
2328 2338 return;
2329 2339 }
2330 2340 mp1->b_cont = mp;
2331 2341 mp1->b_datap->db_type = M_PROTO;
2332 2342 tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2333 2343 mp1->b_wptr = (uchar_t *)tudi + udi_size;
2334 2344 tudi->PRIM_type = T_UNITDATA_IND;
2335 2345 tudi->SRC_length = sizeof (sin_t);
2336 2346 tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2337 2347 tudi->OPT_offset = sizeof (struct T_unitdata_ind) +
2338 2348 sizeof (sin_t);
2339 2349 udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin_t));
2340 2350 tudi->OPT_length = udi_size;
2341 2351 sin = (sin_t *)&tudi[1];
2342 2352 sin->sin_addr.s_addr = ((ipha_t *)rptr)->ipha_src;
2343 2353 sin->sin_port = udpha->uha_src_port;
2344 2354 sin->sin_family = connp->conn_family;
2345 2355 *(uint32_t *)&sin->sin_zero[0] = 0;
2346 2356 *(uint32_t *)&sin->sin_zero[4] = 0;
2347 2357
2348 2358 /*
2349 2359 * Add options if IP_RECVDSTADDR, IP_RECVIF, IP_RECVSLLA or
2350 2360 * IP_RECVTTL has been set.
2351 2361 */
2352 2362 if (udi_size != 0) {
2353 2363 conn_recvancillary_add(connp, recv_ancillary, ira,
2354 2364 &ipps, (uchar_t *)&sin[1], udi_size);
2355 2365 }
2356 2366 } else {
2357 2367 sin6_t *sin6;
2358 2368
2359 2369 /*
2360 2370 * Handle both IPv4 and IPv6 packets for IPv6 sockets.
2361 2371 *
2362 2372 * Normally we only send up the address. If receiving of any
2363 2373 * optional receive side information is enabled, we also send
2364 2374 * that up as options.
2365 2375 */
2366 2376 udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2367 2377
2368 2378 if (recv_ancillary.crb_all != 0) {
2369 2379 udi_size += conn_recvancillary_size(connp,
2370 2380 recv_ancillary, ira, mp, &ipps);
2371 2381 }
2372 2382
2373 2383 mp1 = allocb(udi_size, BPRI_MED);
2374 2384 if (mp1 == NULL) {
2375 2385 freemsg(mp);
2376 2386 UDPS_BUMP_MIB(us, udpInErrors);
2377 2387 return;
2378 2388 }
2379 2389 mp1->b_cont = mp;
2380 2390 mp1->b_datap->db_type = M_PROTO;
2381 2391 tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2382 2392 mp1->b_wptr = (uchar_t *)tudi + udi_size;
2383 2393 tudi->PRIM_type = T_UNITDATA_IND;
2384 2394 tudi->SRC_length = sizeof (sin6_t);
2385 2395 tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2386 2396 tudi->OPT_offset = sizeof (struct T_unitdata_ind) +
2387 2397 sizeof (sin6_t);
2388 2398 udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin6_t));
2389 2399 tudi->OPT_length = udi_size;
2390 2400 sin6 = (sin6_t *)&tudi[1];
2391 2401 if (ira->ira_flags & IRAF_IS_IPV4) {
2392 2402 in6_addr_t v6dst;
2393 2403
2394 2404 IN6_IPADDR_TO_V4MAPPED(((ipha_t *)rptr)->ipha_src,
2395 2405 &sin6->sin6_addr);
2396 2406 IN6_IPADDR_TO_V4MAPPED(((ipha_t *)rptr)->ipha_dst,
2397 2407 &v6dst);
2398 2408 sin6->sin6_flowinfo = 0;
2399 2409 sin6->sin6_scope_id = 0;
2400 2410 sin6->__sin6_src_id = ip_srcid_find_addr(&v6dst,
2401 2411 IPCL_ZONEID(connp), us->us_netstack);
2402 2412 } else {
2403 2413 ip6h = (ip6_t *)rptr;
2404 2414
2405 2415 sin6->sin6_addr = ip6h->ip6_src;
2406 2416 /* No sin6_flowinfo per API */
2407 2417 sin6->sin6_flowinfo = 0;
2408 2418 /* For link-scope pass up scope id */
2409 2419 if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src))
2410 2420 sin6->sin6_scope_id = ira->ira_ruifindex;
2411 2421 else
2412 2422 sin6->sin6_scope_id = 0;
2413 2423 sin6->__sin6_src_id = ip_srcid_find_addr(
2414 2424 &ip6h->ip6_dst, IPCL_ZONEID(connp),
2415 2425 us->us_netstack);
2416 2426 }
2417 2427 sin6->sin6_port = udpha->uha_src_port;
2418 2428 sin6->sin6_family = connp->conn_family;
2419 2429
2420 2430 if (udi_size != 0) {
2421 2431 conn_recvancillary_add(connp, recv_ancillary, ira,
2422 2432 &ipps, (uchar_t *)&sin6[1], udi_size);
2423 2433 }
2424 2434 }
2425 2435
2426 2436 /*
2427 2437 * DTrace this UDP input as udp:::receive (this is for IPv4, IPv6 and
2428 2438 * loopback traffic).
2429 2439 */
2430 2440 DTRACE_UDP5(receive, mblk_t *, NULL, ip_xmit_attr_t *, connp->conn_ixa,
2431 2441 void_ip_t *, rptr, udp_t *, udp, udpha_t *, udpha);
2432 2442
2433 2443 /* Walk past the headers unless IP_RECVHDR was set. */
2434 2444 if (!udp->udp_rcvhdr) {
2435 2445 mp->b_rptr = rptr + hdr_length;
2436 2446 pkt_len -= hdr_length;
2437 2447 }
2438 2448
2439 2449 UDPS_BUMP_MIB(us, udpHCInDatagrams);
2440 2450 udp_ulp_recv(connp, mp1, pkt_len, ira);
2441 2451 return;
2442 2452
2443 2453 tossit:
2444 2454 freemsg(mp);
2445 2455 UDPS_BUMP_MIB(us, udpInErrors);
2446 2456 }
2447 2457
2448 2458 /*
2449 2459 * This routine creates a T_UDERROR_IND message and passes it upstream.
2450 2460 * The address and options are copied from the T_UNITDATA_REQ message
2451 2461 * passed in mp. This message is freed.
2452 2462 */
2453 2463 static void
2454 2464 udp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err)
2455 2465 {
2456 2466 struct T_unitdata_req *tudr;
2457 2467 mblk_t *mp1;
2458 2468 uchar_t *destaddr;
2459 2469 t_scalar_t destlen;
2460 2470 uchar_t *optaddr;
2461 2471 t_scalar_t optlen;
2462 2472
2463 2473 if ((mp->b_wptr < mp->b_rptr) ||
2464 2474 (MBLKL(mp)) < sizeof (struct T_unitdata_req)) {
2465 2475 goto done;
2466 2476 }
2467 2477 tudr = (struct T_unitdata_req *)mp->b_rptr;
2468 2478 destaddr = mp->b_rptr + tudr->DEST_offset;
2469 2479 if (destaddr < mp->b_rptr || destaddr >= mp->b_wptr ||
2470 2480 destaddr + tudr->DEST_length < mp->b_rptr ||
2471 2481 destaddr + tudr->DEST_length > mp->b_wptr) {
2472 2482 goto done;
2473 2483 }
2474 2484 optaddr = mp->b_rptr + tudr->OPT_offset;
2475 2485 if (optaddr < mp->b_rptr || optaddr >= mp->b_wptr ||
2476 2486 optaddr + tudr->OPT_length < mp->b_rptr ||
2477 2487 optaddr + tudr->OPT_length > mp->b_wptr) {
2478 2488 goto done;
2479 2489 }
2480 2490 destlen = tudr->DEST_length;
2481 2491 optlen = tudr->OPT_length;
2482 2492
2483 2493 mp1 = mi_tpi_uderror_ind((char *)destaddr, destlen,
2484 2494 (char *)optaddr, optlen, err);
2485 2495 if (mp1 != NULL)
2486 2496 qreply(q, mp1);
2487 2497
2488 2498 done:
2489 2499 freemsg(mp);
2490 2500 }
2491 2501
2492 2502 /*
2493 2503 * This routine removes a port number association from a stream. It
2494 2504 * is called by udp_wput to handle T_UNBIND_REQ messages.
2495 2505 */
2496 2506 static void
2497 2507 udp_tpi_unbind(queue_t *q, mblk_t *mp)
2498 2508 {
2499 2509 conn_t *connp = Q_TO_CONN(q);
2500 2510 int error;
2501 2511
2502 2512 error = udp_do_unbind(connp);
2503 2513 if (error) {
2504 2514 if (error < 0)
2505 2515 udp_err_ack(q, mp, -error, 0);
2506 2516 else
2507 2517 udp_err_ack(q, mp, TSYSERR, error);
2508 2518 return;
2509 2519 }
2510 2520
2511 2521 mp = mi_tpi_ok_ack_alloc(mp);
2512 2522 ASSERT(mp != NULL);
2513 2523 ASSERT(((struct T_ok_ack *)mp->b_rptr)->PRIM_type == T_OK_ACK);
2514 2524 qreply(q, mp);
2515 2525 }
2516 2526
2517 2527 /*
2518 2528 * Don't let port fall into the privileged range.
2519 2529 * Since the extra privileged ports can be arbitrary we also
2520 2530 * ensure that we exclude those from consideration.
2521 2531 * us->us_epriv_ports is not sorted thus we loop over it until
2522 2532 * there are no changes.
2523 2533 */
2524 2534 static in_port_t
2525 2535 udp_update_next_port(udp_t *udp, in_port_t port, boolean_t random)
2526 2536 {
2527 2537 int i, bump;
2528 2538 in_port_t nextport;
2529 2539 boolean_t restart = B_FALSE;
2530 2540 udp_stack_t *us = udp->udp_us;
2531 2541
2532 2542 if (random && udp_random_anon_port != 0) {
2533 2543 (void) random_get_pseudo_bytes((uint8_t *)&port,
2534 2544 sizeof (in_port_t));
2535 2545 /*
2536 2546 * Unless changed by a sys admin, the smallest anon port
2537 2547 * is 32768 and the largest anon port is 65535. It is
2538 2548 * very likely (50%) for the random port to be smaller
2539 2549 * than the smallest anon port. When that happens,
2540 2550 * add port % (anon port range) to the smallest anon
2541 2551 * port to get the random port. It should fall into the
2542 2552 * valid anon port range.
2543 2553 */
2544 2554 if ((port < us->us_smallest_anon_port) ||
2545 2555 (port > us->us_largest_anon_port)) {
2546 2556 if (us->us_smallest_anon_port ==
2547 2557 us->us_largest_anon_port) {
2548 2558 bump = 0;
2549 2559 } else {
2550 2560 bump = port % (us->us_largest_anon_port -
2551 2561 us->us_smallest_anon_port);
2552 2562 }
2553 2563
2554 2564 port = us->us_smallest_anon_port + bump;
2555 2565 }
2556 2566 }
2557 2567
2558 2568 retry:
2559 2569 if (port < us->us_smallest_anon_port)
2560 2570 port = us->us_smallest_anon_port;
2561 2571
2562 2572 if (port > us->us_largest_anon_port) {
2563 2573 port = us->us_smallest_anon_port;
2564 2574 if (restart)
2565 2575 return (0);
2566 2576 restart = B_TRUE;
2567 2577 }
2568 2578
2569 2579 if (port < us->us_smallest_nonpriv_port)
2570 2580 port = us->us_smallest_nonpriv_port;
2571 2581
2572 2582 for (i = 0; i < us->us_num_epriv_ports; i++) {
2573 2583 if (port == us->us_epriv_ports[i]) {
2574 2584 port++;
2575 2585 /*
2576 2586 * Make sure that the port is in the
2577 2587 * valid range.
2578 2588 */
2579 2589 goto retry;
2580 2590 }
2581 2591 }
2582 2592
2583 2593 if (is_system_labeled() &&
2584 2594 (nextport = tsol_next_port(crgetzone(udp->udp_connp->conn_cred),
2585 2595 port, IPPROTO_UDP, B_TRUE)) != 0) {
2586 2596 port = nextport;
2587 2597 goto retry;
2588 2598 }
2589 2599
2590 2600 return (port);
2591 2601 }
2592 2602
2593 2603 /*
2594 2604 * Handle T_UNITDATA_REQ with options. Both IPv4 and IPv6
2595 2605 * Either tudr_mp or msg is set. If tudr_mp we take ancillary data from
2596 2606 * the TPI options, otherwise we take them from msg_control.
2597 2607 * If both sin and sin6 is set it is a connected socket and we use conn_faddr.
2598 2608 * Always consumes mp; never consumes tudr_mp.
2599 2609 */
2600 2610 static int
2601 2611 udp_output_ancillary(conn_t *connp, sin_t *sin, sin6_t *sin6, mblk_t *mp,
2602 2612 mblk_t *tudr_mp, struct nmsghdr *msg, cred_t *cr, pid_t pid)
2603 2613 {
2604 2614 udp_t *udp = connp->conn_udp;
2605 2615 udp_stack_t *us = udp->udp_us;
2606 2616 int error;
2607 2617 ip_xmit_attr_t *ixa;
2608 2618 ip_pkt_t *ipp;
2609 2619 in6_addr_t v6src;
2610 2620 in6_addr_t v6dst;
2611 2621 in6_addr_t v6nexthop;
2612 2622 in_port_t dstport;
2613 2623 uint32_t flowinfo;
2614 2624 uint_t srcid;
2615 2625 int is_absreq_failure = 0;
2616 2626 conn_opt_arg_t coas, *coa;
2617 2627
2618 2628 ASSERT(tudr_mp != NULL || msg != NULL);
2619 2629
2620 2630 /*
2621 2631 * Get ixa before checking state to handle a disconnect race.
2622 2632 *
2623 2633 * We need an exclusive copy of conn_ixa since the ancillary data
2624 2634 * options might modify it. That copy has no pointers hence we
2625 2635 * need to set them up once we've parsed the ancillary data.
2626 2636 */
2627 2637 ixa = conn_get_ixa_exclusive(connp);
2628 2638 if (ixa == NULL) {
2629 2639 UDPS_BUMP_MIB(us, udpOutErrors);
2630 2640 freemsg(mp);
2631 2641 return (ENOMEM);
2632 2642 }
2633 2643 ASSERT(cr != NULL);
2634 2644 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2635 2645 ixa->ixa_cred = cr;
2636 2646 ixa->ixa_cpid = pid;
2637 2647 if (is_system_labeled()) {
2638 2648 /* We need to restart with a label based on the cred */
2639 2649 ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
2640 2650 }
2641 2651
2642 2652 /* In case previous destination was multicast or multirt */
2643 2653 ip_attr_newdst(ixa);
2644 2654
2645 2655 /* Get a copy of conn_xmit_ipp since the options might change it */
2646 2656 ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
2647 2657 if (ipp == NULL) {
2648 2658 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2649 2659 ixa->ixa_cred = connp->conn_cred; /* Restore */
2650 2660 ixa->ixa_cpid = connp->conn_cpid;
2651 2661 ixa_refrele(ixa);
2652 2662 UDPS_BUMP_MIB(us, udpOutErrors);
2653 2663 freemsg(mp);
2654 2664 return (ENOMEM);
2655 2665 }
2656 2666 mutex_enter(&connp->conn_lock);
2657 2667 error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
2658 2668 mutex_exit(&connp->conn_lock);
2659 2669 if (error != 0) {
2660 2670 UDPS_BUMP_MIB(us, udpOutErrors);
2661 2671 freemsg(mp);
2662 2672 goto done;
2663 2673 }
2664 2674
2665 2675 /*
2666 2676 * Parse the options and update ixa and ipp as a result.
2667 2677 * Note that ixa_tsl can be updated if SCM_UCRED.
2668 2678 * ixa_refrele/ixa_inactivate will release any reference on ixa_tsl.
2669 2679 */
2670 2680
2671 2681 coa = &coas;
2672 2682 coa->coa_connp = connp;
2673 2683 coa->coa_ixa = ixa;
2674 2684 coa->coa_ipp = ipp;
2675 2685 coa->coa_ancillary = B_TRUE;
2676 2686 coa->coa_changed = 0;
2677 2687
2678 2688 if (msg != NULL) {
2679 2689 error = process_auxiliary_options(connp, msg->msg_control,
2680 2690 msg->msg_controllen, coa, &udp_opt_obj, udp_opt_set, cr);
2681 2691 } else {
2682 2692 struct T_unitdata_req *tudr;
2683 2693
2684 2694 tudr = (struct T_unitdata_req *)tudr_mp->b_rptr;
2685 2695 ASSERT(tudr->PRIM_type == T_UNITDATA_REQ);
2686 2696 error = tpi_optcom_buf(connp->conn_wq, tudr_mp,
2687 2697 &tudr->OPT_length, tudr->OPT_offset, cr, &udp_opt_obj,
2688 2698 coa, &is_absreq_failure);
2689 2699 }
2690 2700 if (error != 0) {
2691 2701 /*
2692 2702 * Note: No special action needed in this
2693 2703 * module for "is_absreq_failure"
2694 2704 */
2695 2705 freemsg(mp);
2696 2706 UDPS_BUMP_MIB(us, udpOutErrors);
2697 2707 goto done;
2698 2708 }
2699 2709 ASSERT(is_absreq_failure == 0);
2700 2710
2701 2711 mutex_enter(&connp->conn_lock);
2702 2712 /*
2703 2713 * If laddr is unspecified then we look at sin6_src_id.
2704 2714 * We will give precedence to a source address set with IPV6_PKTINFO
2705 2715 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
2706 2716 * want ip_attr_connect to select a source (since it can fail) when
2707 2717 * IPV6_PKTINFO is specified.
2708 2718 * If this doesn't result in a source address then we get a source
2709 2719 * from ip_attr_connect() below.
2710 2720 */
2711 2721 v6src = connp->conn_saddr_v6;
2712 2722 if (sin != NULL) {
2713 2723 IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
2714 2724 dstport = sin->sin_port;
2715 2725 flowinfo = 0;
2716 2726 ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
2717 2727 ixa->ixa_flags |= IXAF_IS_IPV4;
2718 2728 } else if (sin6 != NULL) {
2719 2729 boolean_t v4mapped;
2720 2730
2721 2731 v6dst = sin6->sin6_addr;
2722 2732 dstport = sin6->sin6_port;
2723 2733 flowinfo = sin6->sin6_flowinfo;
2724 2734 srcid = sin6->__sin6_src_id;
2725 2735 if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
2726 2736 ixa->ixa_scopeid = sin6->sin6_scope_id;
2727 2737 ixa->ixa_flags |= IXAF_SCOPEID_SET;
2728 2738 } else {
2729 2739 ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
2730 2740 }
2731 2741 v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
2732 2742 if (v4mapped)
2733 2743 ixa->ixa_flags |= IXAF_IS_IPV4;
2734 2744 else
2735 2745 ixa->ixa_flags &= ~IXAF_IS_IPV4;
2736 2746 if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
2737 2747 if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
2738 2748 v4mapped, connp->conn_netstack)) {
2739 2749 /* Mismatch - v4mapped/v6 specified by srcid. */
2740 2750 mutex_exit(&connp->conn_lock);
2741 2751 error = EADDRNOTAVAIL;
2742 2752 goto failed; /* Does freemsg() and mib. */
2743 2753 }
2744 2754 }
2745 2755 } else {
2746 2756 /* Connected case */
2747 2757 v6dst = connp->conn_faddr_v6;
2748 2758 dstport = connp->conn_fport;
2749 2759 flowinfo = connp->conn_flowinfo;
2750 2760 }
2751 2761 mutex_exit(&connp->conn_lock);
2752 2762
2753 2763 /* Handle IP_PKTINFO/IPV6_PKTINFO setting source address. */
2754 2764 if (ipp->ipp_fields & IPPF_ADDR) {
2755 2765 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2756 2766 if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
2757 2767 v6src = ipp->ipp_addr;
2758 2768 } else {
2759 2769 if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
2760 2770 v6src = ipp->ipp_addr;
2761 2771 }
2762 2772 }
2763 2773
2764 2774 ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
2765 2775 error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
2766 2776 &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST | IPDF_IPSEC);
2767 2777
2768 2778 switch (error) {
2769 2779 case 0:
2770 2780 break;
2771 2781 case EADDRNOTAVAIL:
2772 2782 /*
2773 2783 * IXAF_VERIFY_SOURCE tells us to pick a better source.
2774 2784 * Don't have the application see that errno
2775 2785 */
2776 2786 error = ENETUNREACH;
2777 2787 goto failed;
2778 2788 case ENETDOWN:
2779 2789 /*
2780 2790 * Have !ipif_addr_ready address; drop packet silently
2781 2791 * until we can get applications to not send until we
2782 2792 * are ready.
2783 2793 */
2784 2794 error = 0;
2785 2795 goto failed;
2786 2796 case EHOSTUNREACH:
2787 2797 case ENETUNREACH:
2788 2798 if (ixa->ixa_ire != NULL) {
2789 2799 /*
2790 2800 * Let conn_ip_output/ire_send_noroute return
2791 2801 * the error and send any local ICMP error.
2792 2802 */
2793 2803 error = 0;
2794 2804 break;
2795 2805 }
2796 2806 /* FALLTHRU */
2797 2807 default:
2798 2808 failed:
2799 2809 freemsg(mp);
2800 2810 UDPS_BUMP_MIB(us, udpOutErrors);
2801 2811 goto done;
2802 2812 }
2803 2813
2804 2814 /*
2805 2815 * We might be going to a different destination than last time,
2806 2816 * thus check that TX allows the communication and compute any
2807 2817 * needed label.
2808 2818 *
2809 2819 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
2810 2820 * don't have to worry about concurrent threads.
2811 2821 */
2812 2822 if (is_system_labeled()) {
2813 2823 /* Using UDP MLP requires SCM_UCRED from user */
2814 2824 if (connp->conn_mlp_type != mlptSingle &&
2815 2825 !((ixa->ixa_flags & IXAF_UCRED_TSL))) {
2816 2826 UDPS_BUMP_MIB(us, udpOutErrors);
2817 2827 error = ECONNREFUSED;
2818 2828 freemsg(mp);
2819 2829 goto done;
2820 2830 }
2821 2831 /*
2822 2832 * Check whether Trusted Solaris policy allows communication
2823 2833 * with this host, and pretend that the destination is
2824 2834 * unreachable if not.
2825 2835 * Compute any needed label and place it in ipp_label_v4/v6.
2826 2836 *
2827 2837 * Later conn_build_hdr_template/conn_prepend_hdr takes
2828 2838 * ipp_label_v4/v6 to form the packet.
2829 2839 *
2830 2840 * Tsol note: We have ipp structure local to this thread so
2831 2841 * no locking is needed.
2832 2842 */
2833 2843 error = conn_update_label(connp, ixa, &v6dst, ipp);
2834 2844 if (error != 0) {
2835 2845 freemsg(mp);
2836 2846 UDPS_BUMP_MIB(us, udpOutErrors);
2837 2847 goto done;
2838 2848 }
2839 2849 }
2840 2850 mp = udp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, dstport,
2841 2851 flowinfo, mp, &error);
2842 2852 if (mp == NULL) {
2843 2853 ASSERT(error != 0);
2844 2854 UDPS_BUMP_MIB(us, udpOutErrors);
2845 2855 goto done;
2846 2856 }
2847 2857 if (ixa->ixa_pktlen > IP_MAXPACKET) {
2848 2858 error = EMSGSIZE;
2849 2859 UDPS_BUMP_MIB(us, udpOutErrors);
2850 2860 freemsg(mp);
2851 2861 goto done;
2852 2862 }
2853 2863 /* We're done. Pass the packet to ip. */
2854 2864 UDPS_BUMP_MIB(us, udpHCOutDatagrams);
2855 2865
2856 2866 DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
2857 2867 void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
2858 2868 &mp->b_rptr[ixa->ixa_ip_hdr_length]);
2859 2869
2860 2870 error = conn_ip_output(mp, ixa);
2861 2871 /* No udpOutErrors if an error since IP increases its error counter */
2862 2872 switch (error) {
2863 2873 case 0:
2864 2874 break;
2865 2875 case EWOULDBLOCK:
2866 2876 (void) ixa_check_drain_insert(connp, ixa);
2867 2877 error = 0;
2868 2878 break;
2869 2879 case EADDRNOTAVAIL:
2870 2880 /*
2871 2881 * IXAF_VERIFY_SOURCE tells us to pick a better source.
2872 2882 * Don't have the application see that errno
2873 2883 */
2874 2884 error = ENETUNREACH;
2875 2885 /* FALLTHRU */
2876 2886 default:
2877 2887 mutex_enter(&connp->conn_lock);
2878 2888 /*
2879 2889 * Clear the source and v6lastdst so we call ip_attr_connect
2880 2890 * for the next packet and try to pick a better source.
2881 2891 */
2882 2892 if (connp->conn_mcbc_bind)
2883 2893 connp->conn_saddr_v6 = ipv6_all_zeros;
2884 2894 else
2885 2895 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
2886 2896 connp->conn_v6lastdst = ipv6_all_zeros;
2887 2897 mutex_exit(&connp->conn_lock);
2888 2898 break;
2889 2899 }
2890 2900 done:
2891 2901 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2892 2902 ixa->ixa_cred = connp->conn_cred; /* Restore */
2893 2903 ixa->ixa_cpid = connp->conn_cpid;
2894 2904 ixa_refrele(ixa);
2895 2905 ip_pkt_free(ipp);
2896 2906 kmem_free(ipp, sizeof (*ipp));
2897 2907 return (error);
2898 2908 }
2899 2909
2900 2910 /*
2901 2911 * Handle sending an M_DATA for a connected socket.
2902 2912 * Handles both IPv4 and IPv6.
2903 2913 */
2904 2914 static int
2905 2915 udp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
2906 2916 {
2907 2917 udp_t *udp = connp->conn_udp;
2908 2918 udp_stack_t *us = udp->udp_us;
2909 2919 int error;
2910 2920 ip_xmit_attr_t *ixa;
2911 2921
2912 2922 /*
2913 2923 * If no other thread is using conn_ixa this just gets a reference to
2914 2924 * conn_ixa. Otherwise we get a safe copy of conn_ixa.
2915 2925 */
2916 2926 ixa = conn_get_ixa(connp, B_FALSE);
2917 2927 if (ixa == NULL) {
2918 2928 UDPS_BUMP_MIB(us, udpOutErrors);
2919 2929 freemsg(mp);
2920 2930 return (ENOMEM);
2921 2931 }
2922 2932
2923 2933 ASSERT(cr != NULL);
2924 2934 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2925 2935 ixa->ixa_cred = cr;
2926 2936 ixa->ixa_cpid = pid;
2927 2937
2928 2938 mutex_enter(&connp->conn_lock);
2929 2939 mp = udp_prepend_header_template(connp, ixa, mp, &connp->conn_saddr_v6,
2930 2940 connp->conn_fport, connp->conn_flowinfo, &error);
2931 2941
2932 2942 if (mp == NULL) {
2933 2943 ASSERT(error != 0);
2934 2944 mutex_exit(&connp->conn_lock);
2935 2945 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2936 2946 ixa->ixa_cred = connp->conn_cred; /* Restore */
2937 2947 ixa->ixa_cpid = connp->conn_cpid;
2938 2948 ixa_refrele(ixa);
2939 2949 UDPS_BUMP_MIB(us, udpOutErrors);
2940 2950 freemsg(mp);
2941 2951 return (error);
2942 2952 }
2943 2953
2944 2954 /*
2945 2955 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
2946 2956 * safe copy, then we need to fill in any pointers in it.
2947 2957 */
2948 2958 if (ixa->ixa_ire == NULL) {
2949 2959 in6_addr_t faddr, saddr;
2950 2960 in6_addr_t nexthop;
2951 2961 in_port_t fport;
2952 2962
2953 2963 saddr = connp->conn_saddr_v6;
2954 2964 faddr = connp->conn_faddr_v6;
2955 2965 fport = connp->conn_fport;
2956 2966 ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &faddr, &nexthop);
2957 2967 mutex_exit(&connp->conn_lock);
2958 2968
2959 2969 error = ip_attr_connect(connp, ixa, &saddr, &faddr, &nexthop,
2960 2970 fport, NULL, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
2961 2971 IPDF_IPSEC);
2962 2972 switch (error) {
2963 2973 case 0:
2964 2974 break;
2965 2975 case EADDRNOTAVAIL:
2966 2976 /*
2967 2977 * IXAF_VERIFY_SOURCE tells us to pick a better source.
2968 2978 * Don't have the application see that errno
2969 2979 */
2970 2980 error = ENETUNREACH;
2971 2981 goto failed;
2972 2982 case ENETDOWN:
2973 2983 /*
2974 2984 * Have !ipif_addr_ready address; drop packet silently
2975 2985 * until we can get applications to not send until we
2976 2986 * are ready.
2977 2987 */
2978 2988 error = 0;
2979 2989 goto failed;
2980 2990 case EHOSTUNREACH:
2981 2991 case ENETUNREACH:
2982 2992 if (ixa->ixa_ire != NULL) {
2983 2993 /*
2984 2994 * Let conn_ip_output/ire_send_noroute return
2985 2995 * the error and send any local ICMP error.
2986 2996 */
2987 2997 error = 0;
2988 2998 break;
2989 2999 }
2990 3000 /* FALLTHRU */
2991 3001 default:
2992 3002 failed:
2993 3003 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2994 3004 ixa->ixa_cred = connp->conn_cred; /* Restore */
2995 3005 ixa->ixa_cpid = connp->conn_cpid;
2996 3006 ixa_refrele(ixa);
2997 3007 freemsg(mp);
2998 3008 UDPS_BUMP_MIB(us, udpOutErrors);
2999 3009 return (error);
3000 3010 }
3001 3011 } else {
3002 3012 /* Done with conn_t */
3003 3013 mutex_exit(&connp->conn_lock);
3004 3014 }
3005 3015 ASSERT(ixa->ixa_ire != NULL);
3006 3016
3007 3017 /* We're done. Pass the packet to ip. */
3008 3018 UDPS_BUMP_MIB(us, udpHCOutDatagrams);
3009 3019
3010 3020 DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
3011 3021 void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
3012 3022 &mp->b_rptr[ixa->ixa_ip_hdr_length]);
3013 3023
3014 3024 error = conn_ip_output(mp, ixa);
3015 3025 /* No udpOutErrors if an error since IP increases its error counter */
3016 3026 switch (error) {
3017 3027 case 0:
3018 3028 break;
3019 3029 case EWOULDBLOCK:
3020 3030 (void) ixa_check_drain_insert(connp, ixa);
3021 3031 error = 0;
3022 3032 break;
3023 3033 case EADDRNOTAVAIL:
3024 3034 /*
3025 3035 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3026 3036 * Don't have the application see that errno
3027 3037 */
3028 3038 error = ENETUNREACH;
3029 3039 break;
3030 3040 }
3031 3041 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3032 3042 ixa->ixa_cred = connp->conn_cred; /* Restore */
3033 3043 ixa->ixa_cpid = connp->conn_cpid;
3034 3044 ixa_refrele(ixa);
3035 3045 return (error);
3036 3046 }
3037 3047
3038 3048 /*
3039 3049 * Handle sending an M_DATA to the last destination.
3040 3050 * Handles both IPv4 and IPv6.
3041 3051 *
3042 3052 * NOTE: The caller must hold conn_lock and we drop it here.
3043 3053 */
3044 3054 static int
3045 3055 udp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid,
3046 3056 ip_xmit_attr_t *ixa)
3047 3057 {
3048 3058 udp_t *udp = connp->conn_udp;
3049 3059 udp_stack_t *us = udp->udp_us;
3050 3060 int error;
3051 3061
3052 3062 ASSERT(MUTEX_HELD(&connp->conn_lock));
3053 3063 ASSERT(ixa != NULL);
3054 3064
3055 3065 ASSERT(cr != NULL);
3056 3066 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3057 3067 ixa->ixa_cred = cr;
3058 3068 ixa->ixa_cpid = pid;
3059 3069
3060 3070 mp = udp_prepend_header_template(connp, ixa, mp, &connp->conn_v6lastsrc,
3061 3071 connp->conn_lastdstport, connp->conn_lastflowinfo, &error);
3062 3072
3063 3073 if (mp == NULL) {
3064 3074 ASSERT(error != 0);
3065 3075 mutex_exit(&connp->conn_lock);
3066 3076 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3067 3077 ixa->ixa_cred = connp->conn_cred; /* Restore */
3068 3078 ixa->ixa_cpid = connp->conn_cpid;
3069 3079 ixa_refrele(ixa);
3070 3080 UDPS_BUMP_MIB(us, udpOutErrors);
3071 3081 freemsg(mp);
3072 3082 return (error);
3073 3083 }
3074 3084
3075 3085 /*
3076 3086 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3077 3087 * safe copy, then we need to fill in any pointers in it.
3078 3088 */
3079 3089 if (ixa->ixa_ire == NULL) {
3080 3090 in6_addr_t lastdst, lastsrc;
3081 3091 in6_addr_t nexthop;
3082 3092 in_port_t lastport;
3083 3093
3084 3094 lastsrc = connp->conn_v6lastsrc;
3085 3095 lastdst = connp->conn_v6lastdst;
3086 3096 lastport = connp->conn_lastdstport;
3087 3097 ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &lastdst, &nexthop);
3088 3098 mutex_exit(&connp->conn_lock);
3089 3099
3090 3100 error = ip_attr_connect(connp, ixa, &lastsrc, &lastdst,
3091 3101 &nexthop, lastport, NULL, NULL, IPDF_ALLOW_MCBC |
3092 3102 IPDF_VERIFY_DST | IPDF_IPSEC);
3093 3103 switch (error) {
3094 3104 case 0:
3095 3105 break;
3096 3106 case EADDRNOTAVAIL:
3097 3107 /*
3098 3108 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3099 3109 * Don't have the application see that errno
3100 3110 */
3101 3111 error = ENETUNREACH;
3102 3112 goto failed;
3103 3113 case ENETDOWN:
3104 3114 /*
3105 3115 * Have !ipif_addr_ready address; drop packet silently
3106 3116 * until we can get applications to not send until we
3107 3117 * are ready.
3108 3118 */
3109 3119 error = 0;
3110 3120 goto failed;
3111 3121 case EHOSTUNREACH:
3112 3122 case ENETUNREACH:
3113 3123 if (ixa->ixa_ire != NULL) {
3114 3124 /*
3115 3125 * Let conn_ip_output/ire_send_noroute return
3116 3126 * the error and send any local ICMP error.
3117 3127 */
3118 3128 error = 0;
3119 3129 break;
3120 3130 }
3121 3131 /* FALLTHRU */
3122 3132 default:
3123 3133 failed:
3124 3134 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3125 3135 ixa->ixa_cred = connp->conn_cred; /* Restore */
3126 3136 ixa->ixa_cpid = connp->conn_cpid;
3127 3137 ixa_refrele(ixa);
3128 3138 freemsg(mp);
3129 3139 UDPS_BUMP_MIB(us, udpOutErrors);
3130 3140 return (error);
3131 3141 }
3132 3142 } else {
3133 3143 /* Done with conn_t */
3134 3144 mutex_exit(&connp->conn_lock);
3135 3145 }
3136 3146
3137 3147 /* We're done. Pass the packet to ip. */
3138 3148 UDPS_BUMP_MIB(us, udpHCOutDatagrams);
3139 3149
3140 3150 DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
3141 3151 void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
3142 3152 &mp->b_rptr[ixa->ixa_ip_hdr_length]);
3143 3153
3144 3154 error = conn_ip_output(mp, ixa);
3145 3155 /* No udpOutErrors if an error since IP increases its error counter */
3146 3156 switch (error) {
3147 3157 case 0:
3148 3158 break;
3149 3159 case EWOULDBLOCK:
3150 3160 (void) ixa_check_drain_insert(connp, ixa);
3151 3161 error = 0;
3152 3162 break;
3153 3163 case EADDRNOTAVAIL:
3154 3164 /*
3155 3165 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3156 3166 * Don't have the application see that errno
3157 3167 */
3158 3168 error = ENETUNREACH;
3159 3169 /* FALLTHRU */
3160 3170 default:
3161 3171 mutex_enter(&connp->conn_lock);
3162 3172 /*
3163 3173 * Clear the source and v6lastdst so we call ip_attr_connect
3164 3174 * for the next packet and try to pick a better source.
3165 3175 */
3166 3176 if (connp->conn_mcbc_bind)
3167 3177 connp->conn_saddr_v6 = ipv6_all_zeros;
3168 3178 else
3169 3179 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3170 3180 connp->conn_v6lastdst = ipv6_all_zeros;
3171 3181 mutex_exit(&connp->conn_lock);
3172 3182 break;
3173 3183 }
3174 3184 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3175 3185 ixa->ixa_cred = connp->conn_cred; /* Restore */
3176 3186 ixa->ixa_cpid = connp->conn_cpid;
3177 3187 ixa_refrele(ixa);
3178 3188 return (error);
3179 3189 }
3180 3190
3181 3191
3182 3192 /*
3183 3193 * Prepend the header template and then fill in the source and
3184 3194 * flowinfo. The caller needs to handle the destination address since
3185 3195 * it's setting is different if rthdr or source route.
3186 3196 *
3187 3197 * Returns NULL is allocation failed or if the packet would exceed IP_MAXPACKET.
3188 3198 * When it returns NULL it sets errorp.
3189 3199 */
3190 3200 static mblk_t *
3191 3201 udp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
3192 3202 const in6_addr_t *v6src, in_port_t dstport, uint32_t flowinfo, int *errorp)
3193 3203 {
3194 3204 udp_t *udp = connp->conn_udp;
3195 3205 udp_stack_t *us = udp->udp_us;
3196 3206 boolean_t insert_spi = udp->udp_nat_t_endpoint;
3197 3207 uint_t pktlen;
3198 3208 uint_t alloclen;
3199 3209 uint_t copylen;
3200 3210 uint8_t *iph;
3201 3211 uint_t ip_hdr_length;
3202 3212 udpha_t *udpha;
3203 3213 uint32_t cksum;
3204 3214 ip_pkt_t *ipp;
3205 3215
3206 3216 ASSERT(MUTEX_HELD(&connp->conn_lock));
3207 3217
3208 3218 /*
3209 3219 * Copy the header template and leave space for an SPI
3210 3220 */
3211 3221 copylen = connp->conn_ht_iphc_len;
3212 3222 alloclen = copylen + (insert_spi ? sizeof (uint32_t) : 0);
3213 3223 pktlen = alloclen + msgdsize(mp);
3214 3224 if (pktlen > IP_MAXPACKET) {
3215 3225 freemsg(mp);
3216 3226 *errorp = EMSGSIZE;
3217 3227 return (NULL);
3218 3228 }
3219 3229 ixa->ixa_pktlen = pktlen;
3220 3230
3221 3231 /* check/fix buffer config, setup pointers into it */
3222 3232 iph = mp->b_rptr - alloclen;
3223 3233 if (DB_REF(mp) != 1 || iph < DB_BASE(mp) || !OK_32PTR(iph)) {
3224 3234 mblk_t *mp1;
3225 3235
3226 3236 mp1 = allocb(alloclen + us->us_wroff_extra, BPRI_MED);
3227 3237 if (mp1 == NULL) {
3228 3238 freemsg(mp);
3229 3239 *errorp = ENOMEM;
3230 3240 return (NULL);
3231 3241 }
3232 3242 mp1->b_wptr = DB_LIM(mp1);
3233 3243 mp1->b_cont = mp;
3234 3244 mp = mp1;
3235 3245 iph = (mp->b_wptr - alloclen);
3236 3246 }
3237 3247 mp->b_rptr = iph;
3238 3248 bcopy(connp->conn_ht_iphc, iph, copylen);
3239 3249 ip_hdr_length = (uint_t)(connp->conn_ht_ulp - connp->conn_ht_iphc);
3240 3250
3241 3251 ixa->ixa_ip_hdr_length = ip_hdr_length;
3242 3252 udpha = (udpha_t *)(iph + ip_hdr_length);
3243 3253
3244 3254 /*
3245 3255 * Setup header length and prepare for ULP checksum done in IP.
3246 3256 * udp_build_hdr_template has already massaged any routing header
3247 3257 * and placed the result in conn_sum.
3248 3258 *
3249 3259 * We make it easy for IP to include our pseudo header
3250 3260 * by putting our length in uha_checksum.
3251 3261 */
3252 3262 cksum = pktlen - ip_hdr_length;
3253 3263 udpha->uha_length = htons(cksum);
3254 3264
3255 3265 cksum += connp->conn_sum;
3256 3266 cksum = (cksum >> 16) + (cksum & 0xFFFF);
3257 3267 ASSERT(cksum < 0x10000);
3258 3268
3259 3269 ipp = &connp->conn_xmit_ipp;
3260 3270 if (ixa->ixa_flags & IXAF_IS_IPV4) {
3261 3271 ipha_t *ipha = (ipha_t *)iph;
3262 3272
3263 3273 ipha->ipha_length = htons((uint16_t)pktlen);
3264 3274
3265 3275 /* IP does the checksum if uha_checksum is non-zero */
3266 3276 if (us->us_do_checksum)
3267 3277 udpha->uha_checksum = htons(cksum);
3268 3278
3269 3279 /* if IP_PKTINFO specified an addres it wins over bind() */
3270 3280 if ((ipp->ipp_fields & IPPF_ADDR) &&
3271 3281 IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
3272 3282 ASSERT(ipp->ipp_addr_v4 != INADDR_ANY);
3273 3283 ipha->ipha_src = ipp->ipp_addr_v4;
3274 3284 } else {
3275 3285 IN6_V4MAPPED_TO_IPADDR(v6src, ipha->ipha_src);
3276 3286 }
3277 3287 } else {
3278 3288 ip6_t *ip6h = (ip6_t *)iph;
3279 3289
3280 3290 ip6h->ip6_plen = htons((uint16_t)(pktlen - IPV6_HDR_LEN));
3281 3291 udpha->uha_checksum = htons(cksum);
3282 3292
3283 3293 /* if IP_PKTINFO specified an addres it wins over bind() */
3284 3294 if ((ipp->ipp_fields & IPPF_ADDR) &&
3285 3295 !IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
3286 3296 ASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ipp->ipp_addr));
3287 3297 ip6h->ip6_src = ipp->ipp_addr;
3288 3298 } else {
3289 3299 ip6h->ip6_src = *v6src;
3290 3300 }
3291 3301 ip6h->ip6_vcf =
3292 3302 (IPV6_DEFAULT_VERS_AND_FLOW & IPV6_VERS_AND_FLOW_MASK) |
3293 3303 (flowinfo & ~IPV6_VERS_AND_FLOW_MASK);
3294 3304 if (ipp->ipp_fields & IPPF_TCLASS) {
3295 3305 /* Overrides the class part of flowinfo */
3296 3306 ip6h->ip6_vcf = IPV6_TCLASS_FLOW(ip6h->ip6_vcf,
3297 3307 ipp->ipp_tclass);
3298 3308 }
3299 3309 }
3300 3310
3301 3311 /* Insert all-0s SPI now. */
3302 3312 if (insert_spi)
3303 3313 *((uint32_t *)(udpha + 1)) = 0;
3304 3314
3305 3315 udpha->uha_dst_port = dstport;
3306 3316 return (mp);
3307 3317 }
3308 3318
3309 3319 /*
3310 3320 * Send a T_UDERR_IND in response to an M_DATA
3311 3321 */
3312 3322 static void
3313 3323 udp_ud_err_connected(conn_t *connp, t_scalar_t error)
3314 3324 {
3315 3325 struct sockaddr_storage ss;
3316 3326 sin_t *sin;
3317 3327 sin6_t *sin6;
3318 3328 struct sockaddr *addr;
3319 3329 socklen_t addrlen;
3320 3330 mblk_t *mp1;
3321 3331
3322 3332 mutex_enter(&connp->conn_lock);
3323 3333 /* Initialize addr and addrlen as if they're passed in */
3324 3334 if (connp->conn_family == AF_INET) {
3325 3335 sin = (sin_t *)&ss;
3326 3336 *sin = sin_null;
3327 3337 sin->sin_family = AF_INET;
3328 3338 sin->sin_port = connp->conn_fport;
3329 3339 sin->sin_addr.s_addr = connp->conn_faddr_v4;
3330 3340 addr = (struct sockaddr *)sin;
3331 3341 addrlen = sizeof (*sin);
3332 3342 } else {
3333 3343 sin6 = (sin6_t *)&ss;
3334 3344 *sin6 = sin6_null;
3335 3345 sin6->sin6_family = AF_INET6;
3336 3346 sin6->sin6_port = connp->conn_fport;
3337 3347 sin6->sin6_flowinfo = connp->conn_flowinfo;
3338 3348 sin6->sin6_addr = connp->conn_faddr_v6;
3339 3349 if (IN6_IS_ADDR_LINKSCOPE(&connp->conn_faddr_v6) &&
3340 3350 (connp->conn_ixa->ixa_flags & IXAF_SCOPEID_SET)) {
3341 3351 sin6->sin6_scope_id = connp->conn_ixa->ixa_scopeid;
3342 3352 } else {
3343 3353 sin6->sin6_scope_id = 0;
3344 3354 }
3345 3355 sin6->__sin6_src_id = 0;
3346 3356 addr = (struct sockaddr *)sin6;
3347 3357 addrlen = sizeof (*sin6);
3348 3358 }
3349 3359 mutex_exit(&connp->conn_lock);
3350 3360
3351 3361 mp1 = mi_tpi_uderror_ind((char *)addr, addrlen, NULL, 0, error);
3352 3362 if (mp1 != NULL)
3353 3363 putnext(connp->conn_rq, mp1);
3354 3364 }
3355 3365
3356 3366 /*
3357 3367 * This routine handles all messages passed downstream. It either
3358 3368 * consumes the message or passes it downstream; it never queues a
3359 3369 * a message.
3360 3370 *
3361 3371 * Also entry point for sockfs when udp is in "direct sockfs" mode. This mode
3362 3372 * is valid when we are directly beneath the stream head, and thus sockfs
3363 3373 * is able to bypass STREAMS and directly call us, passing along the sockaddr
3364 3374 * structure without the cumbersome T_UNITDATA_REQ interface for the case of
3365 3375 * connected endpoints.
3366 3376 */
3367 3377 void
3368 3378 udp_wput(queue_t *q, mblk_t *mp)
3369 3379 {
3370 3380 sin6_t *sin6;
3371 3381 sin_t *sin = NULL;
3372 3382 uint_t srcid;
3373 3383 conn_t *connp = Q_TO_CONN(q);
3374 3384 udp_t *udp = connp->conn_udp;
3375 3385 int error = 0;
3376 3386 struct sockaddr *addr = NULL;
3377 3387 socklen_t addrlen;
3378 3388 udp_stack_t *us = udp->udp_us;
3379 3389 struct T_unitdata_req *tudr;
3380 3390 mblk_t *data_mp;
3381 3391 ushort_t ipversion;
3382 3392 cred_t *cr;
3383 3393 pid_t pid;
3384 3394
3385 3395 /*
3386 3396 * We directly handle several cases here: T_UNITDATA_REQ message
3387 3397 * coming down as M_PROTO/M_PCPROTO and M_DATA messages for connected
3388 3398 * socket.
3389 3399 */
3390 3400 switch (DB_TYPE(mp)) {
3391 3401 case M_DATA:
3392 3402 if (!udp->udp_issocket || udp->udp_state != TS_DATA_XFER) {
3393 3403 /* Not connected; address is required */
3394 3404 UDPS_BUMP_MIB(us, udpOutErrors);
3395 3405 UDP_DBGSTAT(us, udp_data_notconn);
3396 3406 UDP_STAT(us, udp_out_err_notconn);
3397 3407 freemsg(mp);
3398 3408 return;
3399 3409 }
3400 3410 /*
3401 3411 * All Solaris components should pass a db_credp
3402 3412 * for this message, hence we ASSERT.
3403 3413 * On production kernels we return an error to be robust against
3404 3414 * random streams modules sitting on top of us.
3405 3415 */
3406 3416 cr = msg_getcred(mp, &pid);
3407 3417 ASSERT(cr != NULL);
3408 3418 if (cr == NULL) {
3409 3419 UDPS_BUMP_MIB(us, udpOutErrors);
3410 3420 freemsg(mp);
3411 3421 return;
3412 3422 }
3413 3423 ASSERT(udp->udp_issocket);
3414 3424 UDP_DBGSTAT(us, udp_data_conn);
3415 3425 error = udp_output_connected(connp, mp, cr, pid);
3416 3426 if (error != 0) {
3417 3427 UDP_STAT(us, udp_out_err_output);
3418 3428 if (connp->conn_rq != NULL)
3419 3429 udp_ud_err_connected(connp, (t_scalar_t)error);
3420 3430 #ifdef DEBUG
3421 3431 printf("udp_output_connected returned %d\n", error);
3422 3432 #endif
3423 3433 }
3424 3434 return;
3425 3435
3426 3436 case M_PROTO:
3427 3437 case M_PCPROTO:
3428 3438 tudr = (struct T_unitdata_req *)mp->b_rptr;
3429 3439 if (MBLKL(mp) < sizeof (*tudr) ||
3430 3440 ((t_primp_t)mp->b_rptr)->type != T_UNITDATA_REQ) {
3431 3441 udp_wput_other(q, mp);
3432 3442 return;
3433 3443 }
3434 3444 break;
3435 3445
3436 3446 default:
3437 3447 udp_wput_other(q, mp);
3438 3448 return;
3439 3449 }
3440 3450
3441 3451 /* Handle valid T_UNITDATA_REQ here */
3442 3452 data_mp = mp->b_cont;
3443 3453 if (data_mp == NULL) {
3444 3454 error = EPROTO;
3445 3455 goto ud_error2;
3446 3456 }
3447 3457 mp->b_cont = NULL;
3448 3458
3449 3459 if (!MBLKIN(mp, 0, tudr->DEST_offset + tudr->DEST_length)) {
3450 3460 error = EADDRNOTAVAIL;
3451 3461 goto ud_error2;
3452 3462 }
3453 3463
3454 3464 /*
3455 3465 * All Solaris components should pass a db_credp
3456 3466 * for this TPI message, hence we should ASSERT.
3457 3467 * However, RPC (svc_clts_ksend) does this odd thing where it
3458 3468 * passes the options from a T_UNITDATA_IND unchanged in a
3459 3469 * T_UNITDATA_REQ. While that is the right thing to do for
3460 3470 * some options, SCM_UCRED being the key one, this also makes it
3461 3471 * pass down IP_RECVDSTADDR. Hence we can't ASSERT here.
3462 3472 */
3463 3473 cr = msg_getcred(mp, &pid);
3464 3474 if (cr == NULL) {
3465 3475 cr = connp->conn_cred;
3466 3476 pid = connp->conn_cpid;
3467 3477 }
3468 3478
3469 3479 /*
3470 3480 * If a port has not been bound to the stream, fail.
3471 3481 * This is not a problem when sockfs is directly
3472 3482 * above us, because it will ensure that the socket
3473 3483 * is first bound before allowing data to be sent.
3474 3484 */
3475 3485 if (udp->udp_state == TS_UNBND) {
3476 3486 error = EPROTO;
3477 3487 goto ud_error2;
3478 3488 }
3479 3489 addr = (struct sockaddr *)&mp->b_rptr[tudr->DEST_offset];
3480 3490 addrlen = tudr->DEST_length;
3481 3491
3482 3492 switch (connp->conn_family) {
3483 3493 case AF_INET6:
3484 3494 sin6 = (sin6_t *)addr;
3485 3495 if (!OK_32PTR((char *)sin6) || (addrlen != sizeof (sin6_t)) ||
3486 3496 (sin6->sin6_family != AF_INET6)) {
3487 3497 error = EADDRNOTAVAIL;
3488 3498 goto ud_error2;
3489 3499 }
3490 3500
3491 3501 srcid = sin6->__sin6_src_id;
3492 3502 if (!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
3493 3503 /*
3494 3504 * Destination is a non-IPv4-compatible IPv6 address.
3495 3505 * Send out an IPv6 format packet.
3496 3506 */
3497 3507
3498 3508 /*
3499 3509 * If the local address is a mapped address return
3500 3510 * an error.
3501 3511 * It would be possible to send an IPv6 packet but the
3502 3512 * response would never make it back to the application
3503 3513 * since it is bound to a mapped address.
3504 3514 */
3505 3515 if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
3506 3516 error = EADDRNOTAVAIL;
3507 3517 goto ud_error2;
3508 3518 }
3509 3519
3510 3520 UDP_DBGSTAT(us, udp_out_ipv6);
3511 3521
3512 3522 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
3513 3523 sin6->sin6_addr = ipv6_loopback;
3514 3524 ipversion = IPV6_VERSION;
3515 3525 } else {
3516 3526 if (connp->conn_ipv6_v6only) {
3517 3527 error = EADDRNOTAVAIL;
3518 3528 goto ud_error2;
3519 3529 }
3520 3530
3521 3531 /*
3522 3532 * If the local address is not zero or a mapped address
3523 3533 * return an error. It would be possible to send an
3524 3534 * IPv4 packet but the response would never make it
3525 3535 * back to the application since it is bound to a
3526 3536 * non-mapped address.
3527 3537 */
3528 3538 if (!IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6) &&
3529 3539 !IN6_IS_ADDR_UNSPECIFIED(&connp->conn_saddr_v6)) {
3530 3540 error = EADDRNOTAVAIL;
3531 3541 goto ud_error2;
3532 3542 }
3533 3543 UDP_DBGSTAT(us, udp_out_mapped);
3534 3544
3535 3545 if (V4_PART_OF_V6(sin6->sin6_addr) == INADDR_ANY) {
3536 3546 V4_PART_OF_V6(sin6->sin6_addr) =
3537 3547 htonl(INADDR_LOOPBACK);
3538 3548 }
3539 3549 ipversion = IPV4_VERSION;
3540 3550 }
3541 3551
3542 3552 if (tudr->OPT_length != 0) {
3543 3553 /*
3544 3554 * If we are connected then the destination needs to be
3545 3555 * the same as the connected one.
3546 3556 */
3547 3557 if (udp->udp_state == TS_DATA_XFER &&
3548 3558 !conn_same_as_last_v6(connp, sin6)) {
3549 3559 error = EISCONN;
3550 3560 goto ud_error2;
3551 3561 }
3552 3562 UDP_STAT(us, udp_out_opt);
3553 3563 error = udp_output_ancillary(connp, NULL, sin6,
3554 3564 data_mp, mp, NULL, cr, pid);
3555 3565 } else {
3556 3566 ip_xmit_attr_t *ixa;
3557 3567
3558 3568 /*
3559 3569 * We have to allocate an ip_xmit_attr_t before we grab
3560 3570 * conn_lock and we need to hold conn_lock once we've
3561 3571 * checked conn_same_as_last_v6 to handle concurrent
3562 3572 * send* calls on a socket.
3563 3573 */
3564 3574 ixa = conn_get_ixa(connp, B_FALSE);
3565 3575 if (ixa == NULL) {
3566 3576 error = ENOMEM;
3567 3577 goto ud_error2;
3568 3578 }
3569 3579 mutex_enter(&connp->conn_lock);
3570 3580
3571 3581 if (conn_same_as_last_v6(connp, sin6) &&
3572 3582 connp->conn_lastsrcid == srcid &&
3573 3583 ipsec_outbound_policy_current(ixa)) {
3574 3584 UDP_DBGSTAT(us, udp_out_lastdst);
3575 3585 /* udp_output_lastdst drops conn_lock */
3576 3586 error = udp_output_lastdst(connp, data_mp, cr,
3577 3587 pid, ixa);
3578 3588 } else {
3579 3589 UDP_DBGSTAT(us, udp_out_diffdst);
3580 3590 /* udp_output_newdst drops conn_lock */
3581 3591 error = udp_output_newdst(connp, data_mp, NULL,
3582 3592 sin6, ipversion, cr, pid, ixa);
3583 3593 }
3584 3594 ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
3585 3595 }
3586 3596 if (error == 0) {
3587 3597 freeb(mp);
3588 3598 return;
3589 3599 }
3590 3600 break;
3591 3601
3592 3602 case AF_INET:
3593 3603 sin = (sin_t *)addr;
3594 3604 if ((!OK_32PTR((char *)sin) || addrlen != sizeof (sin_t)) ||
3595 3605 (sin->sin_family != AF_INET)) {
3596 3606 error = EADDRNOTAVAIL;
3597 3607 goto ud_error2;
3598 3608 }
3599 3609 UDP_DBGSTAT(us, udp_out_ipv4);
3600 3610 if (sin->sin_addr.s_addr == INADDR_ANY)
3601 3611 sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
3602 3612 ipversion = IPV4_VERSION;
3603 3613
3604 3614 srcid = 0;
3605 3615 if (tudr->OPT_length != 0) {
3606 3616 /*
3607 3617 * If we are connected then the destination needs to be
3608 3618 * the same as the connected one.
3609 3619 */
3610 3620 if (udp->udp_state == TS_DATA_XFER &&
3611 3621 !conn_same_as_last_v4(connp, sin)) {
3612 3622 error = EISCONN;
3613 3623 goto ud_error2;
3614 3624 }
3615 3625 UDP_STAT(us, udp_out_opt);
3616 3626 error = udp_output_ancillary(connp, sin, NULL,
3617 3627 data_mp, mp, NULL, cr, pid);
3618 3628 } else {
3619 3629 ip_xmit_attr_t *ixa;
3620 3630
3621 3631 /*
3622 3632 * We have to allocate an ip_xmit_attr_t before we grab
3623 3633 * conn_lock and we need to hold conn_lock once we've
3624 3634 * checked conn_same_as_last_v4 to handle concurrent
3625 3635 * send* calls on a socket.
3626 3636 */
3627 3637 ixa = conn_get_ixa(connp, B_FALSE);
3628 3638 if (ixa == NULL) {
3629 3639 error = ENOMEM;
3630 3640 goto ud_error2;
3631 3641 }
3632 3642 mutex_enter(&connp->conn_lock);
3633 3643
3634 3644 if (conn_same_as_last_v4(connp, sin) &&
3635 3645 ipsec_outbound_policy_current(ixa)) {
3636 3646 UDP_DBGSTAT(us, udp_out_lastdst);
3637 3647 /* udp_output_lastdst drops conn_lock */
3638 3648 error = udp_output_lastdst(connp, data_mp, cr,
3639 3649 pid, ixa);
3640 3650 } else {
3641 3651 UDP_DBGSTAT(us, udp_out_diffdst);
3642 3652 /* udp_output_newdst drops conn_lock */
3643 3653 error = udp_output_newdst(connp, data_mp, sin,
3644 3654 NULL, ipversion, cr, pid, ixa);
3645 3655 }
3646 3656 ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
3647 3657 }
3648 3658 if (error == 0) {
3649 3659 freeb(mp);
3650 3660 return;
3651 3661 }
3652 3662 break;
3653 3663 }
3654 3664 UDP_STAT(us, udp_out_err_output);
3655 3665 ASSERT(mp != NULL);
3656 3666 /* mp is freed by the following routine */
3657 3667 udp_ud_err(q, mp, (t_scalar_t)error);
3658 3668 return;
3659 3669
3660 3670 ud_error2:
3661 3671 UDPS_BUMP_MIB(us, udpOutErrors);
3662 3672 freemsg(data_mp);
3663 3673 UDP_STAT(us, udp_out_err_output);
3664 3674 ASSERT(mp != NULL);
3665 3675 /* mp is freed by the following routine */
3666 3676 udp_ud_err(q, mp, (t_scalar_t)error);
3667 3677 }
3668 3678
3669 3679 /*
3670 3680 * Handle the case of the IP address, port, flow label being different
3671 3681 * for both IPv4 and IPv6.
3672 3682 *
3673 3683 * NOTE: The caller must hold conn_lock and we drop it here.
3674 3684 */
3675 3685 static int
3676 3686 udp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin, sin6_t *sin6,
3677 3687 ushort_t ipversion, cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa)
3678 3688 {
3679 3689 uint_t srcid;
3680 3690 uint32_t flowinfo;
3681 3691 udp_t *udp = connp->conn_udp;
3682 3692 int error = 0;
3683 3693 ip_xmit_attr_t *oldixa;
3684 3694 udp_stack_t *us = udp->udp_us;
3685 3695 in6_addr_t v6src;
3686 3696 in6_addr_t v6dst;
3687 3697 in6_addr_t v6nexthop;
3688 3698 in_port_t dstport;
3689 3699
3690 3700 ASSERT(MUTEX_HELD(&connp->conn_lock));
3691 3701 ASSERT(ixa != NULL);
3692 3702 /*
3693 3703 * We hold conn_lock across all the use and modifications of
3694 3704 * the conn_lastdst, conn_ixa, and conn_xmit_ipp to ensure that they
3695 3705 * stay consistent.
3696 3706 */
3697 3707
3698 3708 ASSERT(cr != NULL);
3699 3709 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3700 3710 ixa->ixa_cred = cr;
3701 3711 ixa->ixa_cpid = pid;
3702 3712 if (is_system_labeled()) {
3703 3713 /* We need to restart with a label based on the cred */
3704 3714 ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3705 3715 }
3706 3716
3707 3717 /*
3708 3718 * If we are connected then the destination needs to be the
3709 3719 * same as the connected one, which is not the case here since we
3710 3720 * checked for that above.
3711 3721 */
3712 3722 if (udp->udp_state == TS_DATA_XFER) {
3713 3723 mutex_exit(&connp->conn_lock);
3714 3724 error = EISCONN;
3715 3725 goto ud_error;
3716 3726 }
3717 3727
3718 3728 /* In case previous destination was multicast or multirt */
3719 3729 ip_attr_newdst(ixa);
3720 3730
3721 3731 /*
3722 3732 * If laddr is unspecified then we look at sin6_src_id.
3723 3733 * We will give precedence to a source address set with IPV6_PKTINFO
3724 3734 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
3725 3735 * want ip_attr_connect to select a source (since it can fail) when
3726 3736 * IPV6_PKTINFO is specified.
3727 3737 * If this doesn't result in a source address then we get a source
3728 3738 * from ip_attr_connect() below.
3729 3739 */
3730 3740 v6src = connp->conn_saddr_v6;
3731 3741 if (sin != NULL) {
3732 3742 IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
3733 3743 dstport = sin->sin_port;
3734 3744 flowinfo = 0;
3735 3745 /* Don't bother with ip_srcid_find_id(), but indicate anyway. */
3736 3746 srcid = 0;
3737 3747 ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3738 3748 ixa->ixa_flags |= IXAF_IS_IPV4;
3739 3749 } else {
3740 3750 boolean_t v4mapped;
3741 3751
3742 3752 v6dst = sin6->sin6_addr;
3743 3753 dstport = sin6->sin6_port;
3744 3754 flowinfo = sin6->sin6_flowinfo;
3745 3755 srcid = sin6->__sin6_src_id;
3746 3756 if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
3747 3757 ixa->ixa_scopeid = sin6->sin6_scope_id;
3748 3758 ixa->ixa_flags |= IXAF_SCOPEID_SET;
3749 3759 } else {
3750 3760 ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3751 3761 }
3752 3762 v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
3753 3763 if (v4mapped)
3754 3764 ixa->ixa_flags |= IXAF_IS_IPV4;
3755 3765 else
3756 3766 ixa->ixa_flags &= ~IXAF_IS_IPV4;
3757 3767 if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
3758 3768 if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
3759 3769 v4mapped, connp->conn_netstack)) {
3760 3770 /* Mismatched v4mapped/v6 specified by srcid. */
3761 3771 mutex_exit(&connp->conn_lock);
3762 3772 error = EADDRNOTAVAIL;
3763 3773 goto ud_error;
3764 3774 }
3765 3775 }
3766 3776 }
3767 3777 /* Handle IP_PKTINFO/IPV6_PKTINFO setting source address. */
3768 3778 if (connp->conn_xmit_ipp.ipp_fields & IPPF_ADDR) {
3769 3779 ip_pkt_t *ipp = &connp->conn_xmit_ipp;
3770 3780
3771 3781 if (ixa->ixa_flags & IXAF_IS_IPV4) {
3772 3782 if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3773 3783 v6src = ipp->ipp_addr;
3774 3784 } else {
3775 3785 if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3776 3786 v6src = ipp->ipp_addr;
3777 3787 }
3778 3788 }
3779 3789
3780 3790 ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &v6dst, &v6nexthop);
3781 3791 mutex_exit(&connp->conn_lock);
3782 3792
3783 3793 error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
3784 3794 &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST | IPDF_IPSEC);
3785 3795 switch (error) {
3786 3796 case 0:
3787 3797 break;
3788 3798 case EADDRNOTAVAIL:
3789 3799 /*
3790 3800 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3791 3801 * Don't have the application see that errno
3792 3802 */
3793 3803 error = ENETUNREACH;
3794 3804 goto failed;
3795 3805 case ENETDOWN:
3796 3806 /*
3797 3807 * Have !ipif_addr_ready address; drop packet silently
3798 3808 * until we can get applications to not send until we
3799 3809 * are ready.
3800 3810 */
3801 3811 error = 0;
3802 3812 goto failed;
3803 3813 case EHOSTUNREACH:
3804 3814 case ENETUNREACH:
3805 3815 if (ixa->ixa_ire != NULL) {
3806 3816 /*
3807 3817 * Let conn_ip_output/ire_send_noroute return
3808 3818 * the error and send any local ICMP error.
3809 3819 */
3810 3820 error = 0;
3811 3821 break;
3812 3822 }
3813 3823 /* FALLTHRU */
3814 3824 failed:
3815 3825 default:
3816 3826 goto ud_error;
3817 3827 }
3818 3828
3819 3829
3820 3830 /*
3821 3831 * Cluster note: we let the cluster hook know that we are sending to a
3822 3832 * new address and/or port.
3823 3833 */
3824 3834 if (cl_inet_connect2 != NULL) {
3825 3835 CL_INET_UDP_CONNECT(connp, B_TRUE, &v6dst, dstport, error);
3826 3836 if (error != 0) {
3827 3837 error = EHOSTUNREACH;
3828 3838 goto ud_error;
3829 3839 }
3830 3840 }
3831 3841
3832 3842 mutex_enter(&connp->conn_lock);
3833 3843 /*
3834 3844 * While we dropped the lock some other thread might have connected
3835 3845 * this socket. If so we bail out with EISCONN to ensure that the
3836 3846 * connecting thread is the one that updates conn_ixa, conn_ht_*
3837 3847 * and conn_*last*.
3838 3848 */
3839 3849 if (udp->udp_state == TS_DATA_XFER) {
3840 3850 mutex_exit(&connp->conn_lock);
3841 3851 error = EISCONN;
3842 3852 goto ud_error;
3843 3853 }
3844 3854
3845 3855 /*
3846 3856 * We need to rebuild the headers if
3847 3857 * - we are labeling packets (could be different for different
3848 3858 * destinations)
3849 3859 * - we have a source route (or routing header) since we need to
3850 3860 * massage that to get the pseudo-header checksum
3851 3861 * - the IP version is different than the last time
3852 3862 * - a socket option with COA_HEADER_CHANGED has been set which
3853 3863 * set conn_v6lastdst to zero.
3854 3864 *
3855 3865 * Otherwise the prepend function will just update the src, dst,
3856 3866 * dstport, and flow label.
3857 3867 */
3858 3868 if (is_system_labeled()) {
3859 3869 /* TX MLP requires SCM_UCRED and don't have that here */
3860 3870 if (connp->conn_mlp_type != mlptSingle) {
3861 3871 mutex_exit(&connp->conn_lock);
3862 3872 error = ECONNREFUSED;
3863 3873 goto ud_error;
3864 3874 }
3865 3875 /*
3866 3876 * Check whether Trusted Solaris policy allows communication
3867 3877 * with this host, and pretend that the destination is
3868 3878 * unreachable if not.
3869 3879 * Compute any needed label and place it in ipp_label_v4/v6.
3870 3880 *
3871 3881 * Later conn_build_hdr_template/conn_prepend_hdr takes
3872 3882 * ipp_label_v4/v6 to form the packet.
3873 3883 *
3874 3884 * Tsol note: Since we hold conn_lock we know no other
3875 3885 * thread manipulates conn_xmit_ipp.
3876 3886 */
3877 3887 error = conn_update_label(connp, ixa, &v6dst,
3878 3888 &connp->conn_xmit_ipp);
3879 3889 if (error != 0) {
3880 3890 mutex_exit(&connp->conn_lock);
3881 3891 goto ud_error;
3882 3892 }
3883 3893 /* Rebuild the header template */
3884 3894 error = udp_build_hdr_template(connp, &v6src, &v6dst, dstport,
3885 3895 flowinfo);
3886 3896 if (error != 0) {
3887 3897 mutex_exit(&connp->conn_lock);
3888 3898 goto ud_error;
3889 3899 }
3890 3900 } else if ((connp->conn_xmit_ipp.ipp_fields &
3891 3901 (IPPF_IPV4_OPTIONS|IPPF_RTHDR)) ||
3892 3902 ipversion != connp->conn_lastipversion ||
3893 3903 IN6_IS_ADDR_UNSPECIFIED(&connp->conn_v6lastdst)) {
3894 3904 /* Rebuild the header template */
3895 3905 error = udp_build_hdr_template(connp, &v6src, &v6dst, dstport,
3896 3906 flowinfo);
3897 3907 if (error != 0) {
3898 3908 mutex_exit(&connp->conn_lock);
3899 3909 goto ud_error;
3900 3910 }
3901 3911 } else {
3902 3912 /* Simply update the destination address if no source route */
3903 3913 if (ixa->ixa_flags & IXAF_IS_IPV4) {
3904 3914 ipha_t *ipha = (ipha_t *)connp->conn_ht_iphc;
3905 3915
3906 3916 IN6_V4MAPPED_TO_IPADDR(&v6dst, ipha->ipha_dst);
3907 3917 if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
3908 3918 ipha->ipha_fragment_offset_and_flags |=
3909 3919 IPH_DF_HTONS;
3910 3920 } else {
3911 3921 ipha->ipha_fragment_offset_and_flags &=
3912 3922 ~IPH_DF_HTONS;
3913 3923 }
3914 3924 } else {
3915 3925 ip6_t *ip6h = (ip6_t *)connp->conn_ht_iphc;
3916 3926 ip6h->ip6_dst = v6dst;
3917 3927 }
3918 3928 }
3919 3929
3920 3930 /*
3921 3931 * Remember the dst/dstport etc which corresponds to the built header
3922 3932 * template and conn_ixa.
3923 3933 */
3924 3934 oldixa = conn_replace_ixa(connp, ixa);
3925 3935 connp->conn_v6lastdst = v6dst;
3926 3936 connp->conn_lastipversion = ipversion;
3927 3937 connp->conn_lastdstport = dstport;
3928 3938 connp->conn_lastflowinfo = flowinfo;
3929 3939 connp->conn_lastscopeid = ixa->ixa_scopeid;
3930 3940 connp->conn_lastsrcid = srcid;
3931 3941 /* Also remember a source to use together with lastdst */
3932 3942 connp->conn_v6lastsrc = v6src;
3933 3943
3934 3944 data_mp = udp_prepend_header_template(connp, ixa, data_mp, &v6src,
3935 3945 dstport, flowinfo, &error);
3936 3946
3937 3947 /* Done with conn_t */
3938 3948 mutex_exit(&connp->conn_lock);
3939 3949 ixa_refrele(oldixa);
3940 3950
3941 3951 if (data_mp == NULL) {
3942 3952 ASSERT(error != 0);
3943 3953 goto ud_error;
3944 3954 }
3945 3955
3946 3956 /* We're done. Pass the packet to ip. */
3947 3957 UDPS_BUMP_MIB(us, udpHCOutDatagrams);
3948 3958
3949 3959 DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
3950 3960 void_ip_t *, data_mp->b_rptr, udp_t *, udp, udpha_t *,
3951 3961 &data_mp->b_rptr[ixa->ixa_ip_hdr_length]);
3952 3962
3953 3963 error = conn_ip_output(data_mp, ixa);
3954 3964 /* No udpOutErrors if an error since IP increases its error counter */
3955 3965 switch (error) {
3956 3966 case 0:
3957 3967 break;
3958 3968 case EWOULDBLOCK:
3959 3969 (void) ixa_check_drain_insert(connp, ixa);
3960 3970 error = 0;
3961 3971 break;
3962 3972 case EADDRNOTAVAIL:
3963 3973 /*
3964 3974 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3965 3975 * Don't have the application see that errno
3966 3976 */
3967 3977 error = ENETUNREACH;
3968 3978 /* FALLTHRU */
3969 3979 default:
3970 3980 mutex_enter(&connp->conn_lock);
3971 3981 /*
3972 3982 * Clear the source and v6lastdst so we call ip_attr_connect
3973 3983 * for the next packet and try to pick a better source.
3974 3984 */
3975 3985 if (connp->conn_mcbc_bind)
3976 3986 connp->conn_saddr_v6 = ipv6_all_zeros;
3977 3987 else
3978 3988 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3979 3989 connp->conn_v6lastdst = ipv6_all_zeros;
3980 3990 mutex_exit(&connp->conn_lock);
3981 3991 break;
3982 3992 }
3983 3993 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3984 3994 ixa->ixa_cred = connp->conn_cred; /* Restore */
3985 3995 ixa->ixa_cpid = connp->conn_cpid;
3986 3996 ixa_refrele(ixa);
3987 3997 return (error);
3988 3998
3989 3999 ud_error:
3990 4000 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3991 4001 ixa->ixa_cred = connp->conn_cred; /* Restore */
3992 4002 ixa->ixa_cpid = connp->conn_cpid;
3993 4003 ixa_refrele(ixa);
3994 4004
3995 4005 freemsg(data_mp);
3996 4006 UDPS_BUMP_MIB(us, udpOutErrors);
3997 4007 UDP_STAT(us, udp_out_err_output);
3998 4008 return (error);
3999 4009 }
4000 4010
4001 4011 /* ARGSUSED */
4002 4012 static void
4003 4013 udp_wput_fallback(queue_t *wq, mblk_t *mp)
4004 4014 {
4005 4015 #ifdef DEBUG
4006 4016 cmn_err(CE_CONT, "udp_wput_fallback: Message in fallback \n");
4007 4017 #endif
4008 4018 freemsg(mp);
4009 4019 }
4010 4020
4011 4021
4012 4022 /*
4013 4023 * Handle special out-of-band ioctl requests (see PSARC/2008/265).
4014 4024 */
4015 4025 static void
4016 4026 udp_wput_cmdblk(queue_t *q, mblk_t *mp)
4017 4027 {
4018 4028 void *data;
4019 4029 mblk_t *datamp = mp->b_cont;
4020 4030 conn_t *connp = Q_TO_CONN(q);
4021 4031 udp_t *udp = connp->conn_udp;
4022 4032 cmdblk_t *cmdp = (cmdblk_t *)mp->b_rptr;
4023 4033
4024 4034 if (datamp == NULL || MBLKL(datamp) < cmdp->cb_len) {
4025 4035 cmdp->cb_error = EPROTO;
4026 4036 qreply(q, mp);
4027 4037 return;
4028 4038 }
4029 4039 data = datamp->b_rptr;
4030 4040
4031 4041 mutex_enter(&connp->conn_lock);
4032 4042 switch (cmdp->cb_cmd) {
4033 4043 case TI_GETPEERNAME:
4034 4044 if (udp->udp_state != TS_DATA_XFER)
4035 4045 cmdp->cb_error = ENOTCONN;
4036 4046 else
4037 4047 cmdp->cb_error = conn_getpeername(connp, data,
4038 4048 &cmdp->cb_len);
4039 4049 break;
4040 4050 case TI_GETMYNAME:
4041 4051 cmdp->cb_error = conn_getsockname(connp, data, &cmdp->cb_len);
4042 4052 break;
4043 4053 default:
4044 4054 cmdp->cb_error = EINVAL;
4045 4055 break;
4046 4056 }
4047 4057 mutex_exit(&connp->conn_lock);
4048 4058
4049 4059 qreply(q, mp);
4050 4060 }
4051 4061
4052 4062 static void
4053 4063 udp_use_pure_tpi(udp_t *udp)
4054 4064 {
4055 4065 conn_t *connp = udp->udp_connp;
4056 4066
4057 4067 mutex_enter(&connp->conn_lock);
4058 4068 udp->udp_issocket = B_FALSE;
4059 4069 mutex_exit(&connp->conn_lock);
4060 4070 UDP_STAT(udp->udp_us, udp_sock_fallback);
4061 4071 }
4062 4072
4063 4073 static void
4064 4074 udp_wput_other(queue_t *q, mblk_t *mp)
4065 4075 {
4066 4076 uchar_t *rptr = mp->b_rptr;
4067 4077 struct iocblk *iocp;
4068 4078 conn_t *connp = Q_TO_CONN(q);
4069 4079 udp_t *udp = connp->conn_udp;
4070 4080 cred_t *cr;
4071 4081
4072 4082 switch (mp->b_datap->db_type) {
4073 4083 case M_CMD:
4074 4084 udp_wput_cmdblk(q, mp);
4075 4085 return;
4076 4086
4077 4087 case M_PROTO:
4078 4088 case M_PCPROTO:
4079 4089 if (mp->b_wptr - rptr < sizeof (t_scalar_t)) {
4080 4090 /*
4081 4091 * If the message does not contain a PRIM_type,
4082 4092 * throw it away.
4083 4093 */
4084 4094 freemsg(mp);
4085 4095 return;
4086 4096 }
4087 4097 switch (((t_primp_t)rptr)->type) {
4088 4098 case T_ADDR_REQ:
4089 4099 udp_addr_req(q, mp);
4090 4100 return;
4091 4101 case O_T_BIND_REQ:
4092 4102 case T_BIND_REQ:
4093 4103 udp_tpi_bind(q, mp);
4094 4104 return;
4095 4105 case T_CONN_REQ:
4096 4106 udp_tpi_connect(q, mp);
4097 4107 return;
4098 4108 case T_CAPABILITY_REQ:
4099 4109 udp_capability_req(q, mp);
4100 4110 return;
4101 4111 case T_INFO_REQ:
4102 4112 udp_info_req(q, mp);
4103 4113 return;
4104 4114 case T_UNITDATA_REQ:
4105 4115 /*
4106 4116 * If a T_UNITDATA_REQ gets here, the address must
4107 4117 * be bad. Valid T_UNITDATA_REQs are handled
4108 4118 * in udp_wput.
4109 4119 */
4110 4120 udp_ud_err(q, mp, EADDRNOTAVAIL);
4111 4121 return;
4112 4122 case T_UNBIND_REQ:
4113 4123 udp_tpi_unbind(q, mp);
4114 4124 return;
4115 4125 case T_SVR4_OPTMGMT_REQ:
4116 4126 /*
4117 4127 * All Solaris components should pass a db_credp
4118 4128 * for this TPI message, hence we ASSERT.
4119 4129 * But in case there is some other M_PROTO that looks
4120 4130 * like a TPI message sent by some other kernel
4121 4131 * component, we check and return an error.
4122 4132 */
4123 4133 cr = msg_getcred(mp, NULL);
4124 4134 ASSERT(cr != NULL);
4125 4135 if (cr == NULL) {
4126 4136 udp_err_ack(q, mp, TSYSERR, EINVAL);
4127 4137 return;
4128 4138 }
4129 4139 if (!snmpcom_req(q, mp, udp_snmp_set, ip_snmp_get,
4130 4140 cr)) {
4131 4141 svr4_optcom_req(q, mp, cr, &udp_opt_obj);
4132 4142 }
4133 4143 return;
4134 4144
4135 4145 case T_OPTMGMT_REQ:
4136 4146 /*
4137 4147 * All Solaris components should pass a db_credp
4138 4148 * for this TPI message, hence we ASSERT.
4139 4149 * But in case there is some other M_PROTO that looks
4140 4150 * like a TPI message sent by some other kernel
4141 4151 * component, we check and return an error.
4142 4152 */
4143 4153 cr = msg_getcred(mp, NULL);
4144 4154 ASSERT(cr != NULL);
4145 4155 if (cr == NULL) {
4146 4156 udp_err_ack(q, mp, TSYSERR, EINVAL);
4147 4157 return;
4148 4158 }
4149 4159 tpi_optcom_req(q, mp, cr, &udp_opt_obj);
4150 4160 return;
4151 4161
4152 4162 case T_DISCON_REQ:
4153 4163 udp_tpi_disconnect(q, mp);
4154 4164 return;
4155 4165
4156 4166 /* The following TPI message is not supported by udp. */
4157 4167 case O_T_CONN_RES:
4158 4168 case T_CONN_RES:
4159 4169 udp_err_ack(q, mp, TNOTSUPPORT, 0);
4160 4170 return;
4161 4171
4162 4172 /* The following 3 TPI requests are illegal for udp. */
4163 4173 case T_DATA_REQ:
4164 4174 case T_EXDATA_REQ:
4165 4175 case T_ORDREL_REQ:
4166 4176 udp_err_ack(q, mp, TNOTSUPPORT, 0);
4167 4177 return;
4168 4178 default:
4169 4179 break;
4170 4180 }
4171 4181 break;
4172 4182 case M_FLUSH:
4173 4183 if (*rptr & FLUSHW)
4174 4184 flushq(q, FLUSHDATA);
4175 4185 break;
4176 4186 case M_IOCTL:
4177 4187 iocp = (struct iocblk *)mp->b_rptr;
4178 4188 switch (iocp->ioc_cmd) {
4179 4189 case TI_GETPEERNAME:
4180 4190 if (udp->udp_state != TS_DATA_XFER) {
4181 4191 /*
4182 4192 * If a default destination address has not
4183 4193 * been associated with the stream, then we
4184 4194 * don't know the peer's name.
4185 4195 */
4186 4196 iocp->ioc_error = ENOTCONN;
4187 4197 iocp->ioc_count = 0;
4188 4198 mp->b_datap->db_type = M_IOCACK;
4189 4199 qreply(q, mp);
4190 4200 return;
4191 4201 }
4192 4202 /* FALLTHRU */
4193 4203 case TI_GETMYNAME:
4194 4204 /*
4195 4205 * For TI_GETPEERNAME and TI_GETMYNAME, we first
4196 4206 * need to copyin the user's strbuf structure.
4197 4207 * Processing will continue in the M_IOCDATA case
4198 4208 * below.
4199 4209 */
4200 4210 mi_copyin(q, mp, NULL,
4201 4211 SIZEOF_STRUCT(strbuf, iocp->ioc_flag));
4202 4212 return;
4203 4213 case _SIOCSOCKFALLBACK:
4204 4214 /*
4205 4215 * Either sockmod is about to be popped and the
4206 4216 * socket would now be treated as a plain stream,
4207 4217 * or a module is about to be pushed so we have
4208 4218 * to follow pure TPI semantics.
4209 4219 */
4210 4220 if (!udp->udp_issocket) {
4211 4221 DB_TYPE(mp) = M_IOCNAK;
4212 4222 iocp->ioc_error = EINVAL;
4213 4223 } else {
4214 4224 udp_use_pure_tpi(udp);
4215 4225
4216 4226 DB_TYPE(mp) = M_IOCACK;
4217 4227 iocp->ioc_error = 0;
4218 4228 }
4219 4229 iocp->ioc_count = 0;
4220 4230 iocp->ioc_rval = 0;
4221 4231 qreply(q, mp);
4222 4232 return;
4223 4233 default:
4224 4234 break;
4225 4235 }
4226 4236 break;
4227 4237 case M_IOCDATA:
4228 4238 udp_wput_iocdata(q, mp);
4229 4239 return;
4230 4240 default:
4231 4241 /* Unrecognized messages are passed through without change. */
4232 4242 break;
4233 4243 }
4234 4244 ip_wput_nondata(q, mp);
4235 4245 }
4236 4246
4237 4247 /*
4238 4248 * udp_wput_iocdata is called by udp_wput_other to handle all M_IOCDATA
4239 4249 * messages.
4240 4250 */
4241 4251 static void
4242 4252 udp_wput_iocdata(queue_t *q, mblk_t *mp)
4243 4253 {
4244 4254 mblk_t *mp1;
4245 4255 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
4246 4256 STRUCT_HANDLE(strbuf, sb);
4247 4257 uint_t addrlen;
4248 4258 conn_t *connp = Q_TO_CONN(q);
4249 4259 udp_t *udp = connp->conn_udp;
4250 4260
4251 4261 /* Make sure it is one of ours. */
4252 4262 switch (iocp->ioc_cmd) {
4253 4263 case TI_GETMYNAME:
4254 4264 case TI_GETPEERNAME:
4255 4265 break;
4256 4266 default:
4257 4267 ip_wput_nondata(q, mp);
4258 4268 return;
4259 4269 }
4260 4270
4261 4271 switch (mi_copy_state(q, mp, &mp1)) {
4262 4272 case -1:
4263 4273 return;
4264 4274 case MI_COPY_CASE(MI_COPY_IN, 1):
4265 4275 break;
4266 4276 case MI_COPY_CASE(MI_COPY_OUT, 1):
4267 4277 /*
4268 4278 * The address has been copied out, so now
4269 4279 * copyout the strbuf.
4270 4280 */
4271 4281 mi_copyout(q, mp);
4272 4282 return;
4273 4283 case MI_COPY_CASE(MI_COPY_OUT, 2):
4274 4284 /*
4275 4285 * The address and strbuf have been copied out.
4276 4286 * We're done, so just acknowledge the original
4277 4287 * M_IOCTL.
4278 4288 */
4279 4289 mi_copy_done(q, mp, 0);
4280 4290 return;
4281 4291 default:
4282 4292 /*
4283 4293 * Something strange has happened, so acknowledge
4284 4294 * the original M_IOCTL with an EPROTO error.
4285 4295 */
4286 4296 mi_copy_done(q, mp, EPROTO);
4287 4297 return;
4288 4298 }
4289 4299
4290 4300 /*
4291 4301 * Now we have the strbuf structure for TI_GETMYNAME
4292 4302 * and TI_GETPEERNAME. Next we copyout the requested
4293 4303 * address and then we'll copyout the strbuf.
4294 4304 */
4295 4305 STRUCT_SET_HANDLE(sb, iocp->ioc_flag, (void *)mp1->b_rptr);
4296 4306
4297 4307 if (connp->conn_family == AF_INET)
4298 4308 addrlen = sizeof (sin_t);
4299 4309 else
4300 4310 addrlen = sizeof (sin6_t);
4301 4311
4302 4312 if (STRUCT_FGET(sb, maxlen) < addrlen) {
4303 4313 mi_copy_done(q, mp, EINVAL);
4304 4314 return;
4305 4315 }
4306 4316
4307 4317 switch (iocp->ioc_cmd) {
4308 4318 case TI_GETMYNAME:
4309 4319 break;
4310 4320 case TI_GETPEERNAME:
4311 4321 if (udp->udp_state != TS_DATA_XFER) {
4312 4322 mi_copy_done(q, mp, ENOTCONN);
4313 4323 return;
4314 4324 }
4315 4325 break;
4316 4326 }
4317 4327 mp1 = mi_copyout_alloc(q, mp, STRUCT_FGETP(sb, buf), addrlen, B_TRUE);
4318 4328 if (!mp1)
4319 4329 return;
4320 4330
4321 4331 STRUCT_FSET(sb, len, addrlen);
4322 4332 switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4323 4333 case TI_GETMYNAME:
4324 4334 (void) conn_getsockname(connp, (struct sockaddr *)mp1->b_wptr,
4325 4335 &addrlen);
4326 4336 break;
4327 4337 case TI_GETPEERNAME:
4328 4338 (void) conn_getpeername(connp, (struct sockaddr *)mp1->b_wptr,
4329 4339 &addrlen);
4330 4340 break;
4331 4341 }
4332 4342 mp1->b_wptr += addrlen;
4333 4343 /* Copy out the address */
4334 4344 mi_copyout(q, mp);
4335 4345 }
4336 4346
4337 4347 void
4338 4348 udp_ddi_g_init(void)
4339 4349 {
4340 4350 udp_max_optsize = optcom_max_optsize(udp_opt_obj.odb_opt_des_arr,
4341 4351 udp_opt_obj.odb_opt_arr_cnt);
4342 4352
4343 4353 /*
4344 4354 * We want to be informed each time a stack is created or
4345 4355 * destroyed in the kernel, so we can maintain the
4346 4356 * set of udp_stack_t's.
4347 4357 */
4348 4358 netstack_register(NS_UDP, udp_stack_init, NULL, udp_stack_fini);
4349 4359 }
4350 4360
4351 4361 void
4352 4362 udp_ddi_g_destroy(void)
4353 4363 {
4354 4364 netstack_unregister(NS_UDP);
4355 4365 }
4356 4366
4357 4367 #define INET_NAME "ip"
4358 4368
4359 4369 /*
4360 4370 * Initialize the UDP stack instance.
4361 4371 */
4362 4372 static void *
4363 4373 udp_stack_init(netstackid_t stackid, netstack_t *ns)
4364 4374 {
4365 4375 udp_stack_t *us;
4366 4376 int i;
4367 4377 int error = 0;
4368 4378 major_t major;
4369 4379 size_t arrsz;
4370 4380
4371 4381 us = (udp_stack_t *)kmem_zalloc(sizeof (*us), KM_SLEEP);
4372 4382 us->us_netstack = ns;
4373 4383
4374 4384 mutex_init(&us->us_epriv_port_lock, NULL, MUTEX_DEFAULT, NULL);
4375 4385 us->us_num_epriv_ports = UDP_NUM_EPRIV_PORTS;
4376 4386 us->us_epriv_ports[0] = ULP_DEF_EPRIV_PORT1;
4377 4387 us->us_epriv_ports[1] = ULP_DEF_EPRIV_PORT2;
4378 4388
4379 4389 /*
4380 4390 * The smallest anonymous port in the priviledged port range which UDP
4381 4391 * looks for free port. Use in the option UDP_ANONPRIVBIND.
4382 4392 */
4383 4393 us->us_min_anonpriv_port = 512;
4384 4394
4385 4395 us->us_bind_fanout_size = udp_bind_fanout_size;
4386 4396
4387 4397 /* Roundup variable that might have been modified in /etc/system */
4388 4398 if (!ISP2(us->us_bind_fanout_size)) {
4389 4399 /* Not a power of two. Round up to nearest power of two */
4390 4400 for (i = 0; i < 31; i++) {
4391 4401 if (us->us_bind_fanout_size < (1 << i))
4392 4402 break;
4393 4403 }
4394 4404 us->us_bind_fanout_size = 1 << i;
4395 4405 }
4396 4406 us->us_bind_fanout = kmem_zalloc(us->us_bind_fanout_size *
4397 4407 sizeof (udp_fanout_t), KM_SLEEP);
4398 4408 for (i = 0; i < us->us_bind_fanout_size; i++) {
4399 4409 mutex_init(&us->us_bind_fanout[i].uf_lock, NULL, MUTEX_DEFAULT,
4400 4410 NULL);
4401 4411 }
4402 4412
4403 4413 arrsz = udp_propinfo_count * sizeof (mod_prop_info_t);
4404 4414 us->us_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz,
4405 4415 KM_SLEEP);
4406 4416 bcopy(udp_propinfo_tbl, us->us_propinfo_tbl, arrsz);
4407 4417
4408 4418 /* Allocate the per netstack stats */
4409 4419 mutex_enter(&cpu_lock);
4410 4420 us->us_sc_cnt = MAX(ncpus, boot_ncpus);
4411 4421 mutex_exit(&cpu_lock);
4412 4422 us->us_sc = kmem_zalloc(max_ncpus * sizeof (udp_stats_cpu_t *),
4413 4423 KM_SLEEP);
4414 4424 for (i = 0; i < us->us_sc_cnt; i++) {
4415 4425 us->us_sc[i] = kmem_zalloc(sizeof (udp_stats_cpu_t),
4416 4426 KM_SLEEP);
4417 4427 }
4418 4428
4419 4429 us->us_kstat = udp_kstat2_init(stackid);
4420 4430 us->us_mibkp = udp_kstat_init(stackid);
4421 4431
4422 4432 major = mod_name_to_major(INET_NAME);
4423 4433 error = ldi_ident_from_major(major, &us->us_ldi_ident);
4424 4434 ASSERT(error == 0);
4425 4435 return (us);
4426 4436 }
4427 4437
4428 4438 /*
4429 4439 * Free the UDP stack instance.
4430 4440 */
4431 4441 static void
4432 4442 udp_stack_fini(netstackid_t stackid, void *arg)
4433 4443 {
4434 4444 udp_stack_t *us = (udp_stack_t *)arg;
4435 4445 int i;
4436 4446
4437 4447 for (i = 0; i < us->us_bind_fanout_size; i++) {
4438 4448 mutex_destroy(&us->us_bind_fanout[i].uf_lock);
4439 4449 }
4440 4450
4441 4451 kmem_free(us->us_bind_fanout, us->us_bind_fanout_size *
4442 4452 sizeof (udp_fanout_t));
4443 4453
4444 4454 us->us_bind_fanout = NULL;
4445 4455
4446 4456 for (i = 0; i < us->us_sc_cnt; i++)
4447 4457 kmem_free(us->us_sc[i], sizeof (udp_stats_cpu_t));
4448 4458 kmem_free(us->us_sc, max_ncpus * sizeof (udp_stats_cpu_t *));
4449 4459
4450 4460 kmem_free(us->us_propinfo_tbl,
4451 4461 udp_propinfo_count * sizeof (mod_prop_info_t));
4452 4462 us->us_propinfo_tbl = NULL;
4453 4463
4454 4464 udp_kstat_fini(stackid, us->us_mibkp);
4455 4465 us->us_mibkp = NULL;
4456 4466
4457 4467 udp_kstat2_fini(stackid, us->us_kstat);
4458 4468 us->us_kstat = NULL;
4459 4469
4460 4470 mutex_destroy(&us->us_epriv_port_lock);
4461 4471 ldi_ident_release(us->us_ldi_ident);
4462 4472 kmem_free(us, sizeof (*us));
4463 4473 }
4464 4474
4465 4475 static size_t
4466 4476 udp_set_rcv_hiwat(udp_t *udp, size_t size)
4467 4477 {
4468 4478 udp_stack_t *us = udp->udp_us;
4469 4479
4470 4480 /* We add a bit of extra buffering */
4471 4481 size += size >> 1;
4472 4482 if (size > us->us_max_buf)
4473 4483 size = us->us_max_buf;
4474 4484
4475 4485 udp->udp_rcv_hiwat = size;
4476 4486 return (size);
4477 4487 }
4478 4488
4479 4489 /*
4480 4490 * For the lower queue so that UDP can be a dummy mux.
4481 4491 * Nobody should be sending
4482 4492 * packets up this stream
4483 4493 */
4484 4494 static void
4485 4495 udp_lrput(queue_t *q, mblk_t *mp)
4486 4496 {
4487 4497 switch (mp->b_datap->db_type) {
4488 4498 case M_FLUSH:
4489 4499 /* Turn around */
4490 4500 if (*mp->b_rptr & FLUSHW) {
4491 4501 *mp->b_rptr &= ~FLUSHR;
4492 4502 qreply(q, mp);
4493 4503 return;
4494 4504 }
4495 4505 break;
4496 4506 }
4497 4507 freemsg(mp);
4498 4508 }
4499 4509
4500 4510 /*
4501 4511 * For the lower queue so that UDP can be a dummy mux.
4502 4512 * Nobody should be sending packets down this stream.
4503 4513 */
4504 4514 /* ARGSUSED */
4505 4515 void
4506 4516 udp_lwput(queue_t *q, mblk_t *mp)
4507 4517 {
4508 4518 freemsg(mp);
4509 4519 }
4510 4520
4511 4521 /*
4512 4522 * When a CPU is added, we need to allocate the per CPU stats struct.
4513 4523 */
4514 4524 void
4515 4525 udp_stack_cpu_add(udp_stack_t *us, processorid_t cpu_seqid)
4516 4526 {
4517 4527 int i;
4518 4528
4519 4529 if (cpu_seqid < us->us_sc_cnt)
4520 4530 return;
4521 4531 for (i = us->us_sc_cnt; i <= cpu_seqid; i++) {
4522 4532 ASSERT(us->us_sc[i] == NULL);
4523 4533 us->us_sc[i] = kmem_zalloc(sizeof (udp_stats_cpu_t),
4524 4534 KM_SLEEP);
4525 4535 }
4526 4536 membar_producer();
4527 4537 us->us_sc_cnt = cpu_seqid + 1;
4528 4538 }
4529 4539
4530 4540 /*
4531 4541 * Below routines for UDP socket module.
4532 4542 */
4533 4543
4534 4544 static conn_t *
4535 4545 udp_do_open(cred_t *credp, boolean_t isv6, int flags, int *errorp)
4536 4546 {
4537 4547 udp_t *udp;
4538 4548 conn_t *connp;
4539 4549 zoneid_t zoneid;
4540 4550 netstack_t *ns;
4541 4551 udp_stack_t *us;
4542 4552 int len;
4543 4553
4544 4554 ASSERT(errorp != NULL);
4545 4555
4546 4556 if ((*errorp = secpolicy_basic_net_access(credp)) != 0)
4547 4557 return (NULL);
4548 4558
4549 4559 ns = netstack_find_by_cred(credp);
4550 4560 ASSERT(ns != NULL);
4551 4561 us = ns->netstack_udp;
4552 4562 ASSERT(us != NULL);
4553 4563
4554 4564 /*
4555 4565 * For exclusive stacks we set the zoneid to zero
4556 4566 * to make UDP operate as if in the global zone.
4557 4567 */
4558 4568 if (ns->netstack_stackid != GLOBAL_NETSTACKID)
4559 4569 zoneid = GLOBAL_ZONEID;
4560 4570 else
4561 4571 zoneid = crgetzoneid(credp);
4562 4572
4563 4573 ASSERT(flags == KM_SLEEP || flags == KM_NOSLEEP);
4564 4574
4565 4575 connp = ipcl_conn_create(IPCL_UDPCONN, flags, ns);
4566 4576 if (connp == NULL) {
4567 4577 netstack_rele(ns);
4568 4578 *errorp = ENOMEM;
4569 4579 return (NULL);
4570 4580 }
4571 4581 udp = connp->conn_udp;
4572 4582
4573 4583 /*
4574 4584 * ipcl_conn_create did a netstack_hold. Undo the hold that was
4575 4585 * done by netstack_find_by_cred()
4576 4586 */
4577 4587 netstack_rele(ns);
4578 4588
4579 4589 /*
4580 4590 * Since this conn_t/udp_t is not yet visible to anybody else we don't
4581 4591 * need to lock anything.
4582 4592 */
4583 4593 ASSERT(connp->conn_proto == IPPROTO_UDP);
4584 4594 ASSERT(connp->conn_udp == udp);
4585 4595 ASSERT(udp->udp_connp == connp);
4586 4596
4587 4597 /* Set the initial state of the stream and the privilege status. */
4588 4598 udp->udp_state = TS_UNBND;
4589 4599 connp->conn_ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
4590 4600 if (isv6) {
4591 4601 connp->conn_family = AF_INET6;
4592 4602 connp->conn_ipversion = IPV6_VERSION;
4593 4603 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
4594 4604 connp->conn_default_ttl = us->us_ipv6_hoplimit;
4595 4605 len = sizeof (ip6_t) + UDPH_SIZE;
4596 4606 } else {
4597 4607 connp->conn_family = AF_INET;
4598 4608 connp->conn_ipversion = IPV4_VERSION;
4599 4609 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
4600 4610 connp->conn_default_ttl = us->us_ipv4_ttl;
4601 4611 len = sizeof (ipha_t) + UDPH_SIZE;
4602 4612 }
4603 4613
4604 4614 ASSERT(connp->conn_ixa->ixa_protocol == connp->conn_proto);
4605 4615 connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;
4606 4616
4607 4617 connp->conn_ixa->ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
4608 4618 connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
4609 4619 /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
4610 4620 connp->conn_ixa->ixa_zoneid = zoneid;
4611 4621
4612 4622 connp->conn_zoneid = zoneid;
4613 4623
4614 4624 /*
4615 4625 * If the caller has the process-wide flag set, then default to MAC
4616 4626 * exempt mode. This allows read-down to unlabeled hosts.
4617 4627 */
4618 4628 if (getpflags(NET_MAC_AWARE, credp) != 0)
4619 4629 connp->conn_mac_mode = CONN_MAC_AWARE;
4620 4630
4621 4631 connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
4622 4632
4623 4633 udp->udp_us = us;
4624 4634
4625 4635 connp->conn_rcvbuf = us->us_recv_hiwat;
4626 4636 connp->conn_sndbuf = us->us_xmit_hiwat;
4627 4637 connp->conn_sndlowat = us->us_xmit_lowat;
4628 4638 connp->conn_rcvlowat = udp_mod_info.mi_lowat;
4629 4639
4630 4640 connp->conn_wroff = len + us->us_wroff_extra;
4631 4641 connp->conn_so_type = SOCK_DGRAM;
4632 4642
4633 4643 connp->conn_recv = udp_input;
4634 4644 connp->conn_recvicmp = udp_icmp_input;
4635 4645 crhold(credp);
4636 4646 connp->conn_cred = credp;
4637 4647 connp->conn_cpid = curproc->p_pid;
4638 4648 connp->conn_open_time = ddi_get_lbolt64();
4639 4649 /* Cache things in ixa without an extra refhold */
4640 4650 ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
4641 4651 connp->conn_ixa->ixa_cred = connp->conn_cred;
4642 4652 connp->conn_ixa->ixa_cpid = connp->conn_cpid;
4643 4653 if (is_system_labeled())
4644 4654 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
4645 4655
4646 4656 *((sin6_t *)&udp->udp_delayed_addr) = sin6_null;
4647 4657
4648 4658 if (us->us_pmtu_discovery)
4649 4659 connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;
4650 4660
4651 4661 return (connp);
4652 4662 }
4653 4663
4654 4664 sock_lower_handle_t
4655 4665 udp_create(int family, int type, int proto, sock_downcalls_t **sock_downcalls,
4656 4666 uint_t *smodep, int *errorp, int flags, cred_t *credp)
4657 4667 {
4658 4668 udp_t *udp = NULL;
4659 4669 udp_stack_t *us;
4660 4670 conn_t *connp;
4661 4671 boolean_t isv6;
4662 4672
4663 4673 if (type != SOCK_DGRAM || (family != AF_INET && family != AF_INET6) ||
4664 4674 (proto != 0 && proto != IPPROTO_UDP)) {
4665 4675 *errorp = EPROTONOSUPPORT;
4666 4676 return (NULL);
4667 4677 }
4668 4678
4669 4679 if (family == AF_INET6)
4670 4680 isv6 = B_TRUE;
4671 4681 else
4672 4682 isv6 = B_FALSE;
4673 4683
4674 4684 connp = udp_do_open(credp, isv6, flags, errorp);
4675 4685 if (connp == NULL)
4676 4686 return (NULL);
4677 4687
4678 4688 udp = connp->conn_udp;
4679 4689 ASSERT(udp != NULL);
4680 4690 us = udp->udp_us;
4681 4691 ASSERT(us != NULL);
4682 4692
4683 4693 udp->udp_issocket = B_TRUE;
4684 4694 connp->conn_flags |= IPCL_NONSTR;
4685 4695
4686 4696 /*
4687 4697 * Set flow control
4688 4698 * Since this conn_t/udp_t is not yet visible to anybody else we don't
4689 4699 * need to lock anything.
4690 4700 */
4691 4701 (void) udp_set_rcv_hiwat(udp, connp->conn_rcvbuf);
4692 4702 udp->udp_rcv_disply_hiwat = connp->conn_rcvbuf;
4693 4703
4694 4704 connp->conn_flow_cntrld = B_FALSE;
4695 4705
4696 4706 mutex_enter(&connp->conn_lock);
4697 4707 connp->conn_state_flags &= ~CONN_INCIPIENT;
4698 4708 mutex_exit(&connp->conn_lock);
4699 4709
4700 4710 *errorp = 0;
4701 4711 *smodep = SM_ATOMIC;
4702 4712 *sock_downcalls = &sock_udp_downcalls;
4703 4713 return ((sock_lower_handle_t)connp);
4704 4714 }
4705 4715
4706 4716 /* ARGSUSED3 */
4707 4717 void
4708 4718 udp_activate(sock_lower_handle_t proto_handle, sock_upper_handle_t sock_handle,
4709 4719 sock_upcalls_t *sock_upcalls, int flags, cred_t *cr)
4710 4720 {
4711 4721 conn_t *connp = (conn_t *)proto_handle;
4712 4722 struct sock_proto_props sopp;
4713 4723
4714 4724 /* All Solaris components should pass a cred for this operation. */
4715 4725 ASSERT(cr != NULL);
4716 4726
4717 4727 connp->conn_upcalls = sock_upcalls;
4718 4728 connp->conn_upper_handle = sock_handle;
4719 4729
4720 4730 sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT |
4721 4731 SOCKOPT_MAXBLK | SOCKOPT_MAXPSZ | SOCKOPT_MINPSZ;
4722 4732 sopp.sopp_wroff = connp->conn_wroff;
4723 4733 sopp.sopp_maxblk = INFPSZ;
4724 4734 sopp.sopp_rxhiwat = connp->conn_rcvbuf;
4725 4735 sopp.sopp_rxlowat = connp->conn_rcvlowat;
4726 4736 sopp.sopp_maxaddrlen = sizeof (sin6_t);
4727 4737 sopp.sopp_maxpsz =
4728 4738 (connp->conn_family == AF_INET) ? UDP_MAXPACKET_IPV4 :
4729 4739 UDP_MAXPACKET_IPV6;
4730 4740 sopp.sopp_minpsz = (udp_mod_info.mi_minpsz == 1) ? 0 :
4731 4741 udp_mod_info.mi_minpsz;
4732 4742
4733 4743 (*connp->conn_upcalls->su_set_proto_props)(connp->conn_upper_handle,
4734 4744 &sopp);
4735 4745 }
4736 4746
4737 4747 static void
4738 4748 udp_do_close(conn_t *connp)
4739 4749 {
4740 4750 udp_t *udp;
4741 4751
4742 4752 ASSERT(connp != NULL && IPCL_IS_UDP(connp));
4743 4753 udp = connp->conn_udp;
4744 4754
4745 4755 if (cl_inet_unbind != NULL && udp->udp_state == TS_IDLE) {
4746 4756 /*
4747 4757 * Running in cluster mode - register unbind information
4748 4758 */
4749 4759 if (connp->conn_ipversion == IPV4_VERSION) {
4750 4760 (*cl_inet_unbind)(
4751 4761 connp->conn_netstack->netstack_stackid,
4752 4762 IPPROTO_UDP, AF_INET,
4753 4763 (uint8_t *)(&V4_PART_OF_V6(connp->conn_laddr_v6)),
4754 4764 (in_port_t)connp->conn_lport, NULL);
4755 4765 } else {
4756 4766 (*cl_inet_unbind)(
4757 4767 connp->conn_netstack->netstack_stackid,
4758 4768 IPPROTO_UDP, AF_INET6,
4759 4769 (uint8_t *)&(connp->conn_laddr_v6),
4760 4770 (in_port_t)connp->conn_lport, NULL);
4761 4771 }
4762 4772 }
4763 4773
4764 4774 udp_bind_hash_remove(udp, B_FALSE);
4765 4775
4766 4776 ip_quiesce_conn(connp);
4767 4777
4768 4778 if (!IPCL_IS_NONSTR(connp)) {
4769 4779 ASSERT(connp->conn_wq != NULL);
4770 4780 ASSERT(connp->conn_rq != NULL);
4771 4781 qprocsoff(connp->conn_rq);
4772 4782 }
4773 4783
4774 4784 udp_close_free(connp);
4775 4785
4776 4786 /*
4777 4787 * Now we are truly single threaded on this stream, and can
4778 4788 * delete the things hanging off the connp, and finally the connp.
4779 4789 * We removed this connp from the fanout list, it cannot be
4780 4790 * accessed thru the fanouts, and we already waited for the
4781 4791 * conn_ref to drop to 0. We are already in close, so
4782 4792 * there cannot be any other thread from the top. qprocsoff
4783 4793 * has completed, and service has completed or won't run in
4784 4794 * future.
4785 4795 */
4786 4796 ASSERT(connp->conn_ref == 1);
4787 4797
4788 4798 if (!IPCL_IS_NONSTR(connp)) {
4789 4799 inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4790 4800 } else {
4791 4801 ip_free_helper_stream(connp);
4792 4802 }
4793 4803
4794 4804 connp->conn_ref--;
4795 4805 ipcl_conn_destroy(connp);
4796 4806 }
4797 4807
4798 4808 /* ARGSUSED1 */
4799 4809 int
4800 4810 udp_close(sock_lower_handle_t proto_handle, int flags, cred_t *cr)
4801 4811 {
4802 4812 conn_t *connp = (conn_t *)proto_handle;
4803 4813
4804 4814 /* All Solaris components should pass a cred for this operation. */
4805 4815 ASSERT(cr != NULL);
4806 4816
4807 4817 udp_do_close(connp);
4808 4818 return (0);
4809 4819 }
4810 4820
4811 4821 static int
4812 4822 udp_do_bind(conn_t *connp, struct sockaddr *sa, socklen_t len, cred_t *cr,
4813 4823 boolean_t bind_to_req_port_only)
4814 4824 {
4815 4825 sin_t *sin;
4816 4826 sin6_t *sin6;
4817 4827 udp_t *udp = connp->conn_udp;
4818 4828 int error = 0;
4819 4829 ip_laddr_t laddr_type = IPVL_UNICAST_UP; /* INADDR_ANY */
4820 4830 in_port_t port; /* Host byte order */
4821 4831 in_port_t requested_port; /* Host byte order */
4822 4832 int count;
4823 4833 ipaddr_t v4src; /* Set if AF_INET */
4824 4834 in6_addr_t v6src;
4825 4835 int loopmax;
4826 4836 udp_fanout_t *udpf;
4827 4837 in_port_t lport; /* Network byte order */
4828 4838 uint_t scopeid = 0;
4829 4839 zoneid_t zoneid = IPCL_ZONEID(connp);
4830 4840 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
4831 4841 boolean_t is_inaddr_any;
4832 4842 mlp_type_t addrtype, mlptype;
4833 4843 udp_stack_t *us = udp->udp_us;
4834 4844
4835 4845 switch (len) {
4836 4846 case sizeof (sin_t): /* Complete IPv4 address */
4837 4847 sin = (sin_t *)sa;
4838 4848
4839 4849 if (sin == NULL || !OK_32PTR((char *)sin))
4840 4850 return (EINVAL);
4841 4851
4842 4852 if (connp->conn_family != AF_INET ||
4843 4853 sin->sin_family != AF_INET) {
4844 4854 return (EAFNOSUPPORT);
4845 4855 }
4846 4856 v4src = sin->sin_addr.s_addr;
4847 4857 IN6_IPADDR_TO_V4MAPPED(v4src, &v6src);
4848 4858 if (v4src != INADDR_ANY) {
4849 4859 laddr_type = ip_laddr_verify_v4(v4src, zoneid, ipst,
4850 4860 B_TRUE);
4851 4861 }
4852 4862 port = ntohs(sin->sin_port);
4853 4863 break;
4854 4864
4855 4865 case sizeof (sin6_t): /* complete IPv6 address */
4856 4866 sin6 = (sin6_t *)sa;
4857 4867
4858 4868 if (sin6 == NULL || !OK_32PTR((char *)sin6))
4859 4869 return (EINVAL);
4860 4870
4861 4871 if (connp->conn_family != AF_INET6 ||
4862 4872 sin6->sin6_family != AF_INET6) {
4863 4873 return (EAFNOSUPPORT);
4864 4874 }
4865 4875 v6src = sin6->sin6_addr;
4866 4876 if (IN6_IS_ADDR_V4MAPPED(&v6src)) {
4867 4877 if (connp->conn_ipv6_v6only)
4868 4878 return (EADDRNOTAVAIL);
4869 4879
4870 4880 IN6_V4MAPPED_TO_IPADDR(&v6src, v4src);
4871 4881 if (v4src != INADDR_ANY) {
4872 4882 laddr_type = ip_laddr_verify_v4(v4src,
4873 4883 zoneid, ipst, B_FALSE);
4874 4884 }
4875 4885 } else {
4876 4886 if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4877 4887 if (IN6_IS_ADDR_LINKSCOPE(&v6src))
4878 4888 scopeid = sin6->sin6_scope_id;
4879 4889 laddr_type = ip_laddr_verify_v6(&v6src,
4880 4890 zoneid, ipst, B_TRUE, scopeid);
4881 4891 }
4882 4892 }
4883 4893 port = ntohs(sin6->sin6_port);
4884 4894 break;
4885 4895
4886 4896 default: /* Invalid request */
4887 4897 (void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
4888 4898 "udp_bind: bad ADDR_length length %u", len);
4889 4899 return (-TBADADDR);
4890 4900 }
4891 4901
4892 4902 /* Is the local address a valid unicast, multicast, or broadcast? */
4893 4903 if (laddr_type == IPVL_BAD)
4894 4904 return (EADDRNOTAVAIL);
4895 4905
4896 4906 requested_port = port;
4897 4907
4898 4908 if (requested_port == 0 || !bind_to_req_port_only)
4899 4909 bind_to_req_port_only = B_FALSE;
4900 4910 else /* T_BIND_REQ and requested_port != 0 */
4901 4911 bind_to_req_port_only = B_TRUE;
4902 4912
4903 4913 if (requested_port == 0) {
4904 4914 /*
4905 4915 * If the application passed in zero for the port number, it
4906 4916 * doesn't care which port number we bind to. Get one in the
4907 4917 * valid range.
4908 4918 */
4909 4919 if (connp->conn_anon_priv_bind) {
4910 4920 port = udp_get_next_priv_port(udp);
4911 4921 } else {
4912 4922 port = udp_update_next_port(udp,
4913 4923 us->us_next_port_to_try, B_TRUE);
4914 4924 }
4915 4925 } else {
4916 4926 /*
4917 4927 * If the port is in the well-known privileged range,
4918 4928 * make sure the caller was privileged.
4919 4929 */
4920 4930 int i;
4921 4931 boolean_t priv = B_FALSE;
4922 4932
4923 4933 if (port < us->us_smallest_nonpriv_port) {
4924 4934 priv = B_TRUE;
4925 4935 } else {
4926 4936 for (i = 0; i < us->us_num_epriv_ports; i++) {
4927 4937 if (port == us->us_epriv_ports[i]) {
4928 4938 priv = B_TRUE;
4929 4939 break;
4930 4940 }
4931 4941 }
4932 4942 }
4933 4943
4934 4944 if (priv) {
4935 4945 if (secpolicy_net_privaddr(cr, port, IPPROTO_UDP) != 0)
4936 4946 return (-TACCES);
4937 4947 }
4938 4948 }
4939 4949
4940 4950 if (port == 0)
4941 4951 return (-TNOADDR);
4942 4952
4943 4953 /*
4944 4954 * The state must be TS_UNBND. TPI mandates that users must send
4945 4955 * TPI primitives only 1 at a time and wait for the response before
4946 4956 * sending the next primitive.
4947 4957 */
4948 4958 mutex_enter(&connp->conn_lock);
4949 4959 if (udp->udp_state != TS_UNBND) {
4950 4960 mutex_exit(&connp->conn_lock);
4951 4961 (void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
4952 4962 "udp_bind: bad state, %u", udp->udp_state);
4953 4963 return (-TOUTSTATE);
4954 4964 }
4955 4965 /*
4956 4966 * Copy the source address into our udp structure. This address
4957 4967 * may still be zero; if so, IP will fill in the correct address
4958 4968 * each time an outbound packet is passed to it. Since the udp is
4959 4969 * not yet in the bind hash list, we don't grab the uf_lock to
4960 4970 * change conn_ipversion
4961 4971 */
4962 4972 if (connp->conn_family == AF_INET) {
4963 4973 ASSERT(sin != NULL);
4964 4974 ASSERT(connp->conn_ixa->ixa_flags & IXAF_IS_IPV4);
4965 4975 } else {
4966 4976 if (IN6_IS_ADDR_V4MAPPED(&v6src)) {
4967 4977 /*
4968 4978 * no need to hold the uf_lock to set the conn_ipversion
4969 4979 * since we are not yet in the fanout list
4970 4980 */
4971 4981 connp->conn_ipversion = IPV4_VERSION;
4972 4982 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
4973 4983 } else {
4974 4984 connp->conn_ipversion = IPV6_VERSION;
4975 4985 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
4976 4986 }
4977 4987 }
4978 4988
4979 4989 /*
4980 4990 * If conn_reuseaddr is not set, then we have to make sure that
4981 4991 * the IP address and port number the application requested
4982 4992 * (or we selected for the application) is not being used by
4983 4993 * another stream. If another stream is already using the
4984 4994 * requested IP address and port, the behavior depends on
4985 4995 * "bind_to_req_port_only". If set the bind fails; otherwise we
4986 4996 * search for any an unused port to bind to the stream.
4987 4997 *
4988 4998 * As per the BSD semantics, as modified by the Deering multicast
4989 4999 * changes, if udp_reuseaddr is set, then we allow multiple binds
4990 5000 * to the same port independent of the local IP address.
4991 5001 *
4992 5002 * This is slightly different than in SunOS 4.X which did not
4993 5003 * support IP multicast. Note that the change implemented by the
4994 5004 * Deering multicast code effects all binds - not only binding
4995 5005 * to IP multicast addresses.
4996 5006 *
4997 5007 * Note that when binding to port zero we ignore SO_REUSEADDR in
4998 5008 * order to guarantee a unique port.
4999 5009 */
5000 5010
5001 5011 count = 0;
5002 5012 if (connp->conn_anon_priv_bind) {
5003 5013 /*
5004 5014 * loopmax = (IPPORT_RESERVED-1) -
5005 5015 * us->us_min_anonpriv_port + 1
5006 5016 */
5007 5017 loopmax = IPPORT_RESERVED - us->us_min_anonpriv_port;
5008 5018 } else {
5009 5019 loopmax = us->us_largest_anon_port -
5010 5020 us->us_smallest_anon_port + 1;
5011 5021 }
5012 5022
5013 5023 is_inaddr_any = V6_OR_V4_INADDR_ANY(v6src);
5014 5024
5015 5025 for (;;) {
5016 5026 udp_t *udp1;
5017 5027 boolean_t found_exclbind = B_FALSE;
5018 5028 conn_t *connp1;
5019 5029
5020 5030 /*
5021 5031 * Walk through the list of udp streams bound to
5022 5032 * requested port with the same IP address.
5023 5033 */
5024 5034 lport = htons(port);
5025 5035 udpf = &us->us_bind_fanout[UDP_BIND_HASH(lport,
5026 5036 us->us_bind_fanout_size)];
5027 5037 mutex_enter(&udpf->uf_lock);
5028 5038 for (udp1 = udpf->uf_udp; udp1 != NULL;
5029 5039 udp1 = udp1->udp_bind_hash) {
5030 5040 connp1 = udp1->udp_connp;
5031 5041
5032 5042 if (lport != connp1->conn_lport)
5033 5043 continue;
5034 5044
5035 5045 /*
5036 5046 * On a labeled system, we must treat bindings to ports
5037 5047 * on shared IP addresses by sockets with MAC exemption
5038 5048 * privilege as being in all zones, as there's
5039 5049 * otherwise no way to identify the right receiver.
5040 5050 */
5041 5051 if (!IPCL_BIND_ZONE_MATCH(connp1, connp))
5042 5052 continue;
5043 5053
5044 5054 /*
5045 5055 * If UDP_EXCLBIND is set for either the bound or
5046 5056 * binding endpoint, the semantics of bind
5047 5057 * is changed according to the following chart.
5048 5058 *
5049 5059 * spec = specified address (v4 or v6)
5050 5060 * unspec = unspecified address (v4 or v6)
5051 5061 * A = specified addresses are different for endpoints
5052 5062 *
5053 5063 * bound bind to allowed?
5054 5064 * -------------------------------------
5055 5065 * unspec unspec no
5056 5066 * unspec spec no
5057 5067 * spec unspec no
5058 5068 * spec spec yes if A
5059 5069 *
5060 5070 * For labeled systems, SO_MAC_EXEMPT behaves the same
5061 5071 * as UDP_EXCLBIND, except that zoneid is ignored.
5062 5072 */
5063 5073 if (connp1->conn_exclbind || connp->conn_exclbind ||
5064 5074 IPCL_CONNS_MAC(udp1->udp_connp, connp)) {
5065 5075 if (V6_OR_V4_INADDR_ANY(
5066 5076 connp1->conn_bound_addr_v6) ||
5067 5077 is_inaddr_any ||
5068 5078 IN6_ARE_ADDR_EQUAL(
5069 5079 &connp1->conn_bound_addr_v6,
5070 5080 &v6src)) {
5071 5081 found_exclbind = B_TRUE;
5072 5082 break;
5073 5083 }
5074 5084 continue;
5075 5085 }
5076 5086
5077 5087 /*
5078 5088 * Check ipversion to allow IPv4 and IPv6 sockets to
5079 5089 * have disjoint port number spaces.
5080 5090 */
5081 5091 if (connp->conn_ipversion != connp1->conn_ipversion) {
5082 5092
5083 5093 /*
5084 5094 * On the first time through the loop, if the
5085 5095 * the user intentionally specified a
5086 5096 * particular port number, then ignore any
5087 5097 * bindings of the other protocol that may
5088 5098 * conflict. This allows the user to bind IPv6
5089 5099 * alone and get both v4 and v6, or bind both
5090 5100 * both and get each seperately. On subsequent
5091 5101 * times through the loop, we're checking a
5092 5102 * port that we chose (not the user) and thus
5093 5103 * we do not allow casual duplicate bindings.
5094 5104 */
5095 5105 if (count == 0 && requested_port != 0)
5096 5106 continue;
5097 5107 }
5098 5108
5099 5109 /*
5100 5110 * No difference depending on SO_REUSEADDR.
5101 5111 *
5102 5112 * If existing port is bound to a
5103 5113 * non-wildcard IP address and
5104 5114 * the requesting stream is bound to
5105 5115 * a distinct different IP addresses
5106 5116 * (non-wildcard, also), keep going.
5107 5117 */
5108 5118 if (!is_inaddr_any &&
5109 5119 !V6_OR_V4_INADDR_ANY(connp1->conn_bound_addr_v6) &&
5110 5120 !IN6_ARE_ADDR_EQUAL(&connp1->conn_laddr_v6,
5111 5121 &v6src)) {
5112 5122 continue;
5113 5123 }
5114 5124 break;
5115 5125 }
5116 5126
5117 5127 if (!found_exclbind &&
5118 5128 (connp->conn_reuseaddr && requested_port != 0)) {
5119 5129 break;
5120 5130 }
5121 5131
5122 5132 if (udp1 == NULL) {
5123 5133 /*
5124 5134 * No other stream has this IP address
5125 5135 * and port number. We can use it.
5126 5136 */
5127 5137 break;
5128 5138 }
5129 5139 mutex_exit(&udpf->uf_lock);
5130 5140 if (bind_to_req_port_only) {
5131 5141 /*
5132 5142 * We get here only when requested port
5133 5143 * is bound (and only first of the for()
5134 5144 * loop iteration).
5135 5145 *
5136 5146 * The semantics of this bind request
5137 5147 * require it to fail so we return from
5138 5148 * the routine (and exit the loop).
5139 5149 *
5140 5150 */
5141 5151 mutex_exit(&connp->conn_lock);
5142 5152 return (-TADDRBUSY);
5143 5153 }
5144 5154
5145 5155 if (connp->conn_anon_priv_bind) {
5146 5156 port = udp_get_next_priv_port(udp);
5147 5157 } else {
5148 5158 if ((count == 0) && (requested_port != 0)) {
5149 5159 /*
5150 5160 * If the application wants us to find
5151 5161 * a port, get one to start with. Set
5152 5162 * requested_port to 0, so that we will
5153 5163 * update us->us_next_port_to_try below.
5154 5164 */
5155 5165 port = udp_update_next_port(udp,
5156 5166 us->us_next_port_to_try, B_TRUE);
5157 5167 requested_port = 0;
5158 5168 } else {
5159 5169 port = udp_update_next_port(udp, port + 1,
5160 5170 B_FALSE);
5161 5171 }
5162 5172 }
5163 5173
5164 5174 if (port == 0 || ++count >= loopmax) {
5165 5175 /*
5166 5176 * We've tried every possible port number and
5167 5177 * there are none available, so send an error
5168 5178 * to the user.
5169 5179 */
5170 5180 mutex_exit(&connp->conn_lock);
5171 5181 return (-TNOADDR);
5172 5182 }
5173 5183 }
5174 5184
5175 5185 /*
5176 5186 * Copy the source address into our udp structure. This address
5177 5187 * may still be zero; if so, ip_attr_connect will fill in the correct
5178 5188 * address when a packet is about to be sent.
5179 5189 * If we are binding to a broadcast or multicast address then
5180 5190 * we just set the conn_bound_addr since we don't want to use
5181 5191 * that as the source address when sending.
5182 5192 */
5183 5193 connp->conn_bound_addr_v6 = v6src;
5184 5194 connp->conn_laddr_v6 = v6src;
5185 5195 if (scopeid != 0) {
5186 5196 connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
5187 5197 connp->conn_ixa->ixa_scopeid = scopeid;
5188 5198 connp->conn_incoming_ifindex = scopeid;
5189 5199 } else {
5190 5200 connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5191 5201 connp->conn_incoming_ifindex = connp->conn_bound_if;
5192 5202 }
5193 5203
5194 5204 switch (laddr_type) {
5195 5205 case IPVL_UNICAST_UP:
5196 5206 case IPVL_UNICAST_DOWN:
5197 5207 connp->conn_saddr_v6 = v6src;
5198 5208 connp->conn_mcbc_bind = B_FALSE;
5199 5209 break;
5200 5210 case IPVL_MCAST:
5201 5211 case IPVL_BCAST:
5202 5212 /* ip_set_destination will pick a source address later */
5203 5213 connp->conn_saddr_v6 = ipv6_all_zeros;
5204 5214 connp->conn_mcbc_bind = B_TRUE;
5205 5215 break;
5206 5216 }
5207 5217
5208 5218 /* Any errors after this point should use late_error */
5209 5219 connp->conn_lport = lport;
5210 5220
5211 5221 /*
5212 5222 * Now reset the next anonymous port if the application requested
5213 5223 * an anonymous port, or we handed out the next anonymous port.
5214 5224 */
5215 5225 if ((requested_port == 0) && (!connp->conn_anon_priv_bind)) {
5216 5226 us->us_next_port_to_try = port + 1;
5217 5227 }
5218 5228
5219 5229 /* Initialize the T_BIND_ACK. */
5220 5230 if (connp->conn_family == AF_INET) {
5221 5231 sin->sin_port = connp->conn_lport;
5222 5232 } else {
5223 5233 sin6->sin6_port = connp->conn_lport;
5224 5234 }
5225 5235 udp->udp_state = TS_IDLE;
5226 5236 udp_bind_hash_insert(udpf, udp);
5227 5237 mutex_exit(&udpf->uf_lock);
5228 5238 mutex_exit(&connp->conn_lock);
5229 5239
5230 5240 if (cl_inet_bind) {
5231 5241 /*
5232 5242 * Running in cluster mode - register bind information
5233 5243 */
5234 5244 if (connp->conn_ipversion == IPV4_VERSION) {
5235 5245 (*cl_inet_bind)(connp->conn_netstack->netstack_stackid,
5236 5246 IPPROTO_UDP, AF_INET, (uint8_t *)&v4src,
5237 5247 (in_port_t)connp->conn_lport, NULL);
5238 5248 } else {
5239 5249 (*cl_inet_bind)(connp->conn_netstack->netstack_stackid,
5240 5250 IPPROTO_UDP, AF_INET6, (uint8_t *)&v6src,
5241 5251 (in_port_t)connp->conn_lport, NULL);
5242 5252 }
5243 5253 }
5244 5254
5245 5255 mutex_enter(&connp->conn_lock);
5246 5256 connp->conn_anon_port = (is_system_labeled() && requested_port == 0);
5247 5257 if (is_system_labeled() && (!connp->conn_anon_port ||
5248 5258 connp->conn_anon_mlp)) {
5249 5259 uint16_t mlpport;
5250 5260 zone_t *zone;
5251 5261
5252 5262 zone = crgetzone(cr);
5253 5263 connp->conn_mlp_type =
5254 5264 connp->conn_recv_ancillary.crb_recvucred ? mlptBoth :
5255 5265 mlptSingle;
5256 5266 addrtype = tsol_mlp_addr_type(
5257 5267 connp->conn_allzones ? ALL_ZONES : zone->zone_id,
5258 5268 IPV6_VERSION, &v6src, us->us_netstack->netstack_ip);
5259 5269 if (addrtype == mlptSingle) {
5260 5270 error = -TNOADDR;
5261 5271 mutex_exit(&connp->conn_lock);
5262 5272 goto late_error;
5263 5273 }
5264 5274 mlpport = connp->conn_anon_port ? PMAPPORT : port;
5265 5275 mlptype = tsol_mlp_port_type(zone, IPPROTO_UDP, mlpport,
5266 5276 addrtype);
5267 5277
5268 5278 /*
5269 5279 * It is a coding error to attempt to bind an MLP port
5270 5280 * without first setting SOL_SOCKET/SCM_UCRED.
5271 5281 */
5272 5282 if (mlptype != mlptSingle &&
5273 5283 connp->conn_mlp_type == mlptSingle) {
5274 5284 error = EINVAL;
5275 5285 mutex_exit(&connp->conn_lock);
5276 5286 goto late_error;
5277 5287 }
5278 5288
5279 5289 /*
5280 5290 * It is an access violation to attempt to bind an MLP port
5281 5291 * without NET_BINDMLP privilege.
5282 5292 */
5283 5293 if (mlptype != mlptSingle &&
5284 5294 secpolicy_net_bindmlp(cr) != 0) {
5285 5295 if (connp->conn_debug) {
5286 5296 (void) strlog(UDP_MOD_ID, 0, 1,
5287 5297 SL_ERROR|SL_TRACE,
5288 5298 "udp_bind: no priv for multilevel port %d",
5289 5299 mlpport);
5290 5300 }
5291 5301 error = -TACCES;
5292 5302 mutex_exit(&connp->conn_lock);
5293 5303 goto late_error;
5294 5304 }
5295 5305
5296 5306 /*
5297 5307 * If we're specifically binding a shared IP address and the
5298 5308 * port is MLP on shared addresses, then check to see if this
5299 5309 * zone actually owns the MLP. Reject if not.
5300 5310 */
5301 5311 if (mlptype == mlptShared && addrtype == mlptShared) {
5302 5312 /*
5303 5313 * No need to handle exclusive-stack zones since
5304 5314 * ALL_ZONES only applies to the shared stack.
5305 5315 */
5306 5316 zoneid_t mlpzone;
5307 5317
5308 5318 mlpzone = tsol_mlp_findzone(IPPROTO_UDP,
5309 5319 htons(mlpport));
5310 5320 if (connp->conn_zoneid != mlpzone) {
5311 5321 if (connp->conn_debug) {
5312 5322 (void) strlog(UDP_MOD_ID, 0, 1,
5313 5323 SL_ERROR|SL_TRACE,
5314 5324 "udp_bind: attempt to bind port "
5315 5325 "%d on shared addr in zone %d "
5316 5326 "(should be %d)",
5317 5327 mlpport, connp->conn_zoneid,
5318 5328 mlpzone);
5319 5329 }
5320 5330 error = -TACCES;
5321 5331 mutex_exit(&connp->conn_lock);
5322 5332 goto late_error;
5323 5333 }
5324 5334 }
5325 5335 if (connp->conn_anon_port) {
5326 5336 error = tsol_mlp_anon(zone, mlptype, connp->conn_proto,
5327 5337 port, B_TRUE);
5328 5338 if (error != 0) {
5329 5339 if (connp->conn_debug) {
5330 5340 (void) strlog(UDP_MOD_ID, 0, 1,
5331 5341 SL_ERROR|SL_TRACE,
5332 5342 "udp_bind: cannot establish anon "
5333 5343 "MLP for port %d", port);
5334 5344 }
5335 5345 error = -TACCES;
5336 5346 mutex_exit(&connp->conn_lock);
5337 5347 goto late_error;
5338 5348 }
5339 5349 }
5340 5350 connp->conn_mlp_type = mlptype;
5341 5351 }
5342 5352
5343 5353 /*
5344 5354 * We create an initial header template here to make a subsequent
5345 5355 * sendto have a starting point. Since conn_last_dst is zero the
5346 5356 * first sendto will always follow the 'dst changed' code path.
5347 5357 * Note that we defer massaging options and the related checksum
5348 5358 * adjustment until we have a destination address.
5349 5359 */
5350 5360 error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5351 5361 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5352 5362 if (error != 0) {
5353 5363 mutex_exit(&connp->conn_lock);
5354 5364 goto late_error;
5355 5365 }
5356 5366 /* Just in case */
5357 5367 connp->conn_faddr_v6 = ipv6_all_zeros;
5358 5368 connp->conn_fport = 0;
5359 5369 connp->conn_v6lastdst = ipv6_all_zeros;
5360 5370 mutex_exit(&connp->conn_lock);
5361 5371
5362 5372 error = ip_laddr_fanout_insert(connp);
5363 5373 if (error != 0)
5364 5374 goto late_error;
5365 5375
5366 5376 /* Bind succeeded */
5367 5377 return (0);
5368 5378
5369 5379 late_error:
5370 5380 /* We had already picked the port number, and then the bind failed */
5371 5381 mutex_enter(&connp->conn_lock);
5372 5382 udpf = &us->us_bind_fanout[
5373 5383 UDP_BIND_HASH(connp->conn_lport,
5374 5384 us->us_bind_fanout_size)];
5375 5385 mutex_enter(&udpf->uf_lock);
5376 5386 connp->conn_saddr_v6 = ipv6_all_zeros;
5377 5387 connp->conn_bound_addr_v6 = ipv6_all_zeros;
5378 5388 connp->conn_laddr_v6 = ipv6_all_zeros;
5379 5389 if (scopeid != 0) {
5380 5390 connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5381 5391 connp->conn_incoming_ifindex = connp->conn_bound_if;
5382 5392 }
5383 5393 udp->udp_state = TS_UNBND;
5384 5394 udp_bind_hash_remove(udp, B_TRUE);
5385 5395 connp->conn_lport = 0;
5386 5396 mutex_exit(&udpf->uf_lock);
5387 5397 connp->conn_anon_port = B_FALSE;
5388 5398 connp->conn_mlp_type = mlptSingle;
5389 5399
5390 5400 connp->conn_v6lastdst = ipv6_all_zeros;
5391 5401
5392 5402 /* Restore the header that was built above - different source address */
5393 5403 (void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5394 5404 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5395 5405 mutex_exit(&connp->conn_lock);
5396 5406 return (error);
5397 5407 }
5398 5408
5399 5409 int
5400 5410 udp_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5401 5411 socklen_t len, cred_t *cr)
5402 5412 {
5403 5413 int error;
5404 5414 conn_t *connp;
5405 5415
5406 5416 /* All Solaris components should pass a cred for this operation. */
5407 5417 ASSERT(cr != NULL);
5408 5418
5409 5419 connp = (conn_t *)proto_handle;
5410 5420
5411 5421 if (sa == NULL)
5412 5422 error = udp_do_unbind(connp);
5413 5423 else
5414 5424 error = udp_do_bind(connp, sa, len, cr, B_TRUE);
5415 5425
5416 5426 if (error < 0) {
5417 5427 if (error == -TOUTSTATE)
5418 5428 error = EINVAL;
5419 5429 else
5420 5430 error = proto_tlitosyserr(-error);
5421 5431 }
5422 5432
5423 5433 return (error);
5424 5434 }
5425 5435
5426 5436 static int
5427 5437 udp_implicit_bind(conn_t *connp, cred_t *cr)
5428 5438 {
5429 5439 sin6_t sin6addr;
5430 5440 sin_t *sin;
5431 5441 sin6_t *sin6;
5432 5442 socklen_t len;
5433 5443 int error;
5434 5444
5435 5445 /* All Solaris components should pass a cred for this operation. */
5436 5446 ASSERT(cr != NULL);
5437 5447
5438 5448 if (connp->conn_family == AF_INET) {
5439 5449 len = sizeof (struct sockaddr_in);
5440 5450 sin = (sin_t *)&sin6addr;
5441 5451 *sin = sin_null;
5442 5452 sin->sin_family = AF_INET;
5443 5453 sin->sin_addr.s_addr = INADDR_ANY;
5444 5454 } else {
5445 5455 ASSERT(connp->conn_family == AF_INET6);
5446 5456 len = sizeof (sin6_t);
5447 5457 sin6 = (sin6_t *)&sin6addr;
5448 5458 *sin6 = sin6_null;
5449 5459 sin6->sin6_family = AF_INET6;
5450 5460 V6_SET_ZERO(sin6->sin6_addr);
5451 5461 }
5452 5462
5453 5463 error = udp_do_bind(connp, (struct sockaddr *)&sin6addr, len,
5454 5464 cr, B_FALSE);
5455 5465 return ((error < 0) ? proto_tlitosyserr(-error) : error);
5456 5466 }
5457 5467
5458 5468 /*
5459 5469 * This routine removes a port number association from a stream. It
5460 5470 * is called by udp_unbind and udp_tpi_unbind.
5461 5471 */
5462 5472 static int
5463 5473 udp_do_unbind(conn_t *connp)
5464 5474 {
5465 5475 udp_t *udp = connp->conn_udp;
5466 5476 udp_fanout_t *udpf;
5467 5477 udp_stack_t *us = udp->udp_us;
5468 5478
5469 5479 if (cl_inet_unbind != NULL) {
5470 5480 /*
5471 5481 * Running in cluster mode - register unbind information
5472 5482 */
5473 5483 if (connp->conn_ipversion == IPV4_VERSION) {
5474 5484 (*cl_inet_unbind)(
5475 5485 connp->conn_netstack->netstack_stackid,
5476 5486 IPPROTO_UDP, AF_INET,
5477 5487 (uint8_t *)(&V4_PART_OF_V6(connp->conn_laddr_v6)),
5478 5488 (in_port_t)connp->conn_lport, NULL);
5479 5489 } else {
5480 5490 (*cl_inet_unbind)(
5481 5491 connp->conn_netstack->netstack_stackid,
5482 5492 IPPROTO_UDP, AF_INET6,
5483 5493 (uint8_t *)&(connp->conn_laddr_v6),
5484 5494 (in_port_t)connp->conn_lport, NULL);
5485 5495 }
5486 5496 }
5487 5497
5488 5498 mutex_enter(&connp->conn_lock);
5489 5499 /* If a bind has not been done, we can't unbind. */
5490 5500 if (udp->udp_state == TS_UNBND) {
5491 5501 mutex_exit(&connp->conn_lock);
5492 5502 return (-TOUTSTATE);
5493 5503 }
5494 5504 udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
5495 5505 us->us_bind_fanout_size)];
5496 5506 mutex_enter(&udpf->uf_lock);
5497 5507 udp_bind_hash_remove(udp, B_TRUE);
5498 5508 connp->conn_saddr_v6 = ipv6_all_zeros;
5499 5509 connp->conn_bound_addr_v6 = ipv6_all_zeros;
5500 5510 connp->conn_laddr_v6 = ipv6_all_zeros;
5501 5511 connp->conn_mcbc_bind = B_FALSE;
5502 5512 connp->conn_lport = 0;
5503 5513 /* In case we were also connected */
5504 5514 connp->conn_faddr_v6 = ipv6_all_zeros;
5505 5515 connp->conn_fport = 0;
5506 5516 mutex_exit(&udpf->uf_lock);
5507 5517
5508 5518 connp->conn_v6lastdst = ipv6_all_zeros;
5509 5519 udp->udp_state = TS_UNBND;
5510 5520
5511 5521 (void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5512 5522 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5513 5523 mutex_exit(&connp->conn_lock);
5514 5524
5515 5525 ip_unbind(connp);
5516 5526
5517 5527 return (0);
5518 5528 }
5519 5529
5520 5530 /*
5521 5531 * It associates a default destination address with the stream.
5522 5532 */
5523 5533 static int
5524 5534 udp_do_connect(conn_t *connp, const struct sockaddr *sa, socklen_t len,
5525 5535 cred_t *cr, pid_t pid)
5526 5536 {
5527 5537 sin6_t *sin6;
5528 5538 sin_t *sin;
5529 5539 in6_addr_t v6dst;
5530 5540 ipaddr_t v4dst;
5531 5541 uint16_t dstport;
5532 5542 uint32_t flowinfo;
5533 5543 udp_fanout_t *udpf;
5534 5544 udp_t *udp, *udp1;
5535 5545 ushort_t ipversion;
5536 5546 udp_stack_t *us;
5537 5547 int error;
5538 5548 conn_t *connp1;
5539 5549 ip_xmit_attr_t *ixa;
5540 5550 ip_xmit_attr_t *oldixa;
5541 5551 uint_t scopeid = 0;
5542 5552 uint_t srcid = 0;
5543 5553 in6_addr_t v6src = connp->conn_saddr_v6;
5544 5554 boolean_t v4mapped;
5545 5555
5546 5556 udp = connp->conn_udp;
5547 5557 us = udp->udp_us;
5548 5558
5549 5559 /*
5550 5560 * Address has been verified by the caller
5551 5561 */
5552 5562 switch (len) {
5553 5563 default:
5554 5564 /*
5555 5565 * Should never happen
5556 5566 */
5557 5567 return (EINVAL);
5558 5568
5559 5569 case sizeof (sin_t):
5560 5570 sin = (sin_t *)sa;
5561 5571 v4dst = sin->sin_addr.s_addr;
5562 5572 dstport = sin->sin_port;
5563 5573 IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
5564 5574 ASSERT(connp->conn_ipversion == IPV4_VERSION);
5565 5575 ipversion = IPV4_VERSION;
5566 5576 break;
5567 5577
5568 5578 case sizeof (sin6_t):
5569 5579 sin6 = (sin6_t *)sa;
5570 5580 v6dst = sin6->sin6_addr;
5571 5581 dstport = sin6->sin6_port;
5572 5582 srcid = sin6->__sin6_src_id;
5573 5583 v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
5574 5584 if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
5575 5585 if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
5576 5586 v4mapped, connp->conn_netstack)) {
5577 5587 /* Mismatch v4mapped/v6 specified by srcid. */
5578 5588 return (EADDRNOTAVAIL);
5579 5589 }
5580 5590 }
5581 5591 if (v4mapped) {
5582 5592 if (connp->conn_ipv6_v6only)
5583 5593 return (EADDRNOTAVAIL);
5584 5594
5585 5595 /*
5586 5596 * Destination adress is mapped IPv6 address.
5587 5597 * Source bound address should be unspecified or
5588 5598 * IPv6 mapped address as well.
5589 5599 */
5590 5600 if (!IN6_IS_ADDR_UNSPECIFIED(
5591 5601 &connp->conn_bound_addr_v6) &&
5592 5602 !IN6_IS_ADDR_V4MAPPED(&connp->conn_bound_addr_v6)) {
5593 5603 return (EADDRNOTAVAIL);
5594 5604 }
5595 5605 IN6_V4MAPPED_TO_IPADDR(&v6dst, v4dst);
5596 5606 ipversion = IPV4_VERSION;
5597 5607 flowinfo = 0;
5598 5608 } else {
5599 5609 ipversion = IPV6_VERSION;
5600 5610 flowinfo = sin6->sin6_flowinfo;
5601 5611 if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
5602 5612 scopeid = sin6->sin6_scope_id;
5603 5613 }
5604 5614 break;
5605 5615 }
5606 5616
5607 5617 if (dstport == 0)
5608 5618 return (-TBADADDR);
5609 5619
5610 5620 /*
5611 5621 * If there is a different thread using conn_ixa then we get a new
5612 5622 * copy and cut the old one loose from conn_ixa. Otherwise we use
5613 5623 * conn_ixa and prevent any other thread from using/changing it.
5614 5624 * Once connect() is done other threads can use conn_ixa since the
5615 5625 * refcnt will be back at one.
5616 5626 * We defer updating conn_ixa until later to handle any concurrent
5617 5627 * conn_ixa_cleanup thread.
5618 5628 */
5619 5629 ixa = conn_get_ixa(connp, B_FALSE);
5620 5630 if (ixa == NULL)
5621 5631 return (ENOMEM);
5622 5632
5623 5633 mutex_enter(&connp->conn_lock);
5624 5634 /*
5625 5635 * This udp_t must have bound to a port already before doing a connect.
5626 5636 * Reject if a connect is in progress (we drop conn_lock during
5627 5637 * udp_do_connect).
5628 5638 */
5629 5639 if (udp->udp_state == TS_UNBND || udp->udp_state == TS_WCON_CREQ) {
5630 5640 mutex_exit(&connp->conn_lock);
5631 5641 (void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
5632 5642 "udp_connect: bad state, %u", udp->udp_state);
5633 5643 ixa_refrele(ixa);
5634 5644 return (-TOUTSTATE);
5635 5645 }
5636 5646 ASSERT(connp->conn_lport != 0 && udp->udp_ptpbhn != NULL);
5637 5647
5638 5648 udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
5639 5649 us->us_bind_fanout_size)];
5640 5650
5641 5651 mutex_enter(&udpf->uf_lock);
5642 5652 if (udp->udp_state == TS_DATA_XFER) {
5643 5653 /* Already connected - clear out state */
5644 5654 if (connp->conn_mcbc_bind)
5645 5655 connp->conn_saddr_v6 = ipv6_all_zeros;
5646 5656 else
5647 5657 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
5648 5658 connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
5649 5659 connp->conn_faddr_v6 = ipv6_all_zeros;
5650 5660 connp->conn_fport = 0;
5651 5661 udp->udp_state = TS_IDLE;
5652 5662 }
5653 5663
5654 5664 connp->conn_fport = dstport;
5655 5665 connp->conn_ipversion = ipversion;
5656 5666 if (ipversion == IPV4_VERSION) {
5657 5667 /*
5658 5668 * Interpret a zero destination to mean loopback.
5659 5669 * Update the T_CONN_REQ (sin/sin6) since it is used to
5660 5670 * generate the T_CONN_CON.
5661 5671 */
5662 5672 if (v4dst == INADDR_ANY) {
5663 5673 v4dst = htonl(INADDR_LOOPBACK);
5664 5674 IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
5665 5675 if (connp->conn_family == AF_INET) {
5666 5676 sin->sin_addr.s_addr = v4dst;
5667 5677 } else {
5668 5678 sin6->sin6_addr = v6dst;
5669 5679 }
5670 5680 }
5671 5681 connp->conn_faddr_v6 = v6dst;
5672 5682 connp->conn_flowinfo = 0;
5673 5683 } else {
5674 5684 ASSERT(connp->conn_ipversion == IPV6_VERSION);
5675 5685 /*
5676 5686 * Interpret a zero destination to mean loopback.
5677 5687 * Update the T_CONN_REQ (sin/sin6) since it is used to
5678 5688 * generate the T_CONN_CON.
5679 5689 */
5680 5690 if (IN6_IS_ADDR_UNSPECIFIED(&v6dst)) {
5681 5691 v6dst = ipv6_loopback;
5682 5692 sin6->sin6_addr = v6dst;
5683 5693 }
5684 5694 connp->conn_faddr_v6 = v6dst;
5685 5695 connp->conn_flowinfo = flowinfo;
5686 5696 }
5687 5697 mutex_exit(&udpf->uf_lock);
5688 5698
5689 5699 /*
5690 5700 * We update our cred/cpid based on the caller of connect
5691 5701 */
5692 5702 if (connp->conn_cred != cr) {
5693 5703 crhold(cr);
5694 5704 crfree(connp->conn_cred);
5695 5705 connp->conn_cred = cr;
5696 5706 }
5697 5707 connp->conn_cpid = pid;
5698 5708 ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
5699 5709 ixa->ixa_cred = cr;
5700 5710 ixa->ixa_cpid = pid;
5701 5711 if (is_system_labeled()) {
5702 5712 /* We need to restart with a label based on the cred */
5703 5713 ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
5704 5714 }
5705 5715
5706 5716 if (scopeid != 0) {
5707 5717 ixa->ixa_flags |= IXAF_SCOPEID_SET;
5708 5718 ixa->ixa_scopeid = scopeid;
5709 5719 connp->conn_incoming_ifindex = scopeid;
5710 5720 } else {
5711 5721 ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5712 5722 connp->conn_incoming_ifindex = connp->conn_bound_if;
5713 5723 }
5714 5724 /*
5715 5725 * conn_connect will drop conn_lock and reacquire it.
5716 5726 * To prevent a send* from messing with this udp_t while the lock
5717 5727 * is dropped we set udp_state and clear conn_v6lastdst.
5718 5728 * That will make all send* fail with EISCONN.
5719 5729 */
5720 5730 connp->conn_v6lastdst = ipv6_all_zeros;
5721 5731 udp->udp_state = TS_WCON_CREQ;
5722 5732
5723 5733 error = conn_connect(connp, NULL, IPDF_ALLOW_MCBC);
5724 5734 mutex_exit(&connp->conn_lock);
5725 5735 if (error != 0)
5726 5736 goto connect_failed;
5727 5737
5728 5738 /*
5729 5739 * The addresses have been verified. Time to insert in
5730 5740 * the correct fanout list.
5731 5741 */
5732 5742 error = ipcl_conn_insert(connp);
5733 5743 if (error != 0)
5734 5744 goto connect_failed;
5735 5745
5736 5746 mutex_enter(&connp->conn_lock);
5737 5747 error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5738 5748 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5739 5749 if (error != 0) {
5740 5750 mutex_exit(&connp->conn_lock);
5741 5751 goto connect_failed;
5742 5752 }
5743 5753
5744 5754 udp->udp_state = TS_DATA_XFER;
5745 5755 /* Record this as the "last" send even though we haven't sent any */
5746 5756 connp->conn_v6lastdst = connp->conn_faddr_v6;
5747 5757 connp->conn_lastipversion = connp->conn_ipversion;
5748 5758 connp->conn_lastdstport = connp->conn_fport;
5749 5759 connp->conn_lastflowinfo = connp->conn_flowinfo;
5750 5760 connp->conn_lastscopeid = scopeid;
5751 5761 connp->conn_lastsrcid = srcid;
5752 5762 /* Also remember a source to use together with lastdst */
5753 5763 connp->conn_v6lastsrc = v6src;
5754 5764
5755 5765 oldixa = conn_replace_ixa(connp, ixa);
5756 5766 mutex_exit(&connp->conn_lock);
5757 5767 ixa_refrele(oldixa);
5758 5768
5759 5769 /*
5760 5770 * We've picked a source address above. Now we can
5761 5771 * verify that the src/port/dst/port is unique for all
5762 5772 * connections in TS_DATA_XFER, skipping ourselves.
5763 5773 */
5764 5774 mutex_enter(&udpf->uf_lock);
5765 5775 for (udp1 = udpf->uf_udp; udp1 != NULL; udp1 = udp1->udp_bind_hash) {
5766 5776 if (udp1->udp_state != TS_DATA_XFER)
5767 5777 continue;
5768 5778
5769 5779 if (udp1 == udp)
5770 5780 continue;
5771 5781
5772 5782 connp1 = udp1->udp_connp;
5773 5783 if (connp->conn_lport != connp1->conn_lport ||
5774 5784 connp->conn_ipversion != connp1->conn_ipversion ||
5775 5785 dstport != connp1->conn_fport ||
5776 5786 !IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6,
5777 5787 &connp1->conn_laddr_v6) ||
5778 5788 !IN6_ARE_ADDR_EQUAL(&v6dst, &connp1->conn_faddr_v6) ||
5779 5789 !(IPCL_ZONE_MATCH(connp, connp1->conn_zoneid) ||
5780 5790 IPCL_ZONE_MATCH(connp1, connp->conn_zoneid)))
5781 5791 continue;
5782 5792 mutex_exit(&udpf->uf_lock);
5783 5793 error = -TBADADDR;
5784 5794 goto connect_failed;
5785 5795 }
5786 5796 if (cl_inet_connect2 != NULL) {
5787 5797 CL_INET_UDP_CONNECT(connp, B_TRUE, &v6dst, dstport, error);
5788 5798 if (error != 0) {
5789 5799 mutex_exit(&udpf->uf_lock);
5790 5800 error = -TBADADDR;
5791 5801 goto connect_failed;
5792 5802 }
5793 5803 }
5794 5804 mutex_exit(&udpf->uf_lock);
5795 5805
5796 5806 ixa_refrele(ixa);
5797 5807 return (0);
5798 5808
5799 5809 connect_failed:
5800 5810 if (ixa != NULL)
5801 5811 ixa_refrele(ixa);
5802 5812 mutex_enter(&connp->conn_lock);
5803 5813 mutex_enter(&udpf->uf_lock);
5804 5814 udp->udp_state = TS_IDLE;
5805 5815 connp->conn_faddr_v6 = ipv6_all_zeros;
5806 5816 connp->conn_fport = 0;
5807 5817 /* In case the source address was set above */
5808 5818 if (connp->conn_mcbc_bind)
5809 5819 connp->conn_saddr_v6 = ipv6_all_zeros;
5810 5820 else
5811 5821 connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
5812 5822 connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
5813 5823 mutex_exit(&udpf->uf_lock);
5814 5824
5815 5825 connp->conn_v6lastdst = ipv6_all_zeros;
5816 5826 connp->conn_flowinfo = 0;
5817 5827
5818 5828 (void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5819 5829 &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5820 5830 mutex_exit(&connp->conn_lock);
5821 5831 return (error);
5822 5832 }
5823 5833
5824 5834 static int
5825 5835 udp_connect(sock_lower_handle_t proto_handle, const struct sockaddr *sa,
5826 5836 socklen_t len, sock_connid_t *id, cred_t *cr)
5827 5837 {
5828 5838 conn_t *connp = (conn_t *)proto_handle;
5829 5839 udp_t *udp = connp->conn_udp;
5830 5840 int error;
5831 5841 boolean_t did_bind = B_FALSE;
5832 5842 pid_t pid = curproc->p_pid;
5833 5843
5834 5844 /* All Solaris components should pass a cred for this operation. */
5835 5845 ASSERT(cr != NULL);
5836 5846
5837 5847 if (sa == NULL) {
5838 5848 /*
5839 5849 * Disconnect
5840 5850 * Make sure we are connected
5841 5851 */
5842 5852 if (udp->udp_state != TS_DATA_XFER)
5843 5853 return (EINVAL);
5844 5854
5845 5855 error = udp_disconnect(connp);
5846 5856 return (error);
5847 5857 }
5848 5858
5849 5859 error = proto_verify_ip_addr(connp->conn_family, sa, len);
5850 5860 if (error != 0)
5851 5861 goto done;
5852 5862
5853 5863 /* do an implicit bind if necessary */
5854 5864 if (udp->udp_state == TS_UNBND) {
5855 5865 error = udp_implicit_bind(connp, cr);
5856 5866 /*
5857 5867 * We could be racing with an actual bind, in which case
5858 5868 * we would see EPROTO. We cross our fingers and try
5859 5869 * to connect.
5860 5870 */
5861 5871 if (!(error == 0 || error == EPROTO))
5862 5872 goto done;
5863 5873 did_bind = B_TRUE;
5864 5874 }
5865 5875 /*
5866 5876 * set SO_DGRAM_ERRIND
5867 5877 */
5868 5878 connp->conn_dgram_errind = B_TRUE;
5869 5879
5870 5880 error = udp_do_connect(connp, sa, len, cr, pid);
5871 5881
5872 5882 if (error != 0 && did_bind) {
5873 5883 int unbind_err;
5874 5884
5875 5885 unbind_err = udp_do_unbind(connp);
5876 5886 ASSERT(unbind_err == 0);
5877 5887 }
5878 5888
5879 5889 if (error == 0) {
5880 5890 *id = 0;
5881 5891 (*connp->conn_upcalls->su_connected)
5882 5892 (connp->conn_upper_handle, 0, NULL, -1);
5883 5893 } else if (error < 0) {
5884 5894 error = proto_tlitosyserr(-error);
5885 5895 }
5886 5896
5887 5897 done:
5888 5898 if (error != 0 && udp->udp_state == TS_DATA_XFER) {
5889 5899 /*
5890 5900 * No need to hold locks to set state
5891 5901 * after connect failure socket state is undefined
5892 5902 * We set the state only to imitate old sockfs behavior
5893 5903 */
5894 5904 udp->udp_state = TS_IDLE;
5895 5905 }
5896 5906 return (error);
5897 5907 }
5898 5908
5899 5909 int
5900 5910 udp_send(sock_lower_handle_t proto_handle, mblk_t *mp, struct nmsghdr *msg,
5901 5911 cred_t *cr)
5902 5912 {
|
↓ open down ↓ |
4172 lines elided |
↑ open up ↑ |
5903 5913 sin6_t *sin6;
5904 5914 sin_t *sin = NULL;
5905 5915 uint_t srcid;
5906 5916 conn_t *connp = (conn_t *)proto_handle;
5907 5917 udp_t *udp = connp->conn_udp;
5908 5918 int error = 0;
5909 5919 udp_stack_t *us = udp->udp_us;
5910 5920 ushort_t ipversion;
5911 5921 pid_t pid = curproc->p_pid;
5912 5922 ip_xmit_attr_t *ixa;
5923 + boolean_t snd_to_conn;
5913 5924
5914 5925 ASSERT(DB_TYPE(mp) == M_DATA);
5915 5926
5916 5927 /* All Solaris components should pass a cred for this operation. */
5917 5928 ASSERT(cr != NULL);
5918 5929
5919 5930 /* do an implicit bind if necessary */
5920 5931 if (udp->udp_state == TS_UNBND) {
5921 5932 error = udp_implicit_bind(connp, cr);
5922 5933 /*
5923 5934 * We could be racing with an actual bind, in which case
5924 5935 * we would see EPROTO. We cross our fingers and try
5925 5936 * to connect.
5926 5937 */
5927 5938 if (!(error == 0 || error == EPROTO)) {
5928 5939 freemsg(mp);
5929 5940 return (error);
5930 5941 }
5931 5942 }
5932 5943
5933 5944 /* Connected? */
5934 5945 if (msg->msg_name == NULL) {
5935 5946 if (udp->udp_state != TS_DATA_XFER) {
5936 5947 UDPS_BUMP_MIB(us, udpOutErrors);
5937 5948 return (EDESTADDRREQ);
5938 5949 }
5939 5950 if (msg->msg_controllen != 0) {
|
↓ open down ↓ |
17 lines elided |
↑ open up ↑ |
5940 5951 error = udp_output_ancillary(connp, NULL, NULL, mp,
5941 5952 NULL, msg, cr, pid);
5942 5953 } else {
5943 5954 error = udp_output_connected(connp, mp, cr, pid);
5944 5955 }
5945 5956 if (us->us_sendto_ignerr)
5946 5957 return (0);
5947 5958 else
5948 5959 return (error);
5949 5960 }
5950 - if (udp->udp_state == TS_DATA_XFER) {
5961 +
5962 + /*
5963 + * Check if we're allowed to send to a connection on which we've
5964 + * already called 'connect'. The posix spec. allows both behaviors but
5965 + * historically we've returned an error if already connected. The
5966 + * client can allow this via a sockopt.
5967 + */
5968 + mutex_enter(&connp->conn_lock);
5969 + snd_to_conn = (udp->udp_snd_to_conn != 0);
5970 + mutex_exit(&connp->conn_lock);
5971 + if (udp->udp_state == TS_DATA_XFER && !snd_to_conn) {
5951 5972 UDPS_BUMP_MIB(us, udpOutErrors);
5952 5973 return (EISCONN);
5953 5974 }
5975 +
5954 5976 error = proto_verify_ip_addr(connp->conn_family,
5955 5977 (struct sockaddr *)msg->msg_name, msg->msg_namelen);
5956 5978 if (error != 0) {
5957 5979 UDPS_BUMP_MIB(us, udpOutErrors);
5958 5980 return (error);
5959 5981 }
5960 5982 switch (connp->conn_family) {
5961 5983 case AF_INET6:
5962 5984 sin6 = (sin6_t *)msg->msg_name;
5963 5985
5964 5986 srcid = sin6->__sin6_src_id;
5965 5987
5966 5988 if (!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
5967 5989 /*
5968 5990 * Destination is a non-IPv4-compatible IPv6 address.
5969 5991 * Send out an IPv6 format packet.
5970 5992 */
5971 5993
5972 5994 /*
5973 5995 * If the local address is a mapped address return
5974 5996 * an error.
5975 5997 * It would be possible to send an IPv6 packet but the
5976 5998 * response would never make it back to the application
5977 5999 * since it is bound to a mapped address.
5978 6000 */
5979 6001 if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
5980 6002 UDPS_BUMP_MIB(us, udpOutErrors);
5981 6003 return (EADDRNOTAVAIL);
5982 6004 }
5983 6005 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
5984 6006 sin6->sin6_addr = ipv6_loopback;
5985 6007 ipversion = IPV6_VERSION;
5986 6008 } else {
5987 6009 if (connp->conn_ipv6_v6only) {
5988 6010 UDPS_BUMP_MIB(us, udpOutErrors);
5989 6011 return (EADDRNOTAVAIL);
5990 6012 }
5991 6013
5992 6014 /*
5993 6015 * If the local address is not zero or a mapped address
5994 6016 * return an error. It would be possible to send an
5995 6017 * IPv4 packet but the response would never make it
5996 6018 * back to the application since it is bound to a
5997 6019 * non-mapped address.
5998 6020 */
5999 6021 if (!IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6) &&
6000 6022 !IN6_IS_ADDR_UNSPECIFIED(&connp->conn_saddr_v6)) {
6001 6023 UDPS_BUMP_MIB(us, udpOutErrors);
6002 6024 return (EADDRNOTAVAIL);
6003 6025 }
6004 6026
6005 6027 if (V4_PART_OF_V6(sin6->sin6_addr) == INADDR_ANY) {
6006 6028 V4_PART_OF_V6(sin6->sin6_addr) =
6007 6029 htonl(INADDR_LOOPBACK);
6008 6030 }
6009 6031 ipversion = IPV4_VERSION;
6010 6032 }
6011 6033
6012 6034 /*
6013 6035 * We have to allocate an ip_xmit_attr_t before we grab
6014 6036 * conn_lock and we need to hold conn_lock once we've check
6015 6037 * conn_same_as_last_v6 to handle concurrent send* calls on a
6016 6038 * socket.
6017 6039 */
6018 6040 if (msg->msg_controllen == 0) {
6019 6041 ixa = conn_get_ixa(connp, B_FALSE);
6020 6042 if (ixa == NULL) {
6021 6043 UDPS_BUMP_MIB(us, udpOutErrors);
6022 6044 return (ENOMEM);
6023 6045 }
6024 6046 } else {
6025 6047 ixa = NULL;
6026 6048 }
6027 6049 mutex_enter(&connp->conn_lock);
6028 6050 if (udp->udp_delayed_error != 0) {
6029 6051 sin6_t *sin2 = (sin6_t *)&udp->udp_delayed_addr;
6030 6052
6031 6053 error = udp->udp_delayed_error;
6032 6054 udp->udp_delayed_error = 0;
6033 6055
6034 6056 /* Compare IP address, port, and family */
6035 6057
6036 6058 if (sin6->sin6_port == sin2->sin6_port &&
6037 6059 IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
6038 6060 &sin2->sin6_addr) &&
6039 6061 sin6->sin6_family == sin2->sin6_family) {
6040 6062 mutex_exit(&connp->conn_lock);
6041 6063 UDPS_BUMP_MIB(us, udpOutErrors);
6042 6064 if (ixa != NULL)
6043 6065 ixa_refrele(ixa);
6044 6066 return (error);
6045 6067 }
6046 6068 }
6047 6069
6048 6070 if (msg->msg_controllen != 0) {
6049 6071 mutex_exit(&connp->conn_lock);
6050 6072 ASSERT(ixa == NULL);
6051 6073 error = udp_output_ancillary(connp, NULL, sin6, mp,
6052 6074 NULL, msg, cr, pid);
6053 6075 } else if (conn_same_as_last_v6(connp, sin6) &&
6054 6076 connp->conn_lastsrcid == srcid &&
6055 6077 ipsec_outbound_policy_current(ixa)) {
6056 6078 /* udp_output_lastdst drops conn_lock */
6057 6079 error = udp_output_lastdst(connp, mp, cr, pid, ixa);
6058 6080 } else {
6059 6081 /* udp_output_newdst drops conn_lock */
6060 6082 error = udp_output_newdst(connp, mp, NULL, sin6,
6061 6083 ipversion, cr, pid, ixa);
6062 6084 }
6063 6085 ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
6064 6086 if (us->us_sendto_ignerr)
6065 6087 return (0);
6066 6088 else
6067 6089 return (error);
6068 6090 case AF_INET:
6069 6091 sin = (sin_t *)msg->msg_name;
6070 6092
6071 6093 ipversion = IPV4_VERSION;
6072 6094
6073 6095 if (sin->sin_addr.s_addr == INADDR_ANY)
6074 6096 sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
6075 6097
6076 6098 /*
6077 6099 * We have to allocate an ip_xmit_attr_t before we grab
6078 6100 * conn_lock and we need to hold conn_lock once we've check
6079 6101 * conn_same_as_last_v6 to handle concurrent send* on a socket.
6080 6102 */
6081 6103 if (msg->msg_controllen == 0) {
6082 6104 ixa = conn_get_ixa(connp, B_FALSE);
6083 6105 if (ixa == NULL) {
6084 6106 UDPS_BUMP_MIB(us, udpOutErrors);
6085 6107 return (ENOMEM);
6086 6108 }
6087 6109 } else {
6088 6110 ixa = NULL;
6089 6111 }
6090 6112 mutex_enter(&connp->conn_lock);
6091 6113 if (udp->udp_delayed_error != 0) {
6092 6114 sin_t *sin2 = (sin_t *)&udp->udp_delayed_addr;
6093 6115
6094 6116 error = udp->udp_delayed_error;
6095 6117 udp->udp_delayed_error = 0;
6096 6118
6097 6119 /* Compare IP address and port */
6098 6120
6099 6121 if (sin->sin_port == sin2->sin_port &&
6100 6122 sin->sin_addr.s_addr == sin2->sin_addr.s_addr) {
6101 6123 mutex_exit(&connp->conn_lock);
6102 6124 UDPS_BUMP_MIB(us, udpOutErrors);
6103 6125 if (ixa != NULL)
6104 6126 ixa_refrele(ixa);
6105 6127 return (error);
6106 6128 }
6107 6129 }
6108 6130 if (msg->msg_controllen != 0) {
6109 6131 mutex_exit(&connp->conn_lock);
6110 6132 ASSERT(ixa == NULL);
6111 6133 error = udp_output_ancillary(connp, sin, NULL, mp,
6112 6134 NULL, msg, cr, pid);
6113 6135 } else if (conn_same_as_last_v4(connp, sin) &&
6114 6136 ipsec_outbound_policy_current(ixa)) {
6115 6137 /* udp_output_lastdst drops conn_lock */
6116 6138 error = udp_output_lastdst(connp, mp, cr, pid, ixa);
6117 6139 } else {
6118 6140 /* udp_output_newdst drops conn_lock */
6119 6141 error = udp_output_newdst(connp, mp, sin, NULL,
6120 6142 ipversion, cr, pid, ixa);
6121 6143 }
6122 6144 ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
6123 6145 if (us->us_sendto_ignerr)
6124 6146 return (0);
6125 6147 else
6126 6148 return (error);
6127 6149 default:
6128 6150 return (EINVAL);
6129 6151 }
6130 6152 }
6131 6153
6132 6154 int
6133 6155 udp_fallback(sock_lower_handle_t proto_handle, queue_t *q,
6134 6156 boolean_t issocket, so_proto_quiesced_cb_t quiesced_cb,
6135 6157 sock_quiesce_arg_t *arg)
6136 6158 {
6137 6159 conn_t *connp = (conn_t *)proto_handle;
6138 6160 udp_t *udp;
6139 6161 struct T_capability_ack tca;
6140 6162 struct sockaddr_in6 laddr, faddr;
6141 6163 socklen_t laddrlen, faddrlen;
6142 6164 short opts;
6143 6165 struct stroptions *stropt;
6144 6166 mblk_t *mp, *stropt_mp;
6145 6167 int error;
6146 6168
6147 6169 udp = connp->conn_udp;
6148 6170
6149 6171 stropt_mp = allocb_wait(sizeof (*stropt), BPRI_HI, STR_NOSIG, NULL);
6150 6172
6151 6173 /*
6152 6174 * setup the fallback stream that was allocated
6153 6175 */
6154 6176 connp->conn_dev = (dev_t)RD(q)->q_ptr;
6155 6177 connp->conn_minor_arena = WR(q)->q_ptr;
6156 6178
6157 6179 RD(q)->q_ptr = WR(q)->q_ptr = connp;
6158 6180
6159 6181 WR(q)->q_qinfo = &udp_winit;
6160 6182
6161 6183 connp->conn_rq = RD(q);
6162 6184 connp->conn_wq = WR(q);
6163 6185
6164 6186 /* Notify stream head about options before sending up data */
6165 6187 stropt_mp->b_datap->db_type = M_SETOPTS;
6166 6188 stropt_mp->b_wptr += sizeof (*stropt);
6167 6189 stropt = (struct stroptions *)stropt_mp->b_rptr;
6168 6190 stropt->so_flags = SO_WROFF | SO_HIWAT;
6169 6191 stropt->so_wroff = connp->conn_wroff;
6170 6192 stropt->so_hiwat = udp->udp_rcv_disply_hiwat;
6171 6193 putnext(RD(q), stropt_mp);
6172 6194
6173 6195 /*
6174 6196 * Free the helper stream
6175 6197 */
6176 6198 ip_free_helper_stream(connp);
6177 6199
6178 6200 if (!issocket)
6179 6201 udp_use_pure_tpi(udp);
6180 6202
6181 6203 /*
6182 6204 * Collect the information needed to sync with the sonode
6183 6205 */
6184 6206 udp_do_capability_ack(udp, &tca, TC1_INFO);
6185 6207
6186 6208 laddrlen = faddrlen = sizeof (sin6_t);
6187 6209 (void) udp_getsockname((sock_lower_handle_t)connp,
6188 6210 (struct sockaddr *)&laddr, &laddrlen, CRED());
6189 6211 error = udp_getpeername((sock_lower_handle_t)connp,
6190 6212 (struct sockaddr *)&faddr, &faddrlen, CRED());
6191 6213 if (error != 0)
6192 6214 faddrlen = 0;
6193 6215
6194 6216 opts = 0;
6195 6217 if (connp->conn_dgram_errind)
6196 6218 opts |= SO_DGRAM_ERRIND;
6197 6219 if (connp->conn_ixa->ixa_flags & IXAF_DONTROUTE)
6198 6220 opts |= SO_DONTROUTE;
6199 6221
6200 6222 mp = (*quiesced_cb)(connp->conn_upper_handle, arg, &tca,
6201 6223 (struct sockaddr *)&laddr, laddrlen,
6202 6224 (struct sockaddr *)&faddr, faddrlen, opts);
6203 6225
6204 6226 mutex_enter(&udp->udp_recv_lock);
6205 6227 /*
6206 6228 * Attempts to send data up during fallback will result in it being
6207 6229 * queued in udp_t. First push up the datagrams obtained from the
6208 6230 * socket, then any packets queued in udp_t.
6209 6231 */
6210 6232 if (mp != NULL) {
6211 6233 mp->b_next = udp->udp_fallback_queue_head;
6212 6234 udp->udp_fallback_queue_head = mp;
6213 6235 }
6214 6236 while (udp->udp_fallback_queue_head != NULL) {
6215 6237 mp = udp->udp_fallback_queue_head;
6216 6238 udp->udp_fallback_queue_head = mp->b_next;
6217 6239 mutex_exit(&udp->udp_recv_lock);
6218 6240 mp->b_next = NULL;
6219 6241 putnext(RD(q), mp);
6220 6242 mutex_enter(&udp->udp_recv_lock);
6221 6243 }
6222 6244 udp->udp_fallback_queue_tail = udp->udp_fallback_queue_head;
6223 6245 /*
6224 6246 * No longer a streams less socket
6225 6247 */
6226 6248 mutex_enter(&connp->conn_lock);
6227 6249 connp->conn_flags &= ~IPCL_NONSTR;
6228 6250 mutex_exit(&connp->conn_lock);
6229 6251
6230 6252 mutex_exit(&udp->udp_recv_lock);
6231 6253
6232 6254 ASSERT(connp->conn_ref >= 1);
6233 6255
6234 6256 return (0);
6235 6257 }
6236 6258
6237 6259 /* ARGSUSED3 */
6238 6260 int
6239 6261 udp_getpeername(sock_lower_handle_t proto_handle, struct sockaddr *sa,
6240 6262 socklen_t *salenp, cred_t *cr)
6241 6263 {
6242 6264 conn_t *connp = (conn_t *)proto_handle;
6243 6265 udp_t *udp = connp->conn_udp;
6244 6266 int error;
6245 6267
6246 6268 /* All Solaris components should pass a cred for this operation. */
6247 6269 ASSERT(cr != NULL);
6248 6270
6249 6271 mutex_enter(&connp->conn_lock);
6250 6272 if (udp->udp_state != TS_DATA_XFER)
6251 6273 error = ENOTCONN;
6252 6274 else
6253 6275 error = conn_getpeername(connp, sa, salenp);
6254 6276 mutex_exit(&connp->conn_lock);
6255 6277 return (error);
6256 6278 }
6257 6279
6258 6280 /* ARGSUSED3 */
6259 6281 int
6260 6282 udp_getsockname(sock_lower_handle_t proto_handle, struct sockaddr *sa,
6261 6283 socklen_t *salenp, cred_t *cr)
6262 6284 {
6263 6285 conn_t *connp = (conn_t *)proto_handle;
6264 6286 int error;
6265 6287
6266 6288 /* All Solaris components should pass a cred for this operation. */
6267 6289 ASSERT(cr != NULL);
6268 6290
6269 6291 mutex_enter(&connp->conn_lock);
6270 6292 error = conn_getsockname(connp, sa, salenp);
6271 6293 mutex_exit(&connp->conn_lock);
6272 6294 return (error);
6273 6295 }
6274 6296
6275 6297 int
6276 6298 udp_getsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
6277 6299 void *optvalp, socklen_t *optlen, cred_t *cr)
6278 6300 {
6279 6301 conn_t *connp = (conn_t *)proto_handle;
6280 6302 int error;
6281 6303 t_uscalar_t max_optbuf_len;
6282 6304 void *optvalp_buf;
6283 6305 int len;
6284 6306
6285 6307 /* All Solaris components should pass a cred for this operation. */
6286 6308 ASSERT(cr != NULL);
6287 6309
6288 6310 error = proto_opt_check(level, option_name, *optlen, &max_optbuf_len,
6289 6311 udp_opt_obj.odb_opt_des_arr,
6290 6312 udp_opt_obj.odb_opt_arr_cnt,
6291 6313 B_FALSE, B_TRUE, cr);
6292 6314 if (error != 0) {
6293 6315 if (error < 0)
6294 6316 error = proto_tlitosyserr(-error);
6295 6317 return (error);
6296 6318 }
6297 6319
6298 6320 optvalp_buf = kmem_alloc(max_optbuf_len, KM_SLEEP);
6299 6321 len = udp_opt_get(connp, level, option_name, optvalp_buf);
6300 6322 if (len == -1) {
6301 6323 kmem_free(optvalp_buf, max_optbuf_len);
6302 6324 return (EINVAL);
6303 6325 }
6304 6326
6305 6327 /*
6306 6328 * update optlen and copy option value
6307 6329 */
6308 6330 t_uscalar_t size = MIN(len, *optlen);
6309 6331
6310 6332 bcopy(optvalp_buf, optvalp, size);
6311 6333 bcopy(&size, optlen, sizeof (size));
6312 6334
6313 6335 kmem_free(optvalp_buf, max_optbuf_len);
6314 6336 return (0);
6315 6337 }
6316 6338
6317 6339 int
6318 6340 udp_setsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
6319 6341 const void *optvalp, socklen_t optlen, cred_t *cr)
6320 6342 {
6321 6343 conn_t *connp = (conn_t *)proto_handle;
6322 6344 int error;
6323 6345
6324 6346 /* All Solaris components should pass a cred for this operation. */
6325 6347 ASSERT(cr != NULL);
6326 6348
6327 6349 error = proto_opt_check(level, option_name, optlen, NULL,
6328 6350 udp_opt_obj.odb_opt_des_arr,
6329 6351 udp_opt_obj.odb_opt_arr_cnt,
6330 6352 B_TRUE, B_FALSE, cr);
6331 6353
6332 6354 if (error != 0) {
6333 6355 if (error < 0)
6334 6356 error = proto_tlitosyserr(-error);
6335 6357 return (error);
6336 6358 }
6337 6359
6338 6360 error = udp_opt_set(connp, SETFN_OPTCOM_NEGOTIATE, level, option_name,
6339 6361 optlen, (uchar_t *)optvalp, (uint_t *)&optlen, (uchar_t *)optvalp,
6340 6362 NULL, cr);
6341 6363
6342 6364 ASSERT(error >= 0);
6343 6365
6344 6366 return (error);
6345 6367 }
6346 6368
6347 6369 void
6348 6370 udp_clr_flowctrl(sock_lower_handle_t proto_handle)
6349 6371 {
6350 6372 conn_t *connp = (conn_t *)proto_handle;
6351 6373 udp_t *udp = connp->conn_udp;
6352 6374
6353 6375 mutex_enter(&udp->udp_recv_lock);
6354 6376 connp->conn_flow_cntrld = B_FALSE;
6355 6377 mutex_exit(&udp->udp_recv_lock);
6356 6378 }
6357 6379
6358 6380 /* ARGSUSED2 */
6359 6381 int
6360 6382 udp_shutdown(sock_lower_handle_t proto_handle, int how, cred_t *cr)
6361 6383 {
6362 6384 conn_t *connp = (conn_t *)proto_handle;
6363 6385
6364 6386 /* All Solaris components should pass a cred for this operation. */
6365 6387 ASSERT(cr != NULL);
6366 6388
6367 6389 /* shut down the send side */
6368 6390 if (how != SHUT_RD)
6369 6391 (*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
6370 6392 SOCK_OPCTL_SHUT_SEND, 0);
6371 6393 /* shut down the recv side */
6372 6394 if (how != SHUT_WR)
6373 6395 (*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
6374 6396 SOCK_OPCTL_SHUT_RECV, 0);
6375 6397 return (0);
6376 6398 }
6377 6399
6378 6400 int
6379 6401 udp_ioctl(sock_lower_handle_t proto_handle, int cmd, intptr_t arg,
6380 6402 int mode, int32_t *rvalp, cred_t *cr)
6381 6403 {
6382 6404 conn_t *connp = (conn_t *)proto_handle;
6383 6405 int error;
6384 6406
6385 6407 /* All Solaris components should pass a cred for this operation. */
6386 6408 ASSERT(cr != NULL);
6387 6409
6388 6410 /*
6389 6411 * If we don't have a helper stream then create one.
6390 6412 * ip_create_helper_stream takes care of locking the conn_t,
6391 6413 * so this check for NULL is just a performance optimization.
6392 6414 */
6393 6415 if (connp->conn_helper_info == NULL) {
6394 6416 udp_stack_t *us = connp->conn_udp->udp_us;
6395 6417
6396 6418 ASSERT(us->us_ldi_ident != NULL);
6397 6419
6398 6420 /*
6399 6421 * Create a helper stream for non-STREAMS socket.
6400 6422 */
6401 6423 error = ip_create_helper_stream(connp, us->us_ldi_ident);
6402 6424 if (error != 0) {
6403 6425 ip0dbg(("tcp_ioctl: create of IP helper stream "
6404 6426 "failed %d\n", error));
6405 6427 return (error);
6406 6428 }
6407 6429 }
6408 6430
6409 6431 switch (cmd) {
6410 6432 case _SIOCSOCKFALLBACK:
6411 6433 case TI_GETPEERNAME:
6412 6434 case TI_GETMYNAME:
6413 6435 ip1dbg(("udp_ioctl: cmd 0x%x on non streams socket",
6414 6436 cmd));
6415 6437 error = EINVAL;
6416 6438 break;
6417 6439 default:
6418 6440 /*
6419 6441 * Pass on to IP using helper stream
6420 6442 */
6421 6443 error = ldi_ioctl(connp->conn_helper_info->iphs_handle,
6422 6444 cmd, arg, mode, cr, rvalp);
6423 6445 break;
6424 6446 }
6425 6447 return (error);
6426 6448 }
6427 6449
6428 6450 /* ARGSUSED */
6429 6451 int
6430 6452 udp_accept(sock_lower_handle_t lproto_handle,
6431 6453 sock_lower_handle_t eproto_handle, sock_upper_handle_t sock_handle,
6432 6454 cred_t *cr)
6433 6455 {
6434 6456 return (EOPNOTSUPP);
6435 6457 }
6436 6458
6437 6459 /* ARGSUSED */
6438 6460 int
6439 6461 udp_listen(sock_lower_handle_t proto_handle, int backlog, cred_t *cr)
6440 6462 {
6441 6463 return (EOPNOTSUPP);
6442 6464 }
6443 6465
6444 6466 sock_downcalls_t sock_udp_downcalls = {
6445 6467 udp_activate, /* sd_activate */
6446 6468 udp_accept, /* sd_accept */
6447 6469 udp_bind, /* sd_bind */
6448 6470 udp_listen, /* sd_listen */
6449 6471 udp_connect, /* sd_connect */
6450 6472 udp_getpeername, /* sd_getpeername */
6451 6473 udp_getsockname, /* sd_getsockname */
6452 6474 udp_getsockopt, /* sd_getsockopt */
6453 6475 udp_setsockopt, /* sd_setsockopt */
6454 6476 udp_send, /* sd_send */
6455 6477 NULL, /* sd_send_uio */
6456 6478 NULL, /* sd_recv_uio */
6457 6479 NULL, /* sd_poll */
6458 6480 udp_shutdown, /* sd_shutdown */
6459 6481 udp_clr_flowctrl, /* sd_setflowctrl */
6460 6482 udp_ioctl, /* sd_ioctl */
6461 6483 udp_close /* sd_close */
6462 6484 };
|
↓ open down ↓ |
499 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX