author | kais |
Sat, 12 Nov 2005 18:58:05 -0800 | |
changeset 898 | 64b2a371a6bd |
parent 741 | 40027a3621ac |
child 1092 | 11ac87ff53c2 |
permissions | -rw-r--r-- |
0 | 1 |
/* |
2 |
* CDDL HEADER START |
|
3 |
* |
|
4 |
* The contents of this file are subject to the terms of the |
|
5 |
* Common Development and Distribution License, Version 1.0 only |
|
6 |
* (the "License"). You may not use this file except in compliance |
|
7 |
* with the License. |
|
8 |
* |
|
9 |
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE |
|
10 |
* or http://www.opensolaris.org/os/licensing. |
|
11 |
* See the License for the specific language governing permissions |
|
12 |
* and limitations under the License. |
|
13 |
* |
|
14 |
* When distributing Covered Code, include this CDDL HEADER in each |
|
15 |
* file and include the License file at usr/src/OPENSOLARIS.LICENSE. |
|
16 |
* If applicable, add the following below this CDDL HEADER, with the |
|
17 |
* fields enclosed by brackets "[]" replaced with your own identifying |
|
18 |
* information: Portions Copyright [yyyy] [name of copyright owner] |
|
19 |
* |
|
20 |
* CDDL HEADER END |
|
21 |
*/ |
|
22 |
/* |
|
23 |
* Copyright 2005 Sun Microsystems, Inc. All rights reserved. |
|
24 |
* Use is subject to license terms. |
|
25 |
*/ |
|
26 |
||
27 |
#pragma ident "%Z%%M% %I% %E% SMI" |
|
28 |
||
29 |
#include <sys/types.h> |
|
30 |
#include <sys/t_lock.h> |
|
31 |
#include <sys/param.h> |
|
32 |
#include <sys/systm.h> |
|
33 |
#include <sys/buf.h> |
|
34 |
#include <sys/conf.h> |
|
35 |
#include <sys/cred.h> |
|
36 |
#include <sys/kmem.h> |
|
37 |
#include <sys/sysmacros.h> |
|
38 |
#include <sys/vfs.h> |
|
39 |
#include <sys/vnode.h> |
|
40 |
#include <sys/debug.h> |
|
41 |
#include <sys/errno.h> |
|
42 |
#include <sys/time.h> |
|
43 |
#include <sys/file.h> |
|
44 |
#include <sys/open.h> |
|
45 |
#include <sys/user.h> |
|
46 |
#include <sys/termios.h> |
|
47 |
#include <sys/stream.h> |
|
48 |
#include <sys/strsubr.h> |
|
49 |
#include <sys/strsun.h> |
|
50 |
#include <sys/ddi.h> |
|
51 |
#include <sys/esunddi.h> |
|
52 |
#include <sys/flock.h> |
|
53 |
#include <sys/modctl.h> |
|
54 |
#include <sys/vtrace.h> |
|
55 |
#include <sys/cmn_err.h> |
|
56 |
#include <sys/pathname.h> |
|
57 |
||
58 |
#include <sys/socket.h> |
|
59 |
#include <sys/socketvar.h> |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
60 |
#include <sys/sockio.h> |
0 | 61 |
#include <netinet/in.h> |
62 |
#include <sys/un.h> |
|
63 |
#include <sys/strsun.h> |
|
64 |
||
65 |
#include <sys/tiuser.h> |
|
66 |
#define _SUN_TPI_VERSION 2 |
|
67 |
#include <sys/tihdr.h> |
|
68 |
#include <sys/timod.h> /* TI_GETMYNAME, TI_GETPEERNAME */ |
|
69 |
||
70 |
#include <c2/audit.h> |
|
71 |
||
72 |
#include <inet/common.h> |
|
73 |
#include <inet/ip.h> |
|
74 |
#include <inet/ip6.h> |
|
75 |
#include <inet/tcp.h> |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
76 |
#include <inet/udp_impl.h> |
0 | 77 |
|
78 |
#include <fs/sockfs/nl7c.h> |
|
79 |
#include <sys/zone.h> |
|
80 |
||
898 | 81 |
#include <inet/kssl/ksslapi.h> |
82 |
||
0 | 83 |
/* |
84 |
* Possible failures when memory can't be allocated. The documented behavior: |
|
85 |
* |
|
86 |
* 5.5: 4.X: XNET: |
|
87 |
* accept: ENOMEM/ENOSR/EINTR - (EINTR) ENOMEM/ENOBUFS/ENOSR/ |
|
88 |
* EINTR |
|
89 |
* (4.X does not document EINTR but returns it) |
|
90 |
* bind: ENOSR - ENOBUFS/ENOSR |
|
91 |
* connect: EINTR EINTR ENOBUFS/ENOSR/EINTR |
|
92 |
* getpeername: ENOMEM/ENOSR ENOBUFS (-) ENOBUFS/ENOSR |
|
93 |
* getsockname: ENOMEM/ENOSR ENOBUFS (-) ENOBUFS/ENOSR |
|
94 |
* (4.X getpeername and getsockname do not fail in practice) |
|
95 |
* getsockopt: ENOMEM/ENOSR - ENOBUFS/ENOSR |
|
96 |
* listen: - - ENOBUFS |
|
97 |
* recv: ENOMEM/ENOSR/EINTR EINTR ENOBUFS/ENOMEM/ENOSR/ |
|
98 |
* EINTR |
|
99 |
* send: ENOMEM/ENOSR/EINTR ENOBUFS/EINTR ENOBUFS/ENOMEM/ENOSR/ |
|
100 |
* EINTR |
|
101 |
* setsockopt: ENOMEM/ENOSR - ENOBUFS/ENOMEM/ENOSR |
|
102 |
* shutdown: ENOMEM/ENOSR - ENOBUFS/ENOSR |
|
103 |
* socket: ENOMEM/ENOSR ENOBUFS ENOBUFS/ENOMEM/ENOSR |
|
104 |
* socketpair: ENOMEM/ENOSR - ENOBUFS/ENOMEM/ENOSR |
|
105 |
* |
|
106 |
* Resolution. When allocation fails: |
|
107 |
* recv: return EINTR |
|
108 |
* send: return EINTR |
|
109 |
* connect, accept: EINTR |
|
110 |
* bind, listen, shutdown (unbind, unix_close, disconnect): sleep |
|
111 |
* socket, socketpair: ENOBUFS |
|
112 |
* getpeername, getsockname: sleep |
|
113 |
* getsockopt, setsockopt: sleep |
|
114 |
*/ |
|
115 |
||
116 |
#ifdef SOCK_TEST |
|
117 |
/* |
|
118 |
* Variables that make sockfs do something other than the standard TPI |
|
119 |
* for the AF_INET transports. |
|
120 |
* |
|
121 |
* solisten_tpi_tcp: |
|
122 |
* TCP can handle a O_T_BIND_REQ with an increased backlog even though |
|
123 |
* the transport is already bound. This is needed to avoid loosing the |
|
124 |
* port number should listen() do a T_UNBIND_REQ followed by a |
|
125 |
* O_T_BIND_REQ. |
|
126 |
* |
|
127 |
* soconnect_tpi_udp: |
|
128 |
* UDP and ICMP can handle a T_CONN_REQ. |
|
129 |
* This is needed to make the sequence of connect(), getsockname() |
|
130 |
* return the local IP address used to send packets to the connected to |
|
131 |
* destination. |
|
132 |
* |
|
133 |
* soconnect_tpi_tcp: |
|
134 |
* TCP can handle a T_CONN_REQ without seeing a O_T_BIND_REQ. |
|
135 |
* Set this to non-zero to send TPI conformant messages to TCP in this |
|
136 |
* respect. This is a performance optimization. |
|
137 |
* |
|
138 |
* soaccept_tpi_tcp: |
|
139 |
* TCP can handle a T_CONN_REQ without the acceptor being bound. |
|
140 |
* This is a performance optimization that has been picked up in XTI. |
|
141 |
* |
|
142 |
* soaccept_tpi_multioptions: |
|
143 |
* When inheriting SOL_SOCKET options from the listener to the accepting |
|
144 |
* socket send them as a single message for AF_INET{,6}. |
|
145 |
*/ |
|
146 |
int solisten_tpi_tcp = 0; |
|
147 |
int soconnect_tpi_udp = 0; |
|
148 |
int soconnect_tpi_tcp = 0; |
|
149 |
int soaccept_tpi_tcp = 0; |
|
150 |
int soaccept_tpi_multioptions = 1; |
|
151 |
#else /* SOCK_TEST */ |
|
152 |
#define soconnect_tpi_tcp 0 |
|
153 |
#define soconnect_tpi_udp 0 |
|
154 |
#define solisten_tpi_tcp 0 |
|
155 |
#define soaccept_tpi_tcp 0 |
|
156 |
#define soaccept_tpi_multioptions 1 |
|
157 |
#endif /* SOCK_TEST */ |
|
158 |
||
159 |
#ifdef SOCK_TEST |
|
160 |
extern int do_useracc; |
|
161 |
extern clock_t sock_test_timelimit; |
|
162 |
#endif /* SOCK_TEST */ |
|
163 |
||
164 |
/* |
|
165 |
* Some X/Open added checks might have to be backed out to keep SunOS 4.X |
|
166 |
* applications working. Turn on this flag to disable these checks. |
|
167 |
*/ |
|
168 |
int xnet_skip_checks = 0; |
|
169 |
int xnet_check_print = 0; |
|
170 |
int xnet_truncate_print = 0; |
|
171 |
||
172 |
extern void sigintr(k_sigset_t *, int); |
|
173 |
extern void sigunintr(k_sigset_t *); |
|
174 |
||
175 |
extern void *nl7c_lookup_addr(void *, t_uscalar_t); |
|
176 |
extern void *nl7c_add_addr(void *, t_uscalar_t); |
|
177 |
extern void nl7c_listener_addr(void *, queue_t *); |
|
178 |
||
898 | 179 |
/* Sockets acting as an in-kernel SSL proxy */ |
180 |
extern mblk_t *strsock_kssl_input(vnode_t *, mblk_t *, strwakeup_t *, |
|
181 |
strsigset_t *, strsigset_t *, strpollset_t *); |
|
182 |
extern mblk_t *strsock_kssl_output(vnode_t *, mblk_t *, strwakeup_t *, |
|
183 |
strsigset_t *, strsigset_t *, strpollset_t *); |
|
184 |
||
0 | 185 |
static int sotpi_unbind(struct sonode *, int); |
186 |
||
187 |
/* TPI sockfs sonode operations */ |
|
188 |
static int sotpi_accept(struct sonode *, int, struct sonode **); |
|
189 |
static int sotpi_bind(struct sonode *, struct sockaddr *, socklen_t, |
|
190 |
int); |
|
191 |
static int sotpi_connect(struct sonode *, const struct sockaddr *, |
|
192 |
socklen_t, int, int); |
|
193 |
static int sotpi_listen(struct sonode *, int); |
|
194 |
static int sotpi_sendmsg(struct sonode *, struct nmsghdr *, |
|
195 |
struct uio *); |
|
196 |
static int sotpi_shutdown(struct sonode *, int); |
|
197 |
static int sotpi_getsockname(struct sonode *); |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
198 |
static int sosend_dgramcmsg(struct sonode *, struct sockaddr *, socklen_t, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
199 |
struct uio *, void *, t_uscalar_t, int); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
200 |
static int sodgram_direct(struct sonode *, struct sockaddr *, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
201 |
socklen_t, struct uio *, int); |
0 | 202 |
|
203 |
sonodeops_t sotpi_sonodeops = { |
|
204 |
sotpi_accept, /* sop_accept */ |
|
205 |
sotpi_bind, /* sop_bind */ |
|
206 |
sotpi_listen, /* sop_listen */ |
|
207 |
sotpi_connect, /* sop_connect */ |
|
208 |
sotpi_recvmsg, /* sop_recvmsg */ |
|
209 |
sotpi_sendmsg, /* sop_sendmsg */ |
|
210 |
sotpi_getpeername, /* sop_getpeername */ |
|
211 |
sotpi_getsockname, /* sop_getsockname */ |
|
212 |
sotpi_shutdown, /* sop_shutdown */ |
|
213 |
sotpi_getsockopt, /* sop_getsockopt */ |
|
214 |
sotpi_setsockopt /* sop_setsockopt */ |
|
215 |
}; |
|
216 |
||
217 |
/* |
|
218 |
* Common create code for socket and accept. If tso is set the values |
|
219 |
* from that node is used instead of issuing a T_INFO_REQ. |
|
220 |
* |
|
221 |
* Assumes that the caller has a VN_HOLD on accessvp. |
|
222 |
* The VN_RELE will occur either when sotpi_create() fails or when |
|
223 |
* the returned sonode is freed. |
|
224 |
*/ |
|
225 |
struct sonode * |
|
226 |
sotpi_create(vnode_t *accessvp, int domain, int type, int protocol, int version, |
|
227 |
struct sonode *tso, int *errorp) |
|
228 |
{ |
|
229 |
struct sonode *so; |
|
230 |
vnode_t *vp; |
|
231 |
int flags, error; |
|
232 |
||
233 |
ASSERT(accessvp != NULL); |
|
234 |
vp = makesockvp(accessvp, domain, type, protocol); |
|
235 |
ASSERT(vp != NULL); |
|
236 |
so = VTOSO(vp); |
|
237 |
||
238 |
flags = FREAD|FWRITE; |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
239 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
240 |
if ((type == SOCK_STREAM || type == SOCK_DGRAM) && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
241 |
(domain == AF_INET || domain == AF_INET6) && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
242 |
(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP || |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
243 |
protocol == IPPROTO_IP)) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
244 |
/* Tell tcp or udp that it's talking to sockets */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
245 |
flags |= SO_SOCKSTR; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
246 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
247 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
248 |
* Here we indicate to socktpi_open() our attempt to |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
249 |
* make direct calls between sockfs and transport. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
250 |
* The final decision is left to socktpi_open(). |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
251 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
252 |
so->so_state |= SS_DIRECT; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
253 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
254 |
ASSERT(so->so_type != SOCK_DGRAM || tso == NULL); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
255 |
if (so->so_type == SOCK_STREAM && tso != NULL) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
256 |
if (tso->so_state & SS_DIRECT) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
257 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
258 |
* Inherit SS_DIRECT from listener and pass |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
259 |
* SO_ACCEPTOR open flag to tcp, indicating |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
260 |
* that this is an accept fast-path instance. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
261 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
262 |
flags |= SO_ACCEPTOR; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
263 |
} else { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
264 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
265 |
* SS_DIRECT is not set on listener, meaning |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
266 |
* that the listener has been converted from |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
267 |
* a socket to a stream. Ensure that the |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
268 |
* acceptor inherits these settings. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
269 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
270 |
so->so_state &= ~SS_DIRECT; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
271 |
flags &= ~SO_SOCKSTR; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
272 |
} |
0 | 273 |
} |
274 |
} |
|
275 |
||
276 |
/* |
|
277 |
* Tell local transport that it is talking to sockets. |
|
278 |
*/ |
|
279 |
if (so->so_family == AF_UNIX) { |
|
280 |
flags |= SO_SOCKSTR; |
|
281 |
} |
|
282 |
||
283 |
if (error = socktpi_open(&vp, flags, CRED())) { |
|
284 |
VN_RELE(vp); |
|
285 |
*errorp = error; |
|
286 |
return (NULL); |
|
287 |
} |
|
288 |
||
289 |
if (error = so_strinit(so, tso)) { |
|
290 |
(void) VOP_CLOSE(vp, 0, 1, 0, CRED()); |
|
291 |
VN_RELE(vp); |
|
292 |
*errorp = error; |
|
293 |
return (NULL); |
|
294 |
} |
|
295 |
||
296 |
if (version == SOV_DEFAULT) |
|
297 |
version = so_default_version; |
|
298 |
||
299 |
so->so_version = (short)version; |
|
898 | 300 |
|
301 |
/* Initialize the kernel SSL proxy fields */ |
|
302 |
so->so_kssl_type = KSSL_NO_PROXY; |
|
303 |
so->so_kssl_ent = NULL; |
|
304 |
so->so_kssl_ctx = NULL; |
|
305 |
||
0 | 306 |
return (so); |
307 |
} |
|
308 |
||
309 |
/* |
|
310 |
* Bind the socket to an unspecified address in sockfs only. |
|
311 |
* Used for TCP/UDP transports where we know that the O_T_BIND_REQ isn't |
|
312 |
* required in all cases. |
|
313 |
*/ |
|
314 |
static void |
|
315 |
so_automatic_bind(struct sonode *so) |
|
316 |
{ |
|
317 |
ASSERT(so->so_family == AF_INET || so->so_family == AF_INET6); |
|
318 |
||
319 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
320 |
ASSERT(!(so->so_state & SS_ISBOUND)); |
|
321 |
ASSERT(so->so_unbind_mp); |
|
322 |
||
323 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
324 |
bzero(so->so_laddr_sa, so->so_laddr_len); |
|
325 |
so->so_laddr_sa->sa_family = so->so_family; |
|
326 |
so->so_state |= SS_ISBOUND; |
|
327 |
} |
|
328 |
||
329 |
||
330 |
/* |
|
331 |
* bind the socket. |
|
332 |
* |
|
333 |
* If the socket is already bound and none of _SOBIND_SOCKBSD or _SOBIND_XPG4_2 |
|
334 |
* are passed in we allow rebinding. Note that for backwards compatibility |
|
335 |
* even "svr4" sockets pass in _SOBIND_SOCKBSD/SOV_SOCKBSD to sobind/bind. |
|
336 |
* Thus the rebinding code is currently not executed. |
|
337 |
* |
|
338 |
* The constraints for rebinding are: |
|
339 |
* - it is a SOCK_DGRAM, or |
|
340 |
* - it is a SOCK_STREAM/SOCK_SEQPACKET that has not been connected |
|
341 |
* and no listen() has been done. |
|
342 |
* This rebinding code was added based on some language in the XNET book |
|
343 |
* about not returning EINVAL it the protocol allows rebinding. However, |
|
344 |
* this language is not present in the Posix socket draft. Thus maybe the |
|
345 |
* rebinding logic should be deleted from the source. |
|
346 |
* |
|
347 |
* A null "name" can be used to unbind the socket if: |
|
348 |
* - it is a SOCK_DGRAM, or |
|
349 |
* - it is a SOCK_STREAM/SOCK_SEQPACKET that has not been connected |
|
350 |
* and no listen() has been done. |
|
351 |
*/ |
|
352 |
static int |
|
353 |
sotpi_bindlisten(struct sonode *so, struct sockaddr *name, |
|
354 |
socklen_t namelen, int backlog, int flags) |
|
355 |
{ |
|
356 |
struct T_bind_req bind_req; |
|
357 |
struct T_bind_ack *bind_ack; |
|
358 |
int error = 0; |
|
359 |
mblk_t *mp; |
|
360 |
void *addr; |
|
361 |
t_uscalar_t addrlen; |
|
362 |
int unbind_on_err = 1; |
|
363 |
boolean_t clear_acceptconn_on_err = B_FALSE; |
|
364 |
boolean_t restore_backlog_on_err = B_FALSE; |
|
365 |
int save_so_backlog; |
|
366 |
t_scalar_t PRIM_type = O_T_BIND_REQ; |
|
367 |
boolean_t tcp_udp_xport; |
|
368 |
void *nl7c = NULL; |
|
369 |
||
370 |
dprintso(so, 1, ("sotpi_bindlisten(%p, %p, %d, %d, 0x%x) %s\n", |
|
371 |
so, name, namelen, backlog, flags, |
|
372 |
pr_state(so->so_state, so->so_mode))); |
|
373 |
||
374 |
tcp_udp_xport = so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM; |
|
375 |
||
376 |
if (!(flags & _SOBIND_LOCK_HELD)) { |
|
377 |
mutex_enter(&so->so_lock); |
|
378 |
so_lock_single(so); /* Set SOLOCKED */ |
|
379 |
} else { |
|
380 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
381 |
ASSERT(so->so_flag & SOLOCKED); |
|
382 |
} |
|
383 |
||
384 |
/* |
|
385 |
* Make sure that there is a preallocated unbind_req message |
|
386 |
* before binding. This message allocated when the socket is |
|
387 |
* created but it might be have been consumed. |
|
388 |
*/ |
|
389 |
if (so->so_unbind_mp == NULL) { |
|
390 |
dprintso(so, 1, ("sobind: allocating unbind_req\n")); |
|
391 |
/* NOTE: holding so_lock while sleeping */ |
|
392 |
so->so_unbind_mp = |
|
393 |
soallocproto(sizeof (struct T_unbind_req), _ALLOC_SLEEP); |
|
394 |
} |
|
395 |
||
396 |
if (flags & _SOBIND_REBIND) { |
|
397 |
/* |
|
398 |
* Called from solisten after doing an sotpi_unbind() or |
|
399 |
* potentially without the unbind (latter for AF_INET{,6}). |
|
400 |
*/ |
|
401 |
ASSERT(name == NULL && namelen == 0); |
|
402 |
||
403 |
if (so->so_family == AF_UNIX) { |
|
404 |
ASSERT(so->so_ux_bound_vp); |
|
405 |
addr = &so->so_ux_laddr; |
|
406 |
addrlen = (t_uscalar_t)sizeof (so->so_ux_laddr); |
|
407 |
dprintso(so, 1, |
|
408 |
("sobind rebind UNIX: addrlen %d, addr 0x%p, vp %p\n", |
|
409 |
addrlen, |
|
410 |
((struct so_ux_addr *)addr)->soua_vp, |
|
411 |
so->so_ux_bound_vp)); |
|
412 |
} else { |
|
413 |
addr = so->so_laddr_sa; |
|
414 |
addrlen = (t_uscalar_t)so->so_laddr_len; |
|
415 |
} |
|
416 |
} else if (flags & _SOBIND_UNSPEC) { |
|
417 |
ASSERT(name == NULL && namelen == 0); |
|
418 |
||
419 |
/* |
|
420 |
* The caller checked SS_ISBOUND but not necessarily |
|
421 |
* under so_lock |
|
422 |
*/ |
|
423 |
if (so->so_state & SS_ISBOUND) { |
|
424 |
/* No error */ |
|
425 |
goto done; |
|
426 |
} |
|
427 |
||
428 |
/* Set an initial local address */ |
|
429 |
switch (so->so_family) { |
|
430 |
case AF_UNIX: |
|
431 |
/* |
|
432 |
* Use an address with same size as struct sockaddr |
|
433 |
* just like BSD. |
|
434 |
*/ |
|
435 |
so->so_laddr_len = |
|
436 |
(socklen_t)sizeof (struct sockaddr); |
|
437 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
438 |
bzero(so->so_laddr_sa, so->so_laddr_len); |
|
439 |
so->so_laddr_sa->sa_family = so->so_family; |
|
440 |
||
441 |
/* |
|
442 |
* Pass down an address with the implicit bind |
|
443 |
* magic number and the rest all zeros. |
|
444 |
* The transport will return a unique address. |
|
445 |
*/ |
|
446 |
so->so_ux_laddr.soua_vp = NULL; |
|
447 |
so->so_ux_laddr.soua_magic = SOU_MAGIC_IMPLICIT; |
|
448 |
addr = &so->so_ux_laddr; |
|
449 |
addrlen = (t_uscalar_t)sizeof (so->so_ux_laddr); |
|
450 |
break; |
|
451 |
||
452 |
case AF_INET: |
|
453 |
case AF_INET6: |
|
454 |
/* |
|
455 |
* An unspecified bind in TPI has a NULL address. |
|
456 |
* Set the address in sockfs to have the sa_family. |
|
457 |
*/ |
|
458 |
so->so_laddr_len = (so->so_family == AF_INET) ? |
|
459 |
(socklen_t)sizeof (sin_t) : |
|
460 |
(socklen_t)sizeof (sin6_t); |
|
461 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
462 |
bzero(so->so_laddr_sa, so->so_laddr_len); |
|
463 |
so->so_laddr_sa->sa_family = so->so_family; |
|
464 |
addr = NULL; |
|
465 |
addrlen = 0; |
|
466 |
break; |
|
467 |
||
468 |
default: |
|
469 |
/* |
|
470 |
* An unspecified bind in TPI has a NULL address. |
|
471 |
* Set the address in sockfs to be zero length. |
|
472 |
* |
|
473 |
* Can not assume there is a sa_family for all |
|
474 |
* protocol families. For example, AF_X25 does not |
|
475 |
* have a family field. |
|
476 |
*/ |
|
477 |
so->so_laddr_len = 0; /* XXX correct? */ |
|
478 |
bzero(so->so_laddr_sa, so->so_laddr_len); |
|
479 |
addr = NULL; |
|
480 |
addrlen = 0; |
|
481 |
break; |
|
482 |
} |
|
483 |
||
484 |
} else { |
|
485 |
if (so->so_state & SS_ISBOUND) { |
|
486 |
/* |
|
487 |
* If it is ok to rebind the socket, first unbind |
|
488 |
* with the transport. A rebind to the NULL address |
|
489 |
* is interpreted as an unbind. |
|
490 |
* Note that a bind to NULL in BSD does unbind the |
|
491 |
* socket but it fails with EINVAL. |
|
492 |
* Note that regular sockets set SOV_SOCKBSD i.e. |
|
493 |
* _SOBIND_SOCKBSD gets set here hence no type of |
|
494 |
* socket does currently allow rebinding. |
|
495 |
* |
|
496 |
* If the name is NULL just do an unbind. |
|
497 |
*/ |
|
498 |
if (flags & (_SOBIND_SOCKBSD|_SOBIND_XPG4_2) && |
|
499 |
name != NULL) { |
|
500 |
error = EINVAL; |
|
501 |
unbind_on_err = 0; |
|
502 |
eprintsoline(so, error); |
|
503 |
goto done; |
|
504 |
} |
|
505 |
if ((so->so_mode & SM_CONNREQUIRED) && |
|
506 |
(so->so_state & SS_CANTREBIND)) { |
|
507 |
error = EINVAL; |
|
508 |
unbind_on_err = 0; |
|
509 |
eprintsoline(so, error); |
|
510 |
goto done; |
|
511 |
} |
|
512 |
error = sotpi_unbind(so, 0); |
|
513 |
if (error) { |
|
514 |
eprintsoline(so, error); |
|
515 |
goto done; |
|
516 |
} |
|
517 |
ASSERT(!(so->so_state & SS_ISBOUND)); |
|
518 |
if (name == NULL) { |
|
519 |
so->so_state &= |
|
520 |
~(SS_ISCONNECTED|SS_ISCONNECTING); |
|
521 |
goto done; |
|
522 |
} |
|
523 |
} |
|
524 |
/* X/Open requires this check */ |
|
525 |
if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) { |
|
526 |
if (xnet_check_print) { |
|
527 |
printf("sockfs: X/Open bind state check " |
|
528 |
"caused EINVAL\n"); |
|
529 |
} |
|
530 |
error = EINVAL; |
|
531 |
goto done; |
|
532 |
} |
|
533 |
||
534 |
switch (so->so_family) { |
|
535 |
case AF_UNIX: |
|
536 |
/* |
|
537 |
* All AF_UNIX addresses are nul terminated |
|
538 |
* when copied (copyin_name) in so the minimum |
|
539 |
* length is 3 bytes. |
|
540 |
*/ |
|
541 |
if (name == NULL || |
|
542 |
(ssize_t)namelen <= sizeof (short) + 1) { |
|
543 |
error = EISDIR; |
|
544 |
eprintsoline(so, error); |
|
545 |
goto done; |
|
546 |
} |
|
547 |
/* |
|
548 |
* Verify so_family matches the bound family. |
|
549 |
* BSD does not check this for AF_UNIX resulting |
|
550 |
* in funny mknods. |
|
551 |
*/ |
|
552 |
if (name->sa_family != so->so_family) { |
|
553 |
error = EAFNOSUPPORT; |
|
554 |
goto done; |
|
555 |
} |
|
556 |
break; |
|
557 |
case AF_INET: |
|
558 |
if (name == NULL) { |
|
559 |
error = EINVAL; |
|
560 |
eprintsoline(so, error); |
|
561 |
goto done; |
|
562 |
} |
|
563 |
if ((size_t)namelen != sizeof (sin_t)) { |
|
564 |
error = name->sa_family != so->so_family ? |
|
565 |
EAFNOSUPPORT : EINVAL; |
|
566 |
eprintsoline(so, error); |
|
567 |
goto done; |
|
568 |
} |
|
569 |
if ((flags & _SOBIND_XPG4_2) && |
|
570 |
(name->sa_family != so->so_family)) { |
|
571 |
/* |
|
572 |
* This check has to be made for X/Open |
|
573 |
* sockets however application failures have |
|
574 |
* been observed when it is applied to |
|
575 |
* all sockets. |
|
576 |
*/ |
|
577 |
error = EAFNOSUPPORT; |
|
578 |
eprintsoline(so, error); |
|
579 |
goto done; |
|
580 |
} |
|
581 |
/* |
|
582 |
* Force a zero sa_family to match so_family. |
|
583 |
* |
|
584 |
* Some programs like inetd(1M) don't set the |
|
585 |
* family field. Other programs leave |
|
586 |
* sin_family set to garbage - SunOS 4.X does |
|
587 |
* not check the family field on a bind. |
|
588 |
* We use the family field that |
|
589 |
* was passed in to the socket() call. |
|
590 |
*/ |
|
591 |
name->sa_family = so->so_family; |
|
592 |
break; |
|
593 |
||
594 |
case AF_INET6: { |
|
595 |
#ifdef DEBUG |
|
596 |
sin6_t *sin6 = (sin6_t *)name; |
|
597 |
#endif /* DEBUG */ |
|
598 |
||
599 |
if (name == NULL) { |
|
600 |
error = EINVAL; |
|
601 |
eprintsoline(so, error); |
|
602 |
goto done; |
|
603 |
} |
|
604 |
if ((size_t)namelen != sizeof (sin6_t)) { |
|
605 |
error = name->sa_family != so->so_family ? |
|
606 |
EAFNOSUPPORT : EINVAL; |
|
607 |
eprintsoline(so, error); |
|
608 |
goto done; |
|
609 |
} |
|
610 |
if (name->sa_family != so->so_family) { |
|
611 |
/* |
|
612 |
* With IPv6 we require the family to match |
|
613 |
* unlike in IPv4. |
|
614 |
*/ |
|
615 |
error = EAFNOSUPPORT; |
|
616 |
eprintsoline(so, error); |
|
617 |
goto done; |
|
618 |
} |
|
619 |
#ifdef DEBUG |
|
620 |
/* |
|
621 |
* Verify that apps don't forget to clear |
|
622 |
* sin6_scope_id etc |
|
623 |
*/ |
|
624 |
if (sin6->sin6_scope_id != 0 && |
|
625 |
!IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) { |
|
626 |
cmn_err(CE_WARN, |
|
627 |
"bind with uninitialized sin6_scope_id " |
|
628 |
"(%d) on socket. Pid = %d\n", |
|
629 |
(int)sin6->sin6_scope_id, |
|
630 |
(int)curproc->p_pid); |
|
631 |
} |
|
632 |
if (sin6->__sin6_src_id != 0) { |
|
633 |
cmn_err(CE_WARN, |
|
634 |
"bind with uninitialized __sin6_src_id " |
|
635 |
"(%d) on socket. Pid = %d\n", |
|
636 |
(int)sin6->__sin6_src_id, |
|
637 |
(int)curproc->p_pid); |
|
638 |
} |
|
639 |
#endif /* DEBUG */ |
|
640 |
break; |
|
641 |
} |
|
642 |
default: |
|
643 |
/* |
|
644 |
* Don't do any length or sa_family check to allow |
|
645 |
* non-sockaddr style addresses. |
|
646 |
*/ |
|
647 |
if (name == NULL) { |
|
648 |
error = EINVAL; |
|
649 |
eprintsoline(so, error); |
|
650 |
goto done; |
|
651 |
} |
|
652 |
break; |
|
653 |
} |
|
654 |
||
655 |
if (namelen > (t_uscalar_t)so->so_laddr_maxlen) { |
|
656 |
error = ENAMETOOLONG; |
|
657 |
eprintsoline(so, error); |
|
658 |
goto done; |
|
659 |
} |
|
660 |
/* |
|
661 |
* Save local address. |
|
662 |
*/ |
|
663 |
so->so_laddr_len = (socklen_t)namelen; |
|
664 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
665 |
bcopy(name, so->so_laddr_sa, namelen); |
|
666 |
||
667 |
addr = so->so_laddr_sa; |
|
668 |
addrlen = (t_uscalar_t)so->so_laddr_len; |
|
669 |
switch (so->so_family) { |
|
670 |
case AF_INET6: |
|
671 |
case AF_INET: |
|
672 |
break; |
|
673 |
case AF_UNIX: { |
|
674 |
struct sockaddr_un *soun = |
|
675 |
(struct sockaddr_un *)so->so_laddr_sa; |
|
676 |
struct vnode *vp; |
|
677 |
struct vattr vattr; |
|
678 |
||
679 |
ASSERT(so->so_ux_bound_vp == NULL); |
|
680 |
/* |
|
681 |
* Create vnode for the specified path name. |
|
682 |
* Keep vnode held with a reference in so_ux_bound_vp. |
|
683 |
* Use the vnode pointer as the address used in the |
|
684 |
* bind with the transport. |
|
685 |
* |
|
686 |
* Use the same mode as in BSD. In particular this does |
|
687 |
* not observe the umask. |
|
688 |
*/ |
|
689 |
/* MAXPATHLEN + soun_family + nul termination */ |
|
690 |
if (so->so_laddr_len > |
|
691 |
(socklen_t)(MAXPATHLEN + sizeof (short) + 1)) { |
|
692 |
error = ENAMETOOLONG; |
|
693 |
eprintsoline(so, error); |
|
694 |
goto done; |
|
695 |
} |
|
696 |
vattr.va_type = VSOCK; |
|
697 |
vattr.va_mode = 0777 & ~u.u_cmask; |
|
698 |
vattr.va_mask = AT_TYPE|AT_MODE; |
|
699 |
/* NOTE: holding so_lock */ |
|
700 |
error = vn_create(soun->sun_path, UIO_SYSSPACE, &vattr, |
|
701 |
EXCL, 0, &vp, CRMKNOD, 0, 0); |
|
702 |
if (error) { |
|
703 |
if (error == EEXIST) |
|
704 |
error = EADDRINUSE; |
|
705 |
eprintsoline(so, error); |
|
706 |
goto done; |
|
707 |
} |
|
708 |
/* |
|
709 |
* Establish pointer from the underlying filesystem |
|
710 |
* vnode to the socket node. |
|
711 |
* so_ux_bound_vp and v_stream->sd_vnode form the |
|
712 |
* cross-linkage between the underlying filesystem |
|
713 |
* node and the socket node. |
|
714 |
*/ |
|
715 |
ASSERT(SOTOV(so)->v_stream); |
|
716 |
mutex_enter(&vp->v_lock); |
|
717 |
vp->v_stream = SOTOV(so)->v_stream; |
|
718 |
so->so_ux_bound_vp = vp; |
|
719 |
mutex_exit(&vp->v_lock); |
|
720 |
||
721 |
/* |
|
722 |
* Use the vnode pointer value as a unique address |
|
723 |
* (together with the magic number to avoid conflicts |
|
724 |
* with implicit binds) in the transport provider. |
|
725 |
*/ |
|
726 |
so->so_ux_laddr.soua_vp = (void *)so->so_ux_bound_vp; |
|
727 |
so->so_ux_laddr.soua_magic = SOU_MAGIC_EXPLICIT; |
|
728 |
addr = &so->so_ux_laddr; |
|
729 |
addrlen = (t_uscalar_t)sizeof (so->so_ux_laddr); |
|
730 |
dprintso(so, 1, ("sobind UNIX: addrlen %d, addr %p\n", |
|
731 |
addrlen, |
|
732 |
((struct so_ux_addr *)addr)->soua_vp)); |
|
733 |
break; |
|
734 |
} |
|
735 |
} /* end switch (so->so_family) */ |
|
736 |
} |
|
737 |
||
738 |
/* |
|
739 |
* set SS_ACCEPTCONN before sending down O_T_BIND_REQ since |
|
740 |
* the transport can start passing up T_CONN_IND messages |
|
741 |
* as soon as it receives the bind req and strsock_proto() |
|
742 |
* insists that SS_ACCEPTCONN is set when processing T_CONN_INDs. |
|
743 |
*/ |
|
744 |
if (flags & _SOBIND_LISTEN) { |
|
745 |
if ((so->so_state & SS_ACCEPTCONN) == 0) |
|
746 |
clear_acceptconn_on_err = B_TRUE; |
|
747 |
save_so_backlog = so->so_backlog; |
|
748 |
restore_backlog_on_err = B_TRUE; |
|
749 |
so->so_state |= SS_ACCEPTCONN; |
|
750 |
so->so_backlog = backlog; |
|
751 |
} |
|
752 |
||
753 |
/* |
|
754 |
* If NL7C addr(s) have been configured check for addr/port match, |
|
755 |
* or if an implicit NL7C socket via AF_NCA mark socket as NL7C. |
|
756 |
* |
|
757 |
* NL7C supports the TCP transport only so check AF_INET and AF_INET6 |
|
758 |
* family sockets only. If match mark as such. |
|
759 |
*/ |
|
760 |
if ((nl7c_enabled && addr != NULL && |
|
761 |
(so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
762 |
(nl7c = nl7c_lookup_addr(addr, addrlen))) || |
|
763 |
so->so_nl7c_flags == NL7C_AF_NCA) { |
|
764 |
/* |
|
765 |
* NL7C is not supported in non-global zones, |
|
766 |
* we enforce this restriction here. |
|
767 |
*/ |
|
768 |
if (so->so_zoneid == GLOBAL_ZONEID) { |
|
769 |
/* An NL7C socket, mark it */ |
|
770 |
so->so_nl7c_flags |= NL7C_ENABLED; |
|
771 |
} else |
|
772 |
nl7c = NULL; |
|
773 |
} |
|
774 |
/* |
|
775 |
* We send a T_BIND_REQ for TCP/UDP since we know it supports it, |
|
776 |
* for other transports we will send in a O_T_BIND_REQ. |
|
777 |
*/ |
|
778 |
if (tcp_udp_xport && |
|
779 |
(so->so_family == AF_INET || so->so_family == AF_INET6)) |
|
780 |
PRIM_type = T_BIND_REQ; |
|
781 |
||
782 |
bind_req.PRIM_type = PRIM_type; |
|
783 |
bind_req.ADDR_length = addrlen; |
|
784 |
bind_req.ADDR_offset = (t_scalar_t)sizeof (bind_req); |
|
785 |
bind_req.CONIND_number = backlog; |
|
786 |
/* NOTE: holding so_lock while sleeping */ |
|
787 |
mp = soallocproto2(&bind_req, sizeof (bind_req), |
|
788 |
addr, addrlen, 0, _ALLOC_SLEEP); |
|
789 |
so->so_state &= ~SS_LADDR_VALID; |
|
898 | 790 |
|
0 | 791 |
/* Done using so_laddr_sa - can drop the lock */ |
792 |
mutex_exit(&so->so_lock); |
|
793 |
||
898 | 794 |
/* |
795 |
* Intercept the bind_req message here to check if this <address/port> |
|
796 |
* was configured as an SSL proxy server, or if another endpoint was |
|
797 |
* already configured to act as a proxy for us. |
|
798 |
*/ |
|
799 |
if ((so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
800 |
so->so_type == SOCK_STREAM) { |
|
801 |
||
802 |
if (so->so_kssl_ent != NULL) { |
|
803 |
kssl_release_ent(so->so_kssl_ent, so, so->so_kssl_type); |
|
804 |
so->so_kssl_ent = NULL; |
|
805 |
} |
|
806 |
||
807 |
so->so_kssl_type = kssl_check_proxy(mp, so, &so->so_kssl_ent); |
|
808 |
switch (so->so_kssl_type) { |
|
809 |
case KSSL_NO_PROXY: |
|
810 |
break; |
|
811 |
||
812 |
case KSSL_HAS_PROXY: |
|
813 |
mutex_enter(&so->so_lock); |
|
814 |
goto skip_transport; |
|
815 |
||
816 |
case KSSL_IS_PROXY: |
|
817 |
break; |
|
818 |
} |
|
819 |
} |
|
820 |
||
0 | 821 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
822 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
823 |
if (error) { |
|
824 |
eprintsoline(so, error); |
|
825 |
mutex_enter(&so->so_lock); |
|
826 |
goto done; |
|
827 |
} |
|
828 |
||
829 |
mutex_enter(&so->so_lock); |
|
830 |
error = sowaitprim(so, PRIM_type, T_BIND_ACK, |
|
831 |
(t_uscalar_t)sizeof (*bind_ack), &mp, 0); |
|
832 |
if (error) { |
|
833 |
eprintsoline(so, error); |
|
834 |
goto done; |
|
835 |
} |
|
898 | 836 |
skip_transport: |
0 | 837 |
ASSERT(mp); |
838 |
/* |
|
839 |
* Even if some TPI message (e.g. T_DISCON_IND) was received in |
|
840 |
* strsock_proto while the lock was dropped above, the bind |
|
841 |
* is allowed to complete. |
|
842 |
*/ |
|
843 |
||
844 |
/* Mark as bound. This will be undone if we detect errors below. */ |
|
845 |
if (flags & _SOBIND_NOXLATE) { |
|
846 |
ASSERT(so->so_family == AF_UNIX); |
|
847 |
so->so_state |= SS_FADDR_NOXLATE; |
|
848 |
} |
|
849 |
ASSERT(!(so->so_state & SS_ISBOUND) || (flags & _SOBIND_REBIND)); |
|
850 |
so->so_state |= SS_ISBOUND; |
|
851 |
ASSERT(so->so_unbind_mp); |
|
852 |
||
853 |
/* note that we've already set SS_ACCEPTCONN above */ |
|
854 |
||
855 |
/* |
|
856 |
* Recompute addrlen - an unspecied bind sent down an |
|
857 |
* address of length zero but we expect the appropriate length |
|
858 |
* in return. |
|
859 |
*/ |
|
860 |
addrlen = (t_uscalar_t)(so->so_family == AF_UNIX ? |
|
861 |
sizeof (so->so_ux_laddr) : so->so_laddr_len); |
|
862 |
||
863 |
bind_ack = (struct T_bind_ack *)mp->b_rptr; |
|
864 |
/* |
|
865 |
* The alignment restriction is really too strict but |
|
866 |
* we want enough alignment to inspect the fields of |
|
867 |
* a sockaddr_in. |
|
868 |
*/ |
|
869 |
addr = sogetoff(mp, bind_ack->ADDR_offset, |
|
870 |
bind_ack->ADDR_length, |
|
871 |
__TPI_ALIGN_SIZE); |
|
872 |
if (addr == NULL) { |
|
873 |
freemsg(mp); |
|
874 |
error = EPROTO; |
|
875 |
eprintsoline(so, error); |
|
876 |
goto done; |
|
877 |
} |
|
878 |
if (!(flags & _SOBIND_UNSPEC)) { |
|
879 |
/* |
|
880 |
* Verify that the transport didn't return something we |
|
881 |
* did not want e.g. an address other than what we asked for. |
|
882 |
* |
|
883 |
* NOTE: These checks would go away if/when we switch to |
|
884 |
* using the new TPI (in which the transport would fail |
|
885 |
* the request instead of assigning a different address). |
|
886 |
* |
|
887 |
* NOTE2: For protocols that we don't know (i.e. any |
|
888 |
* other than AF_INET6, AF_INET and AF_UNIX), we |
|
889 |
* cannot know if the transport should be expected to |
|
890 |
* return the same address as that requested. |
|
891 |
* |
|
892 |
* NOTE3: For AF_INET and AF_INET6, TCP/UDP, we send |
|
893 |
* down a T_BIND_REQ. We use O_T_BIND_REQ for others. |
|
894 |
* |
|
895 |
* For example, in the case of netatalk it may be |
|
896 |
* inappropriate for the transport to return the |
|
897 |
* requested address (as it may have allocated a local |
|
898 |
* port number in behaviour similar to that of an |
|
899 |
* AF_INET bind request with a port number of zero). |
|
900 |
* |
|
901 |
* Given the definition of O_T_BIND_REQ, where the |
|
902 |
* transport may bind to an address other than the |
|
903 |
* requested address, it's not possible to determine |
|
904 |
* whether a returned address that differs from the |
|
905 |
* requested address is a reason to fail (because the |
|
906 |
* requested address was not available) or succeed |
|
907 |
* (because the transport allocated an appropriate |
|
908 |
* address and/or port). |
|
909 |
* |
|
910 |
* sockfs currently requires that the transport return |
|
911 |
* the requested address in the T_BIND_ACK, unless |
|
912 |
* there is code here to allow for any discrepancy. |
|
913 |
* Such code exists for AF_INET and AF_INET6. |
|
914 |
* |
|
915 |
* Netatalk chooses to return the requested address |
|
916 |
* rather than the (correct) allocated address. This |
|
917 |
* means that netatalk violates the TPI specification |
|
918 |
* (and would not function correctly if used from a |
|
919 |
* TLI application), but it does mean that it works |
|
920 |
* with sockfs. |
|
921 |
* |
|
922 |
* As noted above, using the newer XTI bind primitive |
|
923 |
* (T_BIND_REQ) in preference to O_T_BIND_REQ would |
|
924 |
* allow sockfs to be more sure about whether or not |
|
925 |
* the bind request had succeeded (as transports are |
|
926 |
* not permitted to bind to a different address than |
|
927 |
* that requested - they must return failure). |
|
928 |
* Unfortunately, support for T_BIND_REQ may not be |
|
929 |
* present in all transport implementations (netatalk, |
|
930 |
* for example, doesn't have it), making the |
|
931 |
* transition difficult. |
|
932 |
*/ |
|
933 |
if (bind_ack->ADDR_length != addrlen) { |
|
934 |
/* Assumes that the requested address was in use */ |
|
935 |
freemsg(mp); |
|
936 |
error = EADDRINUSE; |
|
937 |
eprintsoline(so, error); |
|
938 |
goto done; |
|
939 |
} |
|
940 |
||
941 |
switch (so->so_family) { |
|
942 |
case AF_INET6: |
|
943 |
case AF_INET: { |
|
944 |
sin_t *rname, *aname; |
|
945 |
||
946 |
rname = (sin_t *)addr; |
|
947 |
aname = (sin_t *)so->so_laddr_sa; |
|
948 |
||
949 |
/* |
|
950 |
* Take advantage of the alignment |
|
951 |
* of sin_port and sin6_port which fall |
|
952 |
* in the same place in their data structures. |
|
953 |
* Just use sin_port for either address family. |
|
954 |
* |
|
955 |
* This may become a problem if (heaven forbid) |
|
956 |
* there's a separate ipv6port_reserved... :-P |
|
957 |
* |
|
958 |
* Binding to port 0 has the semantics of letting |
|
959 |
* the transport bind to any port. |
|
960 |
* |
|
961 |
* If the transport is TCP or UDP since we had sent |
|
962 |
* a T_BIND_REQ we would not get a port other than |
|
963 |
* what we asked for. |
|
964 |
*/ |
|
965 |
if (tcp_udp_xport) { |
|
966 |
/* |
|
967 |
* Pick up the new port number if we bound to |
|
968 |
* port 0. |
|
969 |
*/ |
|
970 |
if (aname->sin_port == 0) |
|
971 |
aname->sin_port = rname->sin_port; |
|
972 |
so->so_state |= SS_LADDR_VALID; |
|
973 |
break; |
|
974 |
} |
|
975 |
if (aname->sin_port != 0 && |
|
976 |
aname->sin_port != rname->sin_port) { |
|
977 |
freemsg(mp); |
|
978 |
error = EADDRINUSE; |
|
979 |
eprintsoline(so, error); |
|
980 |
goto done; |
|
981 |
} |
|
982 |
/* |
|
983 |
* Pick up the new port number if we bound to port 0. |
|
984 |
*/ |
|
985 |
aname->sin_port = rname->sin_port; |
|
986 |
||
987 |
/* |
|
988 |
* Unfortunately, addresses aren't _quite_ the same. |
|
989 |
*/ |
|
990 |
if (so->so_family == AF_INET) { |
|
991 |
if (aname->sin_addr.s_addr != |
|
992 |
rname->sin_addr.s_addr) { |
|
993 |
freemsg(mp); |
|
994 |
error = EADDRNOTAVAIL; |
|
995 |
eprintsoline(so, error); |
|
996 |
goto done; |
|
997 |
} |
|
998 |
} else { |
|
999 |
sin6_t *rname6 = (sin6_t *)rname; |
|
1000 |
sin6_t *aname6 = (sin6_t *)aname; |
|
1001 |
||
1002 |
if (!IN6_ARE_ADDR_EQUAL(&aname6->sin6_addr, |
|
1003 |
&rname6->sin6_addr)) { |
|
1004 |
freemsg(mp); |
|
1005 |
error = EADDRNOTAVAIL; |
|
1006 |
eprintsoline(so, error); |
|
1007 |
goto done; |
|
1008 |
} |
|
1009 |
} |
|
1010 |
break; |
|
1011 |
} |
|
1012 |
case AF_UNIX: |
|
1013 |
if (bcmp(addr, &so->so_ux_laddr, addrlen) != 0) { |
|
1014 |
freemsg(mp); |
|
1015 |
error = EADDRINUSE; |
|
1016 |
eprintsoline(so, error); |
|
1017 |
eprintso(so, |
|
1018 |
("addrlen %d, addr 0x%x, vp %p\n", |
|
1019 |
addrlen, *((int *)addr), |
|
1020 |
so->so_ux_bound_vp)); |
|
1021 |
goto done; |
|
1022 |
} |
|
1023 |
so->so_state |= SS_LADDR_VALID; |
|
1024 |
break; |
|
1025 |
default: |
|
1026 |
/* |
|
1027 |
* NOTE: This assumes that addresses can be |
|
1028 |
* byte-compared for equivalence. |
|
1029 |
*/ |
|
1030 |
if (bcmp(addr, so->so_laddr_sa, addrlen) != 0) { |
|
1031 |
freemsg(mp); |
|
1032 |
error = EADDRINUSE; |
|
1033 |
eprintsoline(so, error); |
|
1034 |
goto done; |
|
1035 |
} |
|
1036 |
/* |
|
1037 |
* Don't mark SS_LADDR_VALID, as we cannot be |
|
1038 |
* sure that the returned address is the real |
|
1039 |
* bound address when talking to an unknown |
|
1040 |
* transport. |
|
1041 |
*/ |
|
1042 |
break; |
|
1043 |
} |
|
1044 |
} else { |
|
1045 |
/* |
|
1046 |
* Save for returned address for getsockname. |
|
1047 |
* Needed for unspecific bind unless transport supports |
|
1048 |
* the TI_GETMYNAME ioctl. |
|
1049 |
* Do this for AF_INET{,6} even though they do, as |
|
1050 |
* caching info here is much better performance than |
|
1051 |
* a TPI/STREAMS trip to the transport for getsockname. |
|
1052 |
* Any which can't for some reason _must_ _not_ set |
|
1053 |
* LADDR_VALID here for the caching version of getsockname |
|
1054 |
* to not break; |
|
1055 |
*/ |
|
1056 |
switch (so->so_family) { |
|
1057 |
case AF_UNIX: |
|
1058 |
/* |
|
1059 |
* Record the address bound with the transport |
|
1060 |
* for use by socketpair. |
|
1061 |
*/ |
|
1062 |
bcopy(addr, &so->so_ux_laddr, addrlen); |
|
1063 |
so->so_state |= SS_LADDR_VALID; |
|
1064 |
break; |
|
1065 |
case AF_INET: |
|
1066 |
case AF_INET6: |
|
1067 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
1068 |
bcopy(addr, so->so_laddr_sa, so->so_laddr_len); |
|
1069 |
so->so_state |= SS_LADDR_VALID; |
|
1070 |
break; |
|
1071 |
default: |
|
1072 |
/* |
|
1073 |
* Don't mark SS_LADDR_VALID, as we cannot be |
|
1074 |
* sure that the returned address is the real |
|
1075 |
* bound address when talking to an unknown |
|
1076 |
* transport. |
|
1077 |
*/ |
|
1078 |
break; |
|
1079 |
} |
|
1080 |
} |
|
1081 |
||
1082 |
if (nl7c == NULL && (so->so_nl7c_flags & NL7C_AF_NCA) && |
|
1083 |
(so->so_nl7c_flags & NL7C_ENABLED)) { |
|
1084 |
/* |
|
1085 |
* Was an AF_NCA bind() so add it to the addr list for |
|
1086 |
* reporting purposes. |
|
1087 |
*/ |
|
1088 |
nl7c = nl7c_add_addr(addr, addrlen); |
|
1089 |
} |
|
1090 |
if (nl7c != NULL) { |
|
1091 |
nl7c_listener_addr(nl7c, strvp2wq(SOTOV(so))); |
|
1092 |
} |
|
1093 |
||
1094 |
freemsg(mp); |
|
1095 |
||
1096 |
done: |
|
1097 |
if (error) { |
|
1098 |
/* reset state & backlog to values held on entry */ |
|
1099 |
if (clear_acceptconn_on_err == B_TRUE) |
|
1100 |
so->so_state &= ~SS_ACCEPTCONN; |
|
1101 |
if (restore_backlog_on_err == B_TRUE) |
|
1102 |
so->so_backlog = save_so_backlog; |
|
1103 |
||
1104 |
if (unbind_on_err && so->so_state & SS_ISBOUND) { |
|
1105 |
int err; |
|
1106 |
||
1107 |
err = sotpi_unbind(so, 0); |
|
1108 |
/* LINTED - statement has no consequent: if */ |
|
1109 |
if (err) { |
|
1110 |
eprintsoline(so, error); |
|
1111 |
} else { |
|
1112 |
ASSERT(!(so->so_state & SS_ISBOUND)); |
|
1113 |
} |
|
1114 |
} |
|
1115 |
} |
|
1116 |
if (!(flags & _SOBIND_LOCK_HELD)) { |
|
1117 |
so_unlock_single(so, SOLOCKED); |
|
1118 |
mutex_exit(&so->so_lock); |
|
1119 |
} else { |
|
1120 |
/* If the caller held the lock don't release it here */ |
|
1121 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
1122 |
ASSERT(so->so_flag & SOLOCKED); |
|
1123 |
} |
|
1124 |
return (error); |
|
1125 |
} |
|
1126 |
||
1127 |
/* bind the socket */ |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1128 |
static int |
0 | 1129 |
sotpi_bind(struct sonode *so, struct sockaddr *name, socklen_t namelen, |
1130 |
int flags) |
|
1131 |
{ |
|
1132 |
if ((flags & _SOBIND_SOCKETPAIR) == 0) |
|
1133 |
return (sotpi_bindlisten(so, name, namelen, 0, flags)); |
|
1134 |
||
1135 |
flags &= ~_SOBIND_SOCKETPAIR; |
|
1136 |
return (sotpi_bindlisten(so, name, namelen, 1, flags)); |
|
1137 |
} |
|
1138 |
||
1139 |
/* |
|
1140 |
* Unbind a socket - used when bind() fails, when bind() specifies a NULL |
|
1141 |
* address, or when listen needs to unbind and bind. |
|
1142 |
* If the _SOUNBIND_REBIND flag is specified the addresses are retained |
|
1143 |
* so that a sobind can pick them up. |
|
1144 |
*/ |
|
1145 |
static int |
|
1146 |
sotpi_unbind(struct sonode *so, int flags) |
|
1147 |
{ |
|
1148 |
struct T_unbind_req unbind_req; |
|
1149 |
int error = 0; |
|
1150 |
mblk_t *mp; |
|
1151 |
||
1152 |
dprintso(so, 1, ("sotpi_unbind(%p, 0x%x) %s\n", |
|
1153 |
so, flags, pr_state(so->so_state, so->so_mode))); |
|
1154 |
||
1155 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
1156 |
ASSERT(so->so_flag & SOLOCKED); |
|
1157 |
||
1158 |
if (!(so->so_state & SS_ISBOUND)) { |
|
1159 |
error = EINVAL; |
|
1160 |
eprintsoline(so, error); |
|
1161 |
goto done; |
|
1162 |
} |
|
1163 |
||
1164 |
mutex_exit(&so->so_lock); |
|
1165 |
||
1166 |
/* |
|
1167 |
* Flush the read and write side (except stream head read queue) |
|
1168 |
* and send down T_UNBIND_REQ. |
|
1169 |
*/ |
|
1170 |
(void) putnextctl1(strvp2wq(SOTOV(so)), M_FLUSH, FLUSHRW); |
|
1171 |
||
1172 |
unbind_req.PRIM_type = T_UNBIND_REQ; |
|
1173 |
mp = soallocproto1(&unbind_req, sizeof (unbind_req), |
|
1174 |
0, _ALLOC_SLEEP); |
|
1175 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
1176 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
1177 |
mutex_enter(&so->so_lock); |
|
1178 |
if (error) { |
|
1179 |
eprintsoline(so, error); |
|
1180 |
goto done; |
|
1181 |
} |
|
1182 |
||
1183 |
error = sowaitokack(so, T_UNBIND_REQ); |
|
1184 |
if (error) { |
|
1185 |
eprintsoline(so, error); |
|
1186 |
goto done; |
|
1187 |
} |
|
1188 |
||
1189 |
/* |
|
1190 |
* Even if some TPI message (e.g. T_DISCON_IND) was received in |
|
1191 |
* strsock_proto while the lock was dropped above, the unbind |
|
1192 |
* is allowed to complete. |
|
1193 |
*/ |
|
1194 |
if (!(flags & _SOUNBIND_REBIND)) { |
|
1195 |
/* |
|
1196 |
* Clear out bound address. |
|
1197 |
*/ |
|
1198 |
vnode_t *vp; |
|
1199 |
||
1200 |
if ((vp = so->so_ux_bound_vp) != NULL) { |
|
898 | 1201 |
|
1202 |
/* Undo any SSL proxy setup */ |
|
1203 |
if ((so->so_family == AF_INET || |
|
1204 |
so->so_family == AF_INET6) && |
|
1205 |
(so->so_type == SOCK_STREAM) && |
|
1206 |
(so->so_kssl_ent != NULL)) { |
|
1207 |
kssl_release_ent(so->so_kssl_ent, so, |
|
1208 |
so->so_kssl_type); |
|
1209 |
so->so_kssl_ent = NULL; |
|
1210 |
so->so_kssl_type = KSSL_NO_PROXY; |
|
1211 |
} |
|
1212 |
||
0 | 1213 |
so->so_ux_bound_vp = NULL; |
1214 |
vn_rele_stream(vp); |
|
1215 |
} |
|
1216 |
/* Clear out address */ |
|
1217 |
so->so_laddr_len = 0; |
|
1218 |
} |
|
1219 |
so->so_state &= ~(SS_ISBOUND|SS_ACCEPTCONN|SS_LADDR_VALID); |
|
1220 |
done: |
|
898 | 1221 |
|
0 | 1222 |
/* If the caller held the lock don't release it here */ |
1223 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
1224 |
ASSERT(so->so_flag & SOLOCKED); |
|
1225 |
||
1226 |
return (error); |
|
1227 |
} |
|
1228 |
||
1229 |
/* |
|
1230 |
* listen on the socket. |
|
1231 |
* For TPI conforming transports this has to first unbind with the transport |
|
1232 |
* and then bind again using the new backlog. |
|
1233 |
*/ |
|
1234 |
int |
|
1235 |
sotpi_listen(struct sonode *so, int backlog) |
|
1236 |
{ |
|
1237 |
int error = 0; |
|
1238 |
||
1239 |
dprintso(so, 1, ("sotpi_listen(%p, %d) %s\n", |
|
1240 |
so, backlog, pr_state(so->so_state, so->so_mode))); |
|
1241 |
||
1242 |
if (so->so_serv_type == T_CLTS) |
|
1243 |
return (EOPNOTSUPP); |
|
1244 |
||
1245 |
/* |
|
1246 |
* If the socket is ready to accept connections already, then |
|
1247 |
* return without doing anything. This avoids a problem where |
|
1248 |
* a second listen() call fails if a connection is pending and |
|
1249 |
* leaves the socket unbound. Only when we are not unbinding |
|
1250 |
* with the transport can we safely increase the backlog. |
|
1251 |
*/ |
|
1252 |
if (so->so_state & SS_ACCEPTCONN && |
|
1253 |
!((so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
1254 |
/*CONSTCOND*/ |
|
1255 |
!solisten_tpi_tcp)) |
|
1256 |
return (0); |
|
1257 |
||
1258 |
if (so->so_state & SS_ISCONNECTED) |
|
1259 |
return (EINVAL); |
|
1260 |
||
1261 |
mutex_enter(&so->so_lock); |
|
1262 |
so_lock_single(so); /* Set SOLOCKED */ |
|
1263 |
||
1264 |
if (backlog < 0) |
|
1265 |
backlog = 0; |
|
1266 |
/* |
|
1267 |
* Use the same qlimit as in BSD. BSD checks the qlimit |
|
1268 |
* before queuing the next connection implying that a |
|
1269 |
* listen(sock, 0) allows one connection to be queued. |
|
1270 |
* BSD also uses 1.5 times the requested backlog. |
|
1271 |
* |
|
1272 |
* XNS Issue 4 required a strict interpretation of the backlog. |
|
1273 |
* This has been waived subsequently for Issue 4 and the change |
|
1274 |
* incorporated in XNS Issue 5. So we aren't required to do |
|
1275 |
* anything special for XPG apps. |
|
1276 |
*/ |
|
1277 |
if (backlog >= (INT_MAX - 1) / 3) |
|
1278 |
backlog = INT_MAX; |
|
1279 |
else |
|
1280 |
backlog = backlog * 3 / 2 + 1; |
|
1281 |
||
1282 |
/* |
|
1283 |
* If the listen doesn't change the backlog we do nothing. |
|
1284 |
* This avoids an EPROTO error from the transport. |
|
1285 |
*/ |
|
1286 |
if ((so->so_state & SS_ACCEPTCONN) && |
|
1287 |
so->so_backlog == backlog) |
|
1288 |
goto done; |
|
1289 |
||
1290 |
if (!(so->so_state & SS_ISBOUND)) { |
|
1291 |
/* |
|
1292 |
* Must have been explicitly bound in the UNIX domain. |
|
1293 |
*/ |
|
1294 |
if (so->so_family == AF_UNIX) { |
|
1295 |
error = EINVAL; |
|
1296 |
goto done; |
|
1297 |
} |
|
1298 |
error = sotpi_bindlisten(so, NULL, 0, backlog, |
|
1299 |
_SOBIND_UNSPEC|_SOBIND_LOCK_HELD|_SOBIND_LISTEN); |
|
1300 |
} else if (backlog > 0) { |
|
1301 |
/* |
|
1302 |
* AF_INET{,6} hack to avoid losing the port. |
|
1303 |
* Assumes that all AF_INET{,6} transports can handle a |
|
1304 |
* O_T_BIND_REQ with a non-zero CONIND_number when the TPI |
|
1305 |
* has already bound thus it is possible to avoid the unbind. |
|
1306 |
*/ |
|
1307 |
if (!((so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
1308 |
/*CONSTCOND*/ |
|
1309 |
!solisten_tpi_tcp)) { |
|
1310 |
error = sotpi_unbind(so, _SOUNBIND_REBIND); |
|
1311 |
if (error) |
|
1312 |
goto done; |
|
1313 |
} |
|
1314 |
error = sotpi_bindlisten(so, NULL, 0, backlog, |
|
1315 |
_SOBIND_REBIND|_SOBIND_LOCK_HELD|_SOBIND_LISTEN); |
|
1316 |
} else { |
|
1317 |
so->so_state |= SS_ACCEPTCONN; |
|
1318 |
so->so_backlog = backlog; |
|
1319 |
} |
|
1320 |
if (error) |
|
1321 |
goto done; |
|
1322 |
ASSERT(so->so_state & SS_ACCEPTCONN); |
|
1323 |
done: |
|
1324 |
so_unlock_single(so, SOLOCKED); |
|
1325 |
mutex_exit(&so->so_lock); |
|
1326 |
return (error); |
|
1327 |
} |
|
1328 |
||
1329 |
/* |
|
1330 |
* Disconnect either a specified seqno or all (-1). |
|
1331 |
* The former is used on listening sockets only. |
|
1332 |
* |
|
1333 |
* When seqno == -1 sodisconnect could call sotpi_unbind. However, |
|
1334 |
* the current use of sodisconnect(seqno == -1) is only for shutdown |
|
1335 |
* so there is no point (and potentially incorrect) to unbind. |
|
1336 |
*/ |
|
1337 |
int |
|
1338 |
sodisconnect(struct sonode *so, t_scalar_t seqno, int flags) |
|
1339 |
{ |
|
1340 |
struct T_discon_req discon_req; |
|
1341 |
int error = 0; |
|
1342 |
mblk_t *mp; |
|
1343 |
||
1344 |
dprintso(so, 1, ("sodisconnect(%p, %d, 0x%x) %s\n", |
|
1345 |
so, seqno, flags, pr_state(so->so_state, so->so_mode))); |
|
1346 |
||
1347 |
if (!(flags & _SODISCONNECT_LOCK_HELD)) { |
|
1348 |
mutex_enter(&so->so_lock); |
|
1349 |
so_lock_single(so); /* Set SOLOCKED */ |
|
1350 |
} else { |
|
1351 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
1352 |
ASSERT(so->so_flag & SOLOCKED); |
|
1353 |
} |
|
1354 |
||
1355 |
if (!(so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ACCEPTCONN))) { |
|
1356 |
error = EINVAL; |
|
1357 |
eprintsoline(so, error); |
|
1358 |
goto done; |
|
1359 |
} |
|
1360 |
||
1361 |
mutex_exit(&so->so_lock); |
|
1362 |
/* |
|
1363 |
* Flush the write side (unless this is a listener) |
|
1364 |
* and then send down a T_DISCON_REQ. |
|
1365 |
* (Don't flush on listener since it could flush {O_}T_CONN_RES |
|
1366 |
* and other messages.) |
|
1367 |
*/ |
|
1368 |
if (!(so->so_state & SS_ACCEPTCONN)) |
|
1369 |
(void) putnextctl1(strvp2wq(SOTOV(so)), M_FLUSH, FLUSHW); |
|
1370 |
||
1371 |
discon_req.PRIM_type = T_DISCON_REQ; |
|
1372 |
discon_req.SEQ_number = seqno; |
|
1373 |
mp = soallocproto1(&discon_req, sizeof (discon_req), |
|
1374 |
0, _ALLOC_SLEEP); |
|
1375 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
1376 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
1377 |
mutex_enter(&so->so_lock); |
|
1378 |
if (error) { |
|
1379 |
eprintsoline(so, error); |
|
1380 |
goto done; |
|
1381 |
} |
|
1382 |
||
1383 |
error = sowaitokack(so, T_DISCON_REQ); |
|
1384 |
if (error) { |
|
1385 |
eprintsoline(so, error); |
|
1386 |
goto done; |
|
1387 |
} |
|
1388 |
/* |
|
1389 |
* Even if some TPI message (e.g. T_DISCON_IND) was received in |
|
1390 |
* strsock_proto while the lock was dropped above, the disconnect |
|
1391 |
* is allowed to complete. However, it is not possible to |
|
1392 |
* assert that SS_ISCONNECTED|SS_ISCONNECTING are set. |
|
1393 |
*/ |
|
1394 |
so->so_state &= |
|
1395 |
~(SS_ISCONNECTED|SS_ISCONNECTING|SS_LADDR_VALID|SS_FADDR_VALID); |
|
1396 |
done: |
|
1397 |
if (!(flags & _SODISCONNECT_LOCK_HELD)) { |
|
1398 |
so_unlock_single(so, SOLOCKED); |
|
1399 |
mutex_exit(&so->so_lock); |
|
1400 |
} else { |
|
1401 |
/* If the caller held the lock don't release it here */ |
|
1402 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
1403 |
ASSERT(so->so_flag & SOLOCKED); |
|
1404 |
} |
|
1405 |
return (error); |
|
1406 |
} |
|
1407 |
||
1408 |
int |
|
1409 |
sotpi_accept(struct sonode *so, int fflag, struct sonode **nsop) |
|
1410 |
{ |
|
1411 |
struct T_conn_ind *conn_ind; |
|
1412 |
struct T_conn_res *conn_res; |
|
1413 |
int error = 0; |
|
898 | 1414 |
mblk_t *mp, *ctxmp; |
0 | 1415 |
struct sonode *nso; |
1416 |
vnode_t *nvp; |
|
1417 |
void *src; |
|
1418 |
t_uscalar_t srclen; |
|
1419 |
void *opt; |
|
1420 |
t_uscalar_t optlen; |
|
1421 |
t_scalar_t PRIM_type; |
|
1422 |
t_scalar_t SEQ_number; |
|
1423 |
||
1424 |
dprintso(so, 1, ("sotpi_accept(%p, 0x%x, %p) %s\n", |
|
1425 |
so, fflag, nsop, pr_state(so->so_state, so->so_mode))); |
|
1426 |
||
1427 |
/* |
|
1428 |
* Defer single-threading the accepting socket until |
|
1429 |
* the T_CONN_IND has been received and parsed and the |
|
1430 |
* new sonode has been opened. |
|
1431 |
*/ |
|
1432 |
||
1433 |
/* Check that we are not already connected */ |
|
1434 |
if ((so->so_state & SS_ACCEPTCONN) == 0) |
|
1435 |
goto conn_bad; |
|
1436 |
again: |
|
1437 |
if ((error = sowaitconnind(so, fflag, &mp)) != 0) |
|
1438 |
goto e_bad; |
|
1439 |
||
1440 |
ASSERT(mp); |
|
1441 |
conn_ind = (struct T_conn_ind *)mp->b_rptr; |
|
898 | 1442 |
ctxmp = mp->b_cont; |
1443 |
||
0 | 1444 |
/* |
1445 |
* Save SEQ_number for error paths. |
|
1446 |
*/ |
|
1447 |
SEQ_number = conn_ind->SEQ_number; |
|
1448 |
||
1449 |
srclen = conn_ind->SRC_length; |
|
1450 |
src = sogetoff(mp, conn_ind->SRC_offset, srclen, 1); |
|
1451 |
if (src == NULL) { |
|
1452 |
error = EPROTO; |
|
1453 |
freemsg(mp); |
|
1454 |
eprintsoline(so, error); |
|
1455 |
goto disconnect_unlocked; |
|
1456 |
} |
|
1457 |
optlen = conn_ind->OPT_length; |
|
1458 |
switch (so->so_family) { |
|
1459 |
case AF_INET: |
|
1460 |
case AF_INET6: |
|
1461 |
if ((optlen == sizeof (intptr_t)) && |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1462 |
((so->so_state & SS_DIRECT) != 0)) { |
0 | 1463 |
bcopy(mp->b_rptr + conn_ind->OPT_offset, |
1464 |
&opt, conn_ind->OPT_length); |
|
1465 |
} else { |
|
1466 |
/* |
|
1467 |
* The transport (in this case TCP) hasn't sent up |
|
1468 |
* a pointer to an instance for the accept fast-path. |
|
1469 |
* Disable fast-path completely because the call to |
|
1470 |
* sotpi_create() below would otherwise create an |
|
1471 |
* incomplete TCP instance, which would lead to |
|
1472 |
* problems when sockfs sends a normal T_CONN_RES |
|
1473 |
* message down the new stream. |
|
1474 |
*/ |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1475 |
if (so->so_state & SS_DIRECT) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1476 |
int rval; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1477 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1478 |
* For consistency we inform tcp to disable |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1479 |
* direct interface on the listener, though |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1480 |
* we can certainly live without doing this |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1481 |
* because no data will ever travel upstream |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1482 |
* on the listening socket. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1483 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1484 |
so->so_state &= ~SS_DIRECT; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1485 |
(void) strioctl(SOTOV(so), _SIOCSOCKFALLBACK, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1486 |
0, 0, K_TO_K, CRED(), &rval); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1487 |
} |
0 | 1488 |
opt = NULL; |
1489 |
optlen = 0; |
|
1490 |
} |
|
1491 |
break; |
|
1492 |
case AF_UNIX: |
|
1493 |
default: |
|
1494 |
if (optlen != 0) { |
|
1495 |
opt = sogetoff(mp, conn_ind->OPT_offset, optlen, |
|
1496 |
__TPI_ALIGN_SIZE); |
|
1497 |
if (opt == NULL) { |
|
1498 |
error = EPROTO; |
|
1499 |
freemsg(mp); |
|
1500 |
eprintsoline(so, error); |
|
1501 |
goto disconnect_unlocked; |
|
1502 |
} |
|
1503 |
} |
|
1504 |
if (so->so_family == AF_UNIX) { |
|
1505 |
if (!(so->so_state & SS_FADDR_NOXLATE)) { |
|
1506 |
src = NULL; |
|
1507 |
srclen = 0; |
|
1508 |
} |
|
1509 |
/* Extract src address from options */ |
|
1510 |
if (optlen != 0) |
|
1511 |
so_getopt_srcaddr(opt, optlen, &src, &srclen); |
|
1512 |
} |
|
1513 |
break; |
|
1514 |
} |
|
1515 |
||
1516 |
/* |
|
1517 |
* Create the new socket. |
|
1518 |
*/ |
|
1519 |
VN_HOLD(so->so_accessvp); |
|
1520 |
nso = sotpi_create(so->so_accessvp, so->so_family, so->so_type, |
|
1521 |
so->so_protocol, so->so_version, so, &error); |
|
1522 |
if (nso == NULL) { |
|
1523 |
ASSERT(error != 0); |
|
1524 |
/* |
|
1525 |
* Accept can not fail with ENOBUFS. sotpi_create |
|
1526 |
* sleeps waiting for memory until a signal is caught |
|
1527 |
* so return EINTR. |
|
1528 |
*/ |
|
1529 |
freemsg(mp); |
|
1530 |
if (error == ENOBUFS) |
|
1531 |
error = EINTR; |
|
1532 |
goto e_disc_unl; |
|
1533 |
} |
|
1534 |
nvp = SOTOV(nso); |
|
1535 |
||
898 | 1536 |
/* |
1537 |
* If the transport sent up an SSL connection context, then attach |
|
1538 |
* it the new socket, and set the (sd_wputdatafunc)() and |
|
1539 |
* (sd_rputdatafunc)() stream head hooks to intercept and process |
|
1540 |
* SSL records. |
|
1541 |
*/ |
|
1542 |
if (ctxmp != NULL) { |
|
1543 |
/* |
|
1544 |
* This kssl_ctx_t is already held for us by the transport. |
|
1545 |
* So, we don't need to do a kssl_hold_ctx() here. |
|
1546 |
*/ |
|
1547 |
nso->so_kssl_ctx = *((kssl_ctx_t *)ctxmp->b_rptr); |
|
1548 |
freemsg(ctxmp); |
|
1549 |
mp->b_cont = NULL; |
|
1550 |
strsetrwputdatahooks(nvp, strsock_kssl_input, |
|
1551 |
strsock_kssl_output); |
|
1552 |
} |
|
0 | 1553 |
#ifdef DEBUG |
1554 |
/* |
|
1555 |
* SO_DEBUG is used to trigger the dprint* and eprint* macros thus |
|
1556 |
* it's inherited early to allow debugging of the accept code itself. |
|
1557 |
*/ |
|
1558 |
nso->so_options |= so->so_options & SO_DEBUG; |
|
1559 |
#endif /* DEBUG */ |
|
1560 |
||
1561 |
/* |
|
1562 |
* Save the SRC address from the T_CONN_IND |
|
1563 |
* for getpeername to work on AF_UNIX and on transports that do not |
|
1564 |
* support TI_GETPEERNAME. |
|
1565 |
* |
|
1566 |
* NOTE: AF_UNIX NUL termination is ensured by the sender's |
|
1567 |
* copyin_name(). |
|
1568 |
*/ |
|
1569 |
if (srclen > (t_uscalar_t)nso->so_faddr_maxlen) { |
|
1570 |
error = EINVAL; |
|
1571 |
freemsg(mp); |
|
1572 |
eprintsoline(so, error); |
|
1573 |
goto disconnect_vp_unlocked; |
|
1574 |
} |
|
1575 |
nso->so_faddr_len = (socklen_t)srclen; |
|
1576 |
ASSERT(so->so_faddr_len <= so->so_faddr_maxlen); |
|
1577 |
bcopy(src, nso->so_faddr_sa, srclen); |
|
1578 |
nso->so_state |= SS_FADDR_VALID; |
|
1579 |
||
1580 |
if ((DB_REF(mp) > 1) || MBLKSIZE(mp) < |
|
1581 |
(sizeof (struct T_conn_res) + sizeof (intptr_t))) { |
|
1582 |
cred_t *cr; |
|
1583 |
||
1584 |
if ((cr = DB_CRED(mp)) != NULL) { |
|
1585 |
crhold(cr); |
|
1586 |
nso->so_peercred = cr; |
|
1587 |
nso->so_cpid = DB_CPID(mp); |
|
1588 |
} |
|
1589 |
freemsg(mp); |
|
1590 |
||
1591 |
mp = soallocproto1(NULL, sizeof (struct T_conn_res) + |
|
1592 |
sizeof (intptr_t), 0, _ALLOC_INTR); |
|
1593 |
if (mp == NULL) { |
|
1594 |
/* |
|
1595 |
* Accept can not fail with ENOBUFS. |
|
1596 |
* A signal was caught so return EINTR. |
|
1597 |
*/ |
|
1598 |
error = EINTR; |
|
1599 |
eprintsoline(so, error); |
|
1600 |
goto disconnect_vp_unlocked; |
|
1601 |
} |
|
1602 |
conn_res = (struct T_conn_res *)mp->b_rptr; |
|
1603 |
} else { |
|
1604 |
nso->so_peercred = DB_CRED(mp); |
|
1605 |
nso->so_cpid = DB_CPID(mp); |
|
1606 |
DB_CRED(mp) = NULL; |
|
1607 |
||
1608 |
mp->b_rptr = DB_BASE(mp); |
|
1609 |
conn_res = (struct T_conn_res *)mp->b_rptr; |
|
1610 |
mp->b_wptr = mp->b_rptr + sizeof (struct T_conn_res); |
|
1611 |
} |
|
1612 |
||
1613 |
/* |
|
1614 |
* New socket must be bound at least in sockfs and, except for AF_INET, |
|
1615 |
* (or AF_INET6) it also has to be bound in the transport provider. |
|
1616 |
* After accepting the connection on nso so_laddr_sa will be set to |
|
1617 |
* contain the same address as the listener's local address |
|
1618 |
* so the address we bind to isn't important. |
|
1619 |
*/ |
|
1620 |
if ((nso->so_family == AF_INET || nso->so_family == AF_INET6) && |
|
1621 |
/*CONSTCOND*/ |
|
1622 |
nso->so_type == SOCK_STREAM && !soaccept_tpi_tcp) { |
|
1623 |
/* |
|
1624 |
* Optimization for AF_INET{,6} transports |
|
1625 |
* that can handle a T_CONN_RES without being bound. |
|
1626 |
*/ |
|
1627 |
mutex_enter(&nso->so_lock); |
|
1628 |
so_automatic_bind(nso); |
|
1629 |
mutex_exit(&nso->so_lock); |
|
1630 |
} else { |
|
1631 |
/* Perform NULL bind with the transport provider. */ |
|
1632 |
if ((error = sotpi_bind(nso, NULL, 0, _SOBIND_UNSPEC)) != 0) { |
|
1633 |
ASSERT(error != ENOBUFS); |
|
1634 |
freemsg(mp); |
|
1635 |
eprintsoline(nso, error); |
|
1636 |
goto disconnect_vp_unlocked; |
|
1637 |
} |
|
1638 |
} |
|
1639 |
||
1640 |
/* |
|
1641 |
* Inherit SIOCSPGRP, SS_ASYNC before we send the {O_}T_CONN_RES |
|
1642 |
* so that any data arriving on the new socket will cause the |
|
1643 |
* appropriate signals to be delivered for the new socket. |
|
1644 |
* |
|
1645 |
* No other thread (except strsock_proto and strsock_misc) |
|
1646 |
* can access the new socket thus we relax the locking. |
|
1647 |
*/ |
|
1648 |
nso->so_pgrp = so->so_pgrp; |
|
1649 |
nso->so_state |= so->so_state & (SS_ASYNC|SS_FADDR_NOXLATE); |
|
1650 |
||
1651 |
if (nso->so_pgrp != 0) { |
|
1652 |
if ((error = so_set_events(nso, nvp, CRED())) != 0) { |
|
1653 |
eprintsoline(nso, error); |
|
1654 |
error = 0; |
|
1655 |
nso->so_pgrp = 0; |
|
1656 |
} |
|
1657 |
} |
|
1658 |
||
1659 |
/* |
|
1660 |
* Make note of the socket level options. TCP and IP level options |
|
1661 |
* are already inherited. We could do all this after accept is |
|
1662 |
* successful but doing it here simplifies code and no harm done |
|
1663 |
* for error case. |
|
1664 |
*/ |
|
1665 |
nso->so_options = so->so_options & (SO_DEBUG|SO_REUSEADDR|SO_KEEPALIVE| |
|
1666 |
SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK| |
|
1667 |
SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER); |
|
1668 |
nso->so_sndbuf = so->so_sndbuf; |
|
1669 |
nso->so_rcvbuf = so->so_rcvbuf; |
|
1670 |
if (nso->so_options & SO_LINGER) |
|
1671 |
nso->so_linger = so->so_linger; |
|
1672 |
||
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1673 |
if ((so->so_state & SS_DIRECT) != 0) { |
0 | 1674 |
mblk_t *ack_mp; |
1675 |
||
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
1676 |
ASSERT(nso->so_state & SS_DIRECT); |
0 | 1677 |
ASSERT(opt != NULL); |
1678 |
||
1679 |
conn_res->OPT_length = optlen; |
|
1680 |
conn_res->OPT_offset = MBLKL(mp); |
|
1681 |
bcopy(&opt, mp->b_wptr, optlen); |
|
1682 |
mp->b_wptr += optlen; |
|
1683 |
conn_res->PRIM_type = T_CONN_RES; |
|
1684 |
conn_res->ACCEPTOR_id = 0; |
|
1685 |
PRIM_type = T_CONN_RES; |
|
1686 |
||
1687 |
/* Send down the T_CONN_RES on acceptor STREAM */ |
|
1688 |
error = kstrputmsg(SOTOV(nso), mp, NULL, |
|
1689 |
0, 0, MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
1690 |
if (error) { |
|
1691 |
mutex_enter(&so->so_lock); |
|
1692 |
so_lock_single(so); |
|
1693 |
eprintsoline(so, error); |
|
1694 |
goto disconnect_vp; |
|
1695 |
} |
|
1696 |
mutex_enter(&nso->so_lock); |
|
1697 |
error = sowaitprim(nso, T_CONN_RES, T_OK_ACK, |
|
1698 |
(t_uscalar_t)sizeof (struct T_ok_ack), &ack_mp, 0); |
|
1699 |
if (error) { |
|
1700 |
mutex_exit(&nso->so_lock); |
|
1701 |
mutex_enter(&so->so_lock); |
|
1702 |
so_lock_single(so); |
|
1703 |
eprintsoline(so, error); |
|
1704 |
goto disconnect_vp; |
|
1705 |
} |
|
1706 |
if (nso->so_family == AF_INET) { |
|
1707 |
sin_t *sin; |
|
1708 |
||
1709 |
sin = (sin_t *)(ack_mp->b_rptr + |
|
1710 |
sizeof (struct T_ok_ack)); |
|
1711 |
bcopy(sin, nso->so_laddr_sa, sizeof (sin_t)); |
|
1712 |
nso->so_laddr_len = sizeof (sin_t); |
|
1713 |
} else { |
|
1714 |
sin6_t *sin6; |
|
1715 |
||
1716 |
sin6 = (sin6_t *)(ack_mp->b_rptr + |
|
1717 |
sizeof (struct T_ok_ack)); |
|
1718 |
bcopy(sin6, nso->so_laddr_sa, sizeof (sin6_t)); |
|
1719 |
nso->so_laddr_len = sizeof (sin6_t); |
|
1720 |
} |
|
1721 |
freemsg(ack_mp); |
|
1722 |
||
1723 |
nso->so_state |= SS_ISCONNECTED | SS_LADDR_VALID; |
|
1724 |
nso->so_priv = opt; |
|
1725 |
||
1726 |
if (so->so_nl7c_flags & NL7C_ENABLED) { |
|
1727 |
/* |
|
1728 |
* An NL7C marked listen()er so the new socket |
|
1729 |
* inherits the listen()er's NL7C state. |
|
1730 |
* |
|
1731 |
* When calling NL7C to process the new socket |
|
1732 |
* pass the nonblocking i/o state of the listen |
|
1733 |
* socket as this is the context we are in. |
|
1734 |
*/ |
|
1735 |
nso->so_nl7c_flags = so->so_nl7c_flags; |
|
1736 |
if (nl7c_process(nso, |
|
1737 |
(nso->so_state & (SS_NONBLOCK|SS_NDELAY)), |
|
1738 |
(int)((tcp_t *)nso->so_priv)->tcp_mss)) { |
|
1739 |
/* |
|
1740 |
* NL7C has completed processing on the |
|
1741 |
* socket, close the socket and back to |
|
1742 |
* the top to await the next T_CONN_IND. |
|
1743 |
*/ |
|
1744 |
mutex_exit(&nso->so_lock); |
|
1745 |
(void) VOP_CLOSE(nvp, 0, 1, (offset_t)0, |
|
1746 |
CRED()); |
|
1747 |
VN_RELE(nvp); |
|
1748 |
goto again; |
|
1749 |
} |
|
1750 |
/* Pass the new socket out */ |
|
1751 |
} |
|
1752 |
||
1753 |
mutex_exit(&nso->so_lock); |
|
1754 |
||
1755 |
/* |
|
1756 |
* Pass out new socket. |
|
1757 |
*/ |
|
1758 |
if (nsop != NULL) |
|
1759 |
*nsop = nso; |
|
1760 |
||
1761 |
return (0); |
|
1762 |
} |
|
1763 |
||
1764 |
/* |
|
1765 |
* Copy local address from listener. |
|
1766 |
*/ |
|
1767 |
nso->so_laddr_len = so->so_laddr_len; |
|
1768 |
ASSERT(nso->so_laddr_len <= nso->so_laddr_maxlen); |
|
1769 |
bcopy(so->so_laddr_sa, nso->so_laddr_sa, nso->so_laddr_len); |
|
1770 |
nso->so_state |= SS_LADDR_VALID; |
|
1771 |
||
1772 |
/* |
|
1773 |
* This is the non-performance case for sockets (e.g. AF_UNIX sockets) |
|
1774 |
* which don't support the FireEngine accept fast-path. It is also |
|
1775 |
* used when the virtual "sockmod" has been I_POP'd and I_PUSH'd |
|
1776 |
* again. Neither sockfs nor TCP attempt to find out if some other |
|
1777 |
* random module has been inserted in between (in which case we |
|
1778 |
* should follow TLI accept behaviour). We blindly assume the worst |
|
1779 |
* case and revert back to old behaviour i.e. TCP will not send us |
|
1780 |
* any option (eager) and the accept should happen on the listener |
|
1781 |
* queue. Any queued T_conn_ind have already got their options removed |
|
1782 |
* by so_sock2_stream() when "sockmod" was I_POP'd. |
|
1783 |
*/ |
|
1784 |
/* |
|
1785 |
* Fill in the {O_}T_CONN_RES before getting SOLOCKED. |
|
1786 |
*/ |
|
1787 |
if ((nso->so_mode & SM_ACCEPTOR_ID) == 0) { |
|
1788 |
#ifdef _ILP32 |
|
1789 |
queue_t *q; |
|
1790 |
||
1791 |
/* |
|
1792 |
* Find read queue in driver |
|
1793 |
* Can safely do this since we "own" nso/nvp. |
|
1794 |
*/ |
|
1795 |
q = strvp2wq(nvp)->q_next; |
|
1796 |
while (SAMESTR(q)) |
|
1797 |
q = q->q_next; |
|
1798 |
q = RD(q); |
|
1799 |
conn_res->ACCEPTOR_id = (t_uscalar_t)q; |
|
1800 |
#else |
|
1801 |
conn_res->ACCEPTOR_id = (t_uscalar_t)getminor(nvp->v_rdev); |
|
1802 |
#endif /* _ILP32 */ |
|
1803 |
conn_res->PRIM_type = O_T_CONN_RES; |
|
1804 |
PRIM_type = O_T_CONN_RES; |
|
1805 |
} else { |
|
1806 |
conn_res->ACCEPTOR_id = nso->so_acceptor_id; |
|
1807 |
conn_res->PRIM_type = T_CONN_RES; |
|
1808 |
PRIM_type = T_CONN_RES; |
|
1809 |
} |
|
1810 |
conn_res->SEQ_number = SEQ_number; |
|
1811 |
conn_res->OPT_length = 0; |
|
1812 |
conn_res->OPT_offset = 0; |
|
1813 |
||
1814 |
mutex_enter(&so->so_lock); |
|
1815 |
so_lock_single(so); /* Set SOLOCKED */ |
|
1816 |
mutex_exit(&so->so_lock); |
|
1817 |
||
1818 |
error = kstrputmsg(SOTOV(so), mp, NULL, |
|
1819 |
0, 0, MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
1820 |
mutex_enter(&so->so_lock); |
|
1821 |
if (error) { |
|
1822 |
eprintsoline(so, error); |
|
1823 |
goto disconnect_vp; |
|
1824 |
} |
|
1825 |
error = sowaitokack(so, PRIM_type); |
|
1826 |
if (error) { |
|
1827 |
eprintsoline(so, error); |
|
1828 |
goto disconnect_vp; |
|
1829 |
} |
|
1830 |
so_unlock_single(so, SOLOCKED); |
|
1831 |
mutex_exit(&so->so_lock); |
|
1832 |
||
1833 |
nso->so_state |= SS_ISCONNECTED; |
|
1834 |
||
1835 |
/* |
|
1836 |
* Pass out new socket. |
|
1837 |
*/ |
|
1838 |
if (nsop != NULL) |
|
1839 |
*nsop = nso; |
|
1840 |
||
1841 |
return (0); |
|
1842 |
||
1843 |
||
1844 |
eproto_disc_unl: |
|
1845 |
error = EPROTO; |
|
1846 |
e_disc_unl: |
|
1847 |
eprintsoline(so, error); |
|
1848 |
goto disconnect_unlocked; |
|
1849 |
||
1850 |
pr_disc_vp_unl: |
|
1851 |
eprintsoline(so, error); |
|
1852 |
disconnect_vp_unlocked: |
|
1853 |
(void) VOP_CLOSE(nvp, 0, 1, 0, CRED()); |
|
1854 |
VN_RELE(nvp); |
|
1855 |
disconnect_unlocked: |
|
1856 |
(void) sodisconnect(so, SEQ_number, 0); |
|
1857 |
return (error); |
|
1858 |
||
1859 |
pr_disc_vp: |
|
1860 |
eprintsoline(so, error); |
|
1861 |
disconnect_vp: |
|
1862 |
(void) sodisconnect(so, SEQ_number, _SODISCONNECT_LOCK_HELD); |
|
1863 |
so_unlock_single(so, SOLOCKED); |
|
1864 |
mutex_exit(&so->so_lock); |
|
1865 |
(void) VOP_CLOSE(nvp, 0, 1, 0, CRED()); |
|
1866 |
VN_RELE(nvp); |
|
1867 |
return (error); |
|
1868 |
||
1869 |
conn_bad: /* Note: SunOS 4/BSD unconditionally returns EINVAL here */ |
|
1870 |
error = (so->so_type == SOCK_DGRAM || so->so_type == SOCK_RAW) |
|
1871 |
? EOPNOTSUPP : EINVAL; |
|
1872 |
e_bad: |
|
1873 |
eprintsoline(so, error); |
|
1874 |
return (error); |
|
1875 |
} |
|
1876 |
||
1877 |
/* |
|
1878 |
* connect a socket. |
|
1879 |
* |
|
1880 |
* Allow SOCK_DGRAM sockets to reconnect (by specifying a new address) and to |
|
1881 |
* unconnect (by specifying a null address). |
|
1882 |
*/ |
|
1883 |
int |
|
1884 |
sotpi_connect(struct sonode *so, |
|
1885 |
const struct sockaddr *name, |
|
1886 |
socklen_t namelen, |
|
1887 |
int fflag, |
|
1888 |
int flags) |
|
1889 |
{ |
|
1890 |
struct T_conn_req conn_req; |
|
1891 |
int error = 0; |
|
1892 |
mblk_t *mp; |
|
1893 |
void *src; |
|
1894 |
socklen_t srclen; |
|
1895 |
void *addr; |
|
1896 |
socklen_t addrlen; |
|
1897 |
boolean_t need_unlock; |
|
1898 |
||
1899 |
dprintso(so, 1, ("sotpi_connect(%p, %p, %d, 0x%x, 0x%x) %s\n", |
|
1900 |
so, name, namelen, fflag, flags, |
|
1901 |
pr_state(so->so_state, so->so_mode))); |
|
1902 |
||
1903 |
/* |
|
1904 |
* Preallocate the T_CONN_REQ mblk before grabbing SOLOCKED to |
|
1905 |
* avoid sleeping for memory with SOLOCKED held. |
|
1906 |
* We know that the T_CONN_REQ can't be larger than 2 * so_faddr_maxlen |
|
1907 |
* + sizeof (struct T_opthdr). |
|
1908 |
* (the AF_UNIX so_ux_addr_xlate() does not make the address |
|
1909 |
* exceed so_faddr_maxlen). |
|
1910 |
*/ |
|
1911 |
mp = soallocproto(sizeof (struct T_conn_req) + |
|
1912 |
2 * so->so_faddr_maxlen + sizeof (struct T_opthdr), _ALLOC_INTR); |
|
1913 |
if (mp == NULL) { |
|
1914 |
/* |
|
1915 |
* Connect can not fail with ENOBUFS. A signal was |
|
1916 |
* caught so return EINTR. |
|
1917 |
*/ |
|
1918 |
error = EINTR; |
|
1919 |
eprintsoline(so, error); |
|
1920 |
return (error); |
|
1921 |
} |
|
1922 |
||
1923 |
mutex_enter(&so->so_lock); |
|
1924 |
/* |
|
1925 |
* Make sure that there is a preallocated unbind_req |
|
1926 |
* message before any binding. This message allocated when |
|
1927 |
* the socket is created but it might be have been |
|
1928 |
* consumed. |
|
1929 |
*/ |
|
1930 |
if (so->so_unbind_mp == NULL) { |
|
1931 |
dprintso(so, 1, ("sotpi_connect: allocating unbind_req\n")); |
|
1932 |
/* NOTE: holding so_lock while sleeping */ |
|
1933 |
so->so_unbind_mp = |
|
1934 |
soallocproto(sizeof (struct T_unbind_req), _ALLOC_INTR); |
|
1935 |
if (so->so_unbind_mp == NULL) { |
|
1936 |
error = EINTR; |
|
1937 |
need_unlock = B_FALSE; |
|
1938 |
goto done; |
|
1939 |
} |
|
1940 |
} |
|
1941 |
||
1942 |
so_lock_single(so); /* Set SOLOCKED */ |
|
1943 |
need_unlock = B_TRUE; |
|
1944 |
||
1945 |
/* |
|
1946 |
* Can't have done a listen before connecting. |
|
1947 |
*/ |
|
1948 |
if (so->so_state & SS_ACCEPTCONN) { |
|
1949 |
error = EOPNOTSUPP; |
|
1950 |
goto done; |
|
1951 |
} |
|
1952 |
||
1953 |
/* |
|
1954 |
* Must be bound with the transport |
|
1955 |
*/ |
|
1956 |
if (!(so->so_state & SS_ISBOUND)) { |
|
1957 |
if ((so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
1958 |
/*CONSTCOND*/ |
|
1959 |
so->so_type == SOCK_STREAM && !soconnect_tpi_tcp) { |
|
1960 |
/* |
|
1961 |
* Optimization for AF_INET{,6} transports |
|
1962 |
* that can handle a T_CONN_REQ without being bound. |
|
1963 |
*/ |
|
1964 |
so_automatic_bind(so); |
|
1965 |
} else { |
|
1966 |
error = sotpi_bind(so, NULL, 0, |
|
1967 |
_SOBIND_UNSPEC|_SOBIND_LOCK_HELD); |
|
1968 |
if (error) |
|
1969 |
goto done; |
|
1970 |
} |
|
1971 |
ASSERT(so->so_state & SS_ISBOUND); |
|
1972 |
flags |= _SOCONNECT_DID_BIND; |
|
1973 |
} |
|
1974 |
||
1975 |
/* |
|
1976 |
* Handle a connect to a name parameter of type AF_UNSPEC like a |
|
1977 |
* connect to a null address. This is the portable method to |
|
1978 |
* unconnect a socket. |
|
1979 |
*/ |
|
1980 |
if ((namelen >= sizeof (sa_family_t)) && |
|
1981 |
(name->sa_family == AF_UNSPEC)) { |
|
1982 |
name = NULL; |
|
1983 |
namelen = 0; |
|
1984 |
} |
|
1985 |
||
1986 |
/* |
|
1987 |
* Check that we are not already connected. |
|
1988 |
* A connection-oriented socket cannot be reconnected. |
|
1989 |
* A connected connection-less socket can be |
|
1990 |
* - connected to a different address by a subsequent connect |
|
1991 |
* - "unconnected" by a connect to the NULL address |
|
1992 |
*/ |
|
1993 |
if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) { |
|
1994 |
ASSERT(!(flags & _SOCONNECT_DID_BIND)); |
|
1995 |
if (so->so_mode & SM_CONNREQUIRED) { |
|
1996 |
/* Connection-oriented socket */ |
|
1997 |
error = so->so_state & SS_ISCONNECTED ? |
|
1998 |
EISCONN : EALREADY; |
|
1999 |
goto done; |
|
2000 |
} |
|
2001 |
/* Connection-less socket */ |
|
2002 |
if (name == NULL) { |
|
2003 |
/* |
|
2004 |
* Remove the connected state and clear SO_DGRAM_ERRIND |
|
2005 |
* since it was set when the socket was connected. |
|
2006 |
* If this is UDP also send down a T_DISCON_REQ. |
|
2007 |
*/ |
|
2008 |
int val; |
|
2009 |
||
2010 |
if ((so->so_family == AF_INET || |
|
2011 |
so->so_family == AF_INET6) && |
|
2012 |
(so->so_type == SOCK_DGRAM || |
|
2013 |
so->so_type == SOCK_RAW) && |
|
2014 |
/*CONSTCOND*/ |
|
2015 |
!soconnect_tpi_udp) { |
|
2016 |
/* XXX What about implicitly unbinding here? */ |
|
2017 |
error = sodisconnect(so, -1, |
|
2018 |
_SODISCONNECT_LOCK_HELD); |
|
2019 |
} else { |
|
2020 |
so->so_state &= |
|
2021 |
~(SS_ISCONNECTED | SS_ISCONNECTING | |
|
2022 |
SS_FADDR_VALID); |
|
2023 |
so->so_faddr_len = 0; |
|
2024 |
} |
|
2025 |
||
2026 |
so_unlock_single(so, SOLOCKED); |
|
2027 |
mutex_exit(&so->so_lock); |
|
2028 |
||
2029 |
val = 0; |
|
2030 |
(void) sotpi_setsockopt(so, SOL_SOCKET, SO_DGRAM_ERRIND, |
|
2031 |
&val, (t_uscalar_t)sizeof (val)); |
|
2032 |
||
2033 |
mutex_enter(&so->so_lock); |
|
2034 |
so_lock_single(so); /* Set SOLOCKED */ |
|
2035 |
goto done; |
|
2036 |
} |
|
2037 |
} |
|
2038 |
ASSERT(so->so_state & SS_ISBOUND); |
|
2039 |
||
2040 |
if (name == NULL || namelen == 0) { |
|
2041 |
error = EINVAL; |
|
2042 |
goto done; |
|
2043 |
} |
|
2044 |
/* |
|
2045 |
* Mark the socket if so_faddr_sa represents the transport level |
|
2046 |
* address. |
|
2047 |
*/ |
|
2048 |
if (flags & _SOCONNECT_NOXLATE) { |
|
2049 |
struct sockaddr_ux *soaddr_ux; |
|
2050 |
||
2051 |
ASSERT(so->so_family == AF_UNIX); |
|
2052 |
if (namelen != sizeof (struct sockaddr_ux)) { |
|
2053 |
error = EINVAL; |
|
2054 |
goto done; |
|
2055 |
} |
|
2056 |
soaddr_ux = (struct sockaddr_ux *)name; |
|
2057 |
name = (struct sockaddr *)&soaddr_ux->sou_addr; |
|
2058 |
namelen = sizeof (soaddr_ux->sou_addr); |
|
2059 |
so->so_state |= SS_FADDR_NOXLATE; |
|
2060 |
} |
|
2061 |
||
2062 |
/* |
|
2063 |
* Length and family checks. |
|
2064 |
*/ |
|
2065 |
error = so_addr_verify(so, name, namelen); |
|
2066 |
if (error) |
|
2067 |
goto bad; |
|
2068 |
||
2069 |
/* |
|
2070 |
* Save foreign address. Needed for AF_UNIX as well as |
|
2071 |
* transport providers that do not support TI_GETPEERNAME. |
|
2072 |
* Also used for cached foreign address for TCP and UDP. |
|
2073 |
*/ |
|
2074 |
if (namelen > (t_uscalar_t)so->so_faddr_maxlen) { |
|
2075 |
error = EINVAL; |
|
2076 |
goto done; |
|
2077 |
} |
|
2078 |
so->so_faddr_len = (socklen_t)namelen; |
|
2079 |
ASSERT(so->so_faddr_len <= so->so_faddr_maxlen); |
|
2080 |
bcopy(name, so->so_faddr_sa, namelen); |
|
2081 |
so->so_state |= SS_FADDR_VALID; |
|
2082 |
||
2083 |
if (so->so_family == AF_UNIX) { |
|
2084 |
if (so->so_state & SS_FADDR_NOXLATE) { |
|
2085 |
/* |
|
2086 |
* Already have a transport internal address. Do not |
|
2087 |
* pass any (transport internal) source address. |
|
2088 |
*/ |
|
2089 |
addr = so->so_faddr_sa; |
|
2090 |
addrlen = (t_uscalar_t)so->so_faddr_len; |
|
2091 |
src = NULL; |
|
2092 |
srclen = 0; |
|
2093 |
} else { |
|
2094 |
/* |
|
2095 |
* Pass the sockaddr_un source address as an option |
|
2096 |
* and translate the remote address. |
|
2097 |
* Holding so_lock thus so_laddr_sa can not change. |
|
2098 |
*/ |
|
2099 |
src = so->so_laddr_sa; |
|
2100 |
srclen = (t_uscalar_t)so->so_laddr_len; |
|
2101 |
dprintso(so, 1, |
|
2102 |
("sotpi_connect UNIX: srclen %d, src %p\n", |
|
2103 |
srclen, src)); |
|
2104 |
error = so_ux_addr_xlate(so, |
|
2105 |
so->so_faddr_sa, (socklen_t)so->so_faddr_len, |
|
2106 |
(flags & _SOCONNECT_XPG4_2), |
|
2107 |
&addr, &addrlen); |
|
2108 |
if (error) |
|
2109 |
goto bad; |
|
2110 |
} |
|
2111 |
} else { |
|
2112 |
addr = so->so_faddr_sa; |
|
2113 |
addrlen = (t_uscalar_t)so->so_faddr_len; |
|
2114 |
src = NULL; |
|
2115 |
srclen = 0; |
|
2116 |
} |
|
2117 |
/* |
|
2118 |
* When connecting a datagram socket we issue the SO_DGRAM_ERRIND |
|
2119 |
* option which asks the transport provider to send T_UDERR_IND |
|
2120 |
* messages. These T_UDERR_IND messages are used to return connected |
|
2121 |
* style errors (e.g. ECONNRESET) for connected datagram sockets. |
|
2122 |
* |
|
2123 |
* In addition, for UDP (and SOCK_RAW AF_INET{,6} sockets) |
|
2124 |
* we send down a T_CONN_REQ. This is needed to let the |
|
2125 |
* transport assign a local address that is consistent with |
|
2126 |
* the remote address. Applications depend on a getsockname() |
|
2127 |
* after a connect() to retrieve the "source" IP address for |
|
2128 |
* the connected socket. Invalidate the cached local address |
|
2129 |
* to force getsockname() to enquire of the transport. |
|
2130 |
*/ |
|
2131 |
if (!(so->so_mode & SM_CONNREQUIRED)) { |
|
2132 |
/* |
|
2133 |
* Datagram socket. |
|
2134 |
*/ |
|
2135 |
int32_t val; |
|
2136 |
||
2137 |
so_unlock_single(so, SOLOCKED); |
|
2138 |
mutex_exit(&so->so_lock); |
|
2139 |
||
2140 |
val = 1; |
|
2141 |
(void) sotpi_setsockopt(so, SOL_SOCKET, SO_DGRAM_ERRIND, |
|
2142 |
&val, (t_uscalar_t)sizeof (val)); |
|
2143 |
||
2144 |
mutex_enter(&so->so_lock); |
|
2145 |
so_lock_single(so); /* Set SOLOCKED */ |
|
2146 |
if ((so->so_family != AF_INET && so->so_family != AF_INET6) || |
|
2147 |
(so->so_type != SOCK_DGRAM && so->so_type != SOCK_RAW) || |
|
2148 |
soconnect_tpi_udp) { |
|
2149 |
soisconnected(so); |
|
2150 |
goto done; |
|
2151 |
} |
|
2152 |
/* |
|
2153 |
* Send down T_CONN_REQ etc. |
|
2154 |
* Clear fflag to avoid returning EWOULDBLOCK. |
|
2155 |
*/ |
|
2156 |
fflag = 0; |
|
2157 |
ASSERT(so->so_family != AF_UNIX); |
|
2158 |
so->so_state &= ~SS_LADDR_VALID; |
|
2159 |
} else if (so->so_laddr_len != 0) { |
|
2160 |
/* |
|
2161 |
* If the local address or port was "any" then it may be |
|
2162 |
* changed by the transport as a result of the |
|
2163 |
* connect. Invalidate the cached version if we have one. |
|
2164 |
*/ |
|
2165 |
switch (so->so_family) { |
|
2166 |
case AF_INET: |
|
2167 |
ASSERT(so->so_laddr_len == (socklen_t)sizeof (sin_t)); |
|
2168 |
if (((sin_t *)so->so_laddr_sa)->sin_addr.s_addr == |
|
2169 |
INADDR_ANY || |
|
2170 |
((sin_t *)so->so_laddr_sa)->sin_port == 0) |
|
2171 |
so->so_state &= ~SS_LADDR_VALID; |
|
2172 |
break; |
|
2173 |
||
2174 |
case AF_INET6: |
|
2175 |
ASSERT(so->so_laddr_len == (socklen_t)sizeof (sin6_t)); |
|
2176 |
if (IN6_IS_ADDR_UNSPECIFIED( |
|
2177 |
&((sin6_t *)so->so_laddr_sa) ->sin6_addr) || |
|
2178 |
IN6_IS_ADDR_V4MAPPED_ANY( |
|
2179 |
&((sin6_t *)so->so_laddr_sa)->sin6_addr) || |
|
2180 |
((sin6_t *)so->so_laddr_sa)->sin6_port == 0) |
|
2181 |
so->so_state &= ~SS_LADDR_VALID; |
|
2182 |
break; |
|
2183 |
||
2184 |
default: |
|
2185 |
break; |
|
2186 |
} |
|
2187 |
} |
|
2188 |
||
2189 |
/* |
|
2190 |
* Check for failure of an earlier call |
|
2191 |
*/ |
|
2192 |
if (so->so_error != 0) |
|
2193 |
goto so_bad; |
|
2194 |
||
2195 |
/* |
|
2196 |
* Send down T_CONN_REQ. Message was allocated above. |
|
2197 |
*/ |
|
2198 |
conn_req.PRIM_type = T_CONN_REQ; |
|
2199 |
conn_req.DEST_length = addrlen; |
|
2200 |
conn_req.DEST_offset = (t_scalar_t)sizeof (conn_req); |
|
2201 |
if (srclen == 0) { |
|
2202 |
conn_req.OPT_length = 0; |
|
2203 |
conn_req.OPT_offset = 0; |
|
2204 |
soappendmsg(mp, &conn_req, sizeof (conn_req)); |
|
2205 |
soappendmsg(mp, addr, addrlen); |
|
2206 |
} else { |
|
2207 |
/* |
|
2208 |
* There is a AF_UNIX sockaddr_un to include as a source |
|
2209 |
* address option. |
|
2210 |
*/ |
|
2211 |
struct T_opthdr toh; |
|
2212 |
||
2213 |
toh.level = SOL_SOCKET; |
|
2214 |
toh.name = SO_SRCADDR; |
|
2215 |
toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr)); |
|
2216 |
toh.status = 0; |
|
2217 |
conn_req.OPT_length = |
|
2218 |
(t_scalar_t)(sizeof (toh) + _TPI_ALIGN_TOPT(srclen)); |
|
2219 |
conn_req.OPT_offset = (t_scalar_t)(sizeof (conn_req) + |
|
2220 |
_TPI_ALIGN_TOPT(addrlen)); |
|
2221 |
||
2222 |
soappendmsg(mp, &conn_req, sizeof (conn_req)); |
|
2223 |
soappendmsg(mp, addr, addrlen); |
|
2224 |
mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen; |
|
2225 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
2226 |
soappendmsg(mp, src, srclen); |
|
2227 |
mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen; |
|
2228 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
2229 |
} |
|
2230 |
/* |
|
2231 |
* Set SS_ISCONNECTING before sending down the T_CONN_REQ |
|
2232 |
* in order to have the right state when the T_CONN_CON shows up. |
|
2233 |
*/ |
|
2234 |
soisconnecting(so); |
|
2235 |
mutex_exit(&so->so_lock); |
|
2236 |
||
2237 |
#ifdef C2_AUDIT |
|
2238 |
if (audit_active) |
|
2239 |
audit_sock(T_CONN_REQ, strvp2wq(SOTOV(so)), mp, 0); |
|
2240 |
#endif /* C2_AUDIT */ |
|
2241 |
||
2242 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
2243 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0); |
|
2244 |
mp = NULL; |
|
2245 |
mutex_enter(&so->so_lock); |
|
2246 |
if (error != 0) |
|
2247 |
goto bad; |
|
2248 |
||
2249 |
if ((error = sowaitokack(so, T_CONN_REQ)) != 0) |
|
2250 |
goto bad; |
|
2251 |
||
2252 |
/* Allow other threads to access the socket */ |
|
2253 |
so_unlock_single(so, SOLOCKED); |
|
2254 |
need_unlock = B_FALSE; |
|
2255 |
||
2256 |
/* |
|
2257 |
* Wait until we get a T_CONN_CON or an error |
|
2258 |
*/ |
|
2259 |
if ((error = sowaitconnected(so, fflag, 0)) != 0) { |
|
2260 |
so_lock_single(so); /* Set SOLOCKED */ |
|
2261 |
need_unlock = B_TRUE; |
|
2262 |
} |
|
2263 |
||
2264 |
done: |
|
2265 |
freemsg(mp); |
|
2266 |
switch (error) { |
|
2267 |
case EINPROGRESS: |
|
2268 |
case EALREADY: |
|
2269 |
case EISCONN: |
|
2270 |
case EINTR: |
|
2271 |
/* Non-fatal errors */ |
|
2272 |
so->so_state &= ~SS_LADDR_VALID; |
|
2273 |
/* FALLTHRU */ |
|
2274 |
case 0: |
|
2275 |
break; |
|
2276 |
||
2277 |
case EHOSTUNREACH: |
|
2278 |
if (flags & _SOCONNECT_XPG4_2) { |
|
2279 |
/* |
|
2280 |
* X/Open specification contains a requirement that |
|
2281 |
* ENETUNREACH be returned but does not require |
|
2282 |
* EHOSTUNREACH. In order to keep the test suite |
|
2283 |
* happy we mess with the errno here. |
|
2284 |
*/ |
|
2285 |
error = ENETUNREACH; |
|
2286 |
} |
|
2287 |
/* FALLTHRU */ |
|
2288 |
||
2289 |
default: |
|
2290 |
ASSERT(need_unlock); |
|
2291 |
/* |
|
2292 |
* Fatal errors: clear SS_ISCONNECTING in case it was set, |
|
2293 |
* and invalidate local-address cache |
|
2294 |
*/ |
|
2295 |
so->so_state &= ~(SS_ISCONNECTING | SS_LADDR_VALID); |
|
2296 |
/* A discon_ind might have already unbound us */ |
|
2297 |
if ((flags & _SOCONNECT_DID_BIND) && |
|
2298 |
(so->so_state & SS_ISBOUND)) { |
|
2299 |
int err; |
|
2300 |
||
2301 |
err = sotpi_unbind(so, 0); |
|
2302 |
/* LINTED - statement has no conseq */ |
|
2303 |
if (err) { |
|
2304 |
eprintsoline(so, err); |
|
2305 |
} |
|
2306 |
} |
|
2307 |
break; |
|
2308 |
} |
|
2309 |
if (need_unlock) |
|
2310 |
so_unlock_single(so, SOLOCKED); |
|
2311 |
mutex_exit(&so->so_lock); |
|
2312 |
return (error); |
|
2313 |
||
2314 |
so_bad: error = sogeterr(so); |
|
2315 |
bad: eprintsoline(so, error); |
|
2316 |
goto done; |
|
2317 |
} |
|
2318 |
||
2319 |
int |
|
2320 |
sotpi_shutdown(struct sonode *so, int how) |
|
2321 |
{ |
|
2322 |
struct T_ordrel_req ordrel_req; |
|
2323 |
mblk_t *mp; |
|
2324 |
uint_t old_state, state_change; |
|
2325 |
int error = 0; |
|
2326 |
||
2327 |
dprintso(so, 1, ("sotpi_shutdown(%p, %d) %s\n", |
|
2328 |
so, how, pr_state(so->so_state, so->so_mode))); |
|
2329 |
||
2330 |
mutex_enter(&so->so_lock); |
|
2331 |
so_lock_single(so); /* Set SOLOCKED */ |
|
2332 |
||
2333 |
/* |
|
2334 |
* SunOS 4.X has no check for datagram sockets. |
|
2335 |
* 5.X checks that it is connected (ENOTCONN) |
|
2336 |
* X/Open requires that we check the connected state. |
|
2337 |
*/ |
|
2338 |
if (!(so->so_state & SS_ISCONNECTED)) { |
|
2339 |
if (!xnet_skip_checks) { |
|
2340 |
error = ENOTCONN; |
|
2341 |
if (xnet_check_print) { |
|
2342 |
printf("sockfs: X/Open shutdown check " |
|
2343 |
"caused ENOTCONN\n"); |
|
2344 |
} |
|
2345 |
} |
|
2346 |
goto done; |
|
2347 |
} |
|
2348 |
/* |
|
2349 |
* Record the current state and then perform any state changes. |
|
2350 |
* Then use the difference between the old and new states to |
|
2351 |
* determine which messages need to be sent. |
|
2352 |
* This prevents e.g. duplicate T_ORDREL_REQ when there are |
|
2353 |
* duplicate calls to shutdown(). |
|
2354 |
*/ |
|
2355 |
old_state = so->so_state; |
|
2356 |
||
2357 |
switch (how) { |
|
2358 |
case 0: |
|
2359 |
socantrcvmore(so); |
|
2360 |
break; |
|
2361 |
case 1: |
|
2362 |
socantsendmore(so); |
|
2363 |
break; |
|
2364 |
case 2: |
|
2365 |
socantsendmore(so); |
|
2366 |
socantrcvmore(so); |
|
2367 |
break; |
|
2368 |
default: |
|
2369 |
error = EINVAL; |
|
2370 |
goto done; |
|
2371 |
} |
|
2372 |
||
2373 |
/* |
|
2374 |
* Assumes that the SS_CANT* flags are never cleared in the above code. |
|
2375 |
*/ |
|
2376 |
state_change = (so->so_state & (SS_CANTRCVMORE|SS_CANTSENDMORE)) - |
|
2377 |
(old_state & (SS_CANTRCVMORE|SS_CANTSENDMORE)); |
|
2378 |
ASSERT((state_change & ~(SS_CANTRCVMORE|SS_CANTSENDMORE)) == 0); |
|
2379 |
||
2380 |
switch (state_change) { |
|
2381 |
case 0: |
|
2382 |
dprintso(so, 1, |
|
2383 |
("sotpi_shutdown: nothing to send in state 0x%x\n", |
|
2384 |
so->so_state)); |
|
2385 |
goto done; |
|
2386 |
||
2387 |
case SS_CANTRCVMORE: |
|
2388 |
mutex_exit(&so->so_lock); |
|
2389 |
strseteof(SOTOV(so), 1); |
|
2390 |
/* |
|
2391 |
* strseteof takes care of read side wakeups, |
|
2392 |
* pollwakeups, and signals. |
|
2393 |
*/ |
|
2394 |
/* |
|
2395 |
* Get the read lock before flushing data to avoid problems |
|
2396 |
* with the T_EXDATA_IND MSG_PEEK code in sotpi_recvmsg. |
|
2397 |
*/ |
|
2398 |
mutex_enter(&so->so_lock); |
|
2399 |
(void) so_lock_read(so, 0); /* Set SOREADLOCKED */ |
|
2400 |
mutex_exit(&so->so_lock); |
|
2401 |
||
2402 |
/* Flush read side queue */ |
|
2403 |
strflushrq(SOTOV(so), FLUSHALL); |
|
2404 |
||
2405 |
mutex_enter(&so->so_lock); |
|
2406 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
2407 |
break; |
|
2408 |
||
2409 |
case SS_CANTSENDMORE: |
|
2410 |
mutex_exit(&so->so_lock); |
|
2411 |
strsetwerror(SOTOV(so), 0, 0, sogetwrerr); |
|
2412 |
mutex_enter(&so->so_lock); |
|
2413 |
break; |
|
2414 |
||
2415 |
case SS_CANTSENDMORE|SS_CANTRCVMORE: |
|
2416 |
mutex_exit(&so->so_lock); |
|
2417 |
strsetwerror(SOTOV(so), 0, 0, sogetwrerr); |
|
2418 |
strseteof(SOTOV(so), 1); |
|
2419 |
/* |
|
2420 |
* strseteof takes care of read side wakeups, |
|
2421 |
* pollwakeups, and signals. |
|
2422 |
*/ |
|
2423 |
/* |
|
2424 |
* Get the read lock before flushing data to avoid problems |
|
2425 |
* with the T_EXDATA_IND MSG_PEEK code in sotpi_recvmsg. |
|
2426 |
*/ |
|
2427 |
mutex_enter(&so->so_lock); |
|
2428 |
(void) so_lock_read(so, 0); /* Set SOREADLOCKED */ |
|
2429 |
mutex_exit(&so->so_lock); |
|
2430 |
||
2431 |
/* Flush read side queue */ |
|
2432 |
strflushrq(SOTOV(so), FLUSHALL); |
|
2433 |
||
2434 |
mutex_enter(&so->so_lock); |
|
2435 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
2436 |
break; |
|
2437 |
} |
|
2438 |
||
2439 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
2440 |
||
2441 |
/* |
|
2442 |
* If either SS_CANTSENDMORE or SS_CANTRCVMORE or both of them |
|
2443 |
* was set due to this call and the new state has both of them set: |
|
2444 |
* Send the AF_UNIX close indication |
|
2445 |
* For T_COTS send a discon_ind |
|
2446 |
* |
|
2447 |
* If cantsend was set due to this call: |
|
2448 |
* For T_COTSORD send an ordrel_ind |
|
2449 |
* |
|
2450 |
* Note that for T_CLTS there is no message sent here. |
|
2451 |
*/ |
|
2452 |
if ((so->so_state & (SS_CANTRCVMORE|SS_CANTSENDMORE)) == |
|
2453 |
(SS_CANTRCVMORE|SS_CANTSENDMORE)) { |
|
2454 |
/* |
|
2455 |
* For SunOS 4.X compatibility we tell the other end |
|
2456 |
* that we are unable to receive at this point. |
|
2457 |
*/ |
|
2458 |
if (so->so_family == AF_UNIX && so->so_serv_type != T_CLTS) |
|
2459 |
so_unix_close(so); |
|
2460 |
||
2461 |
if (so->so_serv_type == T_COTS) |
|
2462 |
error = sodisconnect(so, -1, _SODISCONNECT_LOCK_HELD); |
|
2463 |
} |
|
2464 |
if ((state_change & SS_CANTSENDMORE) && |
|
2465 |
(so->so_serv_type == T_COTS_ORD)) { |
|
2466 |
/* Send an orderly release */ |
|
2467 |
ordrel_req.PRIM_type = T_ORDREL_REQ; |
|
2468 |
||
2469 |
mutex_exit(&so->so_lock); |
|
2470 |
mp = soallocproto1(&ordrel_req, sizeof (ordrel_req), |
|
2471 |
0, _ALLOC_SLEEP); |
|
2472 |
/* |
|
2473 |
* Send down the T_ORDREL_REQ even if there is flow control. |
|
2474 |
* This prevents shutdown from blocking. |
|
2475 |
* Note that there is no T_OK_ACK for ordrel_req. |
|
2476 |
*/ |
|
2477 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
2478 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0); |
|
2479 |
mutex_enter(&so->so_lock); |
|
2480 |
if (error) { |
|
2481 |
eprintsoline(so, error); |
|
2482 |
goto done; |
|
2483 |
} |
|
2484 |
} |
|
2485 |
||
2486 |
done: |
|
2487 |
so_unlock_single(so, SOLOCKED); |
|
2488 |
mutex_exit(&so->so_lock); |
|
2489 |
return (error); |
|
2490 |
} |
|
2491 |
||
2492 |
/* |
|
2493 |
* For any connected SOCK_STREAM/SOCK_SEQPACKET AF_UNIX socket we send |
|
2494 |
* a zero-length T_OPTDATA_REQ with the SO_UNIX_CLOSE option to inform the peer |
|
2495 |
* that we have closed. |
|
2496 |
* Also, for connected AF_UNIX SOCK_DGRAM sockets we send a zero-length |
|
2497 |
* T_UNITDATA_REQ containing the same option. |
|
2498 |
* |
|
2499 |
* For SOCK_DGRAM half-connections (somebody connected to this end |
|
2500 |
* but this end is not connect) we don't know where to send any |
|
2501 |
* SO_UNIX_CLOSE. |
|
2502 |
* |
|
2503 |
* We have to ignore stream head errors just in case there has been |
|
2504 |
* a shutdown(output). |
|
2505 |
* Ignore any flow control to try to get the message more quickly to the peer. |
|
2506 |
* While locally ignoring flow control solves the problem when there |
|
2507 |
* is only the loopback transport on the stream it would not provide |
|
2508 |
* the correct AF_UNIX socket semantics when one or more modules have |
|
2509 |
* been pushed. |
|
2510 |
*/ |
|
2511 |
void |
|
2512 |
so_unix_close(struct sonode *so) |
|
2513 |
{ |
|
2514 |
int error; |
|
2515 |
struct T_opthdr toh; |
|
2516 |
mblk_t *mp; |
|
2517 |
||
2518 |
ASSERT(MUTEX_HELD(&so->so_lock)); |
|
2519 |
||
2520 |
ASSERT(so->so_family == AF_UNIX); |
|
2521 |
||
2522 |
if ((so->so_state & (SS_ISCONNECTED|SS_ISBOUND)) != |
|
2523 |
(SS_ISCONNECTED|SS_ISBOUND)) |
|
2524 |
return; |
|
2525 |
||
2526 |
dprintso(so, 1, ("so_unix_close(%p) %s\n", |
|
2527 |
so, pr_state(so->so_state, so->so_mode))); |
|
2528 |
||
2529 |
toh.level = SOL_SOCKET; |
|
2530 |
toh.name = SO_UNIX_CLOSE; |
|
2531 |
||
2532 |
/* zero length + header */ |
|
2533 |
toh.len = (t_uscalar_t)sizeof (struct T_opthdr); |
|
2534 |
toh.status = 0; |
|
2535 |
||
2536 |
if (so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) { |
|
2537 |
struct T_optdata_req tdr; |
|
2538 |
||
2539 |
tdr.PRIM_type = T_OPTDATA_REQ; |
|
2540 |
tdr.DATA_flag = 0; |
|
2541 |
||
2542 |
tdr.OPT_length = (t_scalar_t)sizeof (toh); |
|
2543 |
tdr.OPT_offset = (t_scalar_t)sizeof (tdr); |
|
2544 |
||
2545 |
/* NOTE: holding so_lock while sleeping */ |
|
2546 |
mp = soallocproto2(&tdr, sizeof (tdr), |
|
2547 |
&toh, sizeof (toh), 0, _ALLOC_SLEEP); |
|
2548 |
} else { |
|
2549 |
struct T_unitdata_req tudr; |
|
2550 |
void *addr; |
|
2551 |
socklen_t addrlen; |
|
2552 |
void *src; |
|
2553 |
socklen_t srclen; |
|
2554 |
struct T_opthdr toh2; |
|
2555 |
t_scalar_t size; |
|
2556 |
||
2557 |
/* Connecteded DGRAM socket */ |
|
2558 |
||
2559 |
/* |
|
2560 |
* For AF_UNIX the destination address is translated to |
|
2561 |
* an internal name and the source address is passed as |
|
2562 |
* an option. |
|
2563 |
*/ |
|
2564 |
/* |
|
2565 |
* Length and family checks. |
|
2566 |
*/ |
|
2567 |
error = so_addr_verify(so, so->so_faddr_sa, |
|
2568 |
(t_uscalar_t)so->so_faddr_len); |
|
2569 |
if (error) { |
|
2570 |
eprintsoline(so, error); |
|
2571 |
return; |
|
2572 |
} |
|
2573 |
if (so->so_state & SS_FADDR_NOXLATE) { |
|
2574 |
/* |
|
2575 |
* Already have a transport internal address. Do not |
|
2576 |
* pass any (transport internal) source address. |
|
2577 |
*/ |
|
2578 |
addr = so->so_faddr_sa; |
|
2579 |
addrlen = (t_uscalar_t)so->so_faddr_len; |
|
2580 |
src = NULL; |
|
2581 |
srclen = 0; |
|
2582 |
} else { |
|
2583 |
/* |
|
2584 |
* Pass the sockaddr_un source address as an option |
|
2585 |
* and translate the remote address. |
|
2586 |
* Holding so_lock thus so_laddr_sa can not change. |
|
2587 |
*/ |
|
2588 |
src = so->so_laddr_sa; |
|
2589 |
srclen = (socklen_t)so->so_laddr_len; |
|
2590 |
dprintso(so, 1, |
|
2591 |
("so_ux_close: srclen %d, src %p\n", |
|
2592 |
srclen, src)); |
|
2593 |
error = so_ux_addr_xlate(so, |
|
2594 |
so->so_faddr_sa, |
|
2595 |
(socklen_t)so->so_faddr_len, 0, |
|
2596 |
&addr, &addrlen); |
|
2597 |
if (error) { |
|
2598 |
eprintsoline(so, error); |
|
2599 |
return; |
|
2600 |
} |
|
2601 |
} |
|
2602 |
tudr.PRIM_type = T_UNITDATA_REQ; |
|
2603 |
tudr.DEST_length = addrlen; |
|
2604 |
tudr.DEST_offset = (t_scalar_t)sizeof (tudr); |
|
2605 |
if (srclen == 0) { |
|
2606 |
tudr.OPT_length = (t_scalar_t)sizeof (toh); |
|
2607 |
tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) + |
|
2608 |
_TPI_ALIGN_TOPT(addrlen)); |
|
2609 |
||
2610 |
size = tudr.OPT_offset + tudr.OPT_length; |
|
2611 |
/* NOTE: holding so_lock while sleeping */ |
|
2612 |
mp = soallocproto2(&tudr, sizeof (tudr), |
|
2613 |
addr, addrlen, size, _ALLOC_SLEEP); |
|
2614 |
mp->b_wptr += (_TPI_ALIGN_TOPT(addrlen) - addrlen); |
|
2615 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
2616 |
} else { |
|
2617 |
/* |
|
2618 |
* There is a AF_UNIX sockaddr_un to include as a |
|
2619 |
* source address option. |
|
2620 |
*/ |
|
2621 |
tudr.OPT_length = (t_scalar_t)(2 * sizeof (toh) + |
|
2622 |
_TPI_ALIGN_TOPT(srclen)); |
|
2623 |
tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) + |
|
2624 |
_TPI_ALIGN_TOPT(addrlen)); |
|
2625 |
||
2626 |
toh2.level = SOL_SOCKET; |
|
2627 |
toh2.name = SO_SRCADDR; |
|
2628 |
toh2.len = (t_uscalar_t)(srclen + |
|
2629 |
sizeof (struct T_opthdr)); |
|
2630 |
toh2.status = 0; |
|
2631 |
||
2632 |
size = tudr.OPT_offset + tudr.OPT_length; |
|
2633 |
||
2634 |
/* NOTE: holding so_lock while sleeping */ |
|
2635 |
mp = soallocproto2(&tudr, sizeof (tudr), |
|
2636 |
addr, addrlen, size, _ALLOC_SLEEP); |
|
2637 |
mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen; |
|
2638 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
2639 |
soappendmsg(mp, &toh2, sizeof (toh2)); |
|
2640 |
soappendmsg(mp, src, srclen); |
|
2641 |
mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen; |
|
2642 |
} |
|
2643 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
2644 |
} |
|
2645 |
mutex_exit(&so->so_lock); |
|
2646 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
2647 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0); |
|
2648 |
mutex_enter(&so->so_lock); |
|
2649 |
} |
|
2650 |
||
2651 |
/* |
|
2652 |
* Handle recv* calls that set MSG_OOB or MSG_OOB together with MSG_PEEK. |
|
2653 |
*/ |
|
2654 |
int |
|
2655 |
sorecvoob(struct sonode *so, struct nmsghdr *msg, struct uio *uiop, int flags) |
|
2656 |
{ |
|
2657 |
mblk_t *mp, *nmp; |
|
2658 |
int error; |
|
2659 |
||
2660 |
dprintso(so, 1, ("sorecvoob(%p, %p, 0x%x)\n", so, msg, flags)); |
|
2661 |
||
2662 |
/* |
|
2663 |
* There is never any oob data with addresses or control since |
|
2664 |
* the T_EXDATA_IND does not carry any options. |
|
2665 |
*/ |
|
2666 |
msg->msg_controllen = 0; |
|
2667 |
msg->msg_namelen = 0; |
|
2668 |
||
2669 |
mutex_enter(&so->so_lock); |
|
2670 |
ASSERT(so_verify_oobstate(so)); |
|
2671 |
if ((so->so_options & SO_OOBINLINE) || |
|
2672 |
(so->so_state & (SS_OOBPEND|SS_HADOOBDATA)) != SS_OOBPEND) { |
|
2673 |
dprintso(so, 1, ("sorecvoob: inline or data consumed\n")); |
|
2674 |
mutex_exit(&so->so_lock); |
|
2675 |
return (EINVAL); |
|
2676 |
} |
|
2677 |
if (!(so->so_state & SS_HAVEOOBDATA)) { |
|
2678 |
dprintso(so, 1, ("sorecvoob: no data yet\n")); |
|
2679 |
mutex_exit(&so->so_lock); |
|
2680 |
return (EWOULDBLOCK); |
|
2681 |
} |
|
2682 |
ASSERT(so->so_oobmsg != NULL); |
|
2683 |
mp = so->so_oobmsg; |
|
2684 |
if (flags & MSG_PEEK) { |
|
2685 |
/* |
|
2686 |
* Since recv* can not return ENOBUFS we can not use dupmsg. |
|
2687 |
* Instead we revert to the consolidation private |
|
2688 |
* allocb_wait plus bcopy. |
|
2689 |
*/ |
|
2690 |
mblk_t *mp1; |
|
2691 |
||
2692 |
mp1 = allocb_wait(msgdsize(mp), BPRI_MED, STR_NOSIG, NULL); |
|
2693 |
ASSERT(mp1); |
|
2694 |
||
2695 |
while (mp != NULL) { |
|
2696 |
ssize_t size; |
|
2697 |
||
2698 |
size = MBLKL(mp); |
|
2699 |
bcopy(mp->b_rptr, mp1->b_wptr, size); |
|
2700 |
mp1->b_wptr += size; |
|
2701 |
ASSERT(mp1->b_wptr <= mp1->b_datap->db_lim); |
|
2702 |
mp = mp->b_cont; |
|
2703 |
} |
|
2704 |
mp = mp1; |
|
2705 |
} else { |
|
2706 |
/* |
|
2707 |
* Update the state indicating that the data has been consumed. |
|
2708 |
* Keep SS_OOBPEND set until data is consumed past the mark. |
|
2709 |
*/ |
|
2710 |
so->so_oobmsg = NULL; |
|
2711 |
so->so_state ^= SS_HAVEOOBDATA|SS_HADOOBDATA; |
|
2712 |
} |
|
2713 |
dprintso(so, 1, |
|
2714 |
("after recvoob(%p): counts %d/%d state %s\n", |
|
2715 |
so, so->so_oobsigcnt, |
|
2716 |
so->so_oobcnt, pr_state(so->so_state, so->so_mode))); |
|
2717 |
ASSERT(so_verify_oobstate(so)); |
|
2718 |
mutex_exit(&so->so_lock); |
|
2719 |
||
2720 |
error = 0; |
|
2721 |
nmp = mp; |
|
2722 |
while (nmp != NULL && uiop->uio_resid > 0) { |
|
2723 |
ssize_t n = MBLKL(nmp); |
|
2724 |
||
2725 |
n = MIN(n, uiop->uio_resid); |
|
2726 |
if (n > 0) |
|
2727 |
error = uiomove(nmp->b_rptr, n, |
|
2728 |
UIO_READ, uiop); |
|
2729 |
if (error) |
|
2730 |
break; |
|
2731 |
nmp = nmp->b_cont; |
|
2732 |
} |
|
2733 |
freemsg(mp); |
|
2734 |
return (error); |
|
2735 |
} |
|
2736 |
||
2737 |
/* |
|
2738 |
* Called by sotpi_recvmsg when reading a non-zero amount of data. |
|
2739 |
* In addition, the caller typically verifies that there is some |
|
2740 |
* potential state to clear by checking |
|
2741 |
* if (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) |
|
2742 |
* before calling this routine. |
|
2743 |
* Note that such a check can be made without holding so_lock since |
|
2744 |
* sotpi_recvmsg is single-threaded (using SOREADLOCKED) and only sotpi_recvmsg |
|
2745 |
* decrements so_oobsigcnt. |
|
2746 |
* |
|
2747 |
* When data is read *after* the point that all pending |
|
2748 |
* oob data has been consumed the oob indication is cleared. |
|
2749 |
* |
|
2750 |
* This logic keeps select/poll returning POLLRDBAND and |
|
2751 |
* SIOCATMARK returning true until we have read past |
|
2752 |
* the mark. |
|
2753 |
*/ |
|
2754 |
static void |
|
2755 |
sorecv_update_oobstate(struct sonode *so) |
|
2756 |
{ |
|
2757 |
mutex_enter(&so->so_lock); |
|
2758 |
ASSERT(so_verify_oobstate(so)); |
|
2759 |
dprintso(so, 1, |
|
2760 |
("sorecv_update_oobstate: counts %d/%d state %s\n", |
|
2761 |
so->so_oobsigcnt, |
|
2762 |
so->so_oobcnt, pr_state(so->so_state, so->so_mode))); |
|
2763 |
if (so->so_oobsigcnt == 0) { |
|
2764 |
/* No more pending oob indications */ |
|
2765 |
so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK); |
|
2766 |
freemsg(so->so_oobmsg); |
|
2767 |
so->so_oobmsg = NULL; |
|
2768 |
} |
|
2769 |
ASSERT(so_verify_oobstate(so)); |
|
2770 |
mutex_exit(&so->so_lock); |
|
2771 |
} |
|
2772 |
||
2773 |
/* |
|
2774 |
* Handle recv* calls for an so which has NL7C saved recv mblk_t(s). |
|
2775 |
*/ |
|
2776 |
static int |
|
2777 |
nl7c_sorecv(struct sonode *so, mblk_t **rmp, uio_t *uiop, rval_t *rp) |
|
2778 |
{ |
|
2779 |
int error = 0; |
|
2780 |
mblk_t *tmp = NULL; |
|
2781 |
mblk_t *pmp = NULL; |
|
2782 |
mblk_t *nmp = so->so_nl7c_rcv_mp; |
|
2783 |
||
2784 |
ASSERT(nmp != NULL); |
|
2785 |
||
2786 |
while (nmp != NULL && uiop->uio_resid > 0) { |
|
2787 |
ssize_t n; |
|
2788 |
||
2789 |
if (DB_TYPE(nmp) == M_DATA) { |
|
2790 |
/* |
|
2791 |
* We have some data, uiomove up to resid bytes. |
|
2792 |
*/ |
|
2793 |
n = MIN(MBLKL(nmp), uiop->uio_resid); |
|
2794 |
if (n > 0) |
|
2795 |
error = uiomove(nmp->b_rptr, n, UIO_READ, uiop); |
|
2796 |
if (error) |
|
2797 |
break; |
|
2798 |
nmp->b_rptr += n; |
|
2799 |
if (nmp->b_rptr == nmp->b_wptr) { |
|
2800 |
pmp = nmp; |
|
2801 |
nmp = nmp->b_cont; |
|
2802 |
} |
|
2803 |
} else { |
|
2804 |
/* |
|
2805 |
* We only handle data, save for caller to handle. |
|
2806 |
*/ |
|
2807 |
if (pmp != NULL) { |
|
2808 |
pmp->b_cont = nmp->b_cont; |
|
2809 |
} |
|
2810 |
nmp->b_cont = NULL; |
|
2811 |
if (*rmp == NULL) { |
|
2812 |
*rmp = nmp; |
|
2813 |
} else { |
|
2814 |
tmp->b_next = nmp; |
|
2815 |
} |
|
2816 |
nmp = nmp->b_cont; |
|
2817 |
tmp = nmp; |
|
2818 |
} |
|
2819 |
} |
|
2820 |
if (pmp != NULL) { |
|
2821 |
/* Free any mblk_t(s) which we have consumed */ |
|
2822 |
pmp->b_cont = NULL; |
|
2823 |
freemsg(so->so_nl7c_rcv_mp); |
|
2824 |
} |
|
2825 |
if ((so->so_nl7c_rcv_mp = nmp) == NULL) { |
|
2826 |
/* Last mblk_t so return the saved rval from kstrgetmsg() */ |
|
2827 |
rp->r_vals = so->so_nl7c_rcv_rval; |
|
2828 |
so->so_nl7c_rcv_rval = 0; |
|
2829 |
} else { |
|
2830 |
/* More mblk_t(s) to process so no rval to return */ |
|
2831 |
rp->r_vals = 0; |
|
2832 |
} |
|
2833 |
return (error); |
|
2834 |
} |
|
2835 |
||
2836 |
/* |
|
2837 |
* Receive the next message on the queue. |
|
2838 |
* If msg_controllen is non-zero when called the caller is interested in |
|
2839 |
* any received control info (options). |
|
2840 |
* If msg_namelen is non-zero when called the caller is interested in |
|
2841 |
* any received source address. |
|
2842 |
* The routine returns with msg_control and msg_name pointing to |
|
2843 |
* kmem_alloc'ed memory which the caller has to free. |
|
2844 |
*/ |
|
2845 |
int |
|
2846 |
sotpi_recvmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop) |
|
2847 |
{ |
|
2848 |
union T_primitives *tpr; |
|
2849 |
mblk_t *mp; |
|
2850 |
uchar_t pri; |
|
2851 |
int pflag, opflag; |
|
2852 |
void *control; |
|
2853 |
t_uscalar_t controllen; |
|
2854 |
t_uscalar_t namelen; |
|
2855 |
int so_state = so->so_state; /* Snapshot */ |
|
2856 |
ssize_t saved_resid; |
|
2857 |
int error; |
|
2858 |
rval_t rval; |
|
2859 |
int flags; |
|
2860 |
clock_t timout; |
|
2861 |
int first; |
|
2862 |
||
2863 |
flags = msg->msg_flags; |
|
2864 |
msg->msg_flags = 0; |
|
2865 |
||
2866 |
dprintso(so, 1, ("sotpi_recvmsg(%p, %p, 0x%x) state %s err %d\n", |
|
2867 |
so, msg, flags, |
|
2868 |
pr_state(so->so_state, so->so_mode), so->so_error)); |
|
2869 |
||
2870 |
/* |
|
2871 |
* If we are not connected because we have never been connected |
|
2872 |
* we return ENOTCONN. If we have been connected (but are no longer |
|
2873 |
* connected) then SS_CANTRCVMORE is set and we let kstrgetmsg return |
|
2874 |
* the EOF. |
|
2875 |
* |
|
2876 |
* An alternative would be to post an ENOTCONN error in stream head |
|
2877 |
* (read+write) and clear it when we're connected. However, that error |
|
2878 |
* would cause incorrect poll/select behavior! |
|
2879 |
*/ |
|
2880 |
if ((so_state & (SS_ISCONNECTED|SS_CANTRCVMORE)) == 0 && |
|
2881 |
(so->so_mode & SM_CONNREQUIRED)) { |
|
2882 |
return (ENOTCONN); |
|
2883 |
} |
|
2884 |
||
2885 |
/* |
|
2886 |
* Note: SunOS 4.X checks uio_resid == 0 before going to sleep (but |
|
2887 |
* after checking that the read queue is empty) and returns zero. |
|
2888 |
* This implementation will sleep (in kstrgetmsg) even if uio_resid |
|
2889 |
* is zero. |
|
2890 |
*/ |
|
2891 |
||
2892 |
if (flags & MSG_OOB) { |
|
2893 |
/* Check that the transport supports OOB */ |
|
2894 |
if (!(so->so_mode & SM_EXDATA)) |
|
2895 |
return (EOPNOTSUPP); |
|
2896 |
return (sorecvoob(so, msg, uiop, flags)); |
|
2897 |
} |
|
2898 |
||
2899 |
/* |
|
2900 |
* Set msg_controllen and msg_namelen to zero here to make it |
|
2901 |
* simpler in the cases that no control or name is returned. |
|
2902 |
*/ |
|
2903 |
controllen = msg->msg_controllen; |
|
2904 |
namelen = msg->msg_namelen; |
|
2905 |
msg->msg_controllen = 0; |
|
2906 |
msg->msg_namelen = 0; |
|
2907 |
||
2908 |
dprintso(so, 1, ("sotpi_recvmsg: namelen %d controllen %d\n", |
|
2909 |
namelen, controllen)); |
|
2910 |
||
2911 |
/* |
|
2912 |
* If an NL7C enabled socket and not waiting for write data. |
|
2913 |
*/ |
|
2914 |
mutex_enter(&so->so_lock); |
|
2915 |
if ((so->so_nl7c_flags & (NL7C_ENABLED|NL7C_WAITWRITE)) == |
|
2916 |
NL7C_ENABLED) { |
|
2917 |
if (so->so_nl7c_uri) { |
|
2918 |
/* |
|
2919 |
* Close uri processing for a previous request. |
|
2920 |
*/ |
|
2921 |
nl7c_close(so); |
|
2922 |
} |
|
2923 |
if (nl7c_process(so, |
|
2924 |
(so->so_state & (SS_NONBLOCK|SS_NDELAY)), |
|
2925 |
(int)((tcp_t *)so->so_priv)->tcp_mss)) { |
|
2926 |
/* |
|
2927 |
* NL7C has completed processing on the socket, |
|
2928 |
* clear the enabled bit as no further NL7C |
|
2929 |
* processing will be needed. |
|
2930 |
*/ |
|
2931 |
so->so_nl7c_flags = 0; |
|
2932 |
} |
|
2933 |
} |
|
2934 |
||
2935 |
/* |
|
2936 |
* Only one reader is allowed at any given time. This is needed |
|
2937 |
* for T_EXDATA handling and, in the future, MSG_WAITALL. |
|
2938 |
* |
|
2939 |
* This is slightly different that BSD behavior in that it fails with |
|
2940 |
* EWOULDBLOCK when using nonblocking io. In BSD the read queue access |
|
2941 |
* is single-threaded using sblock(), which is dropped while waiting |
|
2942 |
* for data to appear. The difference shows up e.g. if one |
|
2943 |
* file descriptor does not have O_NONBLOCK but a dup'ed file descriptor |
|
2944 |
* does use nonblocking io and different threads are reading each |
|
2945 |
* file descriptor. In BSD there would never be an EWOULDBLOCK error |
|
2946 |
* in this case as long as the read queue doesn't get empty. |
|
2947 |
* In this implementation the thread using nonblocking io can |
|
2948 |
* get an EWOULDBLOCK error due to the blocking thread executing |
|
2949 |
* e.g. in the uiomove in kstrgetmsg. |
|
2950 |
* This difference is not believed to be significant. |
|
2951 |
*/ |
|
2952 |
error = so_lock_read_intr(so, uiop->uio_fmode); /* Set SOREADLOCKED */ |
|
2953 |
mutex_exit(&so->so_lock); |
|
2954 |
if (error) |
|
2955 |
return (error); |
|
2956 |
||
2957 |
/* |
|
2958 |
* Tell kstrgetmsg to not inspect the stream head errors until all |
|
2959 |
* queued data has been consumed. |
|
2960 |
* Use a timeout=-1 to wait forever unless MSG_DONTWAIT is set. |
|
2961 |
* Also, If uio_fmode indicates nonblocking kstrgetmsg will not block. |
|
2962 |
* |
|
2963 |
* MSG_WAITALL only applies to M_DATA and T_DATA_IND messages and |
|
2964 |
* to T_OPTDATA_IND that do not contain any user-visible control msg. |
|
2965 |
* Note that MSG_WAITALL set with MSG_PEEK is a noop. |
|
2966 |
*/ |
|
2967 |
pflag = MSG_ANY | MSG_DELAYERROR; |
|
2968 |
if (flags & MSG_PEEK) { |
|
2969 |
pflag |= MSG_IPEEK; |
|
2970 |
flags &= ~MSG_WAITALL; |
|
2971 |
} |
|
2972 |
if (so->so_mode & SM_ATOMIC) |
|
2973 |
pflag |= MSG_DISCARDTAIL; |
|
2974 |
||
2975 |
if (flags & MSG_DONTWAIT) |
|
2976 |
timout = 0; |
|
2977 |
else |
|
2978 |
timout = -1; |
|
2979 |
opflag = pflag; |
|
2980 |
first = 1; |
|
2981 |
||
2982 |
/* |
|
2983 |
* If so saved NL7C rcv mblk_t(s) uiomove them first |
|
2984 |
* else get'm from the streamhead. |
|
2985 |
*/ |
|
2986 |
retry: |
|
2987 |
saved_resid = uiop->uio_resid; |
|
2988 |
pri = 0; |
|
2989 |
mp = NULL; |
|
2990 |
if (so->so_nl7c_rcv_mp != NULL) { |
|
2991 |
error = nl7c_sorecv(so, &mp, uiop, &rval); |
|
2992 |
} else { |
|
2993 |
error = kstrgetmsg(SOTOV(so), &mp, uiop, &pri, &pflag, |
|
2994 |
timout, &rval); |
|
2995 |
} |
|
2996 |
if (error) { |
|
2997 |
switch (error) { |
|
2998 |
case EINTR: |
|
2999 |
case EWOULDBLOCK: |
|
3000 |
if (!first) |
|
3001 |
error = 0; |
|
3002 |
break; |
|
3003 |
case ETIME: |
|
3004 |
/* Returned from kstrgetmsg when timeout expires */ |
|
3005 |
if (!first) |
|
3006 |
error = 0; |
|
3007 |
else |
|
3008 |
error = EWOULDBLOCK; |
|
3009 |
break; |
|
3010 |
default: |
|
3011 |
eprintsoline(so, error); |
|
3012 |
break; |
|
3013 |
} |
|
3014 |
mutex_enter(&so->so_lock); |
|
3015 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3016 |
mutex_exit(&so->so_lock); |
|
3017 |
return (error); |
|
3018 |
} |
|
3019 |
/* |
|
3020 |
* For datagrams the MOREDATA flag is used to set MSG_TRUNC. |
|
3021 |
* For non-datagrams MOREDATA is used to set MSG_EOR. |
|
3022 |
*/ |
|
3023 |
ASSERT(!(rval.r_val1 & MORECTL)); |
|
3024 |
if ((rval.r_val1 & MOREDATA) && (so->so_mode & SM_ATOMIC)) |
|
3025 |
msg->msg_flags |= MSG_TRUNC; |
|
3026 |
||
3027 |
if (mp == NULL) { |
|
3028 |
dprintso(so, 1, ("sotpi_recvmsg: got M_DATA\n")); |
|
3029 |
/* |
|
3030 |
* 4.3BSD and 4.4BSD clears the mark when peeking across it. |
|
3031 |
* The draft Posix socket spec states that the mark should |
|
3032 |
* not be cleared when peeking. We follow the latter. |
|
3033 |
*/ |
|
3034 |
if ((so->so_state & |
|
3035 |
(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) && |
|
3036 |
(uiop->uio_resid != saved_resid) && |
|
3037 |
!(flags & MSG_PEEK)) { |
|
3038 |
sorecv_update_oobstate(so); |
|
3039 |
} |
|
3040 |
||
3041 |
mutex_enter(&so->so_lock); |
|
3042 |
/* Set MSG_EOR based on MOREDATA */ |
|
3043 |
if (!(rval.r_val1 & MOREDATA)) { |
|
3044 |
if (so->so_state & SS_SAVEDEOR) { |
|
3045 |
msg->msg_flags |= MSG_EOR; |
|
3046 |
so->so_state &= ~SS_SAVEDEOR; |
|
3047 |
} |
|
3048 |
} |
|
3049 |
/* |
|
3050 |
* If some data was received (i.e. not EOF) and the |
|
3051 |
* read/recv* has not been satisfied wait for some more. |
|
3052 |
*/ |
|
3053 |
if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) && |
|
3054 |
uiop->uio_resid != saved_resid && uiop->uio_resid > 0) { |
|
3055 |
mutex_exit(&so->so_lock); |
|
3056 |
first = 0; |
|
3057 |
pflag = opflag | MSG_NOMARK; |
|
3058 |
goto retry; |
|
3059 |
} |
|
3060 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3061 |
mutex_exit(&so->so_lock); |
|
3062 |
return (0); |
|
3063 |
} |
|
3064 |
||
3065 |
/* strsock_proto has already verified length and alignment */ |
|
3066 |
tpr = (union T_primitives *)mp->b_rptr; |
|
3067 |
dprintso(so, 1, ("sotpi_recvmsg: type %d\n", tpr->type)); |
|
3068 |
||
3069 |
switch (tpr->type) { |
|
3070 |
case T_DATA_IND: { |
|
3071 |
if ((so->so_state & |
|
3072 |
(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) && |
|
3073 |
(uiop->uio_resid != saved_resid) && |
|
3074 |
!(flags & MSG_PEEK)) { |
|
3075 |
sorecv_update_oobstate(so); |
|
3076 |
} |
|
3077 |
||
3078 |
/* |
|
3079 |
* Set msg_flags to MSG_EOR based on |
|
3080 |
* MORE_flag and MOREDATA. |
|
3081 |
*/ |
|
3082 |
mutex_enter(&so->so_lock); |
|
3083 |
so->so_state &= ~SS_SAVEDEOR; |
|
3084 |
if (!(tpr->data_ind.MORE_flag & 1)) { |
|
3085 |
if (!(rval.r_val1 & MOREDATA)) |
|
3086 |
msg->msg_flags |= MSG_EOR; |
|
3087 |
else |
|
3088 |
so->so_state |= SS_SAVEDEOR; |
|
3089 |
} |
|
3090 |
freemsg(mp); |
|
3091 |
/* |
|
3092 |
* If some data was received (i.e. not EOF) and the |
|
3093 |
* read/recv* has not been satisfied wait for some more. |
|
3094 |
*/ |
|
3095 |
if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) && |
|
3096 |
uiop->uio_resid != saved_resid && uiop->uio_resid > 0) { |
|
3097 |
mutex_exit(&so->so_lock); |
|
3098 |
first = 0; |
|
3099 |
pflag = opflag | MSG_NOMARK; |
|
3100 |
goto retry; |
|
3101 |
} |
|
3102 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3103 |
mutex_exit(&so->so_lock); |
|
3104 |
return (0); |
|
3105 |
} |
|
3106 |
case T_UNITDATA_IND: { |
|
3107 |
void *addr; |
|
3108 |
t_uscalar_t addrlen; |
|
3109 |
void *abuf; |
|
3110 |
t_uscalar_t optlen; |
|
3111 |
void *opt; |
|
3112 |
||
3113 |
if ((so->so_state & |
|
3114 |
(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) && |
|
3115 |
(uiop->uio_resid != saved_resid) && |
|
3116 |
!(flags & MSG_PEEK)) { |
|
3117 |
sorecv_update_oobstate(so); |
|
3118 |
} |
|
3119 |
||
3120 |
if (namelen != 0) { |
|
3121 |
/* Caller wants source address */ |
|
3122 |
addrlen = tpr->unitdata_ind.SRC_length; |
|
3123 |
addr = sogetoff(mp, |
|
3124 |
tpr->unitdata_ind.SRC_offset, |
|
3125 |
addrlen, 1); |
|
3126 |
if (addr == NULL) { |
|
3127 |
freemsg(mp); |
|
3128 |
error = EPROTO; |
|
3129 |
eprintsoline(so, error); |
|
3130 |
goto err; |
|
3131 |
} |
|
3132 |
if (so->so_family == AF_UNIX) { |
|
3133 |
/* |
|
3134 |
* Can not use the transport level address. |
|
3135 |
* If there is a SO_SRCADDR option carrying |
|
3136 |
* the socket level address it will be |
|
3137 |
* extracted below. |
|
3138 |
*/ |
|
3139 |
addr = NULL; |
|
3140 |
addrlen = 0; |
|
3141 |
} |
|
3142 |
} |
|
3143 |
optlen = tpr->unitdata_ind.OPT_length; |
|
3144 |
if (optlen != 0) { |
|
3145 |
t_uscalar_t ncontrollen; |
|
3146 |
||
3147 |
/* |
|
3148 |
* Extract any source address option. |
|
3149 |
* Determine how large cmsg buffer is needed. |
|
3150 |
*/ |
|
3151 |
opt = sogetoff(mp, |
|
3152 |
tpr->unitdata_ind.OPT_offset, |
|
3153 |
optlen, __TPI_ALIGN_SIZE); |
|
3154 |
||
3155 |
if (opt == NULL) { |
|
3156 |
freemsg(mp); |
|
3157 |
error = EPROTO; |
|
3158 |
eprintsoline(so, error); |
|
3159 |
goto err; |
|
3160 |
} |
|
3161 |
if (so->so_family == AF_UNIX) |
|
3162 |
so_getopt_srcaddr(opt, optlen, &addr, &addrlen); |
|
3163 |
ncontrollen = so_cmsglen(mp, opt, optlen, |
|
3164 |
!(flags & MSG_XPG4_2)); |
|
3165 |
if (controllen != 0) |
|
3166 |
controllen = ncontrollen; |
|
3167 |
else if (ncontrollen != 0) |
|
3168 |
msg->msg_flags |= MSG_CTRUNC; |
|
3169 |
} else { |
|
3170 |
controllen = 0; |
|
3171 |
} |
|
3172 |
||
3173 |
if (namelen != 0) { |
|
3174 |
/* |
|
3175 |
* Return address to caller. |
|
3176 |
* Caller handles truncation if length |
|
3177 |
* exceeds msg_namelen. |
|
3178 |
* NOTE: AF_UNIX NUL termination is ensured by |
|
3179 |
* the sender's copyin_name(). |
|
3180 |
*/ |
|
3181 |
abuf = kmem_alloc(addrlen, KM_SLEEP); |
|
3182 |
||
3183 |
bcopy(addr, abuf, addrlen); |
|
3184 |
msg->msg_name = abuf; |
|
3185 |
msg->msg_namelen = addrlen; |
|
3186 |
} |
|
3187 |
||
3188 |
if (controllen != 0) { |
|
3189 |
/* |
|
3190 |
* Return control msg to caller. |
|
3191 |
* Caller handles truncation if length |
|
3192 |
* exceeds msg_controllen. |
|
3193 |
*/ |
|
3194 |
control = kmem_alloc(controllen, KM_SLEEP); |
|
3195 |
||
3196 |
error = so_opt2cmsg(mp, opt, optlen, |
|
3197 |
!(flags & MSG_XPG4_2), |
|
3198 |
control, controllen); |
|
3199 |
if (error) { |
|
3200 |
freemsg(mp); |
|
3201 |
if (msg->msg_namelen != 0) |
|
3202 |
kmem_free(msg->msg_name, |
|
3203 |
msg->msg_namelen); |
|
3204 |
kmem_free(control, controllen); |
|
3205 |
eprintsoline(so, error); |
|
3206 |
goto err; |
|
3207 |
} |
|
3208 |
msg->msg_control = control; |
|
3209 |
msg->msg_controllen = controllen; |
|
3210 |
} |
|
3211 |
||
3212 |
freemsg(mp); |
|
3213 |
mutex_enter(&so->so_lock); |
|
3214 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3215 |
mutex_exit(&so->so_lock); |
|
3216 |
return (0); |
|
3217 |
} |
|
3218 |
case T_OPTDATA_IND: { |
|
3219 |
struct T_optdata_req *tdr; |
|
3220 |
void *opt; |
|
3221 |
t_uscalar_t optlen; |
|
3222 |
||
3223 |
if ((so->so_state & |
|
3224 |
(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) && |
|
3225 |
(uiop->uio_resid != saved_resid) && |
|
3226 |
!(flags & MSG_PEEK)) { |
|
3227 |
sorecv_update_oobstate(so); |
|
3228 |
} |
|
3229 |
||
3230 |
tdr = (struct T_optdata_req *)mp->b_rptr; |
|
3231 |
optlen = tdr->OPT_length; |
|
3232 |
if (optlen != 0) { |
|
3233 |
t_uscalar_t ncontrollen; |
|
3234 |
/* |
|
3235 |
* Determine how large cmsg buffer is needed. |
|
3236 |
*/ |
|
3237 |
opt = sogetoff(mp, |
|
3238 |
tpr->optdata_ind.OPT_offset, |
|
3239 |
optlen, __TPI_ALIGN_SIZE); |
|
3240 |
||
3241 |
if (opt == NULL) { |
|
3242 |
freemsg(mp); |
|
3243 |
error = EPROTO; |
|
3244 |
eprintsoline(so, error); |
|
3245 |
goto err; |
|
3246 |
} |
|
3247 |
||
3248 |
ncontrollen = so_cmsglen(mp, opt, optlen, |
|
3249 |
!(flags & MSG_XPG4_2)); |
|
3250 |
if (controllen != 0) |
|
3251 |
controllen = ncontrollen; |
|
3252 |
else if (ncontrollen != 0) |
|
3253 |
msg->msg_flags |= MSG_CTRUNC; |
|
3254 |
} else { |
|
3255 |
controllen = 0; |
|
3256 |
} |
|
3257 |
||
3258 |
if (controllen != 0) { |
|
3259 |
/* |
|
3260 |
* Return control msg to caller. |
|
3261 |
* Caller handles truncation if length |
|
3262 |
* exceeds msg_controllen. |
|
3263 |
*/ |
|
3264 |
control = kmem_alloc(controllen, KM_SLEEP); |
|
3265 |
||
3266 |
error = so_opt2cmsg(mp, opt, optlen, |
|
3267 |
!(flags & MSG_XPG4_2), |
|
3268 |
control, controllen); |
|
3269 |
if (error) { |
|
3270 |
freemsg(mp); |
|
3271 |
kmem_free(control, controllen); |
|
3272 |
eprintsoline(so, error); |
|
3273 |
goto err; |
|
3274 |
} |
|
3275 |
msg->msg_control = control; |
|
3276 |
msg->msg_controllen = controllen; |
|
3277 |
} |
|
3278 |
||
3279 |
/* |
|
3280 |
* Set msg_flags to MSG_EOR based on |
|
3281 |
* DATA_flag and MOREDATA. |
|
3282 |
*/ |
|
3283 |
mutex_enter(&so->so_lock); |
|
3284 |
so->so_state &= ~SS_SAVEDEOR; |
|
3285 |
if (!(tpr->data_ind.MORE_flag & 1)) { |
|
3286 |
if (!(rval.r_val1 & MOREDATA)) |
|
3287 |
msg->msg_flags |= MSG_EOR; |
|
3288 |
else |
|
3289 |
so->so_state |= SS_SAVEDEOR; |
|
3290 |
} |
|
3291 |
freemsg(mp); |
|
3292 |
/* |
|
3293 |
* If some data was received (i.e. not EOF) and the |
|
3294 |
* read/recv* has not been satisfied wait for some more. |
|
3295 |
* Not possible to wait if control info was received. |
|
3296 |
*/ |
|
3297 |
if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) && |
|
3298 |
controllen == 0 && |
|
3299 |
uiop->uio_resid != saved_resid && uiop->uio_resid > 0) { |
|
3300 |
mutex_exit(&so->so_lock); |
|
3301 |
first = 0; |
|
3302 |
pflag = opflag | MSG_NOMARK; |
|
3303 |
goto retry; |
|
3304 |
} |
|
3305 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3306 |
mutex_exit(&so->so_lock); |
|
3307 |
return (0); |
|
3308 |
} |
|
3309 |
case T_EXDATA_IND: { |
|
3310 |
dprintso(so, 1, |
|
3311 |
("sotpi_recvmsg: EXDATA_IND counts %d/%d consumed %ld " |
|
3312 |
"state %s\n", |
|
3313 |
so->so_oobsigcnt, so->so_oobcnt, |
|
3314 |
saved_resid - uiop->uio_resid, |
|
3315 |
pr_state(so->so_state, so->so_mode))); |
|
3316 |
/* |
|
3317 |
* kstrgetmsg handles MSGMARK so there is nothing to |
|
3318 |
* inspect in the T_EXDATA_IND. |
|
3319 |
* strsock_proto makes the stream head queue the T_EXDATA_IND |
|
3320 |
* as a separate message with no M_DATA component. Furthermore, |
|
3321 |
* the stream head does not consolidate M_DATA messages onto |
|
3322 |
* an MSGMARK'ed message ensuring that the T_EXDATA_IND |
|
3323 |
* remains a message by itself. This is needed since MSGMARK |
|
3324 |
* marks both the whole message as well as the last byte |
|
3325 |
* of the message. |
|
3326 |
*/ |
|
3327 |
freemsg(mp); |
|
3328 |
ASSERT(uiop->uio_resid == saved_resid); /* No data */ |
|
3329 |
if (flags & MSG_PEEK) { |
|
3330 |
/* |
|
3331 |
* Even though we are peeking we consume the |
|
3332 |
* T_EXDATA_IND thereby moving the mark information |
|
3333 |
* to SS_RCVATMARK. Then the oob code below will |
|
3334 |
* retry the peeking kstrgetmsg. |
|
3335 |
* Note that the stream head read queue is |
|
3336 |
* never flushed without holding SOREADLOCKED |
|
3337 |
* thus the T_EXDATA_IND can not disappear |
|
3338 |
* underneath us. |
|
3339 |
*/ |
|
3340 |
dprintso(so, 1, |
|
3341 |
("sotpi_recvmsg: consume EXDATA_IND " |
|
3342 |
"counts %d/%d state %s\n", |
|
3343 |
so->so_oobsigcnt, |
|
3344 |
so->so_oobcnt, |
|
3345 |
pr_state(so->so_state, so->so_mode))); |
|
3346 |
||
3347 |
pflag = MSG_ANY | MSG_DELAYERROR; |
|
3348 |
if (so->so_mode & SM_ATOMIC) |
|
3349 |
pflag |= MSG_DISCARDTAIL; |
|
3350 |
||
3351 |
pri = 0; |
|
3352 |
mp = NULL; |
|
3353 |
||
3354 |
error = kstrgetmsg(SOTOV(so), &mp, uiop, |
|
3355 |
&pri, &pflag, (clock_t)-1, &rval); |
|
3356 |
ASSERT(uiop->uio_resid == saved_resid); |
|
3357 |
||
3358 |
if (error) { |
|
3359 |
#ifdef SOCK_DEBUG |
|
3360 |
if (error != EWOULDBLOCK && error != EINTR) { |
|
3361 |
eprintsoline(so, error); |
|
3362 |
} |
|
3363 |
#endif /* SOCK_DEBUG */ |
|
3364 |
mutex_enter(&so->so_lock); |
|
3365 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3366 |
mutex_exit(&so->so_lock); |
|
3367 |
return (error); |
|
3368 |
} |
|
3369 |
ASSERT(mp); |
|
3370 |
tpr = (union T_primitives *)mp->b_rptr; |
|
3371 |
ASSERT(tpr->type == T_EXDATA_IND); |
|
3372 |
freemsg(mp); |
|
3373 |
} /* end "if (flags & MSG_PEEK)" */ |
|
3374 |
||
3375 |
/* |
|
3376 |
* Decrement the number of queued and pending oob. |
|
3377 |
* |
|
3378 |
* SS_RCVATMARK is cleared when we read past a mark. |
|
3379 |
* SS_HAVEOOBDATA is cleared when we've read past the |
|
3380 |
* last mark. |
|
3381 |
* SS_OOBPEND is cleared if we've read past the last |
|
3382 |
* mark and no (new) SIGURG has been posted. |
|
3383 |
*/ |
|
3384 |
mutex_enter(&so->so_lock); |
|
3385 |
ASSERT(so_verify_oobstate(so)); |
|
3386 |
ASSERT(so->so_oobsigcnt >= so->so_oobcnt); |
|
3387 |
ASSERT(so->so_oobsigcnt > 0); |
|
3388 |
so->so_oobsigcnt--; |
|
3389 |
ASSERT(so->so_oobcnt > 0); |
|
3390 |
so->so_oobcnt--; |
|
3391 |
/* |
|
3392 |
* Since the T_EXDATA_IND has been removed from the stream |
|
3393 |
* head, but we have not read data past the mark, |
|
3394 |
* sockfs needs to track that the socket is still at the mark. |
|
3395 |
* |
|
3396 |
* Since no data was received call kstrgetmsg again to wait |
|
3397 |
* for data. |
|
3398 |
*/ |
|
3399 |
so->so_state |= SS_RCVATMARK; |
|
3400 |
mutex_exit(&so->so_lock); |
|
3401 |
dprintso(so, 1, |
|
3402 |
("sotpi_recvmsg: retry EXDATA_IND counts %d/%d state %s\n", |
|
3403 |
so->so_oobsigcnt, so->so_oobcnt, |
|
3404 |
pr_state(so->so_state, so->so_mode))); |
|
3405 |
pflag = opflag; |
|
3406 |
goto retry; |
|
3407 |
} |
|
3408 |
default: |
|
3409 |
ASSERT(0); |
|
3410 |
freemsg(mp); |
|
3411 |
error = EPROTO; |
|
3412 |
eprintsoline(so, error); |
|
3413 |
goto err; |
|
3414 |
} |
|
3415 |
/* NOTREACHED */ |
|
3416 |
err: |
|
3417 |
mutex_enter(&so->so_lock); |
|
3418 |
so_unlock_read(so); /* Clear SOREADLOCKED */ |
|
3419 |
mutex_exit(&so->so_lock); |
|
3420 |
return (error); |
|
3421 |
} |
|
3422 |
||
3423 |
/* |
|
3424 |
* Sending data with options on a datagram socket. |
|
3425 |
* Assumes caller has verified that SS_ISBOUND etc. are set. |
|
3426 |
*/ |
|
3427 |
static int |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3428 |
sosend_dgramcmsg(struct sonode *so, struct sockaddr *name, socklen_t namelen, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3429 |
struct uio *uiop, void *control, t_uscalar_t controllen, int flags) |
0 | 3430 |
{ |
3431 |
struct T_unitdata_req tudr; |
|
3432 |
mblk_t *mp; |
|
3433 |
int error; |
|
3434 |
void *addr; |
|
3435 |
socklen_t addrlen; |
|
3436 |
void *src; |
|
3437 |
socklen_t srclen; |
|
3438 |
ssize_t len; |
|
3439 |
int size; |
|
3440 |
struct T_opthdr toh; |
|
3441 |
struct fdbuf *fdbuf; |
|
3442 |
t_uscalar_t optlen; |
|
3443 |
void *fds; |
|
3444 |
int fdlen; |
|
3445 |
||
3446 |
ASSERT(name && namelen); |
|
3447 |
ASSERT(control && controllen); |
|
3448 |
||
3449 |
len = uiop->uio_resid; |
|
3450 |
if (len > (ssize_t)so->so_tidu_size) { |
|
3451 |
return (EMSGSIZE); |
|
3452 |
} |
|
3453 |
||
3454 |
/* |
|
3455 |
* For AF_UNIX the destination address is translated to an internal |
|
3456 |
* name and the source address is passed as an option. |
|
3457 |
* Also, file descriptors are passed as file pointers in an |
|
3458 |
* option. |
|
3459 |
*/ |
|
3460 |
||
3461 |
/* |
|
3462 |
* Length and family checks. |
|
3463 |
*/ |
|
3464 |
error = so_addr_verify(so, name, namelen); |
|
3465 |
if (error) { |
|
3466 |
eprintsoline(so, error); |
|
3467 |
return (error); |
|
3468 |
} |
|
3469 |
if (so->so_family == AF_UNIX) { |
|
3470 |
if (so->so_state & SS_FADDR_NOXLATE) { |
|
3471 |
/* |
|
3472 |
* Already have a transport internal address. Do not |
|
3473 |
* pass any (transport internal) source address. |
|
3474 |
*/ |
|
3475 |
addr = name; |
|
3476 |
addrlen = namelen; |
|
3477 |
src = NULL; |
|
3478 |
srclen = 0; |
|
3479 |
} else { |
|
3480 |
/* |
|
3481 |
* Pass the sockaddr_un source address as an option |
|
3482 |
* and translate the remote address. |
|
3483 |
* |
|
3484 |
* Note that this code does not prevent so_laddr_sa |
|
3485 |
* from changing while it is being used. Thus |
|
3486 |
* if an unbind+bind occurs concurrently with this |
|
3487 |
* send the peer might see a partially new and a |
|
3488 |
* partially old "from" address. |
|
3489 |
*/ |
|
3490 |
src = so->so_laddr_sa; |
|
3491 |
srclen = (t_uscalar_t)so->so_laddr_len; |
|
3492 |
dprintso(so, 1, |
|
3493 |
("sosend_dgramcmsg UNIX: srclen %d, src %p\n", |
|
3494 |
srclen, src)); |
|
3495 |
error = so_ux_addr_xlate(so, name, namelen, |
|
3496 |
(flags & MSG_XPG4_2), |
|
3497 |
&addr, &addrlen); |
|
3498 |
if (error) { |
|
3499 |
eprintsoline(so, error); |
|
3500 |
return (error); |
|
3501 |
} |
|
3502 |
} |
|
3503 |
} else { |
|
3504 |
addr = name; |
|
3505 |
addrlen = namelen; |
|
3506 |
src = NULL; |
|
3507 |
srclen = 0; |
|
3508 |
} |
|
3509 |
optlen = so_optlen(control, controllen, |
|
3510 |
!(flags & MSG_XPG4_2)); |
|
3511 |
tudr.PRIM_type = T_UNITDATA_REQ; |
|
3512 |
tudr.DEST_length = addrlen; |
|
3513 |
tudr.DEST_offset = (t_scalar_t)sizeof (tudr); |
|
3514 |
if (srclen != 0) |
|
3515 |
tudr.OPT_length = (t_scalar_t)(optlen + sizeof (toh) + |
|
3516 |
_TPI_ALIGN_TOPT(srclen)); |
|
3517 |
else |
|
3518 |
tudr.OPT_length = optlen; |
|
3519 |
tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) + |
|
3520 |
_TPI_ALIGN_TOPT(addrlen)); |
|
3521 |
||
3522 |
size = tudr.OPT_offset + tudr.OPT_length; |
|
3523 |
||
3524 |
/* |
|
3525 |
* File descriptors only when SM_FDPASSING set. |
|
3526 |
*/ |
|
3527 |
error = so_getfdopt(control, controllen, |
|
3528 |
!(flags & MSG_XPG4_2), &fds, &fdlen); |
|
3529 |
if (error) |
|
3530 |
return (error); |
|
3531 |
if (fdlen != -1) { |
|
3532 |
if (!(so->so_mode & SM_FDPASSING)) |
|
3533 |
return (EOPNOTSUPP); |
|
3534 |
||
3535 |
error = fdbuf_create(fds, fdlen, &fdbuf); |
|
3536 |
if (error) |
|
3537 |
return (error); |
|
3538 |
mp = fdbuf_allocmsg(size, fdbuf); |
|
3539 |
} else { |
|
3540 |
mp = soallocproto(size, _ALLOC_INTR); |
|
455 | 3541 |
if (mp == NULL) { |
3542 |
/* |
|
3543 |
* Caught a signal waiting for memory. |
|
3544 |
* Let send* return EINTR. |
|
3545 |
*/ |
|
3546 |
return (EINTR); |
|
3547 |
} |
|
0 | 3548 |
} |
3549 |
soappendmsg(mp, &tudr, sizeof (tudr)); |
|
3550 |
soappendmsg(mp, addr, addrlen); |
|
3551 |
mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen; |
|
3552 |
||
3553 |
if (fdlen != -1) { |
|
3554 |
ASSERT(fdbuf != NULL); |
|
3555 |
toh.level = SOL_SOCKET; |
|
3556 |
toh.name = SO_FILEP; |
|
3557 |
toh.len = fdbuf->fd_size + |
|
3558 |
(t_uscalar_t)sizeof (struct T_opthdr); |
|
3559 |
toh.status = 0; |
|
3560 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
3561 |
soappendmsg(mp, fdbuf, fdbuf->fd_size); |
|
3562 |
ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr)); |
|
3563 |
} |
|
3564 |
if (srclen != 0) { |
|
3565 |
/* |
|
3566 |
* There is a AF_UNIX sockaddr_un to include as a source |
|
3567 |
* address option. |
|
3568 |
*/ |
|
3569 |
toh.level = SOL_SOCKET; |
|
3570 |
toh.name = SO_SRCADDR; |
|
3571 |
toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr)); |
|
3572 |
toh.status = 0; |
|
3573 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
3574 |
soappendmsg(mp, src, srclen); |
|
3575 |
mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen; |
|
3576 |
ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr)); |
|
3577 |
} |
|
3578 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
3579 |
so_cmsg2opt(control, controllen, !(flags & MSG_XPG4_2), mp); |
|
3580 |
/* At most 3 bytes left in the message */ |
|
3581 |
ASSERT(MBLKL(mp) > (ssize_t)(size - __TPI_ALIGN_SIZE)); |
|
3582 |
ASSERT(MBLKL(mp) <= (ssize_t)size); |
|
3583 |
||
3584 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
3585 |
#ifdef C2_AUDIT |
|
3586 |
if (audit_active) |
|
3587 |
audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0); |
|
3588 |
#endif /* C2_AUDIT */ |
|
3589 |
||
3590 |
error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0); |
|
3591 |
#ifdef SOCK_DEBUG |
|
3592 |
if (error) { |
|
3593 |
eprintsoline(so, error); |
|
3594 |
} |
|
3595 |
#endif /* SOCK_DEBUG */ |
|
3596 |
return (error); |
|
3597 |
} |
|
3598 |
||
3599 |
/* |
|
3600 |
* Sending data with options on a connected stream socket. |
|
3601 |
* Assumes caller has verified that SS_ISCONNECTED is set. |
|
3602 |
*/ |
|
3603 |
static int |
|
3604 |
sosend_svccmsg(struct sonode *so, |
|
3605 |
struct uio *uiop, |
|
3606 |
int more, |
|
3607 |
void *control, |
|
3608 |
t_uscalar_t controllen, |
|
3609 |
int flags) |
|
3610 |
{ |
|
3611 |
struct T_optdata_req tdr; |
|
3612 |
mblk_t *mp; |
|
3613 |
int error; |
|
3614 |
ssize_t iosize; |
|
3615 |
int first = 1; |
|
3616 |
int size; |
|
3617 |
struct fdbuf *fdbuf; |
|
3618 |
t_uscalar_t optlen; |
|
3619 |
void *fds; |
|
3620 |
int fdlen; |
|
3621 |
struct T_opthdr toh; |
|
3622 |
||
3623 |
dprintso(so, 1, |
|
3624 |
("sosend_svccmsg: resid %ld bytes\n", uiop->uio_resid)); |
|
3625 |
||
3626 |
/* |
|
3627 |
* Has to be bound and connected. However, since no locks are |
|
3628 |
* held the state could have changed after sotpi_sendmsg checked it |
|
3629 |
* thus it is not possible to ASSERT on the state. |
|
3630 |
*/ |
|
3631 |
||
3632 |
/* Options on connection-oriented only when SM_OPTDATA set. */ |
|
3633 |
if (!(so->so_mode & SM_OPTDATA)) |
|
3634 |
return (EOPNOTSUPP); |
|
3635 |
||
3636 |
do { |
|
3637 |
/* |
|
3638 |
* Set the MORE flag if uio_resid does not fit in this |
|
3639 |
* message or if the caller passed in "more". |
|
3640 |
* Error for transports with zero tidu_size. |
|
3641 |
*/ |
|
3642 |
tdr.PRIM_type = T_OPTDATA_REQ; |
|
3643 |
iosize = so->so_tidu_size; |
|
3644 |
if (iosize <= 0) |
|
3645 |
return (EMSGSIZE); |
|
3646 |
if (uiop->uio_resid > iosize) { |
|
3647 |
tdr.DATA_flag = 1; |
|
3648 |
} else { |
|
3649 |
if (more) |
|
3650 |
tdr.DATA_flag = 1; |
|
3651 |
else |
|
3652 |
tdr.DATA_flag = 0; |
|
3653 |
iosize = uiop->uio_resid; |
|
3654 |
} |
|
3655 |
dprintso(so, 1, ("sosend_svccmsg: sending %d, %ld bytes\n", |
|
3656 |
tdr.DATA_flag, iosize)); |
|
3657 |
||
3658 |
optlen = so_optlen(control, controllen, !(flags & MSG_XPG4_2)); |
|
3659 |
tdr.OPT_length = optlen; |
|
3660 |
tdr.OPT_offset = (t_scalar_t)sizeof (tdr); |
|
3661 |
||
3662 |
size = (int)sizeof (tdr) + optlen; |
|
3663 |
/* |
|
3664 |
* File descriptors only when SM_FDPASSING set. |
|
3665 |
*/ |
|
3666 |
error = so_getfdopt(control, controllen, |
|
3667 |
!(flags & MSG_XPG4_2), &fds, &fdlen); |
|
3668 |
if (error) |
|
3669 |
return (error); |
|
3670 |
if (fdlen != -1) { |
|
3671 |
if (!(so->so_mode & SM_FDPASSING)) |
|
3672 |
return (EOPNOTSUPP); |
|
3673 |
||
3674 |
error = fdbuf_create(fds, fdlen, &fdbuf); |
|
3675 |
if (error) |
|
3676 |
return (error); |
|
3677 |
mp = fdbuf_allocmsg(size, fdbuf); |
|
3678 |
} else { |
|
3679 |
mp = soallocproto(size, _ALLOC_INTR); |
|
455 | 3680 |
if (mp == NULL) { |
3681 |
/* |
|
3682 |
* Caught a signal waiting for memory. |
|
3683 |
* Let send* return EINTR. |
|
3684 |
*/ |
|
3685 |
return (first ? EINTR : 0); |
|
3686 |
} |
|
0 | 3687 |
} |
3688 |
soappendmsg(mp, &tdr, sizeof (tdr)); |
|
3689 |
||
3690 |
if (fdlen != -1) { |
|
3691 |
ASSERT(fdbuf != NULL); |
|
3692 |
toh.level = SOL_SOCKET; |
|
3693 |
toh.name = SO_FILEP; |
|
3694 |
toh.len = fdbuf->fd_size + |
|
3695 |
(t_uscalar_t)sizeof (struct T_opthdr); |
|
3696 |
toh.status = 0; |
|
3697 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
3698 |
soappendmsg(mp, fdbuf, fdbuf->fd_size); |
|
3699 |
ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr)); |
|
3700 |
} |
|
3701 |
so_cmsg2opt(control, controllen, !(flags & MSG_XPG4_2), mp); |
|
3702 |
/* At most 3 bytes left in the message */ |
|
3703 |
ASSERT(MBLKL(mp) > (ssize_t)(size - __TPI_ALIGN_SIZE)); |
|
3704 |
ASSERT(MBLKL(mp) <= (ssize_t)size); |
|
3705 |
||
3706 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
3707 |
||
3708 |
error = kstrputmsg(SOTOV(so), mp, uiop, iosize, |
|
3709 |
0, MSG_BAND, 0); |
|
3710 |
if (error) { |
|
3711 |
if (!first && error == EWOULDBLOCK) |
|
3712 |
return (0); |
|
3713 |
eprintsoline(so, error); |
|
3714 |
return (error); |
|
3715 |
} |
|
3716 |
control = NULL; |
|
3717 |
first = 0; |
|
3718 |
if (uiop->uio_resid > 0) { |
|
3719 |
/* |
|
3720 |
* Recheck for fatal errors. Fail write even though |
|
3721 |
* some data have been written. This is consistent |
|
3722 |
* with strwrite semantics and BSD sockets semantics. |
|
3723 |
*/ |
|
3724 |
if (so->so_state & SS_CANTSENDMORE) { |
|
3725 |
tsignal(curthread, SIGPIPE); |
|
3726 |
eprintsoline(so, error); |
|
3727 |
return (EPIPE); |
|
3728 |
} |
|
3729 |
if (so->so_error != 0) { |
|
3730 |
mutex_enter(&so->so_lock); |
|
3731 |
error = sogeterr(so); |
|
3732 |
mutex_exit(&so->so_lock); |
|
3733 |
if (error != 0) { |
|
3734 |
eprintsoline(so, error); |
|
3735 |
return (error); |
|
3736 |
} |
|
3737 |
} |
|
3738 |
} |
|
3739 |
} while (uiop->uio_resid > 0); |
|
3740 |
return (0); |
|
3741 |
} |
|
3742 |
||
3743 |
/* |
|
3744 |
* Sending data on a datagram socket. |
|
3745 |
* Assumes caller has verified that SS_ISBOUND etc. are set. |
|
3746 |
* |
|
3747 |
* For AF_UNIX the destination address is translated to an internal |
|
3748 |
* name and the source address is passed as an option. |
|
3749 |
*/ |
|
3750 |
int |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3751 |
sosend_dgram(struct sonode *so, struct sockaddr *name, socklen_t namelen, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3752 |
struct uio *uiop, int flags) |
0 | 3753 |
{ |
3754 |
struct T_unitdata_req tudr; |
|
3755 |
mblk_t *mp; |
|
3756 |
int error; |
|
3757 |
void *addr; |
|
3758 |
socklen_t addrlen; |
|
3759 |
void *src; |
|
3760 |
socklen_t srclen; |
|
3761 |
ssize_t len; |
|
3762 |
||
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3763 |
ASSERT(name != NULL && namelen != 0); |
0 | 3764 |
|
3765 |
len = uiop->uio_resid; |
|
3766 |
if (len > so->so_tidu_size) { |
|
3767 |
error = EMSGSIZE; |
|
3768 |
goto done; |
|
3769 |
} |
|
3770 |
||
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3771 |
/* Length and family checks */ |
0 | 3772 |
error = so_addr_verify(so, name, namelen); |
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3773 |
if (error != 0) |
0 | 3774 |
goto done; |
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3775 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3776 |
if (so->so_state & SS_DIRECT) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3777 |
return (sodgram_direct(so, name, namelen, uiop, flags)); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
3778 |
|
0 | 3779 |
if (so->so_family == AF_UNIX) { |
3780 |
if (so->so_state & SS_FADDR_NOXLATE) { |
|
3781 |
/* |
|
3782 |
* Already have a transport internal address. Do not |
|
3783 |
* pass any (transport internal) source address. |
|
3784 |
*/ |
|
3785 |
addr = name; |
|
3786 |
addrlen = namelen; |
|
3787 |
src = NULL; |
|
3788 |
srclen = 0; |
|
3789 |
} else { |
|
3790 |
/* |
|
3791 |
* Pass the sockaddr_un source address as an option |
|
3792 |
* and translate the remote address. |
|
3793 |
* |
|
3794 |
* Note that this code does not prevent so_laddr_sa |
|
3795 |
* from changing while it is being used. Thus |
|
3796 |
* if an unbind+bind occurs concurrently with this |
|
3797 |
* send the peer might see a partially new and a |
|
3798 |
* partially old "from" address. |
|
3799 |
*/ |
|
3800 |
src = so->so_laddr_sa; |
|
3801 |
srclen = (socklen_t)so->so_laddr_len; |
|
3802 |
dprintso(so, 1, |
|
3803 |
("sosend_dgram UNIX: srclen %d, src %p\n", |
|
3804 |
srclen, src)); |
|
3805 |
error = so_ux_addr_xlate(so, name, namelen, |
|
3806 |
(flags & MSG_XPG4_2), |
|
3807 |
&addr, &addrlen); |
|
3808 |
if (error) { |
|
3809 |
eprintsoline(so, error); |
|
3810 |
goto done; |
|
3811 |
} |
|
3812 |
} |
|
3813 |
} else { |
|
3814 |
addr = name; |
|
3815 |
addrlen = namelen; |
|
3816 |
src = NULL; |
|
3817 |
srclen = 0; |
|
3818 |
} |
|
3819 |
tudr.PRIM_type = T_UNITDATA_REQ; |
|
3820 |
tudr.DEST_length = addrlen; |
|
3821 |
tudr.DEST_offset = (t_scalar_t)sizeof (tudr); |
|
3822 |
if (srclen == 0) { |
|
3823 |
tudr.OPT_length = 0; |
|
3824 |
tudr.OPT_offset = 0; |
|
3825 |
||
3826 |
mp = soallocproto2(&tudr, sizeof (tudr), |
|
3827 |
addr, addrlen, 0, _ALLOC_INTR); |
|
3828 |
if (mp == NULL) { |
|
3829 |
/* |
|
3830 |
* Caught a signal waiting for memory. |
|
3831 |
* Let send* return EINTR. |
|
3832 |
*/ |
|
3833 |
error = EINTR; |
|
3834 |
goto done; |
|
3835 |
} |
|
3836 |
} else { |
|
3837 |
/* |
|
3838 |
* There is a AF_UNIX sockaddr_un to include as a source |
|
3839 |
* address option. |
|
3840 |
*/ |
|
3841 |
struct T_opthdr toh; |
|
3842 |
ssize_t size; |
|
3843 |
||
3844 |
tudr.OPT_length = (t_scalar_t)(sizeof (toh) + |
|
3845 |
_TPI_ALIGN_TOPT(srclen)); |
|
3846 |
tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) + |
|
3847 |
_TPI_ALIGN_TOPT(addrlen)); |
|
3848 |
||
3849 |
toh.level = SOL_SOCKET; |
|
3850 |
toh.name = SO_SRCADDR; |
|
3851 |
toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr)); |
|
3852 |
toh.status = 0; |
|
3853 |
||
3854 |
size = tudr.OPT_offset + tudr.OPT_length; |
|
3855 |
mp = soallocproto2(&tudr, sizeof (tudr), |
|
3856 |
addr, addrlen, size, _ALLOC_INTR); |
|
3857 |
if (mp == NULL) { |
|
3858 |
/* |
|
3859 |
* Caught a signal waiting for memory. |
|
3860 |
* Let send* return EINTR. |
|
3861 |
*/ |
|
3862 |
error = EINTR; |
|
3863 |
goto done; |
|
3864 |
} |
|
3865 |
mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen; |
|
3866 |
soappendmsg(mp, &toh, sizeof (toh)); |
|
3867 |
soappendmsg(mp, src, srclen); |
|
3868 |
mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen; |
|
3869 |
ASSERT(mp->b_wptr <= mp->b_datap->db_lim); |
|
3870 |
} |
|
3871 |
||
3872 |
#ifdef C2_AUDIT |
|
3873 |
if (audit_active) |
|
3874 |
audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0); |
|
3875 |
#endif /* C2_AUDIT */ |
|
3876 |
||
3877 |
error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0); |
|
3878 |
done: |
|
3879 |
#ifdef SOCK_DEBUG |
|
3880 |
if (error) { |
|
3881 |
eprintsoline(so, error); |
|
3882 |
} |
|
3883 |
#endif /* SOCK_DEBUG */ |
|
3884 |
return (error); |
|
3885 |
} |
|
3886 |
||
3887 |
/* |
|
3888 |
* Sending data on a connected stream socket. |
|
3889 |
* Assumes caller has verified that SS_ISCONNECTED is set. |
|
3890 |
*/ |
|
3891 |
int |
|
3892 |
sosend_svc(struct sonode *so, |
|
3893 |
struct uio *uiop, |
|
3894 |
t_scalar_t prim, |
|
3895 |
int more, |
|
3896 |
int sflag) |
|
3897 |
{ |
|
3898 |
struct T_data_req tdr; |
|
3899 |
mblk_t *mp; |
|
3900 |
int error; |
|
3901 |
ssize_t iosize; |
|
3902 |
int first = 1; |
|
3903 |
||
3904 |
dprintso(so, 1, |
|
3905 |
("sosend_svc: %p, resid %ld bytes, prim %d, sflag 0x%x\n", |
|
3906 |
so, uiop->uio_resid, prim, sflag)); |
|
3907 |
||
3908 |
/* |
|
3909 |
* Has to be bound and connected. However, since no locks are |
|
3910 |
* held the state could have changed after sotpi_sendmsg checked it |
|
3911 |
* thus it is not possible to ASSERT on the state. |
|
3912 |
*/ |
|
3913 |
||
3914 |
do { |
|
3915 |
/* |
|
3916 |
* Set the MORE flag if uio_resid does not fit in this |
|
3917 |
* message or if the caller passed in "more". |
|
3918 |
* Error for transports with zero tidu_size. |
|
3919 |
*/ |
|
3920 |
tdr.PRIM_type = prim; |
|
3921 |
iosize = so->so_tidu_size; |
|
3922 |
if (iosize <= 0) |
|
3923 |
return (EMSGSIZE); |
|
3924 |
if (uiop->uio_resid > iosize) { |
|
3925 |
tdr.MORE_flag = 1; |
|
3926 |
} else { |
|
3927 |
if (more) |
|
3928 |
tdr.MORE_flag = 1; |
|
3929 |
else |
|
3930 |
tdr.MORE_flag = 0; |
|
3931 |
iosize = uiop->uio_resid; |
|
3932 |
} |
|
3933 |
dprintso(so, 1, ("sosend_svc: sending 0x%x %d, %ld bytes\n", |
|
3934 |
prim, tdr.MORE_flag, iosize)); |
|
3935 |
mp = soallocproto1(&tdr, sizeof (tdr), 0, _ALLOC_INTR); |
|
3936 |
if (mp == NULL) { |
|
3937 |
/* |
|
3938 |
* Caught a signal waiting for memory. |
|
3939 |
* Let send* return EINTR. |
|
3940 |
*/ |
|
3941 |
if (first) |
|
3942 |
return (EINTR); |
|
3943 |
else |
|
3944 |
return (0); |
|
3945 |
} |
|
3946 |
||
3947 |
error = kstrputmsg(SOTOV(so), mp, uiop, iosize, |
|
3948 |
0, sflag | MSG_BAND, 0); |
|
3949 |
if (error) { |
|
3950 |
if (!first && error == EWOULDBLOCK) |
|
3951 |
return (0); |
|
3952 |
eprintsoline(so, error); |
|
3953 |
return (error); |
|
3954 |
} |
|
3955 |
first = 0; |
|
3956 |
if (uiop->uio_resid > 0) { |
|
3957 |
/* |
|
3958 |
* Recheck for fatal errors. Fail write even though |
|
3959 |
* some data have been written. This is consistent |
|
3960 |
* with strwrite semantics and BSD sockets semantics. |
|
3961 |
*/ |
|
3962 |
if (so->so_state & SS_CANTSENDMORE) { |
|
3963 |
tsignal(curthread, SIGPIPE); |
|
3964 |
eprintsoline(so, error); |
|
3965 |
return (EPIPE); |
|
3966 |
} |
|
3967 |
if (so->so_error != 0) { |
|
3968 |
mutex_enter(&so->so_lock); |
|
3969 |
error = sogeterr(so); |
|
3970 |
mutex_exit(&so->so_lock); |
|
3971 |
if (error != 0) { |
|
3972 |
eprintsoline(so, error); |
|
3973 |
return (error); |
|
3974 |
} |
|
3975 |
} |
|
3976 |
} |
|
3977 |
} while (uiop->uio_resid > 0); |
|
3978 |
return (0); |
|
3979 |
} |
|
3980 |
||
3981 |
/* |
|
3982 |
* Check the state for errors and call the appropriate send function. |
|
3983 |
* |
|
3984 |
* If MSG_DONTROUTE is set (and SO_DONTROUTE isn't already set) |
|
3985 |
* this function issues a setsockopt to toggle SO_DONTROUTE before and |
|
3986 |
* after sending the message. |
|
3987 |
*/ |
|
3988 |
static int |
|
3989 |
sotpi_sendmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop) |
|
3990 |
{ |
|
3991 |
int so_state; |
|
3992 |
int so_mode; |
|
3993 |
int error; |
|
3994 |
struct sockaddr *name; |
|
3995 |
t_uscalar_t namelen; |
|
3996 |
int dontroute; |
|
3997 |
int flags; |
|
3998 |
||
3999 |
dprintso(so, 1, ("sotpi_sendmsg(%p, %p, 0x%x) state %s, error %d\n", |
|
4000 |
so, msg, msg->msg_flags, |
|
4001 |
pr_state(so->so_state, so->so_mode), so->so_error)); |
|
4002 |
||
4003 |
mutex_enter(&so->so_lock); |
|
4004 |
so_state = so->so_state; |
|
4005 |
||
4006 |
if (so_state & SS_CANTSENDMORE) { |
|
4007 |
mutex_exit(&so->so_lock); |
|
4008 |
tsignal(curthread, SIGPIPE); |
|
4009 |
return (EPIPE); |
|
4010 |
} |
|
4011 |
||
4012 |
if (so->so_error != 0) { |
|
4013 |
error = sogeterr(so); |
|
4014 |
if (error != 0) { |
|
4015 |
mutex_exit(&so->so_lock); |
|
4016 |
return (error); |
|
4017 |
} |
|
4018 |
} |
|
4019 |
||
4020 |
name = (struct sockaddr *)msg->msg_name; |
|
4021 |
namelen = msg->msg_namelen; |
|
4022 |
||
4023 |
so_mode = so->so_mode; |
|
4024 |
||
4025 |
if (name == NULL) { |
|
4026 |
if (!(so_state & SS_ISCONNECTED)) { |
|
4027 |
mutex_exit(&so->so_lock); |
|
4028 |
if (so_mode & SM_CONNREQUIRED) |
|
4029 |
return (ENOTCONN); |
|
4030 |
else |
|
4031 |
return (EDESTADDRREQ); |
|
4032 |
} |
|
4033 |
if (so_mode & SM_CONNREQUIRED) { |
|
4034 |
name = NULL; |
|
4035 |
namelen = 0; |
|
4036 |
} else { |
|
4037 |
/* |
|
4038 |
* Note that this code does not prevent so_faddr_sa |
|
4039 |
* from changing while it is being used. Thus |
|
4040 |
* if an "unconnect"+connect occurs concurrently with |
|
4041 |
* this send the datagram might be delivered to a |
|
4042 |
* garbaled address. |
|
4043 |
*/ |
|
4044 |
ASSERT(so->so_faddr_sa); |
|
4045 |
name = so->so_faddr_sa; |
|
4046 |
namelen = (t_uscalar_t)so->so_faddr_len; |
|
4047 |
} |
|
4048 |
} else { |
|
4049 |
if (!(so_state & SS_ISCONNECTED) && |
|
4050 |
(so_mode & SM_CONNREQUIRED)) { |
|
4051 |
/* Required but not connected */ |
|
4052 |
mutex_exit(&so->so_lock); |
|
4053 |
return (ENOTCONN); |
|
4054 |
} |
|
4055 |
/* |
|
4056 |
* Ignore the address on connection-oriented sockets. |
|
4057 |
* Just like BSD this code does not generate an error for |
|
4058 |
* TCP (a CONNREQUIRED socket) when sending to an address |
|
4059 |
* passed in with sendto/sendmsg. Instead the data is |
|
4060 |
* delivered on the connection as if no address had been |
|
4061 |
* supplied. |
|
4062 |
*/ |
|
4063 |
if ((so_state & SS_ISCONNECTED) && |
|
4064 |
!(so_mode & SM_CONNREQUIRED)) { |
|
4065 |
mutex_exit(&so->so_lock); |
|
4066 |
return (EISCONN); |
|
4067 |
} |
|
4068 |
if (!(so_state & SS_ISBOUND)) { |
|
4069 |
so_lock_single(so); /* Set SOLOCKED */ |
|
4070 |
error = sotpi_bind(so, NULL, 0, |
|
4071 |
_SOBIND_UNSPEC|_SOBIND_LOCK_HELD); |
|
4072 |
so_unlock_single(so, SOLOCKED); |
|
4073 |
if (error) { |
|
4074 |
mutex_exit(&so->so_lock); |
|
4075 |
eprintsoline(so, error); |
|
4076 |
return (error); |
|
4077 |
} |
|
4078 |
} |
|
4079 |
/* |
|
4080 |
* Handle delayed datagram errors. These are only queued |
|
4081 |
* when the application sets SO_DGRAM_ERRIND. |
|
4082 |
* Return the error if we are sending to the address |
|
4083 |
* that was returned in the last T_UDERROR_IND. |
|
4084 |
* If sending to some other address discard the delayed |
|
4085 |
* error indication. |
|
4086 |
*/ |
|
4087 |
if (so->so_delayed_error) { |
|
4088 |
struct T_uderror_ind *tudi; |
|
4089 |
void *addr; |
|
4090 |
t_uscalar_t addrlen; |
|
4091 |
boolean_t match = B_FALSE; |
|
4092 |
||
4093 |
ASSERT(so->so_eaddr_mp); |
|
4094 |
error = so->so_delayed_error; |
|
4095 |
so->so_delayed_error = 0; |
|
4096 |
tudi = (struct T_uderror_ind *)so->so_eaddr_mp->b_rptr; |
|
4097 |
addrlen = tudi->DEST_length; |
|
4098 |
addr = sogetoff(so->so_eaddr_mp, |
|
4099 |
tudi->DEST_offset, |
|
4100 |
addrlen, 1); |
|
4101 |
ASSERT(addr); /* Checked by strsock_proto */ |
|
4102 |
switch (so->so_family) { |
|
4103 |
case AF_INET: { |
|
4104 |
/* Compare just IP address and port */ |
|
4105 |
sin_t *sin1 = (sin_t *)name; |
|
4106 |
sin_t *sin2 = (sin_t *)addr; |
|
4107 |
||
4108 |
if (addrlen == sizeof (sin_t) && |
|
4109 |
namelen == addrlen && |
|
4110 |
sin1->sin_port == sin2->sin_port && |
|
4111 |
sin1->sin_addr.s_addr == |
|
4112 |
sin2->sin_addr.s_addr) |
|
4113 |
match = B_TRUE; |
|
4114 |
break; |
|
4115 |
} |
|
4116 |
case AF_INET6: { |
|
4117 |
/* Compare just IP address and port. Not flow */ |
|
4118 |
sin6_t *sin1 = (sin6_t *)name; |
|
4119 |
sin6_t *sin2 = (sin6_t *)addr; |
|
4120 |
||
4121 |
if (addrlen == sizeof (sin6_t) && |
|
4122 |
namelen == addrlen && |
|
4123 |
sin1->sin6_port == sin2->sin6_port && |
|
4124 |
IN6_ARE_ADDR_EQUAL(&sin1->sin6_addr, |
|
4125 |
&sin2->sin6_addr)) |
|
4126 |
match = B_TRUE; |
|
4127 |
break; |
|
4128 |
} |
|
4129 |
case AF_UNIX: |
|
4130 |
default: |
|
4131 |
if (namelen == addrlen && |
|
4132 |
bcmp(name, addr, namelen) == 0) |
|
4133 |
match = B_TRUE; |
|
4134 |
} |
|
4135 |
if (match) { |
|
4136 |
freemsg(so->so_eaddr_mp); |
|
4137 |
so->so_eaddr_mp = NULL; |
|
4138 |
mutex_exit(&so->so_lock); |
|
4139 |
#ifdef DEBUG |
|
4140 |
dprintso(so, 0, |
|
4141 |
("sockfs delayed error %d for %s\n", |
|
4142 |
error, |
|
4143 |
pr_addr(so->so_family, name, namelen))); |
|
4144 |
#endif /* DEBUG */ |
|
4145 |
return (error); |
|
4146 |
} |
|
4147 |
freemsg(so->so_eaddr_mp); |
|
4148 |
so->so_eaddr_mp = NULL; |
|
4149 |
} |
|
4150 |
} |
|
4151 |
mutex_exit(&so->so_lock); |
|
4152 |
||
4153 |
flags = msg->msg_flags; |
|
4154 |
dontroute = 0; |
|
4155 |
if ((flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE)) { |
|
4156 |
uint32_t val; |
|
4157 |
||
4158 |
val = 1; |
|
4159 |
error = sotpi_setsockopt(so, SOL_SOCKET, SO_DONTROUTE, |
|
4160 |
&val, (t_uscalar_t)sizeof (val)); |
|
4161 |
if (error) |
|
4162 |
return (error); |
|
4163 |
dontroute = 1; |
|
4164 |
} |
|
4165 |
||
4166 |
if ((flags & MSG_OOB) && !(so_mode & SM_EXDATA)) { |
|
4167 |
error = EOPNOTSUPP; |
|
4168 |
goto done; |
|
4169 |
} |
|
4170 |
if (msg->msg_controllen != 0) { |
|
4171 |
if (!(so_mode & SM_CONNREQUIRED)) { |
|
4172 |
error = sosend_dgramcmsg(so, name, namelen, uiop, |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4173 |
msg->msg_control, msg->msg_controllen, flags); |
0 | 4174 |
} else { |
4175 |
if (flags & MSG_OOB) { |
|
4176 |
/* Can't generate T_EXDATA_REQ with options */ |
|
4177 |
error = EOPNOTSUPP; |
|
4178 |
goto done; |
|
4179 |
} |
|
4180 |
error = sosend_svccmsg(so, uiop, |
|
4181 |
!(flags & MSG_EOR), |
|
4182 |
msg->msg_control, msg->msg_controllen, |
|
4183 |
flags); |
|
4184 |
} |
|
4185 |
goto done; |
|
4186 |
} |
|
4187 |
||
4188 |
if (!(so_mode & SM_CONNREQUIRED)) { |
|
4189 |
/* |
|
4190 |
* If there is no SO_DONTROUTE to turn off return immediately |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4191 |
* from send_dgram. This can allow tail-call optimizations. |
0 | 4192 |
*/ |
4193 |
if (!dontroute) { |
|
4194 |
return (sosend_dgram(so, name, namelen, uiop, flags)); |
|
4195 |
} |
|
4196 |
error = sosend_dgram(so, name, namelen, uiop, flags); |
|
4197 |
} else { |
|
4198 |
t_scalar_t prim; |
|
4199 |
int sflag; |
|
4200 |
||
4201 |
/* Ignore msg_name in the connected state */ |
|
4202 |
if (flags & MSG_OOB) { |
|
4203 |
prim = T_EXDATA_REQ; |
|
4204 |
/* |
|
4205 |
* Send down T_EXDATA_REQ even if there is flow |
|
4206 |
* control for data. |
|
4207 |
*/ |
|
4208 |
sflag = MSG_IGNFLOW; |
|
4209 |
} else { |
|
4210 |
if (so_mode & SM_BYTESTREAM) { |
|
4211 |
/* Byte stream transport - use write */ |
|
4212 |
||
4213 |
dprintso(so, 1, ("sotpi_sendmsg: write\n")); |
|
4214 |
/* |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4215 |
* If there is no SO_DONTROUTE to turn off, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4216 |
* SS_DIRECT is on, and there is no flow |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4217 |
* control, we can take the fast path. |
0 | 4218 |
*/ |
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4219 |
if (!dontroute && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4220 |
(so_state & SS_DIRECT) && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4221 |
canputnext(SOTOV(so)->v_stream->sd_wrq)) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4222 |
return (sostream_direct(so, uiop, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4223 |
NULL, CRED())); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4224 |
} |
0 | 4225 |
error = strwrite(SOTOV(so), uiop, CRED()); |
4226 |
goto done; |
|
4227 |
} |
|
4228 |
prim = T_DATA_REQ; |
|
4229 |
sflag = 0; |
|
4230 |
} |
|
4231 |
/* |
|
4232 |
* If there is no SO_DONTROUTE to turn off return immediately |
|
4233 |
* from sosend_svc. This can allow tail-call optimizations. |
|
4234 |
*/ |
|
4235 |
if (!dontroute) |
|
4236 |
return (sosend_svc(so, uiop, prim, |
|
4237 |
!(flags & MSG_EOR), sflag)); |
|
4238 |
error = sosend_svc(so, uiop, prim, |
|
4239 |
!(flags & MSG_EOR), sflag); |
|
4240 |
} |
|
4241 |
ASSERT(dontroute); |
|
4242 |
done: |
|
4243 |
if (dontroute) { |
|
4244 |
uint32_t val; |
|
4245 |
||
4246 |
val = 0; |
|
4247 |
(void) sotpi_setsockopt(so, SOL_SOCKET, SO_DONTROUTE, |
|
4248 |
&val, (t_uscalar_t)sizeof (val)); |
|
4249 |
} |
|
4250 |
return (error); |
|
4251 |
} |
|
4252 |
||
4253 |
/* |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4254 |
* Sending data on a datagram socket. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4255 |
* Assumes caller has verified that SS_ISBOUND etc. are set. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4256 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4257 |
/* ARGSUSED */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4258 |
static int |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4259 |
sodgram_direct(struct sonode *so, struct sockaddr *name, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4260 |
socklen_t namelen, struct uio *uiop, int flags) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4261 |
{ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4262 |
struct T_unitdata_req tudr; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4263 |
mblk_t *mp; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4264 |
int error = 0; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4265 |
void *addr; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4266 |
socklen_t addrlen; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4267 |
ssize_t len; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4268 |
struct stdata *stp = SOTOV(so)->v_stream; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4269 |
int so_state; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4270 |
queue_t *udp_wq; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4271 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4272 |
ASSERT(name != NULL && namelen != 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4273 |
ASSERT(!(so->so_mode & SM_CONNREQUIRED)); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4274 |
ASSERT(!(so->so_mode & SM_EXDATA)); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4275 |
ASSERT(so->so_family == AF_INET || so->so_family == AF_INET6); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4276 |
ASSERT(SOTOV(so)->v_type == VSOCK); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4277 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4278 |
/* Caller checked for proper length */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4279 |
len = uiop->uio_resid; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4280 |
ASSERT(len <= so->so_tidu_size); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4281 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4282 |
/* Length and family checks have been done by caller */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4283 |
ASSERT(name->sa_family == so->so_family); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4284 |
ASSERT(so->so_family == AF_INET || |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4285 |
(namelen == (socklen_t)sizeof (struct sockaddr_in6))); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4286 |
ASSERT(so->so_family == AF_INET6 || |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4287 |
(namelen == (socklen_t)sizeof (struct sockaddr_in))); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4288 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4289 |
addr = name; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4290 |
addrlen = namelen; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4291 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4292 |
if (stp->sd_sidp != NULL && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4293 |
(error = straccess(stp, JCWRITE)) != 0) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4294 |
goto done; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4295 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4296 |
so_state = so->so_state; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4297 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4298 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4299 |
* For UDP we don't break up the copyin into smaller pieces |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4300 |
* as in the TCP case. That means if ENOMEM is returned by |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4301 |
* mcopyinuio() then the uio vector has not been modified at |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4302 |
* all and we fallback to either strwrite() or kstrputmsg() |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4303 |
* below. Note also that we never generate priority messages |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4304 |
* from here. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4305 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4306 |
udp_wq = stp->sd_wrq->q_next; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4307 |
if (canput(udp_wq) && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4308 |
(mp = mcopyinuio(stp, uiop, -1, -1, &error)) != NULL) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4309 |
ASSERT(DB_TYPE(mp) == M_DATA); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4310 |
ASSERT(uiop->uio_resid == 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4311 |
#ifdef C2_AUDIT |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4312 |
if (audit_active) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4313 |
audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4314 |
#endif /* C2_AUDIT */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4315 |
udp_wput_data(udp_wq, mp, addr, addrlen); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4316 |
return (0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4317 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4318 |
if (error != 0 && error != ENOMEM) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4319 |
return (error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4320 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4321 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4322 |
* For connected, let strwrite() handle the blocking case. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4323 |
* Otherwise we fall thru and use kstrputmsg(). |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4324 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4325 |
if (so_state & SS_ISCONNECTED) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4326 |
return (strwrite(SOTOV(so), uiop, CRED())); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4327 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4328 |
tudr.PRIM_type = T_UNITDATA_REQ; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4329 |
tudr.DEST_length = addrlen; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4330 |
tudr.DEST_offset = (t_scalar_t)sizeof (tudr); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4331 |
tudr.OPT_length = 0; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4332 |
tudr.OPT_offset = 0; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4333 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4334 |
mp = soallocproto2(&tudr, sizeof (tudr), addr, addrlen, 0, _ALLOC_INTR); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4335 |
if (mp == NULL) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4336 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4337 |
* Caught a signal waiting for memory. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4338 |
* Let send* return EINTR. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4339 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4340 |
error = EINTR; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4341 |
goto done; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4342 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4343 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4344 |
#ifdef C2_AUDIT |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4345 |
if (audit_active) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4346 |
audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4347 |
#endif /* C2_AUDIT */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4348 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4349 |
error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4350 |
done: |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4351 |
#ifdef SOCK_DEBUG |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4352 |
if (error != 0) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4353 |
eprintsoline(so, error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4354 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4355 |
#endif /* SOCK_DEBUG */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4356 |
return (error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4357 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4358 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4359 |
int |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4360 |
sostream_direct(struct sonode *so, struct uio *uiop, mblk_t *mp, cred_t *cr) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4361 |
{ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4362 |
struct stdata *stp = SOTOV(so)->v_stream; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4363 |
ssize_t iosize, rmax, maxblk; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4364 |
queue_t *tcp_wq = stp->sd_wrq->q_next; |
898 | 4365 |
mblk_t *newmp; |
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4366 |
int error = 0, wflag = 0; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4367 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4368 |
ASSERT(so->so_mode & SM_BYTESTREAM); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4369 |
ASSERT(SOTOV(so)->v_type == VSOCK); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4370 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4371 |
if (stp->sd_sidp != NULL && |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4372 |
(error = straccess(stp, JCWRITE)) != 0) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4373 |
return (error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4374 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4375 |
if (uiop == NULL) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4376 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4377 |
* kstrwritemp() should have checked sd_flag and |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4378 |
* flow-control before coming here. If we end up |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4379 |
* here it means that we can simply pass down the |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4380 |
* data to tcp. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4381 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4382 |
ASSERT(mp != NULL); |
898 | 4383 |
if (stp->sd_wputdatafunc != NULL) { |
4384 |
newmp = (stp->sd_wputdatafunc)(SOTOV(so), mp, NULL, |
|
4385 |
NULL, NULL, NULL); |
|
4386 |
if (newmp == NULL) { |
|
4387 |
/* The caller will free mp */ |
|
4388 |
return (ECOMM); |
|
4389 |
} |
|
4390 |
mp = newmp; |
|
4391 |
} |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4392 |
tcp_wput(tcp_wq, mp); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4393 |
return (0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4394 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4395 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4396 |
/* Fallback to strwrite() to do proper error handling */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4397 |
if (stp->sd_flag & (STWRERR|STRHUP|STPLEX|STRDELIM|OLDNDELAY)) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4398 |
return (strwrite(SOTOV(so), uiop, cr)); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4399 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4400 |
rmax = stp->sd_qn_maxpsz; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4401 |
ASSERT(rmax >= 0 || rmax == INFPSZ); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4402 |
if (rmax == 0 || uiop->uio_resid <= 0) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4403 |
return (0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4404 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4405 |
if (rmax == INFPSZ) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4406 |
rmax = uiop->uio_resid; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4407 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4408 |
maxblk = stp->sd_maxblk; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4409 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4410 |
for (;;) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4411 |
iosize = MIN(uiop->uio_resid, rmax); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4412 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4413 |
mp = mcopyinuio(stp, uiop, iosize, maxblk, &error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4414 |
if (mp == NULL) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4415 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4416 |
* Fallback to strwrite() for ENOMEM; if this |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4417 |
* is our first time in this routine and the uio |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4418 |
* vector has not been modified, we will end up |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4419 |
* calling strwrite() without any flag set. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4420 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4421 |
if (error == ENOMEM) |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4422 |
goto slow_send; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4423 |
else |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4424 |
return (error); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4425 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4426 |
ASSERT(uiop->uio_resid >= 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4427 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4428 |
* If mp is non-NULL and ENOMEM is set, it means that |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4429 |
* mcopyinuio() was able to break down some of the user |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4430 |
* data into one or more mblks. Send the partial data |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4431 |
* to tcp and let the rest be handled in strwrite(). |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4432 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4433 |
ASSERT(error == 0 || error == ENOMEM); |
898 | 4434 |
if (stp->sd_wputdatafunc != NULL) { |
4435 |
newmp = (stp->sd_wputdatafunc)(SOTOV(so), mp, NULL, |
|
4436 |
NULL, NULL, NULL); |
|
4437 |
if (newmp == NULL) { |
|
4438 |
/* The caller will free mp */ |
|
4439 |
return (ECOMM); |
|
4440 |
} |
|
4441 |
mp = newmp; |
|
4442 |
} |
|
741
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4443 |
tcp_wput(tcp_wq, mp); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4444 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4445 |
wflag |= NOINTR; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4446 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4447 |
if (uiop->uio_resid == 0) { /* No more data; we're done */ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4448 |
ASSERT(error == 0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4449 |
break; |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4450 |
} else if (error == ENOMEM || !canput(tcp_wq) || (stp->sd_flag & |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4451 |
(STWRERR|STRHUP|STPLEX|STRDELIM|OLDNDELAY))) { |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4452 |
slow_send: |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4453 |
/* |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4454 |
* We were able to send down partial data using |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4455 |
* the direct call interface, but are now relying |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4456 |
* on strwrite() to handle the non-fastpath cases. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4457 |
* If the socket is blocking we will sleep in |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4458 |
* strwaitq() until write is permitted, otherwise, |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4459 |
* we will need to return the amount of bytes |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4460 |
* written so far back to the app. This is the |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4461 |
* reason why we pass NOINTR flag to strwrite() |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4462 |
* for non-blocking socket, because we don't want |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4463 |
* to return EAGAIN when portion of the user data |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4464 |
* has actually been sent down. |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4465 |
*/ |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4466 |
return (strwrite_common(SOTOV(so), uiop, cr, wflag)); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4467 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4468 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4469 |
return (0); |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4470 |
} |
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4471 |
|
40027a3621ac
PSARC 2005/082 Yosemite: UDP Performance Enhancement
masputra
parents:
455
diff
changeset
|
4472 |
/* |
0 | 4473 |
* Update so_faddr by asking the transport (unless AF_UNIX). |
4474 |
*/ |
|
4475 |
int |
|
4476 |
sotpi_getpeername(struct sonode *so) |
|
4477 |
{ |
|
4478 |
struct strbuf strbuf; |
|
4479 |
int error = 0, res; |
|
4480 |
void *addr; |
|
4481 |
t_uscalar_t addrlen; |
|
4482 |
k_sigset_t smask; |
|
4483 |
||
4484 |
dprintso(so, 1, ("sotpi_getpeername(%p) %s\n", |
|
4485 |
so, pr_state(so->so_state, so->so_mode))); |
|
4486 |
||
4487 |
mutex_enter(&so->so_lock); |
|
4488 |
so_lock_single(so); /* Set SOLOCKED */ |
|
4489 |
if (!(so->so_state & SS_ISCONNECTED)) { |
|
4490 |
error = ENOTCONN; |
|
4491 |
goto done; |
|
4492 |
} |
|
4493 |
/* Added this check for X/Open */ |
|
4494 |
if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) { |
|
4495 |
error = EINVAL; |
|
4496 |
if (xnet_check_print) { |
|
4497 |
printf("sockfs: X/Open getpeername check => EINVAL\n"); |
|
4498 |
} |
|
4499 |
goto done; |
|
4500 |
} |
|
4501 |
#ifdef DEBUG |
|
4502 |
dprintso(so, 1, ("sotpi_getpeername (local): %s\n", |
|
4503 |
pr_addr(so->so_family, so->so_faddr_sa, |
|
4504 |
(t_uscalar_t)so->so_faddr_len))); |
|
4505 |
#endif /* DEBUG */ |
|
4506 |
||
4507 |
if (so->so_family == AF_UNIX || so->so_family == AF_NCA) { |
|
4508 |
/* Transport has different name space - return local info */ |
|
4509 |
error = 0; |
|
4510 |
goto done; |
|
4511 |
} |
|
4512 |
||
4513 |
ASSERT(so->so_faddr_sa); |
|
4514 |
/* Allocate local buffer to use with ioctl */ |
|
4515 |
addrlen = (t_uscalar_t)so->so_faddr_maxlen; |
|
4516 |
mutex_exit(&so->so_lock); |
|
4517 |
addr = kmem_alloc(addrlen, KM_SLEEP); |
|
4518 |
||
4519 |
/* |
|
4520 |
* Issue TI_GETPEERNAME with signals masked. |
|
4521 |
* Put the result in so_faddr_sa so that getpeername works after |
|
4522 |
* a shutdown(output). |
|
4523 |
* If the ioctl fails (e.g. due to a ECONNRESET) the error is reposted |
|
4524 |
* back to the socket. |
|
4525 |
*/ |
|
4526 |
strbuf.buf = addr; |
|
4527 |
strbuf.maxlen = addrlen; |
|
4528 |
strbuf.len = 0; |
|
4529 |
||
4530 |
sigintr(&smask, 0); |
|
4531 |
res = 0; |
|
4532 |
ASSERT(CRED()); |
|
4533 |
error = strioctl(SOTOV(so), TI_GETPEERNAME, (intptr_t)&strbuf, |
|
4534 |
0, K_TO_K, CRED(), &res); |
|
4535 |
sigunintr(&smask); |
|
4536 |
||
4537 |
mutex_enter(&so->so_lock); |
|
4538 |
/* |
|
4539 |
* If there is an error record the error in so_error put don't fail |
|
4540 |
* the getpeername. Instead fallback on the recorded |
|
4541 |
* so->so_faddr_sa. |
|
4542 |
*/ |
|
4543 |
if (error) { |
|
4544 |
/* |
|
4545 |
* Various stream head errors can be returned to the ioctl. |
|
4546 |
* However, it is impossible to determine which ones of |
|
4547 |
* these are really socket level errors that were incorrectly |
|
4548 |
* consumed by the ioctl. Thus this code silently ignores the |
|
4549 |
* error - to code explicitly does not reinstate the error |
|
4550 |
* using soseterror(). |
|
4551 |
* Experiments have shows that at least this set of |
|
4552 |
* errors are reported and should not be reinstated on the |
|
4553 |
* socket: |
|
4554 |
* EINVAL E.g. if an I_LINK was in effect when |
|
4555 |
* getpeername was called. |
|
4556 |
* EPIPE The ioctl error semantics prefer the write |
|
4557 |
* side error over the read side error. |
|
4558 |
* ENOTCONN The transport just got disconnected but |
|
4559 |
* sockfs had not yet seen the T_DISCON_IND |
|
4560 |
* when issuing the ioctl. |
|
4561 |
*/ |
|
4562 |
error = 0; |
|
4563 |
} else if (res == 0 && strbuf.len > 0 && |
|
4564 |
(so->so_state & SS_ISCONNECTED)) { |
|
4565 |
ASSERT(strbuf.len <= (int)so->so_faddr_maxlen); |
|
4566 |
so->so_faddr_len = (socklen_t)strbuf.len; |
|
4567 |
bcopy(addr, so->so_faddr_sa, so->so_faddr_len); |
|
4568 |
so->so_state |= SS_FADDR_VALID; |
|
4569 |
} |
|
4570 |
kmem_free(addr, addrlen); |
|
4571 |
#ifdef DEBUG |
|
4572 |
dprintso(so, 1, ("sotpi_getpeername (tp): %s\n", |
|
4573 |
pr_addr(so->so_family, so->so_faddr_sa, |
|
4574 |
(t_uscalar_t)so->so_faddr_len))); |
|
4575 |
#endif /* DEBUG */ |
|
4576 |
done: |
|
4577 |
so_unlock_single(so, SOLOCKED); |
|
4578 |
mutex_exit(&so->so_lock); |
|
4579 |
return (error); |
|
4580 |
} |
|
4581 |
||
4582 |
/* |
|
4583 |
* Update so_laddr by asking the transport (unless AF_UNIX). |
|
4584 |
*/ |
|
4585 |
int |
|
4586 |
sotpi_getsockname(struct sonode *so) |
|
4587 |
{ |
|
4588 |
struct strbuf strbuf; |
|
4589 |
int error = 0, res; |
|
4590 |
void *addr; |
|
4591 |
t_uscalar_t addrlen; |
|
4592 |
k_sigset_t smask; |
|
4593 |
||
4594 |
dprintso(so, 1, ("sotpi_getsockname(%p) %s\n", |
|
4595 |
so, pr_state(so->so_state, so->so_mode))); |
|
4596 |
||
4597 |
mutex_enter(&so->so_lock); |
|
4598 |
so_lock_single(so); /* Set SOLOCKED */ |
|
4599 |
if (!(so->so_state & SS_ISBOUND) && so->so_family != AF_UNIX) { |
|
4600 |
/* Return an all zero address except for the family */ |
|
4601 |
if (so->so_family == AF_INET) |
|
4602 |
so->so_laddr_len = (socklen_t)sizeof (sin_t); |
|
4603 |
else if (so->so_family == AF_INET6) |
|
4604 |
so->so_laddr_len = (socklen_t)sizeof (sin6_t); |
|
4605 |
ASSERT(so->so_laddr_len <= so->so_laddr_maxlen); |
|
4606 |
bzero(so->so_laddr_sa, so->so_laddr_len); |
|
4607 |
/* |
|
4608 |
* Can not assume there is a sa_family for all |
|
4609 |
* protocol families. |
|
4610 |
*/ |
|
4611 |
if (so->so_family == AF_INET || so->so_family == AF_INET6) |
|
4612 |
so->so_laddr_sa->sa_family = so->so_family; |
|
4613 |
} |
|
4614 |
#ifdef DEBUG |
|
4615 |
dprintso(so, 1, ("sotpi_getsockname (local): %s\n", |
|
4616 |
pr_addr(so->so_family, so->so_laddr_sa, |
|
4617 |
(t_uscalar_t)so->so_laddr_len))); |
|
4618 |
#endif /* DEBUG */ |
|
4619 |
if (so->so_family == AF_UNIX) { |
|
4620 |
/* Transport has different name space - return local info */ |
|
4621 |
error = 0; |
|
4622 |
goto done; |
|
4623 |
} |
|
4624 |
/* Allocate local buffer to use with ioctl */ |
|
4625 |
addrlen = (t_uscalar_t)so->so_laddr_maxlen; |
|
4626 |
mutex_exit(&so->so_lock); |
|
4627 |
addr = kmem_alloc(addrlen, KM_SLEEP); |
|
4628 |
||
4629 |
/* |
|
4630 |
* Issue TI_GETMYNAME with signals masked. |
|
4631 |
* Put the result in so_laddr_sa so that getsockname works after |
|
4632 |
* a shutdown(output). |
|
4633 |
* If the ioctl fails (e.g. due to a ECONNRESET) the error is reposted |
|
4634 |
* back to the socket. |
|
4635 |
*/ |
|
4636 |
strbuf.buf = addr; |
|
4637 |
strbuf.maxlen = addrlen; |
|
4638 |
strbuf.len = 0; |
|
4639 |
||
4640 |
sigintr(&smask, 0); |
|
4641 |
res = 0; |
|
4642 |
ASSERT(CRED()); |
|
4643 |
error = strioctl(SOTOV(so), TI_GETMYNAME, (intptr_t)&strbuf, |
|
4644 |
0, K_TO_K, CRED(), &res); |
|
4645 |
sigunintr(&smask); |
|
4646 |
||
4647 |
mutex_enter(&so->so_lock); |
|
4648 |
/* |
|
4649 |
* If there is an error record the error in so_error put don't fail |
|
4650 |
* the getsockname. Instead fallback on the recorded |
|
4651 |
* so->so_laddr_sa. |
|
4652 |
*/ |
|
4653 |
if (error) { |
|
4654 |
/* |
|
4655 |
* Various stream head errors can be returned to the ioctl. |
|
4656 |
* However, it is impossible to determine which ones of |
|
4657 |
* these are really socket level errors that were incorrectly |
|
4658 |
* consumed by the ioctl. Thus this code silently ignores the |
|
4659 |
* error - to code explicitly does not reinstate the error |
|
4660 |
* using soseterror(). |
|
4661 |
* Experiments have shows that at least this set of |
|
4662 |
* errors are reported and should not be reinstated on the |
|
4663 |
* socket: |
|
4664 |
* EINVAL E.g. if an I_LINK was in effect when |
|
4665 |
* getsockname was called. |
|
4666 |
* EPIPE The ioctl error semantics prefer the write |
|
4667 |
* side error over the read side error. |
|
4668 |
*/ |
|
4669 |
error = 0; |
|
4670 |
} else if (res == 0 && strbuf.len > 0 && |
|
4671 |
(so->so_state & SS_ISBOUND)) { |
|
4672 |
ASSERT(strbuf.len <= (int)so->so_laddr_maxlen); |
|
4673 |
so->so_laddr_len = (socklen_t)strbuf.len; |
|
4674 |
bcopy(addr, so->so_laddr_sa, so->so_laddr_len); |
|
4675 |
so->so_state |= SS_LADDR_VALID; |
|
4676 |
} |
|
4677 |
kmem_free(addr, addrlen); |
|
4678 |
#ifdef DEBUG |
|
4679 |
dprintso(so, 1, ("sotpi_getsockname (tp): %s\n", |
|
4680 |
pr_addr(so->so_family, so->so_laddr_sa, |
|
4681 |
(t_uscalar_t)so->so_laddr_len))); |
|
4682 |
#endif /* DEBUG */ |
|
4683 |
done: |
|
4684 |
so_unlock_single(so, SOLOCKED); |
|
4685 |
mutex_exit(&so->so_lock); |
|
4686 |
return (error); |
|
4687 |
} |
|
4688 |
||
4689 |
/* |
|
4690 |
* Get socket options. For SOL_SOCKET options some options are handled |
|
4691 |
* by the sockfs while others use the value recorded in the sonode as a |
|
4692 |
* fallback should the T_SVR4_OPTMGMT_REQ fail. |
|
4693 |
* |
|
4694 |
* On the return most *optlenp bytes are copied to optval. |
|
4695 |
*/ |
|
4696 |
int |
|
4697 |
sotpi_getsockopt(struct sonode *so, int level, int option_name, |
|
4698 |
void *optval, socklen_t *optlenp, int flags) |
|
4699 |
{ |
|
4700 |
struct T_optmgmt_req optmgmt_req; |
|
4701 |
struct T_optmgmt_ack *optmgmt_ack; |
|
4702 |
struct opthdr oh; |
|
4703 |
struct opthdr *opt_res; |
|
4704 |
mblk_t *mp = NULL; |
|
4705 |
int error = 0; |
|
4706 |
void *option = NULL; /* Set if fallback value */ |
|
4707 |
t_uscalar_t maxlen = *optlenp; |
|
4708 |
t_uscalar_t len; |
|
4709 |
uint32_t value; |
|
4710 |
||
4711 |
dprintso(so, 1, ("sotpi_getsockopt(%p, 0x%x, 0x%x, %p, %p) %s\n", |
|
4712 |
so, level, option_name, optval, optlenp, |
|
4713 |
pr_state(so->so_state, so->so_mode))); |
|
4714 |
||
4715 |
mutex_enter(&so->so_lock); |
|
4716 |
so_lock_single(so); /* Set SOLOCKED */ |
|
4717 |
||
4718 |
/* |
|
4719 |
* Check for SOL_SOCKET options. |
|
4720 |
* Certain SOL_SOCKET options are returned directly whereas |
|
4721 |
* others only provide a default (fallback) value should |
|
4722 |
* the T_SVR4_OPTMGMT_REQ fail. |
|
4723 |
*/ |
|
4724 |
if (level == SOL_SOCKET) { |
|
4725 |
/* Check parameters */ |
|
4726 |
switch (option_name) { |
|
4727 |
case SO_TYPE: |
|
4728 |
case SO_ERROR: |
|
4729 |
case SO_DEBUG: |
|
4730 |
case SO_ACCEPTCONN: |
|
4731 |
case SO_REUSEADDR: |
|
4732 |
case SO_KEEPALIVE: |
|
4733 |
case SO_DONTROUTE: |
|
4734 |
case SO_BROADCAST: |
|
4735 |
case SO_USELOOPBACK: |
|
4736 |
case SO_OOBINLINE: |
|
4737 |
case SO_SNDBUF: |
|
4738 |
case SO_RCVBUF: |
|
4739 |
#ifdef notyet |
|
4740 |
case SO_SNDLOWAT: |
|
4741 |
case SO_RCVLOWAT: |
|
4742 |
case SO_SNDTIMEO: |
|
4743 |
case SO_RCVTIMEO: |
|
4744 |
#endif /* notyet */ |
|
4745 |
case SO_DGRAM_ERRIND: |
|
4746 |
if (maxlen < (t_uscalar_t)sizeof (int32_t)) { |
|
4747 |
error = EINVAL; |
|
4748 |
eprintsoline(so, error); |
|
4749 |
goto done2; |
|
4750 |
} |
|
4751 |
break; |
|
4752 |
case SO_LINGER: |
|
4753 |
if (maxlen < (t_uscalar_t)sizeof (struct linger)) { |
|
4754 |
error = EINVAL; |
|
4755 |
eprintsoline(so, error); |
|
4756 |
goto done2; |
|
4757 |
} |
|
4758 |
break; |
|
4759 |
} |
|
4760 |
||
4761 |
len = (t_uscalar_t)sizeof (uint32_t); /* Default */ |
|
4762 |
||
4763 |
switch (option_name) { |
|
4764 |
case SO_TYPE: |
|
4765 |
value = so->so_type; |
|
4766 |
option = &value; |
|
4767 |
goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */ |
|
4768 |
||
4769 |
case SO_ERROR: |
|
4770 |
value = sogeterr(so); |
|
4771 |
option = &value; |
|
4772 |
goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */ |
|
4773 |
||
4774 |
case SO_ACCEPTCONN: |
|
4775 |
if (so->so_state & SS_ACCEPTCONN) |
|
4776 |
value = SO_ACCEPTCONN; |
|
4777 |
else |
|
4778 |
value = 0; |
|
4779 |
#ifdef DEBUG |
|
4780 |
if (value) { |
|
4781 |
dprintso(so, 1, |
|
4782 |
("sotpi_getsockopt: 0x%x is set\n", |
|
4783 |
option_name)); |
|
4784 |
} else { |
|
4785 |
dprintso(so, 1, |
|
4786 |
("sotpi_getsockopt: 0x%x not set\n", |
|
4787 |
option_name)); |
|
4788 |
} |
|
4789 |
#endif /* DEBUG */ |
|
4790 |
option = &value; |
|
4791 |
goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */ |
|
4792 |
||
4793 |
case SO_DEBUG: |
|
4794 |
case SO_REUSEADDR: |
|
4795 |
case SO_KEEPALIVE: |
|
4796 |
case SO_DONTROUTE: |
|
4797 |
case SO_BROADCAST: |
|
4798 |
case SO_USELOOPBACK: |
|
4799 |
case SO_OOBINLINE: |
|
4800 |
case SO_DGRAM_ERRIND: |
|
4801 |
value = (so->so_options & option_name); |
|
4802 |
#ifdef DEBUG |
|
4803 |
if (value) { |
|
4804 |
dprintso(so, 1, |
|
4805 |
("sotpi_getsockopt: 0x%x is set\n", |
|
4806 |
option_name)); |
|
4807 |
} else { |
|
4808 |
dprintso(so, 1, |
|
4809 |
("sotpi_getsockopt: 0x%x not set\n", |
|
4810 |
option_name)); |
|
4811 |
} |
|
4812 |
#endif /* DEBUG */ |
|
4813 |
option = &value; |
|
4814 |
goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */ |
|
4815 |
||
4816 |
/* |
|
4817 |
* The following options are only returned by sockfs when the |
|
4818 |
* T_SVR4_OPTMGMT_REQ fails. |
|
4819 |
*/ |
|
4820 |
case SO_LINGER: |
|
4821 |
option = &so->so_linger; |
|
4822 |
len = (t_uscalar_t)sizeof (struct linger); |
|
4823 |
break; |
|
4824 |
case SO_SNDBUF: { |
|
4825 |
ssize_t lvalue; |
|
4826 |
||
4827 |
/* |
|
4828 |
* If the option has not been set then get a default |
|
4829 |
* value from the read queue. This value is |
|
4830 |
* returned if the transport fails |
|
4831 |
* the T_SVR4_OPTMGMT_REQ. |
|
4832 |
*/ |
|
4833 |
lvalue = so->so_sndbuf; |
|
4834 |
if (lvalue == 0) { |
|
4835 |
mutex_exit(&so->so_lock); |
|
4836 |
(void) strqget(strvp2wq(SOTOV(so))->q_next, |
|
4837 |
QHIWAT, 0, &lvalue); |
|
4838 |
mutex_enter(&so->so_lock); |
|
4839 |
dprintso(so, 1, |
|
4840 |
("got SO_SNDBUF %ld from q\n", lvalue)); |
|
4841 |
} |
|
4842 |
value = (int)lvalue; |
|
4843 |
option = &value; |
|
4844 |
len = (t_uscalar_t)sizeof (so->so_sndbuf); |
|
4845 |
break; |
|
4846 |
} |
|
4847 |
case SO_RCVBUF: { |
|
4848 |
ssize_t lvalue; |
|
4849 |
||
4850 |
/* |
|
4851 |
* If the option has not been set then get a default |
|
4852 |
* value from the read queue. This value is |
|
4853 |
* returned if the transport fails |
|
4854 |
* the T_SVR4_OPTMGMT_REQ. |
|
4855 |
* |
|
4856 |
* XXX If SO_RCVBUF has been set and this is an |
|
4857 |
* XPG 4.2 application then do not ask the transport |
|
4858 |
* since the transport might adjust the value and not |
|
4859 |
* return exactly what was set by the application. |
|
4860 |
* For non-XPG 4.2 application we return the value |
|
4861 |
* that the transport is actually using. |
|
4862 |
*/ |
|
4863 |
lvalue = so->so_rcvbuf; |
|
4864 |
if (lvalue == 0) { |
|
4865 |
mutex_exit(&so->so_lock); |
|
4866 |
(void) strqget(RD(strvp2wq(SOTOV(so))), |
|
4867 |
QHIWAT, 0, &lvalue); |
|
4868 |
mutex_enter(&so->so_lock); |
|
4869 |
dprintso(so, 1, |
|
4870 |
("got SO_RCVBUF %ld from q\n", lvalue)); |
|
4871 |
} else if (flags & _SOGETSOCKOPT_XPG4_2) { |
|
4872 |
value = (int)lvalue; |
|
4873 |
option = &value; |
|
4874 |
goto copyout; /* skip asking transport */ |
|
4875 |
} |
|
4876 |
value = (int)lvalue; |
|
4877 |
option = &value; |
|
4878 |
len = (t_uscalar_t)sizeof (so->so_rcvbuf); |
|
4879 |
break; |
|
4880 |
} |
|
4881 |
#ifdef notyet |
|
4882 |
/* |
|
4883 |
* We do not implement the semantics of these options |
|
4884 |
* thus we shouldn't implement the options either. |
|
4885 |
*/ |
|
4886 |
case SO_SNDLOWAT: |
|
4887 |
value = so->so_sndlowat; |
|
4888 |
option = &value; |
|
4889 |
break; |
|
4890 |
case SO_RCVLOWAT: |
|
4891 |
value = so->so_rcvlowat; |
|
4892 |
option = &value; |
|
4893 |
break; |
|
4894 |
case SO_SNDTIMEO: |
|
4895 |
value = so->so_sndtimeo; |
|
4896 |
option = &value; |
|
4897 |
break; |
|
4898 |
case SO_RCVTIMEO: |
|
4899 |
value = so->so_rcvtimeo; |
|
4900 |
option = &value; |
|
4901 |
break; |
|
4902 |
#endif /* notyet */ |
|
4903 |
} |
|
4904 |
} |
|
4905 |
||
4906 |
if (so->so_family == AF_NCA) { |
|
4907 |
goto done2; |
|
4908 |
} |
|
4909 |
||
4910 |
mutex_exit(&so->so_lock); |
|
4911 |
||
4912 |
/* Send request */ |
|
4913 |
optmgmt_req.PRIM_type = T_SVR4_OPTMGMT_REQ; |
|
4914 |
optmgmt_req.MGMT_flags = T_CHECK; |
|
4915 |
optmgmt_req.OPT_length = (t_scalar_t)(sizeof (oh) + maxlen); |
|
4916 |
optmgmt_req.OPT_offset = (t_scalar_t)sizeof (optmgmt_req); |
|
4917 |
||
4918 |
oh.level = level; |
|
4919 |
oh.name = option_name; |
|
4920 |
oh.len = maxlen; |
|
4921 |
||
4922 |
mp = soallocproto3(&optmgmt_req, sizeof (optmgmt_req), |
|
4923 |
&oh, sizeof (oh), NULL, maxlen, 0, _ALLOC_SLEEP); |
|
4924 |
/* Let option management work in the presence of data flow control */ |
|
4925 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
4926 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0); |
|
4927 |
mp = NULL; |
|
4928 |
mutex_enter(&so->so_lock); |
|
4929 |
if (error) { |
|
4930 |
eprintsoline(so, error); |
|
4931 |
goto done2; |
|
4932 |
} |
|
4933 |
error = sowaitprim(so, T_SVR4_OPTMGMT_REQ, T_OPTMGMT_ACK, |
|
4934 |
(t_uscalar_t)(sizeof (*optmgmt_ack) + sizeof (*opt_res)), &mp, 0); |
|
4935 |
if (error) { |
|
4936 |
if (option != NULL) { |
|
4937 |
/* We have a fallback value */ |
|
4938 |
error = 0; |
|
4939 |
goto copyout; |
|
4940 |
} |
|
4941 |
eprintsoline(so, error); |
|
4942 |
goto done2; |
|
4943 |
} |
|
4944 |
ASSERT(mp); |
|
4945 |
optmgmt_ack = (struct T_optmgmt_ack *)mp->b_rptr; |
|
4946 |
opt_res = (struct opthdr *)sogetoff(mp, optmgmt_ack->OPT_offset, |
|
4947 |
optmgmt_ack->OPT_length, __TPI_ALIGN_SIZE); |
|
4948 |
if (opt_res == NULL) { |
|
4949 |
if (option != NULL) { |
|
4950 |
/* We have a fallback value */ |
|
4951 |
error = 0; |
|
4952 |
goto copyout; |
|
4953 |
} |
|
4954 |
error = EPROTO; |
|
4955 |
eprintsoline(so, error); |
|
4956 |
goto done; |
|
4957 |
} |
|
4958 |
option = &opt_res[1]; |
|
4959 |
||
4960 |
/* check to ensure that the option is within bounds */ |
|
4961 |
if (((uintptr_t)option + opt_res->len < (uintptr_t)option) || |
|
4962 |
(uintptr_t)option + opt_res->len > (uintptr_t)mp->b_wptr) { |
|
4963 |
if (option != NULL) { |
|
4964 |
/* We have a fallback value */ |
|
4965 |
error = 0; |
|
4966 |
goto copyout; |
|
4967 |
} |
|
4968 |
error = EPROTO; |
|
4969 |
eprintsoline(so, error); |
|
4970 |
goto done; |
|
4971 |
} |
|
4972 |
||
4973 |
len = opt_res->len; |
|
4974 |
||
4975 |
copyout: { |
|
4976 |
t_uscalar_t size = MIN(len, maxlen); |
|
4977 |
bcopy(option, optval, size); |
|
4978 |
bcopy(&size, optlenp, sizeof (size)); |
|
4979 |
} |
|
4980 |
done: |
|
4981 |
freemsg(mp); |
|
4982 |
done2: |
|
4983 |
so_unlock_single(so, SOLOCKED); |
|
4984 |
mutex_exit(&so->so_lock); |
|
4985 |
return (error); |
|
4986 |
} |
|
4987 |
||
4988 |
/* |
|
4989 |
* Set socket options. All options are passed down in a T_SVR4_OPTMGMT_REQ. |
|
4990 |
* SOL_SOCKET options are also recorded in the sonode. A setsockopt for |
|
4991 |
* SOL_SOCKET options will not fail just because the T_SVR4_OPTMGMT_REQ fails - |
|
4992 |
* setsockopt has to work even if the transport does not support the option. |
|
4993 |
*/ |
|
4994 |
int |
|
4995 |
sotpi_setsockopt(struct sonode *so, int level, int option_name, |
|
4996 |
const void *optval, t_uscalar_t optlen) |
|
4997 |
{ |
|
4998 |
struct T_optmgmt_req optmgmt_req; |
|
4999 |
struct opthdr oh; |
|
5000 |
mblk_t *mp; |
|
5001 |
int error = 0; |
|
5002 |
boolean_t handled = B_FALSE; |
|
5003 |
||
5004 |
dprintso(so, 1, ("sotpi_setsockopt(%p, 0x%x, 0x%x, %p, %d) %s\n", |
|
5005 |
so, level, option_name, optval, optlen, |
|
5006 |
pr_state(so->so_state, so->so_mode))); |
|
5007 |
||
5008 |
||
5009 |
/* X/Open requires this check */ |
|
5010 |
if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) { |
|
5011 |
if (xnet_check_print) |
|
5012 |
printf("sockfs: X/Open setsockopt check => EINVAL\n"); |
|
5013 |
return (EINVAL); |
|
5014 |
} |
|
5015 |
||
5016 |
/* Caller allocates aligned optval, or passes null */ |
|
5017 |
ASSERT(((uintptr_t)optval & (sizeof (t_scalar_t) - 1)) == 0); |
|
5018 |
/* If optval is null optlen is 0, and vice-versa */ |
|
5019 |
ASSERT(optval != NULL || optlen == 0); |
|
5020 |
ASSERT(optlen != 0 || optval == NULL); |
|
5021 |
||
5022 |
mutex_enter(&so->so_lock); |
|
5023 |
so_lock_single(so); /* Set SOLOCKED */ |
|
5024 |
mutex_exit(&so->so_lock); |
|
5025 |
||
5026 |
if (so->so_family == AF_NCA) { |
|
5027 |
/* Ignore any flow control problems with the transport. */ |
|
5028 |
mutex_enter(&so->so_lock); |
|
5029 |
goto done; |
|
5030 |
} |
|
5031 |
||
5032 |
/* |
|
5033 |
* For SOCKET or TCP level options, try to set it here itself |
|
5034 |
* provided socket has not been popped and we know the tcp |
|
5035 |
* structure (stored in so_priv). |
|
5036 |
*/ |
|
5037 |
if ((level == SOL_SOCKET || level == IPPROTO_TCP) && |
|
5038 |
(so->so_family == AF_INET || so->so_family == AF_INET6) && |
|
5039 |
(so->so_version == SOV_SOCKSTREAM) && (so->so_priv != NULL)) { |
|
5040 |
tcp_t *tcp = so->so_priv; |
|
5041 |
boolean_t onoff; |
|
5042 |
||
5043 |
#define intvalue (*(int32_t *)optval) |
|
5044 |
||
5045 |
switch (level) { |
|
5046 |
case SOL_SOCKET: |
|
5047 |
switch (option_name) { /* Check length param */ |
|
5048 |
case SO_DEBUG: |
|
5049 |
case SO_REUSEADDR: |
|
5050 |
case SO_DONTROUTE: |
|
5051 |
case SO_BROADCAST: |
|
5052 |
case SO_USELOOPBACK: |
|
5053 |
case SO_OOBINLINE: |
|
5054 |
case SO_DGRAM_ERRIND: |
|
5055 |
if (optlen != (t_uscalar_t)sizeof (int32_t)) { |
|
5056 |
error = EINVAL; |
|
5057 |
eprintsoline(so, error); |
|
5058 |
mutex_enter(&so->so_lock); |
|
5059 |
goto done2; |
|
5060 |
} |
|
5061 |
ASSERT(optval); |
|
5062 |
onoff = intvalue != 0; |
|
5063 |
handled = B_TRUE; |
|
5064 |
break; |
|
5065 |
case SO_LINGER: |
|
5066 |
if (optlen != |
|
5067 |
(t_uscalar_t)sizeof (struct linger)) { |
|
5068 |
error = EINVAL; |
|
5069 |
eprintsoline(so, error); |
|
5070 |
mutex_enter(&so->so_lock); |
|
5071 |
goto done2; |
|
5072 |
} |
|
5073 |
ASSERT(optval); |
|
5074 |
handled = B_TRUE; |
|
5075 |
break; |
|
5076 |
} |
|
5077 |
||
5078 |
switch (option_name) { /* Do actions */ |
|
5079 |
case SO_LINGER: { |
|
5080 |
struct linger *lgr = (struct linger *)optval; |
|
5081 |
||
5082 |
if (lgr->l_onoff) { |
|
5083 |
tcp->tcp_linger = 1; |
|
5084 |
tcp->tcp_lingertime = lgr->l_linger; |
|
5085 |
so->so_linger.l_onoff = SO_LINGER; |
|
5086 |
so->so_options |= SO_LINGER; |
|
5087 |
} else { |
|
5088 |
tcp->tcp_linger = 0; |
|
5089 |
tcp->tcp_lingertime = 0; |
|
5090 |
so->so_linger.l_onoff = 0; |
|
5091 |
so->so_options &= ~SO_LINGER; |
|
5092 |
} |
|
5093 |
so->so_linger.l_linger = lgr->l_linger; |
|
5094 |
handled = B_TRUE; |
|
5095 |
break; |
|
5096 |
} |
|
5097 |
case SO_DEBUG: |
|
5098 |
tcp->tcp_debug = onoff; |
|
5099 |
#ifdef SOCK_TEST |
|
5100 |
if (intvalue & 2) |
|
5101 |
sock_test_timelimit = 10 * hz; |
|
5102 |
else |
|
5103 |
sock_test_timelimit = 0; |
|
5104 |
||
5105 |
if (intvalue & 4) |
|
5106 |
do_useracc = 0; |
|
5107 |
else |
|
5108 |
do_useracc = 1; |
|
5109 |
#endif /* SOCK_TEST */ |
|
5110 |
break; |
|
5111 |
case SO_DONTROUTE: |
|
5112 |
/* |
|
5113 |
* SO_DONTROUTE, SO_USELOOPBACK and |
|
5114 |
* SO_BROADCAST are only of interest to IP. |
|
5115 |
* We track them here only so |
|
5116 |
* that we can report their current value. |
|
5117 |
*/ |
|
5118 |
tcp->tcp_dontroute = onoff; |
|
5119 |
if (onoff) |
|
5120 |
so->so_options |= option_name; |
|
5121 |
else |
|
5122 |
so->so_options &= ~option_name; |
|
5123 |
break; |
|
5124 |
case SO_USELOOPBACK: |
|
5125 |
tcp->tcp_useloopback = onoff; |
|
5126 |
if (onoff) |
|
5127 |
so->so_options |= option_name; |
|
5128 |
else |
|
5129 |
so->so_options &= ~option_name; |
|
5130 |
break; |
|
5131 |
case SO_BROADCAST: |
|
5132 |
tcp->tcp_broadcast = onoff; |
|
5133 |
if (onoff) |
|
5134 |
so->so_options |= option_name; |
|
5135 |
else |
|
5136 |
so->so_options &= ~option_name; |
|
5137 |
break; |
|
5138 |
case SO_REUSEADDR: |
|
5139 |
tcp->tcp_reuseaddr = onoff; |
|
5140 |
if (onoff) |
|
5141 |
so->so_options |= option_name; |
|
5142 |
else |
|
5143 |
so->so_options &= ~option_name; |
|
5144 |
break; |
|
5145 |
case SO_OOBINLINE: |
|
5146 |
tcp->tcp_oobinline = onoff; |
|
5147 |
if (onoff) |
|
5148 |
so->so_options |= option_name; |
|
5149 |
else |
|
5150 |
so->so_options &= ~option_name; |
|
5151 |
break; |
|
5152 |
case SO_DGRAM_ERRIND: |
|
5153 |
tcp->tcp_dgram_errind = onoff; |
|
5154 |
if (onoff) |
|
5155 |
so->so_options |= option_name; |
|
5156 |
else |
|
5157 |
so->so_options &= ~option_name; |
|
5158 |
break; |
|
5159 |
} |
|
5160 |
break; |
|
5161 |
case IPPROTO_TCP: |
|
5162 |
switch (option_name) { |
|
5163 |
case TCP_NODELAY: |
|
5164 |
if (optlen != (t_uscalar_t)sizeof (int32_t)) { |
|
5165 |
error = EINVAL; |
|
5166 |
eprintsoline(so, error); |
|
5167 |
mutex_enter(&so->so_lock); |
|
5168 |
goto done2; |
|
5169 |
} |
|
5170 |
ASSERT(optval); |
|
5171 |
tcp->tcp_naglim = intvalue ? 1 : tcp->tcp_mss; |
|
5172 |
handled = B_TRUE; |
|
5173 |
break; |
|
5174 |
} |
|
5175 |
break; |
|
5176 |
default: |
|
5177 |
handled = B_FALSE; |
|
5178 |
break; |
|
5179 |
} |
|
5180 |
} |
|
5181 |
||
5182 |
if (handled) { |
|
5183 |
mutex_enter(&so->so_lock); |
|
5184 |
goto done2; |
|
5185 |
} |
|
5186 |
||
5187 |
optmgmt_req.PRIM_type = T_SVR4_OPTMGMT_REQ; |
|
5188 |
optmgmt_req.MGMT_flags = T_NEGOTIATE; |
|
5189 |
optmgmt_req.OPT_length = (t_scalar_t)sizeof (oh) + optlen; |
|
5190 |
optmgmt_req.OPT_offset = (t_scalar_t)sizeof (optmgmt_req); |
|
5191 |
||
5192 |
oh.level = level; |
|
5193 |
oh.name = option_name; |
|
5194 |
oh.len = optlen; |
|
5195 |
||
5196 |
mp = soallocproto3(&optmgmt_req, sizeof (optmgmt_req), |
|
5197 |
&oh, sizeof (oh), optval, optlen, 0, _ALLOC_SLEEP); |
|
5198 |
/* Let option management work in the presence of data flow control */ |
|
5199 |
error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0, |
|
5200 |
MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0); |
|
5201 |
mp = NULL; |
|
5202 |
mutex_enter(&so->so_lock); |
|
5203 |
if (error) { |
|
5204 |
eprintsoline(so, error); |
|
5205 |
goto done; |
|
5206 |
} |
|
5207 |
error = sowaitprim(so, T_SVR4_OPTMGMT_REQ, T_OPTMGMT_ACK, |
|
5208 |
(t_uscalar_t)sizeof (struct T_optmgmt_ack), &mp, 0); |
|
5209 |
if (error) { |
|
5210 |
eprintsoline(so, error); |
|
5211 |
goto done; |
|
5212 |
} |
|
5213 |
ASSERT(mp); |
|
5214 |
/* No need to verify T_optmgmt_ack */ |
|
5215 |
freemsg(mp); |
|
5216 |
done: |
|
5217 |
/* |
|
5218 |
* Check for SOL_SOCKET options and record their values. |
|
5219 |
* If we know about a SOL_SOCKET parameter and the transport |
|
5220 |
* failed it with TBADOPT or TOUTSTATE (i.e. ENOPROTOOPT or |
|
5221 |
* EPROTO) we let the setsockopt succeed. |
|
5222 |
*/ |
|
5223 |
if (level == SOL_SOCKET) { |
|
5224 |
/* Check parameters */ |
|
5225 |
switch (option_name) { |
|
5226 |
case SO_DEBUG: |
|
5227 |
case SO_REUSEADDR: |
|
5228 |
case SO_KEEPALIVE: |
|
5229 |
case SO_DONTROUTE: |
|
5230 |
case SO_BROADCAST: |
|
5231 |
case SO_USELOOPBACK: |
|
5232 |
case SO_OOBINLINE: |
|
5233 |
case SO_SNDBUF: |
|
5234 |
case SO_RCVBUF: |
|
5235 |
#ifdef notyet |
|
5236 |
case SO_SNDLOWAT: |
|
5237 |
case SO_RCVLOWAT: |
|
5238 |
case SO_SNDTIMEO: |
|
5239 |
case SO_RCVTIMEO: |
|
5240 |
#endif /* notyet */ |
|
5241 |
case SO_DGRAM_ERRIND: |
|
5242 |
if (optlen != (t_uscalar_t)sizeof (int32_t)) { |
|
5243 |
error = EINVAL; |
|
5244 |
eprintsoline(so, error); |
|
5245 |
goto done2; |
|
5246 |
} |
|
5247 |
ASSERT(optval); |
|
5248 |
handled = B_TRUE; |
|
5249 |
break; |
|
5250 |
case SO_LINGER: |
|
5251 |
if (optlen != (t_uscalar_t)sizeof (struct linger)) { |
|
5252 |
error = EINVAL; |
|
5253 |
eprintsoline(so, error); |
|
5254 |
goto done2; |
|
5255 |
} |
|
5256 |
ASSERT(optval); |
|
5257 |
handled = B_TRUE; |
|
5258 |
break; |
|
5259 |
} |
|
5260 |
||
5261 |
#define intvalue (*(int32_t *)optval) |
|
5262 |
||
5263 |
switch (option_name) { |
|
5264 |
case SO_TYPE: |
|
5265 |
case SO_ERROR: |
|
5266 |
case SO_ACCEPTCONN: |
|
5267 |
/* Can't be set */ |
|
5268 |
error = ENOPROTOOPT; |
|
5269 |
goto done2; |
|
5270 |
case SO_LINGER: { |
|
5271 |
struct linger *l = (struct linger *)optval; |
|
5272 |
||
5273 |
so->so_linger.l_linger = l->l_linger; |
|
5274 |
if (l->l_onoff) { |
|
5275 |
so->so_linger.l_onoff = SO_LINGER; |
|
5276 |
so->so_options |= SO_LINGER; |
|
5277 |
} else { |
|
5278 |
so->so_linger.l_onoff = 0; |
|
5279 |
so->so_options &= ~SO_LINGER; |
|
5280 |
} |
|
5281 |
break; |
|
5282 |
} |
|
5283 |
||
5284 |
case SO_DEBUG: |
|
5285 |
#ifdef SOCK_TEST |
|
5286 |
if (intvalue & 2) |
|
5287 |
sock_test_timelimit = 10 * hz; |
|
5288 |
else |
|
5289 |
sock_test_timelimit = 0; |
|
5290 |
||
5291 |
if (intvalue & 4) |
|
5292 |
do_useracc = 0; |
|
5293 |
else |
|
5294 |
do_useracc = 1; |
|
5295 |
#endif /* SOCK_TEST */ |
|
5296 |
/* FALLTHRU */ |
|
5297 |
case SO_REUSEADDR: |
|
5298 |
case SO_KEEPALIVE: |
|
5299 |
case SO_DONTROUTE: |
|
5300 |
case SO_BROADCAST: |
|
5301 |
case SO_USELOOPBACK: |
|
5302 |
case SO_OOBINLINE: |
|
5303 |
case SO_DGRAM_ERRIND: |
|
5304 |
if (intvalue != 0) { |
|
5305 |
dprintso(so, 1, |
|
5306 |
("sotpi_setsockopt: setting 0x%x\n", |
|
5307 |
option_name)); |
|
5308 |
so->so_options |= option_name; |
|
5309 |
} else { |
|
5310 |
dprintso(so, 1, |
|
5311 |
("sotpi_setsockopt: clearing 0x%x\n", |
|
5312 |
option_name)); |
|
5313 |
so->so_options &= ~option_name; |
|
5314 |
} |
|
5315 |
break; |
|
5316 |
/* |
|
5317 |
* The following options are only returned by us when the |
|
5318 |
* T_SVR4_OPTMGMT_REQ fails. |
|
5319 |
* XXX XPG 4.2 applications retrieve SO_RCVBUF from sockfs |
|
5320 |
* since the transport might adjust the value and not |
|
5321 |
* return exactly what was set by the application. |
|
5322 |
*/ |
|
5323 |
case SO_SNDBUF: |
|
5324 |
so->so_sndbuf = intvalue; |
|
5325 |
break; |
|
5326 |
case SO_RCVBUF: |
|
5327 |
so->so_rcvbuf = intvalue; |
|
5328 |
break; |
|
5329 |
#ifdef notyet |
|
5330 |
/* |
|
5331 |
* We do not implement the semantics of these options |
|
5332 |
* thus we shouldn't implement the options either. |
|
5333 |
*/ |
|
5334 |
case SO_SNDLOWAT: |
|
5335 |
so->so_sndlowat = intvalue; |
|
5336 |
break; |
|
5337 |
case SO_RCVLOWAT: |
|
5338 |
so->so_rcvlowat = intvalue; |
|
5339 |
break; |
|
5340 |
case SO_SNDTIMEO: |
|
5341 |
so->so_sndtimeo = intvalue; |
|
5342 |
break; |
|
5343 |
case SO_RCVTIMEO: |
|
5344 |
so->so_rcvtimeo = intvalue; |
|
5345 |
break; |
|
5346 |
#endif /* notyet */ |
|
5347 |
} |
|
5348 |
#undef intvalue |
|
5349 |
||
5350 |
if (error) { |
|
5351 |
if ((error == ENOPROTOOPT || error == EPROTO || |
|
5352 |
error == EINVAL) && handled) { |
|
5353 |
dprintso(so, 1, |
|
5354 |
("setsockopt: ignoring error %d for 0x%x\n", |
|
5355 |
error, option_name)); |
|
5356 |
error = 0; |
|
5357 |
} |
|
5358 |
} |
|
5359 |
} |
|
5360 |
done2: |
|
5361 |
ret: |
|
5362 |
so_unlock_single(so, SOLOCKED); |
|
5363 |
mutex_exit(&so->so_lock); |
|
5364 |
return (error); |
|
5365 |
} |