Annotation of src/sys/kern/uipc_socket.c, Revision 1.205
1.205 ! bouyer 1: /* $NetBSD: uipc_socket.c,v 1.204 2011/06/26 16:42:42 christos Exp $ */
1.64 thorpej 2:
3: /*-
1.188 ad 4: * Copyright (c) 2002, 2007, 2008, 2009 The NetBSD Foundation, Inc.
1.64 thorpej 5: * All rights reserved.
6: *
7: * This code is derived from software contributed to The NetBSD Foundation
1.188 ad 8: * by Jason R. Thorpe of Wasabi Systems, Inc, and by Andrew Doran.
1.64 thorpej 9: *
10: * Redistribution and use in source and binary forms, with or without
11: * modification, are permitted provided that the following conditions
12: * are met:
13: * 1. Redistributions of source code must retain the above copyright
14: * notice, this list of conditions and the following disclaimer.
15: * 2. Redistributions in binary form must reproduce the above copyright
16: * notice, this list of conditions and the following disclaimer in the
17: * documentation and/or other materials provided with the distribution.
18: *
19: * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20: * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21: * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22: * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23: * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24: * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25: * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26: * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27: * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28: * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29: * POSSIBILITY OF SUCH DAMAGE.
30: */
1.16 cgd 31:
1.1 cgd 32: /*
1.159 ad 33: * Copyright (c) 2004 The FreeBSD Foundation
34: * Copyright (c) 2004 Robert Watson
1.15 mycroft 35: * Copyright (c) 1982, 1986, 1988, 1990, 1993
36: * The Regents of the University of California. All rights reserved.
1.1 cgd 37: *
38: * Redistribution and use in source and binary forms, with or without
39: * modification, are permitted provided that the following conditions
40: * are met:
41: * 1. Redistributions of source code must retain the above copyright
42: * notice, this list of conditions and the following disclaimer.
43: * 2. Redistributions in binary form must reproduce the above copyright
44: * notice, this list of conditions and the following disclaimer in the
45: * documentation and/or other materials provided with the distribution.
1.85 agc 46: * 3. Neither the name of the University nor the names of its contributors
1.1 cgd 47: * may be used to endorse or promote products derived from this software
48: * without specific prior written permission.
49: *
50: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60: * SUCH DAMAGE.
61: *
1.32 fvdl 62: * @(#)uipc_socket.c 8.6 (Berkeley) 5/2/95
1.1 cgd 63: */
1.59 lukem 64:
65: #include <sys/cdefs.h>
1.205 ! bouyer 66: __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.204 2011/06/26 16:42:42 christos Exp $");
1.64 thorpej 67:
1.179 christos 68: #include "opt_compat_netbsd.h"
1.64 thorpej 69: #include "opt_sock_counters.h"
70: #include "opt_sosend_loan.h"
1.81 martin 71: #include "opt_mbuftrace.h"
1.84 ragge 72: #include "opt_somaxkva.h"
1.167 ad 73: #include "opt_multiprocessor.h" /* XXX */
1.1 cgd 74:
1.9 mycroft 75: #include <sys/param.h>
76: #include <sys/systm.h>
77: #include <sys/proc.h>
78: #include <sys/file.h>
1.142 dyoung 79: #include <sys/filedesc.h>
1.173 plunky 80: #include <sys/kmem.h>
1.9 mycroft 81: #include <sys/mbuf.h>
82: #include <sys/domain.h>
83: #include <sys/kernel.h>
84: #include <sys/protosw.h>
85: #include <sys/socket.h>
86: #include <sys/socketvar.h>
1.21 christos 87: #include <sys/signalvar.h>
1.9 mycroft 88: #include <sys/resourcevar.h>
1.174 pooka 89: #include <sys/uidinfo.h>
1.72 jdolecek 90: #include <sys/event.h>
1.89 christos 91: #include <sys/poll.h>
1.118 elad 92: #include <sys/kauth.h>
1.136 ad 93: #include <sys/mutex.h>
94: #include <sys/condvar.h>
1.205 ! bouyer 95: #include <sys/kthread.h>
1.37 thorpej 96:
1.179 christos 97: #ifdef COMPAT_50
98: #include <compat/sys/time.h>
1.184 christos 99: #include <compat/sys/socket.h>
1.179 christos 100: #endif
101:
1.202 uebayasi 102: #include <uvm/uvm_extern.h>
103: #include <uvm/uvm_loan.h>
104: #include <uvm/uvm_page.h>
1.64 thorpej 105:
1.77 thorpej 106: MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
107: MALLOC_DEFINE(M_SONAME, "soname", "socket name");
1.37 thorpej 108:
1.142 dyoung 109: extern const struct fileops socketops;
110:
1.54 lukem 111: extern int somaxconn; /* patchable (XXX sysctl) */
112: int somaxconn = SOMAXCONN;
1.160 ad 113: kmutex_t *softnet_lock;
1.49 jonathan 114:
1.64 thorpej 115: #ifdef SOSEND_COUNTERS
116: #include <sys/device.h>
117:
1.113 thorpej 118: static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 119: NULL, "sosend", "loan big");
1.113 thorpej 120: static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 121: NULL, "sosend", "copy big");
1.113 thorpej 122: static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 123: NULL, "sosend", "copy small");
1.113 thorpej 124: static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 125: NULL, "sosend", "kva limit");
126:
127: #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++
128:
1.101 matt 129: EVCNT_ATTACH_STATIC(sosend_loan_big);
130: EVCNT_ATTACH_STATIC(sosend_copy_big);
131: EVCNT_ATTACH_STATIC(sosend_copy_small);
132: EVCNT_ATTACH_STATIC(sosend_kvalimit);
1.64 thorpej 133: #else
134:
135: #define SOSEND_COUNTER_INCR(ev) /* nothing */
136:
137: #endif /* SOSEND_COUNTERS */
138:
1.119 yamt 139: static struct callback_entry sokva_reclaimerentry;
1.1 cgd 140:
1.167 ad 141: #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
1.121 yamt 142: int sock_loan_thresh = -1;
1.71 thorpej 143: #else
1.121 yamt 144: int sock_loan_thresh = 4096;
1.65 thorpej 145: #endif
1.64 thorpej 146:
1.136 ad 147: static kmutex_t so_pendfree_lock;
1.205 ! bouyer 148: static struct mbuf *so_pendfree = NULL;
1.64 thorpej 149:
1.84 ragge 150: #ifndef SOMAXKVA
151: #define SOMAXKVA (16 * 1024 * 1024)
152: #endif
153: int somaxkva = SOMAXKVA;
1.113 thorpej 154: static int socurkva;
1.136 ad 155: static kcondvar_t socurkva_cv;
1.64 thorpej 156:
1.191 elad 157: static kauth_listener_t socket_listener;
158:
1.64 thorpej 159: #define SOCK_LOAN_CHUNK 65536
160:
1.205 ! bouyer 161: static void sopendfree_thread(void *);
! 162: static kcondvar_t pendfree_thread_cv;
! 163: static lwp_t *sopendfree_lwp;
1.93 yamt 164:
1.178 pooka 165: static void sysctl_kern_somaxkva_setup(void);
166: static struct sysctllog *socket_sysctllog;
167:
1.113 thorpej 168: static vsize_t
1.129 yamt 169: sokvareserve(struct socket *so, vsize_t len)
1.80 yamt 170: {
1.98 christos 171: int error;
1.80 yamt 172:
1.136 ad 173: mutex_enter(&so_pendfree_lock);
1.80 yamt 174: while (socurkva + len > somaxkva) {
175: SOSEND_COUNTER_INCR(&sosend_kvalimit);
1.136 ad 176: error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
1.98 christos 177: if (error) {
178: len = 0;
179: break;
180: }
1.80 yamt 181: }
1.93 yamt 182: socurkva += len;
1.136 ad 183: mutex_exit(&so_pendfree_lock);
1.98 christos 184: return len;
1.95 yamt 185: }
186:
1.113 thorpej 187: static void
1.95 yamt 188: sokvaunreserve(vsize_t len)
189: {
190:
1.136 ad 191: mutex_enter(&so_pendfree_lock);
1.95 yamt 192: socurkva -= len;
1.136 ad 193: cv_broadcast(&socurkva_cv);
194: mutex_exit(&so_pendfree_lock);
1.95 yamt 195: }
196:
197: /*
198: * sokvaalloc: allocate kva for loan.
199: */
200:
201: vaddr_t
202: sokvaalloc(vsize_t len, struct socket *so)
203: {
204: vaddr_t lva;
205:
206: /*
207: * reserve kva.
208: */
209:
1.98 christos 210: if (sokvareserve(so, len) == 0)
211: return 0;
1.93 yamt 212:
213: /*
214: * allocate kva.
215: */
1.80 yamt 216:
1.109 yamt 217: lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
1.95 yamt 218: if (lva == 0) {
219: sokvaunreserve(len);
1.80 yamt 220: return (0);
1.95 yamt 221: }
1.80 yamt 222:
223: return lva;
224: }
225:
1.93 yamt 226: /*
227: * sokvafree: free kva for loan.
228: */
229:
1.80 yamt 230: void
231: sokvafree(vaddr_t sva, vsize_t len)
232: {
1.93 yamt 233:
234: /*
235: * free kva.
236: */
1.80 yamt 237:
1.109 yamt 238: uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
1.93 yamt 239:
240: /*
241: * unreserve kva.
242: */
243:
1.95 yamt 244: sokvaunreserve(len);
1.80 yamt 245: }
246:
1.64 thorpej 247: static void
1.134 christos 248: sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
1.64 thorpej 249: {
1.156 yamt 250: vaddr_t sva, eva;
1.64 thorpej 251: vsize_t len;
1.156 yamt 252: int npgs;
253:
254: KASSERT(pgs != NULL);
1.64 thorpej 255:
256: eva = round_page((vaddr_t) buf + size);
257: sva = trunc_page((vaddr_t) buf);
258: len = eva - sva;
259: npgs = len >> PAGE_SHIFT;
260:
261: pmap_kremove(sva, len);
262: pmap_update(pmap_kernel());
263: uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
1.80 yamt 264: sokvafree(sva, len);
1.64 thorpej 265: }
266:
1.93 yamt 267: /*
1.205 ! bouyer 268: * sopendfree_thread: free mbufs on "pendfree" list.
1.136 ad 269: * unlock and relock so_pendfree_lock when freeing mbufs.
1.93 yamt 270: */
271:
1.205 ! bouyer 272: static void
! 273: sopendfree_thread(void *v)
1.93 yamt 274: {
1.137 ad 275: struct mbuf *m, *next;
1.205 ! bouyer 276: size_t rv;
1.93 yamt 277:
1.205 ! bouyer 278: mutex_enter(&so_pendfree_lock);
1.64 thorpej 279:
1.205 ! bouyer 280: for (;;) {
! 281: rv = 0;
! 282: while (so_pendfree != NULL) {
! 283: m = so_pendfree;
! 284: so_pendfree = NULL;
! 285: mutex_exit(&so_pendfree_lock);
! 286:
! 287: for (; m != NULL; m = next) {
! 288: next = m->m_next;
! 289: KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
! 290: KASSERT(m->m_ext.ext_refcnt == 0);
! 291:
! 292: rv += m->m_ext.ext_size;
! 293: sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
! 294: m->m_ext.ext_size);
! 295: pool_cache_put(mb_cache, m);
! 296: }
1.93 yamt 297:
1.205 ! bouyer 298: mutex_enter(&so_pendfree_lock);
1.93 yamt 299: }
1.205 ! bouyer 300: if (rv)
! 301: cv_broadcast(&socurkva_cv);
! 302: cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
1.64 thorpej 303: }
1.205 ! bouyer 304: panic("sopendfree_thread");
! 305: /* NOTREACHED */
1.64 thorpej 306: }
307:
1.80 yamt 308: void
1.134 christos 309: soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
1.64 thorpej 310: {
311:
1.156 yamt 312: KASSERT(m != NULL);
1.64 thorpej 313:
1.93 yamt 314: /*
315: * postpone freeing mbuf.
316: *
317: * we can't do it in interrupt context
318: * because we need to put kva back to kernel_map.
319: */
320:
1.136 ad 321: mutex_enter(&so_pendfree_lock);
1.92 yamt 322: m->m_next = so_pendfree;
323: so_pendfree = m;
1.205 ! bouyer 324: cv_signal(&pendfree_thread_cv);
1.136 ad 325: mutex_exit(&so_pendfree_lock);
1.64 thorpej 326: }
327:
328: static long
329: sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
330: {
331: struct iovec *iov = uio->uio_iov;
332: vaddr_t sva, eva;
333: vsize_t len;
1.156 yamt 334: vaddr_t lva;
335: int npgs, error;
336: vaddr_t va;
337: int i;
1.64 thorpej 338:
1.116 yamt 339: if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
1.64 thorpej 340: return (0);
341:
342: if (iov->iov_len < (size_t) space)
343: space = iov->iov_len;
344: if (space > SOCK_LOAN_CHUNK)
345: space = SOCK_LOAN_CHUNK;
346:
347: eva = round_page((vaddr_t) iov->iov_base + space);
348: sva = trunc_page((vaddr_t) iov->iov_base);
349: len = eva - sva;
350: npgs = len >> PAGE_SHIFT;
351:
1.79 thorpej 352: KASSERT(npgs <= M_EXT_MAXPAGES);
353:
1.80 yamt 354: lva = sokvaalloc(len, so);
1.64 thorpej 355: if (lva == 0)
1.80 yamt 356: return 0;
1.64 thorpej 357:
1.116 yamt 358: error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
1.79 thorpej 359: m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
1.64 thorpej 360: if (error) {
1.80 yamt 361: sokvafree(lva, len);
1.64 thorpej 362: return (0);
363: }
364:
365: for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
1.79 thorpej 366: pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
1.194 cegger 367: VM_PROT_READ, 0);
1.64 thorpej 368: pmap_update(pmap_kernel());
369:
370: lva += (vaddr_t) iov->iov_base & PAGE_MASK;
371:
1.134 christos 372: MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
1.79 thorpej 373: m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
1.64 thorpej 374:
375: uio->uio_resid -= space;
376: /* uio_offset not updated, not set/used for write(2) */
1.134 christos 377: uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
1.64 thorpej 378: uio->uio_iov->iov_len -= space;
379: if (uio->uio_iov->iov_len == 0) {
380: uio->uio_iov++;
381: uio->uio_iovcnt--;
382: }
383:
384: return (space);
385: }
386:
1.119 yamt 387: static int
1.129 yamt 388: sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.119 yamt 389: {
390:
391: KASSERT(ce == &sokva_reclaimerentry);
392: KASSERT(obj == NULL);
393:
394: if (!vm_map_starved_p(kernel_map)) {
395: return CALLBACK_CHAIN_ABORT;
396: }
397: return CALLBACK_CHAIN_CONTINUE;
398: }
399:
1.142 dyoung 400: struct mbuf *
1.147 dyoung 401: getsombuf(struct socket *so, int type)
1.142 dyoung 402: {
403: struct mbuf *m;
404:
1.147 dyoung 405: m = m_get(M_WAIT, type);
1.142 dyoung 406: MCLAIM(m, so->so_mowner);
407: return m;
408: }
409:
1.191 elad 410: static int
411: socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
412: void *arg0, void *arg1, void *arg2, void *arg3)
413: {
414: int result;
415: enum kauth_network_req req;
416:
417: result = KAUTH_RESULT_DEFER;
418: req = (enum kauth_network_req)arg0;
419:
1.193 elad 420: if ((action != KAUTH_NETWORK_SOCKET) &&
421: (action != KAUTH_NETWORK_BIND))
1.191 elad 422: return result;
423:
424: switch (req) {
1.193 elad 425: case KAUTH_REQ_NETWORK_BIND_PORT:
426: result = KAUTH_RESULT_ALLOW;
427: break;
428:
1.191 elad 429: case KAUTH_REQ_NETWORK_SOCKET_DROP: {
430: /* Normal users can only drop their own connections. */
431: struct socket *so = (struct socket *)arg1;
432:
1.199 elad 433: if (proc_uidmatch(cred, so->so_cred))
1.191 elad 434: result = KAUTH_RESULT_ALLOW;
435:
436: break;
437: }
438:
439: case KAUTH_REQ_NETWORK_SOCKET_OPEN:
440: /* We allow "raw" routing/bluetooth sockets to anyone. */
1.203 matt 441: if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
442: || (u_long)arg1 == PF_BLUETOOTH) {
1.191 elad 443: result = KAUTH_RESULT_ALLOW;
1.203 matt 444: } else {
1.191 elad 445: /* Privileged, let secmodel handle this. */
446: if ((u_long)arg2 == SOCK_RAW)
447: break;
448: }
449:
450: result = KAUTH_RESULT_ALLOW;
451:
452: break;
453:
1.192 elad 454: case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
455: result = KAUTH_RESULT_ALLOW;
456:
457: break;
458:
1.191 elad 459: default:
460: break;
461: }
462:
463: return result;
464: }
465:
1.119 yamt 466: void
467: soinit(void)
468: {
469:
1.178 pooka 470: sysctl_kern_somaxkva_setup();
471:
1.148 ad 472: mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
1.160 ad 473: softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
1.136 ad 474: cv_init(&socurkva_cv, "sokva");
1.205 ! bouyer 475: cv_init(&pendfree_thread_cv, "sopendfr");
1.166 ad 476: soinit2();
1.136 ad 477:
1.119 yamt 478: /* Set the initial adjusted socket buffer size. */
479: if (sb_max_set(sb_max))
480: panic("bad initial sb_max value: %lu", sb_max);
481:
482: callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
483: &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
1.191 elad 484:
485: socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
486: socket_listener_cb, NULL);
1.119 yamt 487: }
488:
1.205 ! bouyer 489: void
! 490: soinit1(void)
! 491: {
! 492: int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
! 493: sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
! 494: if (error)
! 495: panic("soinit1 %d", error);
! 496: }
! 497:
1.1 cgd 498: /*
499: * Socket operation routines.
500: * These routines are called by the routines in
501: * sys_socket.c or from a system process, and
502: * implement the semantics of socket operations by
503: * switching out to the protocol specific routines.
504: */
505: /*ARGSUSED*/
1.3 andrew 506: int
1.160 ad 507: socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
508: struct socket *lockso)
1.1 cgd 509: {
1.99 matt 510: const struct protosw *prp;
1.54 lukem 511: struct socket *so;
1.115 yamt 512: uid_t uid;
1.160 ad 513: int error;
514: kmutex_t *lock;
1.1 cgd 515:
1.132 elad 516: error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
517: KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
518: KAUTH_ARG(proto));
1.140 dyoung 519: if (error != 0)
520: return error;
1.127 elad 521:
1.1 cgd 522: if (proto)
523: prp = pffindproto(dom, proto, type);
524: else
525: prp = pffindtype(dom, type);
1.140 dyoung 526: if (prp == NULL) {
1.120 ginsbach 527: /* no support for domain */
528: if (pffinddomain(dom) == 0)
1.140 dyoung 529: return EAFNOSUPPORT;
1.120 ginsbach 530: /* no support for socket type */
531: if (proto == 0 && type != 0)
1.140 dyoung 532: return EPROTOTYPE;
533: return EPROTONOSUPPORT;
1.120 ginsbach 534: }
1.140 dyoung 535: if (prp->pr_usrreq == NULL)
536: return EPROTONOSUPPORT;
1.1 cgd 537: if (prp->pr_type != type)
1.140 dyoung 538: return EPROTOTYPE;
1.160 ad 539:
540: so = soget(true);
1.1 cgd 541: so->so_type = type;
542: so->so_proto = prp;
1.33 matt 543: so->so_send = sosend;
544: so->so_receive = soreceive;
1.78 matt 545: #ifdef MBUFTRACE
546: so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
547: so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
548: so->so_mowner = &prp->pr_domain->dom_mowner;
549: #endif
1.138 rmind 550: uid = kauth_cred_geteuid(l->l_cred);
1.115 yamt 551: so->so_uidinfo = uid_find(uid);
1.168 yamt 552: so->so_cpid = l->l_proc->p_pid;
1.160 ad 553: if (lockso != NULL) {
554: /* Caller wants us to share a lock. */
555: lock = lockso->so_lock;
556: so->so_lock = lock;
557: mutex_obj_hold(lock);
558: mutex_enter(lock);
559: } else {
560: /* Lock assigned and taken during PRU_ATTACH. */
561: }
1.140 dyoung 562: error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
563: (struct mbuf *)(long)proto, NULL, l);
1.160 ad 564: KASSERT(solocked(so));
1.140 dyoung 565: if (error != 0) {
1.1 cgd 566: so->so_state |= SS_NOFDREF;
567: sofree(so);
1.140 dyoung 568: return error;
1.1 cgd 569: }
1.198 elad 570: so->so_cred = kauth_cred_dup(l->l_cred);
1.160 ad 571: sounlock(so);
1.1 cgd 572: *aso = so;
1.140 dyoung 573: return 0;
1.1 cgd 574: }
575:
1.142 dyoung 576: /* On success, write file descriptor to fdout and return zero. On
577: * failure, return non-zero; *fdout will be undefined.
578: */
579: int
580: fsocreate(int domain, struct socket **sop, int type, int protocol,
581: struct lwp *l, int *fdout)
582: {
583: struct socket *so;
584: struct file *fp;
585: int fd, error;
1.204 christos 586: int flags = type & SOCK_FLAGS_MASK;
1.142 dyoung 587:
1.204 christos 588: type &= ~SOCK_FLAGS_MASK;
1.155 ad 589: if ((error = fd_allocfile(&fp, &fd)) != 0)
1.204 christos 590: return error;
591: fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
592: fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0);
1.142 dyoung 593: fp->f_type = DTYPE_SOCKET;
594: fp->f_ops = &socketops;
1.160 ad 595: error = socreate(domain, &so, type, protocol, l, NULL);
1.142 dyoung 596: if (error != 0) {
1.155 ad 597: fd_abort(curproc, fp, fd);
1.142 dyoung 598: } else {
599: if (sop != NULL)
600: *sop = so;
601: fp->f_data = so;
1.155 ad 602: fd_affix(curproc, fp, fd);
1.142 dyoung 603: *fdout = fd;
604: }
605: return error;
606: }
607:
1.3 andrew 608: int
1.190 dyoung 609: sofamily(const struct socket *so)
610: {
611: const struct protosw *pr;
612: const struct domain *dom;
613:
614: if ((pr = so->so_proto) == NULL)
615: return AF_UNSPEC;
616: if ((dom = pr->pr_domain) == NULL)
617: return AF_UNSPEC;
618: return dom->dom_family;
619: }
620:
621: int
1.114 christos 622: sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 623: {
1.160 ad 624: int error;
1.1 cgd 625:
1.160 ad 626: solock(so);
1.140 dyoung 627: error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
1.160 ad 628: sounlock(so);
1.140 dyoung 629: return error;
1.1 cgd 630: }
631:
1.3 andrew 632: int
1.150 elad 633: solisten(struct socket *so, int backlog, struct lwp *l)
1.1 cgd 634: {
1.160 ad 635: int error;
1.1 cgd 636:
1.160 ad 637: solock(so);
1.158 ad 638: if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1.163 ad 639: SS_ISDISCONNECTING)) != 0) {
640: sounlock(so);
1.158 ad 641: return (EOPNOTSUPP);
1.163 ad 642: }
1.140 dyoung 643: error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
1.150 elad 644: NULL, NULL, l);
1.140 dyoung 645: if (error != 0) {
1.160 ad 646: sounlock(so);
1.140 dyoung 647: return error;
1.1 cgd 648: }
1.63 matt 649: if (TAILQ_EMPTY(&so->so_q))
1.1 cgd 650: so->so_options |= SO_ACCEPTCONN;
651: if (backlog < 0)
652: backlog = 0;
1.49 jonathan 653: so->so_qlimit = min(backlog, somaxconn);
1.160 ad 654: sounlock(so);
1.140 dyoung 655: return 0;
1.1 cgd 656: }
657:
1.21 christos 658: void
1.54 lukem 659: sofree(struct socket *so)
1.1 cgd 660: {
1.161 ad 661: u_int refs;
1.1 cgd 662:
1.160 ad 663: KASSERT(solocked(so));
664:
665: if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
666: sounlock(so);
1.1 cgd 667: return;
1.160 ad 668: }
1.43 mycroft 669: if (so->so_head) {
670: /*
671: * We must not decommission a socket that's on the accept(2)
672: * queue. If we do, then accept(2) may hang after select(2)
673: * indicated that the listening socket was ready.
674: */
1.160 ad 675: if (!soqremque(so, 0)) {
676: sounlock(so);
1.43 mycroft 677: return;
1.160 ad 678: }
1.43 mycroft 679: }
1.98 christos 680: if (so->so_rcv.sb_hiwat)
1.110 christos 681: (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
1.98 christos 682: RLIM_INFINITY);
683: if (so->so_snd.sb_hiwat)
1.110 christos 684: (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
1.98 christos 685: RLIM_INFINITY);
686: sbrelease(&so->so_snd, so);
1.160 ad 687: KASSERT(!cv_has_waiters(&so->so_cv));
688: KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
689: KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
1.1 cgd 690: sorflush(so);
1.161 ad 691: refs = so->so_aborting; /* XXX */
1.177 ad 692: /* Remove acccept filter if one is present. */
1.170 tls 693: if (so->so_accf != NULL)
1.177 ad 694: (void)accept_filt_clear(so);
1.160 ad 695: sounlock(so);
1.161 ad 696: if (refs == 0) /* XXX */
697: soput(so);
1.1 cgd 698: }
699:
700: /*
701: * Close a socket on last file table reference removal.
702: * Initiate disconnect if connected.
703: * Free socket when disconnect complete.
704: */
1.3 andrew 705: int
1.54 lukem 706: soclose(struct socket *so)
1.1 cgd 707: {
1.54 lukem 708: struct socket *so2;
1.160 ad 709: int error;
710: int error2;
1.1 cgd 711:
1.54 lukem 712: error = 0;
1.160 ad 713: solock(so);
1.1 cgd 714: if (so->so_options & SO_ACCEPTCONN) {
1.172 ad 715: for (;;) {
716: if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
1.160 ad 717: KASSERT(solocked2(so, so2));
718: (void) soqremque(so2, 0);
719: /* soabort drops the lock. */
720: (void) soabort(so2);
721: solock(so);
1.172 ad 722: continue;
1.160 ad 723: }
1.172 ad 724: if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
1.160 ad 725: KASSERT(solocked2(so, so2));
726: (void) soqremque(so2, 1);
727: /* soabort drops the lock. */
728: (void) soabort(so2);
729: solock(so);
1.172 ad 730: continue;
1.160 ad 731: }
1.172 ad 732: break;
733: }
1.1 cgd 734: }
735: if (so->so_pcb == 0)
736: goto discard;
737: if (so->so_state & SS_ISCONNECTED) {
738: if ((so->so_state & SS_ISDISCONNECTING) == 0) {
739: error = sodisconnect(so);
740: if (error)
741: goto drop;
742: }
743: if (so->so_options & SO_LINGER) {
1.151 ad 744: if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
1.1 cgd 745: goto drop;
1.21 christos 746: while (so->so_state & SS_ISCONNECTED) {
1.185 yamt 747: error = sowait(so, true, so->so_linger * hz);
1.21 christos 748: if (error)
1.1 cgd 749: break;
1.21 christos 750: }
1.1 cgd 751: }
752: }
1.54 lukem 753: drop:
1.1 cgd 754: if (so->so_pcb) {
1.160 ad 755: error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
1.140 dyoung 756: NULL, NULL, NULL, NULL);
1.1 cgd 757: if (error == 0)
758: error = error2;
759: }
1.54 lukem 760: discard:
1.1 cgd 761: if (so->so_state & SS_NOFDREF)
762: panic("soclose: NOFDREF");
1.198 elad 763: kauth_cred_free(so->so_cred);
1.1 cgd 764: so->so_state |= SS_NOFDREF;
765: sofree(so);
766: return (error);
767: }
768:
769: /*
1.160 ad 770: * Must be called with the socket locked.. Will return with it unlocked.
1.1 cgd 771: */
1.3 andrew 772: int
1.54 lukem 773: soabort(struct socket *so)
1.1 cgd 774: {
1.161 ad 775: u_int refs;
1.139 yamt 776: int error;
1.160 ad 777:
778: KASSERT(solocked(so));
779: KASSERT(so->so_head == NULL);
1.1 cgd 780:
1.161 ad 781: so->so_aborting++; /* XXX */
1.140 dyoung 782: error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
783: NULL, NULL, NULL);
1.161 ad 784: refs = --so->so_aborting; /* XXX */
1.164 drochner 785: if (error || (refs == 0)) {
1.139 yamt 786: sofree(so);
1.160 ad 787: } else {
788: sounlock(so);
1.139 yamt 789: }
790: return error;
1.1 cgd 791: }
792:
1.3 andrew 793: int
1.54 lukem 794: soaccept(struct socket *so, struct mbuf *nam)
1.1 cgd 795: {
1.160 ad 796: int error;
797:
798: KASSERT(solocked(so));
1.1 cgd 799:
1.54 lukem 800: error = 0;
1.1 cgd 801: if ((so->so_state & SS_NOFDREF) == 0)
802: panic("soaccept: !NOFDREF");
803: so->so_state &= ~SS_NOFDREF;
1.55 thorpej 804: if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
805: (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
1.41 mycroft 806: error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
1.140 dyoung 807: NULL, nam, NULL, NULL);
1.41 mycroft 808: else
1.53 itojun 809: error = ECONNABORTED;
1.52 itojun 810:
1.1 cgd 811: return (error);
812: }
813:
1.3 andrew 814: int
1.114 christos 815: soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 816: {
1.160 ad 817: int error;
818:
819: KASSERT(solocked(so));
1.1 cgd 820:
821: if (so->so_options & SO_ACCEPTCONN)
822: return (EOPNOTSUPP);
823: /*
824: * If protocol is connection-based, can only connect once.
825: * Otherwise, if connected, try to disconnect first.
826: * This allows user to disconnect by connecting to, e.g.,
827: * a null address.
828: */
829: if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
830: ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
831: (error = sodisconnect(so))))
832: error = EISCONN;
833: else
834: error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
1.140 dyoung 835: NULL, nam, NULL, l);
1.1 cgd 836: return (error);
837: }
838:
1.3 andrew 839: int
1.54 lukem 840: soconnect2(struct socket *so1, struct socket *so2)
1.1 cgd 841: {
1.160 ad 842: int error;
843:
844: KASSERT(solocked2(so1, so2));
1.1 cgd 845:
1.22 mycroft 846: error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
1.140 dyoung 847: NULL, (struct mbuf *)so2, NULL, NULL);
1.1 cgd 848: return (error);
849: }
850:
1.3 andrew 851: int
1.54 lukem 852: sodisconnect(struct socket *so)
1.1 cgd 853: {
1.160 ad 854: int error;
855:
856: KASSERT(solocked(so));
1.1 cgd 857:
858: if ((so->so_state & SS_ISCONNECTED) == 0) {
859: error = ENOTCONN;
1.160 ad 860: } else if (so->so_state & SS_ISDISCONNECTING) {
1.1 cgd 861: error = EALREADY;
1.160 ad 862: } else {
863: error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
864: NULL, NULL, NULL, NULL);
1.1 cgd 865: }
866: return (error);
867: }
868:
1.15 mycroft 869: #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1.1 cgd 870: /*
871: * Send on a socket.
872: * If send must go all at once and message is larger than
873: * send buffering, then hard error.
874: * Lock against other senders.
875: * If must go all at once and not enough room now, then
876: * inform user that this would block and do nothing.
877: * Otherwise, if nonblocking, send as much as possible.
878: * The data to be sent is described by "uio" if nonzero,
879: * otherwise by the mbuf chain "top" (which must be null
880: * if uio is not). Data provided in mbuf chain must be small
881: * enough to send all at once.
882: *
883: * Returns nonzero on error, timeout or signal; callers
884: * must check for short counts if EINTR/ERESTART are returned.
885: * Data and control buffers are freed on return.
886: */
1.3 andrew 887: int
1.54 lukem 888: sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
1.114 christos 889: struct mbuf *control, int flags, struct lwp *l)
1.1 cgd 890: {
1.54 lukem 891: struct mbuf **mp, *m;
1.114 christos 892: struct proc *p;
1.58 jdolecek 893: long space, len, resid, clen, mlen;
894: int error, s, dontroute, atomic;
1.196 dsl 895: short wakeup_state = 0;
1.54 lukem 896:
1.114 christos 897: p = l->l_proc;
1.160 ad 898: clen = 0;
1.64 thorpej 899:
1.160 ad 900: /*
901: * solock() provides atomicity of access. splsoftnet() prevents
902: * protocol processing soft interrupts from interrupting us and
903: * blocking (expensive).
904: */
905: s = splsoftnet();
906: solock(so);
1.54 lukem 907: atomic = sosendallatonce(so) || top;
1.1 cgd 908: if (uio)
909: resid = uio->uio_resid;
910: else
911: resid = top->m_pkthdr.len;
1.7 cgd 912: /*
913: * In theory resid should be unsigned.
914: * However, space must be signed, as it might be less than 0
915: * if we over-committed, and we must use a signed comparison
916: * of space and resid. On the other hand, a negative resid
917: * causes us to loop sending 0-length segments to the protocol.
918: */
1.29 mycroft 919: if (resid < 0) {
920: error = EINVAL;
921: goto out;
922: }
1.1 cgd 923: dontroute =
924: (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
925: (so->so_proto->pr_flags & PR_ATOMIC);
1.165 christos 926: l->l_ru.ru_msgsnd++;
1.1 cgd 927: if (control)
928: clen = control->m_len;
1.54 lukem 929: restart:
1.21 christos 930: if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
1.1 cgd 931: goto out;
932: do {
1.160 ad 933: if (so->so_state & SS_CANTSENDMORE) {
934: error = EPIPE;
935: goto release;
936: }
1.48 thorpej 937: if (so->so_error) {
938: error = so->so_error;
939: so->so_error = 0;
940: goto release;
941: }
1.1 cgd 942: if ((so->so_state & SS_ISCONNECTED) == 0) {
943: if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
944: if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1.160 ad 945: !(resid == 0 && clen != 0)) {
946: error = ENOTCONN;
947: goto release;
948: }
949: } else if (addr == 0) {
950: error = EDESTADDRREQ;
951: goto release;
952: }
1.1 cgd 953: }
954: space = sbspace(&so->so_snd);
955: if (flags & MSG_OOB)
956: space += 1024;
1.21 christos 957: if ((atomic && resid > so->so_snd.sb_hiwat) ||
1.160 ad 958: clen > so->so_snd.sb_hiwat) {
959: error = EMSGSIZE;
960: goto release;
961: }
1.96 mycroft 962: if (space < resid + clen &&
1.1 cgd 963: (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1.160 ad 964: if (so->so_nbio) {
965: error = EWOULDBLOCK;
966: goto release;
967: }
1.1 cgd 968: sbunlock(&so->so_snd);
1.196 dsl 969: if (wakeup_state & SS_RESTARTSYS) {
970: error = ERESTART;
971: goto out;
972: }
1.1 cgd 973: error = sbwait(&so->so_snd);
974: if (error)
975: goto out;
1.196 dsl 976: wakeup_state = so->so_state;
1.1 cgd 977: goto restart;
978: }
1.196 dsl 979: wakeup_state = 0;
1.1 cgd 980: mp = ⊤
981: space -= clen;
982: do {
1.45 tv 983: if (uio == NULL) {
984: /*
985: * Data is prepackaged in "top".
986: */
987: resid = 0;
988: if (flags & MSG_EOR)
989: top->m_flags |= M_EOR;
990: } else do {
1.160 ad 991: sounlock(so);
992: splx(s);
1.144 dyoung 993: if (top == NULL) {
1.78 matt 994: m = m_gethdr(M_WAIT, MT_DATA);
1.45 tv 995: mlen = MHLEN;
996: m->m_pkthdr.len = 0;
1.140 dyoung 997: m->m_pkthdr.rcvif = NULL;
1.45 tv 998: } else {
1.78 matt 999: m = m_get(M_WAIT, MT_DATA);
1.45 tv 1000: mlen = MLEN;
1001: }
1.78 matt 1002: MCLAIM(m, so->so_snd.sb_mowner);
1.121 yamt 1003: if (sock_loan_thresh >= 0 &&
1004: uio->uio_iov->iov_len >= sock_loan_thresh &&
1005: space >= sock_loan_thresh &&
1.64 thorpej 1006: (len = sosend_loan(so, uio, m,
1007: space)) != 0) {
1008: SOSEND_COUNTER_INCR(&sosend_loan_big);
1009: space -= len;
1010: goto have_data;
1011: }
1.45 tv 1012: if (resid >= MINCLSIZE && space >= MCLBYTES) {
1.64 thorpej 1013: SOSEND_COUNTER_INCR(&sosend_copy_big);
1.201 oki 1014: m_clget(m, M_DONTWAIT);
1.45 tv 1015: if ((m->m_flags & M_EXT) == 0)
1016: goto nopages;
1017: mlen = MCLBYTES;
1018: if (atomic && top == 0) {
1.58 jdolecek 1019: len = lmin(MCLBYTES - max_hdr,
1.54 lukem 1020: resid);
1.45 tv 1021: m->m_data += max_hdr;
1022: } else
1.58 jdolecek 1023: len = lmin(MCLBYTES, resid);
1.45 tv 1024: space -= len;
1025: } else {
1.64 thorpej 1026: nopages:
1027: SOSEND_COUNTER_INCR(&sosend_copy_small);
1.58 jdolecek 1028: len = lmin(lmin(mlen, resid), space);
1.45 tv 1029: space -= len;
1030: /*
1031: * For datagram protocols, leave room
1032: * for protocol headers in first mbuf.
1033: */
1034: if (atomic && top == 0 && len < mlen)
1035: MH_ALIGN(m, len);
1036: }
1.144 dyoung 1037: error = uiomove(mtod(m, void *), (int)len, uio);
1.64 thorpej 1038: have_data:
1.45 tv 1039: resid = uio->uio_resid;
1040: m->m_len = len;
1041: *mp = m;
1042: top->m_pkthdr.len += len;
1.160 ad 1043: s = splsoftnet();
1044: solock(so);
1.144 dyoung 1045: if (error != 0)
1.45 tv 1046: goto release;
1047: mp = &m->m_next;
1048: if (resid <= 0) {
1049: if (flags & MSG_EOR)
1050: top->m_flags |= M_EOR;
1051: break;
1052: }
1053: } while (space > 0 && atomic);
1.108 perry 1054:
1.160 ad 1055: if (so->so_state & SS_CANTSENDMORE) {
1056: error = EPIPE;
1057: goto release;
1058: }
1.45 tv 1059: if (dontroute)
1060: so->so_options |= SO_DONTROUTE;
1061: if (resid > 0)
1062: so->so_state |= SS_MORETOCOME;
1.46 sommerfe 1063: error = (*so->so_proto->pr_usrreq)(so,
1064: (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1.160 ad 1065: top, addr, control, curlwp);
1.45 tv 1066: if (dontroute)
1067: so->so_options &= ~SO_DONTROUTE;
1068: if (resid > 0)
1069: so->so_state &= ~SS_MORETOCOME;
1070: clen = 0;
1.144 dyoung 1071: control = NULL;
1072: top = NULL;
1.45 tv 1073: mp = ⊤
1.144 dyoung 1074: if (error != 0)
1.1 cgd 1075: goto release;
1076: } while (resid && space > 0);
1077: } while (resid);
1078:
1.54 lukem 1079: release:
1.1 cgd 1080: sbunlock(&so->so_snd);
1.54 lukem 1081: out:
1.160 ad 1082: sounlock(so);
1083: splx(s);
1.1 cgd 1084: if (top)
1085: m_freem(top);
1086: if (control)
1087: m_freem(control);
1088: return (error);
1089: }
1090:
1091: /*
1.159 ad 1092: * Following replacement or removal of the first mbuf on the first
1093: * mbuf chain of a socket buffer, push necessary state changes back
1094: * into the socket buffer so that other consumers see the values
1095: * consistently. 'nextrecord' is the callers locally stored value of
1096: * the original value of sb->sb_mb->m_nextpkt which must be restored
1097: * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL.
1098: */
1099: static void
1100: sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1101: {
1102:
1.160 ad 1103: KASSERT(solocked(sb->sb_so));
1104:
1.159 ad 1105: /*
1106: * First, update for the new value of nextrecord. If necessary,
1107: * make it the first record.
1108: */
1109: if (sb->sb_mb != NULL)
1110: sb->sb_mb->m_nextpkt = nextrecord;
1111: else
1112: sb->sb_mb = nextrecord;
1113:
1114: /*
1115: * Now update any dependent socket buffer fields to reflect
1116: * the new state. This is an inline of SB_EMPTY_FIXUP, with
1117: * the addition of a second clause that takes care of the
1118: * case where sb_mb has been updated, but remains the last
1119: * record.
1120: */
1121: if (sb->sb_mb == NULL) {
1122: sb->sb_mbtail = NULL;
1123: sb->sb_lastrecord = NULL;
1124: } else if (sb->sb_mb->m_nextpkt == NULL)
1125: sb->sb_lastrecord = sb->sb_mb;
1126: }
1127:
1128: /*
1.1 cgd 1129: * Implement receive operations on a socket.
1130: * We depend on the way that records are added to the sockbuf
1131: * by sbappend*. In particular, each record (mbufs linked through m_next)
1132: * must begin with an address if the protocol so specifies,
1133: * followed by an optional mbuf or mbufs containing ancillary data,
1134: * and then zero or more mbufs of data.
1135: * In order to avoid blocking network interrupts for the entire time here,
1136: * we splx() while doing the actual copy to user space.
1137: * Although the sockbuf is locked, new data may still be appended,
1138: * and thus we must maintain consistency of the sockbuf during that time.
1139: *
1140: * The caller may receive the data as a single mbuf chain by supplying
1141: * an mbuf **mp0 for use in returning the chain. The uio is then used
1142: * only for the count in uio_resid.
1143: */
1.3 andrew 1144: int
1.54 lukem 1145: soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1146: struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1.1 cgd 1147: {
1.116 yamt 1148: struct lwp *l = curlwp;
1.160 ad 1149: struct mbuf *m, **mp, *mt;
1.146 dyoung 1150: int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1.99 matt 1151: const struct protosw *pr;
1.54 lukem 1152: struct mbuf *nextrecord;
1.67 he 1153: int mbuf_removed = 0;
1.146 dyoung 1154: const struct domain *dom;
1.196 dsl 1155: short wakeup_state = 0;
1.64 thorpej 1156:
1.54 lukem 1157: pr = so->so_proto;
1.146 dyoung 1158: atomic = pr->pr_flags & PR_ATOMIC;
1159: dom = pr->pr_domain;
1.1 cgd 1160: mp = mp0;
1.54 lukem 1161: type = 0;
1162: orig_resid = uio->uio_resid;
1.102 jonathan 1163:
1.144 dyoung 1164: if (paddr != NULL)
1165: *paddr = NULL;
1166: if (controlp != NULL)
1167: *controlp = NULL;
1168: if (flagsp != NULL)
1.1 cgd 1169: flags = *flagsp &~ MSG_EOR;
1170: else
1171: flags = 0;
1.66 enami 1172:
1.1 cgd 1173: if (flags & MSG_OOB) {
1174: m = m_get(M_WAIT, MT_DATA);
1.160 ad 1175: solock(so);
1.17 cgd 1176: error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1.140 dyoung 1177: (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1.160 ad 1178: sounlock(so);
1.1 cgd 1179: if (error)
1180: goto bad;
1181: do {
1.134 christos 1182: error = uiomove(mtod(m, void *),
1.1 cgd 1183: (int) min(uio->uio_resid, m->m_len), uio);
1184: m = m_free(m);
1.144 dyoung 1185: } while (uio->uio_resid > 0 && error == 0 && m);
1.54 lukem 1186: bad:
1.144 dyoung 1187: if (m != NULL)
1.1 cgd 1188: m_freem(m);
1.144 dyoung 1189: return error;
1.1 cgd 1190: }
1.144 dyoung 1191: if (mp != NULL)
1.140 dyoung 1192: *mp = NULL;
1.160 ad 1193:
1194: /*
1195: * solock() provides atomicity of access. splsoftnet() prevents
1196: * protocol processing soft interrupts from interrupting us and
1197: * blocking (expensive).
1198: */
1199: s = splsoftnet();
1200: solock(so);
1.1 cgd 1201: if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1.140 dyoung 1202: (*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1.1 cgd 1203:
1.54 lukem 1204: restart:
1.160 ad 1205: if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1206: sounlock(so);
1207: splx(s);
1.144 dyoung 1208: return error;
1.160 ad 1209: }
1.1 cgd 1210:
1211: m = so->so_rcv.sb_mb;
1212: /*
1213: * If we have less data than requested, block awaiting more
1214: * (subject to any timeout) if:
1.15 mycroft 1215: * 1. the current count is less than the low water mark,
1.1 cgd 1216: * 2. MSG_WAITALL is set, and it is possible to do the entire
1.15 mycroft 1217: * receive operation at once if we block (resid <= hiwat), or
1218: * 3. MSG_DONTWAIT is not set.
1.1 cgd 1219: * If MSG_WAITALL is set but resid is larger than the receive buffer,
1220: * we have to do the receive in sections, and thus risk returning
1221: * a short count if a timeout or signal occurs after we start.
1222: */
1.144 dyoung 1223: if (m == NULL ||
1224: ((flags & MSG_DONTWAIT) == 0 &&
1225: so->so_rcv.sb_cc < uio->uio_resid &&
1226: (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1227: ((flags & MSG_WAITALL) &&
1228: uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1.146 dyoung 1229: m->m_nextpkt == NULL && !atomic)) {
1.1 cgd 1230: #ifdef DIAGNOSTIC
1.144 dyoung 1231: if (m == NULL && so->so_rcv.sb_cc)
1.1 cgd 1232: panic("receive 1");
1233: #endif
1234: if (so->so_error) {
1.144 dyoung 1235: if (m != NULL)
1.15 mycroft 1236: goto dontblock;
1.1 cgd 1237: error = so->so_error;
1238: if ((flags & MSG_PEEK) == 0)
1239: so->so_error = 0;
1240: goto release;
1241: }
1242: if (so->so_state & SS_CANTRCVMORE) {
1.144 dyoung 1243: if (m != NULL)
1.15 mycroft 1244: goto dontblock;
1.1 cgd 1245: else
1246: goto release;
1247: }
1.144 dyoung 1248: for (; m != NULL; m = m->m_next)
1.1 cgd 1249: if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1250: m = so->so_rcv.sb_mb;
1251: goto dontblock;
1252: }
1253: if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1254: (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1255: error = ENOTCONN;
1256: goto release;
1257: }
1258: if (uio->uio_resid == 0)
1259: goto release;
1.151 ad 1260: if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1.1 cgd 1261: error = EWOULDBLOCK;
1262: goto release;
1263: }
1.69 thorpej 1264: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1265: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1.1 cgd 1266: sbunlock(&so->so_rcv);
1.196 dsl 1267: if (wakeup_state & SS_RESTARTSYS)
1268: error = ERESTART;
1269: else
1270: error = sbwait(&so->so_rcv);
1.160 ad 1271: if (error != 0) {
1272: sounlock(so);
1273: splx(s);
1.144 dyoung 1274: return error;
1.160 ad 1275: }
1.196 dsl 1276: wakeup_state = so->so_state;
1.1 cgd 1277: goto restart;
1278: }
1.54 lukem 1279: dontblock:
1.69 thorpej 1280: /*
1281: * On entry here, m points to the first record of the socket buffer.
1.159 ad 1282: * From this point onward, we maintain 'nextrecord' as a cache of the
1283: * pointer to the next record in the socket buffer. We must keep the
1284: * various socket buffer pointers and local stack versions of the
1285: * pointers in sync, pushing out modifications before dropping the
1.160 ad 1286: * socket lock, and re-reading them when picking it up.
1.159 ad 1287: *
1288: * Otherwise, we will race with the network stack appending new data
1289: * or records onto the socket buffer by using inconsistent/stale
1290: * versions of the field, possibly resulting in socket buffer
1291: * corruption.
1292: *
1293: * By holding the high-level sblock(), we prevent simultaneous
1294: * readers from pulling off the front of the socket buffer.
1.69 thorpej 1295: */
1.144 dyoung 1296: if (l != NULL)
1.157 ad 1297: l->l_ru.ru_msgrcv++;
1.69 thorpej 1298: KASSERT(m == so->so_rcv.sb_mb);
1299: SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1300: SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1.1 cgd 1301: nextrecord = m->m_nextpkt;
1302: if (pr->pr_flags & PR_ADDR) {
1303: #ifdef DIAGNOSTIC
1304: if (m->m_type != MT_SONAME)
1305: panic("receive 1a");
1306: #endif
1.3 andrew 1307: orig_resid = 0;
1.1 cgd 1308: if (flags & MSG_PEEK) {
1309: if (paddr)
1310: *paddr = m_copy(m, 0, m->m_len);
1311: m = m->m_next;
1312: } else {
1313: sbfree(&so->so_rcv, m);
1.67 he 1314: mbuf_removed = 1;
1.144 dyoung 1315: if (paddr != NULL) {
1.1 cgd 1316: *paddr = m;
1317: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1318: m->m_next = NULL;
1.1 cgd 1319: m = so->so_rcv.sb_mb;
1320: } else {
1321: MFREE(m, so->so_rcv.sb_mb);
1322: m = so->so_rcv.sb_mb;
1323: }
1.159 ad 1324: sbsync(&so->so_rcv, nextrecord);
1.1 cgd 1325: }
1326: }
1.159 ad 1327:
1328: /*
1329: * Process one or more MT_CONTROL mbufs present before any data mbufs
1330: * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1331: * just copy the data; if !MSG_PEEK, we call into the protocol to
1332: * perform externalization (or freeing if controlp == NULL).
1333: */
1334: if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1335: struct mbuf *cm = NULL, *cmn;
1336: struct mbuf **cme = &cm;
1337:
1338: do {
1339: if (flags & MSG_PEEK) {
1340: if (controlp != NULL) {
1341: *controlp = m_copy(m, 0, m->m_len);
1342: controlp = &(*controlp)->m_next;
1343: }
1344: m = m->m_next;
1345: } else {
1346: sbfree(&so->so_rcv, m);
1.1 cgd 1347: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1348: m->m_next = NULL;
1.159 ad 1349: *cme = m;
1350: cme = &(*cme)->m_next;
1.1 cgd 1351: m = so->so_rcv.sb_mb;
1.159 ad 1352: }
1353: } while (m != NULL && m->m_type == MT_CONTROL);
1354: if ((flags & MSG_PEEK) == 0)
1355: sbsync(&so->so_rcv, nextrecord);
1356: for (; cm != NULL; cm = cmn) {
1357: cmn = cm->m_next;
1358: cm->m_next = NULL;
1359: type = mtod(cm, struct cmsghdr *)->cmsg_type;
1360: if (controlp != NULL) {
1361: if (dom->dom_externalize != NULL &&
1362: type == SCM_RIGHTS) {
1.160 ad 1363: sounlock(so);
1.159 ad 1364: splx(s);
1.204 christos 1365: error = (*dom->dom_externalize)(cm, l,
1366: (flags & MSG_CMSG_CLOEXEC) ?
1367: O_CLOEXEC : 0);
1.159 ad 1368: s = splsoftnet();
1.160 ad 1369: solock(so);
1.159 ad 1370: }
1371: *controlp = cm;
1372: while (*controlp != NULL)
1373: controlp = &(*controlp)->m_next;
1.1 cgd 1374: } else {
1.106 itojun 1375: /*
1376: * Dispose of any SCM_RIGHTS message that went
1377: * through the read path rather than recv.
1378: */
1.159 ad 1379: if (dom->dom_dispose != NULL &&
1380: type == SCM_RIGHTS) {
1.160 ad 1381: sounlock(so);
1.159 ad 1382: (*dom->dom_dispose)(cm);
1.160 ad 1383: solock(so);
1.159 ad 1384: }
1385: m_freem(cm);
1.1 cgd 1386: }
1387: }
1.159 ad 1388: if (m != NULL)
1389: nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1390: else
1391: nextrecord = so->so_rcv.sb_mb;
1392: orig_resid = 0;
1.1 cgd 1393: }
1.69 thorpej 1394:
1.159 ad 1395: /* If m is non-NULL, we have some data to read. */
1396: if (__predict_true(m != NULL)) {
1.1 cgd 1397: type = m->m_type;
1398: if (type == MT_OOBDATA)
1399: flags |= MSG_OOB;
1400: }
1.69 thorpej 1401: SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1402: SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1403:
1.1 cgd 1404: moff = 0;
1405: offset = 0;
1.144 dyoung 1406: while (m != NULL && uio->uio_resid > 0 && error == 0) {
1.1 cgd 1407: if (m->m_type == MT_OOBDATA) {
1408: if (type != MT_OOBDATA)
1409: break;
1410: } else if (type == MT_OOBDATA)
1411: break;
1412: #ifdef DIAGNOSTIC
1413: else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1414: panic("receive 3");
1415: #endif
1416: so->so_state &= ~SS_RCVATMARK;
1.196 dsl 1417: wakeup_state = 0;
1.1 cgd 1418: len = uio->uio_resid;
1419: if (so->so_oobmark && len > so->so_oobmark - offset)
1420: len = so->so_oobmark - offset;
1421: if (len > m->m_len - moff)
1422: len = m->m_len - moff;
1423: /*
1424: * If mp is set, just pass back the mbufs.
1425: * Otherwise copy them out via the uio, then free.
1426: * Sockbuf must be consistent here (points to current mbuf,
1427: * it points to next record) when we drop priority;
1428: * we must note any additions to the sockbuf when we
1429: * block interrupts again.
1430: */
1.144 dyoung 1431: if (mp == NULL) {
1.69 thorpej 1432: SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1433: SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1.160 ad 1434: sounlock(so);
1.1 cgd 1435: splx(s);
1.134 christos 1436: error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1.20 mycroft 1437: s = splsoftnet();
1.160 ad 1438: solock(so);
1.144 dyoung 1439: if (error != 0) {
1.67 he 1440: /*
1441: * If any part of the record has been removed
1442: * (such as the MT_SONAME mbuf, which will
1443: * happen when PR_ADDR, and thus also
1444: * PR_ATOMIC, is set), then drop the entire
1445: * record to maintain the atomicity of the
1446: * receive operation.
1447: *
1448: * This avoids a later panic("receive 1a")
1449: * when compiled with DIAGNOSTIC.
1450: */
1.146 dyoung 1451: if (m && mbuf_removed && atomic)
1.67 he 1452: (void) sbdroprecord(&so->so_rcv);
1453:
1.57 jdolecek 1454: goto release;
1.67 he 1455: }
1.1 cgd 1456: } else
1457: uio->uio_resid -= len;
1458: if (len == m->m_len - moff) {
1459: if (m->m_flags & M_EOR)
1460: flags |= MSG_EOR;
1461: if (flags & MSG_PEEK) {
1462: m = m->m_next;
1463: moff = 0;
1464: } else {
1465: nextrecord = m->m_nextpkt;
1466: sbfree(&so->so_rcv, m);
1467: if (mp) {
1468: *mp = m;
1469: mp = &m->m_next;
1470: so->so_rcv.sb_mb = m = m->m_next;
1.140 dyoung 1471: *mp = NULL;
1.1 cgd 1472: } else {
1473: MFREE(m, so->so_rcv.sb_mb);
1474: m = so->so_rcv.sb_mb;
1475: }
1.69 thorpej 1476: /*
1477: * If m != NULL, we also know that
1478: * so->so_rcv.sb_mb != NULL.
1479: */
1480: KASSERT(so->so_rcv.sb_mb == m);
1481: if (m) {
1.1 cgd 1482: m->m_nextpkt = nextrecord;
1.69 thorpej 1483: if (nextrecord == NULL)
1484: so->so_rcv.sb_lastrecord = m;
1485: } else {
1486: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1487: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1488: }
1489: SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1490: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1.1 cgd 1491: }
1.144 dyoung 1492: } else if (flags & MSG_PEEK)
1493: moff += len;
1494: else {
1.160 ad 1495: if (mp != NULL) {
1496: mt = m_copym(m, 0, len, M_NOWAIT);
1497: if (__predict_false(mt == NULL)) {
1498: sounlock(so);
1499: mt = m_copym(m, 0, len, M_WAIT);
1500: solock(so);
1501: }
1502: *mp = mt;
1503: }
1.144 dyoung 1504: m->m_data += len;
1505: m->m_len -= len;
1506: so->so_rcv.sb_cc -= len;
1.1 cgd 1507: }
1508: if (so->so_oobmark) {
1509: if ((flags & MSG_PEEK) == 0) {
1510: so->so_oobmark -= len;
1511: if (so->so_oobmark == 0) {
1512: so->so_state |= SS_RCVATMARK;
1513: break;
1514: }
1.7 cgd 1515: } else {
1.1 cgd 1516: offset += len;
1.7 cgd 1517: if (offset == so->so_oobmark)
1518: break;
1519: }
1.1 cgd 1520: }
1521: if (flags & MSG_EOR)
1522: break;
1523: /*
1524: * If the MSG_WAITALL flag is set (for non-atomic socket),
1525: * we must not quit until "uio->uio_resid == 0" or an error
1526: * termination. If a signal/timeout occurs, return
1527: * with a short count but without error.
1528: * Keep sockbuf locked against other readers.
1529: */
1.144 dyoung 1530: while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1.3 andrew 1531: !sosendallatonce(so) && !nextrecord) {
1.1 cgd 1532: if (so->so_error || so->so_state & SS_CANTRCVMORE)
1533: break;
1.68 matt 1534: /*
1535: * If we are peeking and the socket receive buffer is
1536: * full, stop since we can't get more data to peek at.
1537: */
1538: if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1539: break;
1540: /*
1541: * If we've drained the socket buffer, tell the
1542: * protocol in case it needs to do something to
1543: * get it filled again.
1544: */
1545: if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1546: (*pr->pr_usrreq)(so, PRU_RCVD,
1.140 dyoung 1547: NULL, (struct mbuf *)(long)flags, NULL, l);
1.69 thorpej 1548: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1549: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1.196 dsl 1550: if (wakeup_state & SS_RESTARTSYS)
1551: error = ERESTART;
1552: else
1553: error = sbwait(&so->so_rcv);
1.144 dyoung 1554: if (error != 0) {
1.1 cgd 1555: sbunlock(&so->so_rcv);
1.160 ad 1556: sounlock(so);
1.1 cgd 1557: splx(s);
1.144 dyoung 1558: return 0;
1.1 cgd 1559: }
1.21 christos 1560: if ((m = so->so_rcv.sb_mb) != NULL)
1.1 cgd 1561: nextrecord = m->m_nextpkt;
1.196 dsl 1562: wakeup_state = so->so_state;
1.1 cgd 1563: }
1564: }
1.3 andrew 1565:
1.146 dyoung 1566: if (m && atomic) {
1.3 andrew 1567: flags |= MSG_TRUNC;
1568: if ((flags & MSG_PEEK) == 0)
1569: (void) sbdroprecord(&so->so_rcv);
1570: }
1.1 cgd 1571: if ((flags & MSG_PEEK) == 0) {
1.144 dyoung 1572: if (m == NULL) {
1.69 thorpej 1573: /*
1.70 thorpej 1574: * First part is an inline SB_EMPTY_FIXUP(). Second
1.69 thorpej 1575: * part makes sure sb_lastrecord is up-to-date if
1576: * there is still data in the socket buffer.
1577: */
1.1 cgd 1578: so->so_rcv.sb_mb = nextrecord;
1.69 thorpej 1579: if (so->so_rcv.sb_mb == NULL) {
1580: so->so_rcv.sb_mbtail = NULL;
1581: so->so_rcv.sb_lastrecord = NULL;
1582: } else if (nextrecord->m_nextpkt == NULL)
1583: so->so_rcv.sb_lastrecord = nextrecord;
1584: }
1585: SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1586: SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1.1 cgd 1587: if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1.140 dyoung 1588: (*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1589: (struct mbuf *)(long)flags, NULL, l);
1.1 cgd 1590: }
1.3 andrew 1591: if (orig_resid == uio->uio_resid && orig_resid &&
1592: (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1593: sbunlock(&so->so_rcv);
1594: goto restart;
1595: }
1.108 perry 1596:
1.144 dyoung 1597: if (flagsp != NULL)
1.1 cgd 1598: *flagsp |= flags;
1.54 lukem 1599: release:
1.1 cgd 1600: sbunlock(&so->so_rcv);
1.160 ad 1601: sounlock(so);
1.1 cgd 1602: splx(s);
1.144 dyoung 1603: return error;
1.1 cgd 1604: }
1605:
1.14 mycroft 1606: int
1.54 lukem 1607: soshutdown(struct socket *so, int how)
1.1 cgd 1608: {
1.99 matt 1609: const struct protosw *pr;
1.160 ad 1610: int error;
1611:
1612: KASSERT(solocked(so));
1.34 kleink 1613:
1.54 lukem 1614: pr = so->so_proto;
1.34 kleink 1615: if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1616: return (EINVAL);
1.1 cgd 1617:
1.160 ad 1618: if (how == SHUT_RD || how == SHUT_RDWR) {
1.1 cgd 1619: sorflush(so);
1.160 ad 1620: error = 0;
1621: }
1.34 kleink 1622: if (how == SHUT_WR || how == SHUT_RDWR)
1.160 ad 1623: error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1.140 dyoung 1624: NULL, NULL, NULL);
1.160 ad 1625:
1626: return error;
1.1 cgd 1627: }
1628:
1.195 dsl 1629: void
1.196 dsl 1630: sorestart(struct socket *so)
1.188 ad 1631: {
1.196 dsl 1632: /*
1633: * An application has called close() on an fd on which another
1634: * of its threads has called a socket system call.
1635: * Mark this and wake everyone up, and code that would block again
1636: * instead returns ERESTART.
1637: * On system call re-entry the fd is validated and EBADF returned.
1638: * Any other fd will block again on the 2nd syscall.
1639: */
1.188 ad 1640: solock(so);
1.196 dsl 1641: so->so_state |= SS_RESTARTSYS;
1.188 ad 1642: cv_broadcast(&so->so_cv);
1.196 dsl 1643: cv_broadcast(&so->so_snd.sb_cv);
1644: cv_broadcast(&so->so_rcv.sb_cv);
1.188 ad 1645: sounlock(so);
1646: }
1647:
1.14 mycroft 1648: void
1.54 lukem 1649: sorflush(struct socket *so)
1.1 cgd 1650: {
1.54 lukem 1651: struct sockbuf *sb, asb;
1.99 matt 1652: const struct protosw *pr;
1.160 ad 1653:
1654: KASSERT(solocked(so));
1.1 cgd 1655:
1.54 lukem 1656: sb = &so->so_rcv;
1657: pr = so->so_proto;
1.160 ad 1658: socantrcvmore(so);
1.1 cgd 1659: sb->sb_flags |= SB_NOINTR;
1.160 ad 1660: (void )sblock(sb, M_WAITOK);
1.1 cgd 1661: sbunlock(sb);
1662: asb = *sb;
1.86 wrstuden 1663: /*
1664: * Clear most of the sockbuf structure, but leave some of the
1665: * fields valid.
1666: */
1667: memset(&sb->sb_startzero, 0,
1668: sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1.160 ad 1669: if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1670: sounlock(so);
1.1 cgd 1671: (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1.160 ad 1672: solock(so);
1673: }
1.98 christos 1674: sbrelease(&asb, so);
1.1 cgd 1675: }
1676:
1.171 plunky 1677: /*
1678: * internal set SOL_SOCKET options
1679: */
1.142 dyoung 1680: static int
1.171 plunky 1681: sosetopt1(struct socket *so, const struct sockopt *sopt)
1.1 cgd 1682: {
1.182 christos 1683: int error = EINVAL, optval, opt;
1.171 plunky 1684: struct linger l;
1685: struct timeval tv;
1.142 dyoung 1686:
1.179 christos 1687: switch ((opt = sopt->sopt_name)) {
1.142 dyoung 1688:
1.170 tls 1689: case SO_ACCEPTFILTER:
1.177 ad 1690: error = accept_filt_setopt(so, sopt);
1691: KASSERT(solocked(so));
1.170 tls 1692: break;
1693:
1.171 plunky 1694: case SO_LINGER:
1695: error = sockopt_get(sopt, &l, sizeof(l));
1.177 ad 1696: solock(so);
1.171 plunky 1697: if (error)
1.177 ad 1698: break;
1.171 plunky 1699: if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1.177 ad 1700: l.l_linger > (INT_MAX / hz)) {
1701: error = EDOM;
1702: break;
1703: }
1.171 plunky 1704: so->so_linger = l.l_linger;
1705: if (l.l_onoff)
1706: so->so_options |= SO_LINGER;
1707: else
1708: so->so_options &= ~SO_LINGER;
1.177 ad 1709: break;
1.1 cgd 1710:
1.142 dyoung 1711: case SO_DEBUG:
1712: case SO_KEEPALIVE:
1713: case SO_DONTROUTE:
1714: case SO_USELOOPBACK:
1715: case SO_BROADCAST:
1716: case SO_REUSEADDR:
1717: case SO_REUSEPORT:
1718: case SO_OOBINLINE:
1719: case SO_TIMESTAMP:
1.184 christos 1720: #ifdef SO_OTIMESTAMP
1721: case SO_OTIMESTAMP:
1722: #endif
1.171 plunky 1723: error = sockopt_getint(sopt, &optval);
1.177 ad 1724: solock(so);
1.171 plunky 1725: if (error)
1.177 ad 1726: break;
1.171 plunky 1727: if (optval)
1.179 christos 1728: so->so_options |= opt;
1.142 dyoung 1729: else
1.179 christos 1730: so->so_options &= ~opt;
1.142 dyoung 1731: break;
1732:
1733: case SO_SNDBUF:
1734: case SO_RCVBUF:
1735: case SO_SNDLOWAT:
1736: case SO_RCVLOWAT:
1.171 plunky 1737: error = sockopt_getint(sopt, &optval);
1.177 ad 1738: solock(so);
1.171 plunky 1739: if (error)
1.177 ad 1740: break;
1.1 cgd 1741:
1.142 dyoung 1742: /*
1743: * Values < 1 make no sense for any of these
1744: * options, so disallow them.
1745: */
1.177 ad 1746: if (optval < 1) {
1747: error = EINVAL;
1748: break;
1749: }
1.1 cgd 1750:
1.179 christos 1751: switch (opt) {
1.171 plunky 1752: case SO_SNDBUF:
1.177 ad 1753: if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1754: error = ENOBUFS;
1755: break;
1756: }
1.171 plunky 1757: so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1758: break;
1.1 cgd 1759:
1760: case SO_RCVBUF:
1.177 ad 1761: if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1762: error = ENOBUFS;
1763: break;
1764: }
1.171 plunky 1765: so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1.142 dyoung 1766: break;
1767:
1768: /*
1769: * Make sure the low-water is never greater than
1770: * the high-water.
1771: */
1.1 cgd 1772: case SO_SNDLOWAT:
1.171 plunky 1773: if (optval > so->so_snd.sb_hiwat)
1774: optval = so->so_snd.sb_hiwat;
1775:
1776: so->so_snd.sb_lowat = optval;
1.142 dyoung 1777: break;
1.171 plunky 1778:
1.1 cgd 1779: case SO_RCVLOWAT:
1.171 plunky 1780: if (optval > so->so_rcv.sb_hiwat)
1781: optval = so->so_rcv.sb_hiwat;
1782:
1783: so->so_rcv.sb_lowat = optval;
1.142 dyoung 1784: break;
1785: }
1786: break;
1.28 thorpej 1787:
1.179 christos 1788: #ifdef COMPAT_50
1789: case SO_OSNDTIMEO:
1790: case SO_ORCVTIMEO: {
1791: struct timeval50 otv;
1792: error = sockopt_get(sopt, &otv, sizeof(otv));
1.186 pooka 1793: if (error) {
1794: solock(so);
1.183 christos 1795: break;
1.186 pooka 1796: }
1.179 christos 1797: timeval50_to_timeval(&otv, &tv);
1798: opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1.182 christos 1799: error = 0;
1.179 christos 1800: /*FALLTHROUGH*/
1801: }
1802: #endif /* COMPAT_50 */
1803:
1.142 dyoung 1804: case SO_SNDTIMEO:
1805: case SO_RCVTIMEO:
1.182 christos 1806: if (error)
1.179 christos 1807: error = sockopt_get(sopt, &tv, sizeof(tv));
1.177 ad 1808: solock(so);
1.171 plunky 1809: if (error)
1.177 ad 1810: break;
1.171 plunky 1811:
1.177 ad 1812: if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1813: error = EDOM;
1814: break;
1815: }
1.28 thorpej 1816:
1.171 plunky 1817: optval = tv.tv_sec * hz + tv.tv_usec / tick;
1818: if (optval == 0 && tv.tv_usec != 0)
1819: optval = 1;
1.28 thorpej 1820:
1.179 christos 1821: switch (opt) {
1.142 dyoung 1822: case SO_SNDTIMEO:
1.171 plunky 1823: so->so_snd.sb_timeo = optval;
1.1 cgd 1824: break;
1825: case SO_RCVTIMEO:
1.171 plunky 1826: so->so_rcv.sb_timeo = optval;
1.142 dyoung 1827: break;
1828: }
1829: break;
1.1 cgd 1830:
1.142 dyoung 1831: default:
1.177 ad 1832: solock(so);
1833: error = ENOPROTOOPT;
1834: break;
1.142 dyoung 1835: }
1.177 ad 1836: KASSERT(solocked(so));
1837: return error;
1.142 dyoung 1838: }
1.1 cgd 1839:
1.142 dyoung 1840: int
1.171 plunky 1841: sosetopt(struct socket *so, struct sockopt *sopt)
1.142 dyoung 1842: {
1843: int error, prerr;
1.1 cgd 1844:
1.177 ad 1845: if (sopt->sopt_level == SOL_SOCKET) {
1.171 plunky 1846: error = sosetopt1(so, sopt);
1.177 ad 1847: KASSERT(solocked(so));
1848: } else {
1.142 dyoung 1849: error = ENOPROTOOPT;
1.177 ad 1850: solock(so);
1851: }
1.1 cgd 1852:
1.142 dyoung 1853: if ((error == 0 || error == ENOPROTOOPT) &&
1854: so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1855: /* give the protocol stack a shot */
1.171 plunky 1856: prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1.142 dyoung 1857: if (prerr == 0)
1858: error = 0;
1859: else if (prerr != ENOPROTOOPT)
1860: error = prerr;
1.171 plunky 1861: }
1.160 ad 1862: sounlock(so);
1.142 dyoung 1863: return error;
1.1 cgd 1864: }
1865:
1.171 plunky 1866: /*
1867: * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1868: */
1869: int
1870: so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1871: const void *val, size_t valsize)
1872: {
1873: struct sockopt sopt;
1874: int error;
1875:
1876: KASSERT(valsize == 0 || val != NULL);
1877:
1878: sockopt_init(&sopt, level, name, valsize);
1879: sockopt_set(&sopt, val, valsize);
1880:
1881: error = sosetopt(so, &sopt);
1882:
1883: sockopt_destroy(&sopt);
1884:
1885: return error;
1886: }
1887:
1888: /*
1889: * internal get SOL_SOCKET options
1890: */
1891: static int
1892: sogetopt1(struct socket *so, struct sockopt *sopt)
1893: {
1.179 christos 1894: int error, optval, opt;
1.171 plunky 1895: struct linger l;
1896: struct timeval tv;
1897:
1.179 christos 1898: switch ((opt = sopt->sopt_name)) {
1.171 plunky 1899:
1900: case SO_ACCEPTFILTER:
1.177 ad 1901: error = accept_filt_getopt(so, sopt);
1.171 plunky 1902: break;
1903:
1904: case SO_LINGER:
1905: l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1906: l.l_linger = so->so_linger;
1907:
1908: error = sockopt_set(sopt, &l, sizeof(l));
1909: break;
1910:
1911: case SO_USELOOPBACK:
1912: case SO_DONTROUTE:
1913: case SO_DEBUG:
1914: case SO_KEEPALIVE:
1915: case SO_REUSEADDR:
1916: case SO_REUSEPORT:
1917: case SO_BROADCAST:
1918: case SO_OOBINLINE:
1919: case SO_TIMESTAMP:
1.184 christos 1920: #ifdef SO_OTIMESTAMP
1921: case SO_OTIMESTAMP:
1922: #endif
1.179 christos 1923: error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1.171 plunky 1924: break;
1925:
1926: case SO_TYPE:
1927: error = sockopt_setint(sopt, so->so_type);
1928: break;
1929:
1930: case SO_ERROR:
1931: error = sockopt_setint(sopt, so->so_error);
1932: so->so_error = 0;
1933: break;
1934:
1935: case SO_SNDBUF:
1936: error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1937: break;
1938:
1939: case SO_RCVBUF:
1940: error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1941: break;
1942:
1943: case SO_SNDLOWAT:
1944: error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1945: break;
1946:
1947: case SO_RCVLOWAT:
1948: error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1949: break;
1950:
1.179 christos 1951: #ifdef COMPAT_50
1952: case SO_OSNDTIMEO:
1953: case SO_ORCVTIMEO: {
1954: struct timeval50 otv;
1955:
1956: optval = (opt == SO_OSNDTIMEO ?
1957: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1958:
1959: otv.tv_sec = optval / hz;
1960: otv.tv_usec = (optval % hz) * tick;
1961:
1962: error = sockopt_set(sopt, &otv, sizeof(otv));
1963: break;
1964: }
1965: #endif /* COMPAT_50 */
1966:
1.171 plunky 1967: case SO_SNDTIMEO:
1968: case SO_RCVTIMEO:
1.179 christos 1969: optval = (opt == SO_SNDTIMEO ?
1.171 plunky 1970: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1971:
1972: tv.tv_sec = optval / hz;
1973: tv.tv_usec = (optval % hz) * tick;
1974:
1975: error = sockopt_set(sopt, &tv, sizeof(tv));
1976: break;
1977:
1978: case SO_OVERFLOWED:
1979: error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1980: break;
1981:
1982: default:
1983: error = ENOPROTOOPT;
1984: break;
1985: }
1986:
1987: return (error);
1988: }
1989:
1.14 mycroft 1990: int
1.171 plunky 1991: sogetopt(struct socket *so, struct sockopt *sopt)
1.1 cgd 1992: {
1.160 ad 1993: int error;
1.1 cgd 1994:
1.160 ad 1995: solock(so);
1.171 plunky 1996: if (sopt->sopt_level != SOL_SOCKET) {
1.1 cgd 1997: if (so->so_proto && so->so_proto->pr_ctloutput) {
1.160 ad 1998: error = ((*so->so_proto->pr_ctloutput)
1.171 plunky 1999: (PRCO_GETOPT, so, sopt));
1.1 cgd 2000: } else
1.160 ad 2001: error = (ENOPROTOOPT);
1.1 cgd 2002: } else {
1.171 plunky 2003: error = sogetopt1(so, sopt);
2004: }
2005: sounlock(so);
2006: return (error);
2007: }
2008:
2009: /*
2010: * alloc sockopt data buffer buffer
2011: * - will be released at destroy
2012: */
1.176 plunky 2013: static int
2014: sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
1.171 plunky 2015: {
2016:
2017: KASSERT(sopt->sopt_size == 0);
2018:
1.176 plunky 2019: if (len > sizeof(sopt->sopt_buf)) {
2020: sopt->sopt_data = kmem_zalloc(len, kmflag);
2021: if (sopt->sopt_data == NULL)
2022: return ENOMEM;
2023: } else
1.171 plunky 2024: sopt->sopt_data = sopt->sopt_buf;
2025:
2026: sopt->sopt_size = len;
1.176 plunky 2027: return 0;
1.171 plunky 2028: }
2029:
2030: /*
2031: * initialise sockopt storage
1.176 plunky 2032: * - MAY sleep during allocation
1.171 plunky 2033: */
2034: void
2035: sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2036: {
1.1 cgd 2037:
1.171 plunky 2038: memset(sopt, 0, sizeof(*sopt));
1.1 cgd 2039:
1.171 plunky 2040: sopt->sopt_level = level;
2041: sopt->sopt_name = name;
1.176 plunky 2042: (void)sockopt_alloc(sopt, size, KM_SLEEP);
1.171 plunky 2043: }
2044:
2045: /*
2046: * destroy sockopt storage
2047: * - will release any held memory references
2048: */
2049: void
2050: sockopt_destroy(struct sockopt *sopt)
2051: {
2052:
2053: if (sopt->sopt_data != sopt->sopt_buf)
1.173 plunky 2054: kmem_free(sopt->sopt_data, sopt->sopt_size);
1.171 plunky 2055:
2056: memset(sopt, 0, sizeof(*sopt));
2057: }
2058:
2059: /*
2060: * set sockopt value
2061: * - value is copied into sockopt
1.176 plunky 2062: * - memory is allocated when necessary, will not sleep
1.171 plunky 2063: */
2064: int
2065: sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2066: {
1.176 plunky 2067: int error;
1.171 plunky 2068:
1.176 plunky 2069: if (sopt->sopt_size == 0) {
2070: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2071: if (error)
2072: return error;
2073: }
1.171 plunky 2074:
2075: KASSERT(sopt->sopt_size == len);
2076: memcpy(sopt->sopt_data, buf, len);
2077: return 0;
2078: }
2079:
2080: /*
2081: * common case of set sockopt integer value
2082: */
2083: int
2084: sockopt_setint(struct sockopt *sopt, int val)
2085: {
2086:
2087: return sockopt_set(sopt, &val, sizeof(int));
2088: }
2089:
2090: /*
2091: * get sockopt value
2092: * - correct size must be given
2093: */
2094: int
2095: sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2096: {
1.170 tls 2097:
1.171 plunky 2098: if (sopt->sopt_size != len)
2099: return EINVAL;
1.1 cgd 2100:
1.171 plunky 2101: memcpy(buf, sopt->sopt_data, len);
2102: return 0;
2103: }
1.1 cgd 2104:
1.171 plunky 2105: /*
2106: * common case of get sockopt integer value
2107: */
2108: int
2109: sockopt_getint(const struct sockopt *sopt, int *valp)
2110: {
1.1 cgd 2111:
1.171 plunky 2112: return sockopt_get(sopt, valp, sizeof(int));
2113: }
1.1 cgd 2114:
1.171 plunky 2115: /*
2116: * set sockopt value from mbuf
2117: * - ONLY for legacy code
2118: * - mbuf is released by sockopt
1.176 plunky 2119: * - will not sleep
1.171 plunky 2120: */
2121: int
2122: sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2123: {
2124: size_t len;
1.176 plunky 2125: int error;
1.1 cgd 2126:
1.171 plunky 2127: len = m_length(m);
1.1 cgd 2128:
1.176 plunky 2129: if (sopt->sopt_size == 0) {
2130: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2131: if (error)
2132: return error;
2133: }
1.1 cgd 2134:
1.171 plunky 2135: KASSERT(sopt->sopt_size == len);
2136: m_copydata(m, 0, len, sopt->sopt_data);
2137: m_freem(m);
1.1 cgd 2138:
1.171 plunky 2139: return 0;
2140: }
1.1 cgd 2141:
1.171 plunky 2142: /*
2143: * get sockopt value into mbuf
2144: * - ONLY for legacy code
2145: * - mbuf to be released by the caller
1.176 plunky 2146: * - will not sleep
1.171 plunky 2147: */
2148: struct mbuf *
2149: sockopt_getmbuf(const struct sockopt *sopt)
2150: {
2151: struct mbuf *m;
1.107 darrenr 2152:
1.176 plunky 2153: if (sopt->sopt_size > MCLBYTES)
2154: return NULL;
2155:
2156: m = m_get(M_DONTWAIT, MT_SOOPTS);
1.171 plunky 2157: if (m == NULL)
2158: return NULL;
2159:
1.176 plunky 2160: if (sopt->sopt_size > MLEN) {
2161: MCLGET(m, M_DONTWAIT);
2162: if ((m->m_flags & M_EXT) == 0) {
2163: m_free(m);
2164: return NULL;
2165: }
1.1 cgd 2166: }
1.176 plunky 2167:
2168: memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2169: m->m_len = sopt->sopt_size;
1.160 ad 2170:
1.171 plunky 2171: return m;
1.1 cgd 2172: }
2173:
1.14 mycroft 2174: void
1.54 lukem 2175: sohasoutofband(struct socket *so)
1.1 cgd 2176: {
1.153 rmind 2177:
1.90 christos 2178: fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1.189 ad 2179: selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
1.1 cgd 2180: }
1.72 jdolecek 2181:
2182: static void
2183: filt_sordetach(struct knote *kn)
2184: {
2185: struct socket *so;
2186:
1.155 ad 2187: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2188: solock(so);
1.73 christos 2189: SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2190: if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1.72 jdolecek 2191: so->so_rcv.sb_flags &= ~SB_KNOTE;
1.160 ad 2192: sounlock(so);
1.72 jdolecek 2193: }
2194:
2195: /*ARGSUSED*/
2196: static int
1.129 yamt 2197: filt_soread(struct knote *kn, long hint)
1.72 jdolecek 2198: {
2199: struct socket *so;
1.160 ad 2200: int rv;
1.72 jdolecek 2201:
1.155 ad 2202: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2203: if (hint != NOTE_SUBMIT)
2204: solock(so);
1.72 jdolecek 2205: kn->kn_data = so->so_rcv.sb_cc;
2206: if (so->so_state & SS_CANTRCVMORE) {
1.108 perry 2207: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2208: kn->kn_fflags = so->so_error;
1.160 ad 2209: rv = 1;
2210: } else if (so->so_error) /* temporary udp error */
2211: rv = 1;
2212: else if (kn->kn_sfflags & NOTE_LOWAT)
2213: rv = (kn->kn_data >= kn->kn_sdata);
2214: else
2215: rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2216: if (hint != NOTE_SUBMIT)
2217: sounlock(so);
2218: return rv;
1.72 jdolecek 2219: }
2220:
2221: static void
2222: filt_sowdetach(struct knote *kn)
2223: {
2224: struct socket *so;
2225:
1.155 ad 2226: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2227: solock(so);
1.73 christos 2228: SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2229: if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1.72 jdolecek 2230: so->so_snd.sb_flags &= ~SB_KNOTE;
1.160 ad 2231: sounlock(so);
1.72 jdolecek 2232: }
2233:
2234: /*ARGSUSED*/
2235: static int
1.129 yamt 2236: filt_sowrite(struct knote *kn, long hint)
1.72 jdolecek 2237: {
2238: struct socket *so;
1.160 ad 2239: int rv;
1.72 jdolecek 2240:
1.155 ad 2241: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2242: if (hint != NOTE_SUBMIT)
2243: solock(so);
1.72 jdolecek 2244: kn->kn_data = sbspace(&so->so_snd);
2245: if (so->so_state & SS_CANTSENDMORE) {
1.108 perry 2246: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2247: kn->kn_fflags = so->so_error;
1.160 ad 2248: rv = 1;
2249: } else if (so->so_error) /* temporary udp error */
2250: rv = 1;
2251: else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1.72 jdolecek 2252: (so->so_proto->pr_flags & PR_CONNREQUIRED))
1.160 ad 2253: rv = 0;
2254: else if (kn->kn_sfflags & NOTE_LOWAT)
2255: rv = (kn->kn_data >= kn->kn_sdata);
2256: else
2257: rv = (kn->kn_data >= so->so_snd.sb_lowat);
2258: if (hint != NOTE_SUBMIT)
2259: sounlock(so);
2260: return rv;
1.72 jdolecek 2261: }
2262:
2263: /*ARGSUSED*/
2264: static int
1.129 yamt 2265: filt_solisten(struct knote *kn, long hint)
1.72 jdolecek 2266: {
2267: struct socket *so;
1.160 ad 2268: int rv;
1.72 jdolecek 2269:
1.155 ad 2270: so = ((file_t *)kn->kn_obj)->f_data;
1.72 jdolecek 2271:
2272: /*
2273: * Set kn_data to number of incoming connections, not
2274: * counting partial (incomplete) connections.
1.108 perry 2275: */
1.160 ad 2276: if (hint != NOTE_SUBMIT)
2277: solock(so);
1.72 jdolecek 2278: kn->kn_data = so->so_qlen;
1.160 ad 2279: rv = (kn->kn_data > 0);
2280: if (hint != NOTE_SUBMIT)
2281: sounlock(so);
2282: return rv;
1.72 jdolecek 2283: }
2284:
2285: static const struct filterops solisten_filtops =
2286: { 1, NULL, filt_sordetach, filt_solisten };
2287: static const struct filterops soread_filtops =
2288: { 1, NULL, filt_sordetach, filt_soread };
2289: static const struct filterops sowrite_filtops =
2290: { 1, NULL, filt_sowdetach, filt_sowrite };
2291:
2292: int
1.129 yamt 2293: soo_kqfilter(struct file *fp, struct knote *kn)
1.72 jdolecek 2294: {
2295: struct socket *so;
2296: struct sockbuf *sb;
2297:
1.155 ad 2298: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2299: solock(so);
1.72 jdolecek 2300: switch (kn->kn_filter) {
2301: case EVFILT_READ:
2302: if (so->so_options & SO_ACCEPTCONN)
2303: kn->kn_fop = &solisten_filtops;
2304: else
2305: kn->kn_fop = &soread_filtops;
2306: sb = &so->so_rcv;
2307: break;
2308: case EVFILT_WRITE:
2309: kn->kn_fop = &sowrite_filtops;
2310: sb = &so->so_snd;
2311: break;
2312: default:
1.160 ad 2313: sounlock(so);
1.149 pooka 2314: return (EINVAL);
1.72 jdolecek 2315: }
1.73 christos 2316: SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1.72 jdolecek 2317: sb->sb_flags |= SB_KNOTE;
1.160 ad 2318: sounlock(so);
1.72 jdolecek 2319: return (0);
2320: }
2321:
1.154 ad 2322: static int
2323: sodopoll(struct socket *so, int events)
2324: {
2325: int revents;
2326:
2327: revents = 0;
2328:
2329: if (events & (POLLIN | POLLRDNORM))
2330: if (soreadable(so))
2331: revents |= events & (POLLIN | POLLRDNORM);
2332:
2333: if (events & (POLLOUT | POLLWRNORM))
2334: if (sowritable(so))
2335: revents |= events & (POLLOUT | POLLWRNORM);
2336:
2337: if (events & (POLLPRI | POLLRDBAND))
2338: if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2339: revents |= events & (POLLPRI | POLLRDBAND);
2340:
2341: return revents;
2342: }
2343:
2344: int
2345: sopoll(struct socket *so, int events)
2346: {
2347: int revents = 0;
2348:
1.160 ad 2349: #ifndef DIAGNOSTIC
2350: /*
2351: * Do a quick, unlocked check in expectation that the socket
2352: * will be ready for I/O. Don't do this check if DIAGNOSTIC,
2353: * as the solocked() assertions will fail.
2354: */
1.154 ad 2355: if ((revents = sodopoll(so, events)) != 0)
2356: return revents;
1.160 ad 2357: #endif
1.154 ad 2358:
1.160 ad 2359: solock(so);
1.154 ad 2360: if ((revents = sodopoll(so, events)) == 0) {
2361: if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2362: selrecord(curlwp, &so->so_rcv.sb_sel);
1.160 ad 2363: so->so_rcv.sb_flags |= SB_NOTIFY;
1.154 ad 2364: }
2365:
2366: if (events & (POLLOUT | POLLWRNORM)) {
2367: selrecord(curlwp, &so->so_snd.sb_sel);
1.160 ad 2368: so->so_snd.sb_flags |= SB_NOTIFY;
1.154 ad 2369: }
2370: }
1.160 ad 2371: sounlock(so);
1.154 ad 2372:
2373: return revents;
2374: }
2375:
2376:
1.94 yamt 2377: #include <sys/sysctl.h>
2378:
2379: static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2380:
2381: /*
2382: * sysctl helper routine for kern.somaxkva. ensures that the given
2383: * value is not too small.
2384: * (XXX should we maybe make sure it's not too large as well?)
2385: */
2386: static int
2387: sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2388: {
2389: int error, new_somaxkva;
2390: struct sysctlnode node;
2391:
2392: new_somaxkva = somaxkva;
2393: node = *rnode;
2394: node.sysctl_data = &new_somaxkva;
2395: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2396: if (error || newp == NULL)
2397: return (error);
2398:
2399: if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2400: return (EINVAL);
2401:
1.136 ad 2402: mutex_enter(&so_pendfree_lock);
1.94 yamt 2403: somaxkva = new_somaxkva;
1.136 ad 2404: cv_broadcast(&socurkva_cv);
2405: mutex_exit(&so_pendfree_lock);
1.94 yamt 2406:
2407: return (error);
2408: }
2409:
1.178 pooka 2410: static void
1.187 cegger 2411: sysctl_kern_somaxkva_setup(void)
1.94 yamt 2412: {
2413:
1.178 pooka 2414: KASSERT(socket_sysctllog == NULL);
2415: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2416: CTLFLAG_PERMANENT,
2417: CTLTYPE_NODE, "kern", NULL,
2418: NULL, 0, NULL, 0,
2419: CTL_KERN, CTL_EOL);
2420:
1.178 pooka 2421: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2422: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1.103 atatat 2423: CTLTYPE_INT, "somaxkva",
2424: SYSCTL_DESCR("Maximum amount of kernel memory to be "
2425: "used for socket buffers"),
1.94 yamt 2426: sysctl_kern_somaxkva, 0, NULL, 0,
2427: CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2428: }
CVSweb <webmaster@jp.NetBSD.org>