Annotation of src/sys/kern/uipc_socket.c, Revision 1.194
1.194 ! cegger 1: /* $NetBSD: uipc_socket.c,v 1.193 2009/10/03 03:59:39 elad 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>
1.194 ! cegger 66: __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.193 2009/10/03 03:59:39 elad 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>
1.37 thorpej 95:
1.179 christos 96: #ifdef COMPAT_50
97: #include <compat/sys/time.h>
1.184 christos 98: #include <compat/sys/socket.h>
1.179 christos 99: #endif
100:
1.64 thorpej 101: #include <uvm/uvm.h>
102:
1.77 thorpej 103: MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
104: MALLOC_DEFINE(M_SONAME, "soname", "socket name");
1.37 thorpej 105:
1.142 dyoung 106: extern const struct fileops socketops;
107:
1.54 lukem 108: extern int somaxconn; /* patchable (XXX sysctl) */
109: int somaxconn = SOMAXCONN;
1.160 ad 110: kmutex_t *softnet_lock;
1.49 jonathan 111:
1.64 thorpej 112: #ifdef SOSEND_COUNTERS
113: #include <sys/device.h>
114:
1.113 thorpej 115: static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 116: NULL, "sosend", "loan big");
1.113 thorpej 117: static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 118: NULL, "sosend", "copy big");
1.113 thorpej 119: static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 120: NULL, "sosend", "copy small");
1.113 thorpej 121: static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 122: NULL, "sosend", "kva limit");
123:
124: #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++
125:
1.101 matt 126: EVCNT_ATTACH_STATIC(sosend_loan_big);
127: EVCNT_ATTACH_STATIC(sosend_copy_big);
128: EVCNT_ATTACH_STATIC(sosend_copy_small);
129: EVCNT_ATTACH_STATIC(sosend_kvalimit);
1.64 thorpej 130: #else
131:
132: #define SOSEND_COUNTER_INCR(ev) /* nothing */
133:
134: #endif /* SOSEND_COUNTERS */
135:
1.119 yamt 136: static struct callback_entry sokva_reclaimerentry;
1.1 cgd 137:
1.167 ad 138: #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
1.121 yamt 139: int sock_loan_thresh = -1;
1.71 thorpej 140: #else
1.121 yamt 141: int sock_loan_thresh = 4096;
1.65 thorpej 142: #endif
1.64 thorpej 143:
1.136 ad 144: static kmutex_t so_pendfree_lock;
1.113 thorpej 145: static struct mbuf *so_pendfree;
1.64 thorpej 146:
1.84 ragge 147: #ifndef SOMAXKVA
148: #define SOMAXKVA (16 * 1024 * 1024)
149: #endif
150: int somaxkva = SOMAXKVA;
1.113 thorpej 151: static int socurkva;
1.136 ad 152: static kcondvar_t socurkva_cv;
1.64 thorpej 153:
1.191 elad 154: static kauth_listener_t socket_listener;
155:
1.64 thorpej 156: #define SOCK_LOAN_CHUNK 65536
157:
1.117 yamt 158: static size_t sodopendfree(void);
159: static size_t sodopendfreel(void);
1.93 yamt 160:
1.178 pooka 161: static void sysctl_kern_somaxkva_setup(void);
162: static struct sysctllog *socket_sysctllog;
163:
1.113 thorpej 164: static vsize_t
1.129 yamt 165: sokvareserve(struct socket *so, vsize_t len)
1.80 yamt 166: {
1.98 christos 167: int error;
1.80 yamt 168:
1.136 ad 169: mutex_enter(&so_pendfree_lock);
1.80 yamt 170: while (socurkva + len > somaxkva) {
1.93 yamt 171: size_t freed;
172:
173: /*
174: * try to do pendfree.
175: */
176:
1.117 yamt 177: freed = sodopendfreel();
1.93 yamt 178:
179: /*
180: * if some kva was freed, try again.
181: */
182:
183: if (freed)
1.80 yamt 184: continue;
1.93 yamt 185:
1.80 yamt 186: SOSEND_COUNTER_INCR(&sosend_kvalimit);
1.136 ad 187: error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
1.98 christos 188: if (error) {
189: len = 0;
190: break;
191: }
1.80 yamt 192: }
1.93 yamt 193: socurkva += len;
1.136 ad 194: mutex_exit(&so_pendfree_lock);
1.98 christos 195: return len;
1.95 yamt 196: }
197:
1.113 thorpej 198: static void
1.95 yamt 199: sokvaunreserve(vsize_t len)
200: {
201:
1.136 ad 202: mutex_enter(&so_pendfree_lock);
1.95 yamt 203: socurkva -= len;
1.136 ad 204: cv_broadcast(&socurkva_cv);
205: mutex_exit(&so_pendfree_lock);
1.95 yamt 206: }
207:
208: /*
209: * sokvaalloc: allocate kva for loan.
210: */
211:
212: vaddr_t
213: sokvaalloc(vsize_t len, struct socket *so)
214: {
215: vaddr_t lva;
216:
217: /*
218: * reserve kva.
219: */
220:
1.98 christos 221: if (sokvareserve(so, len) == 0)
222: return 0;
1.93 yamt 223:
224: /*
225: * allocate kva.
226: */
1.80 yamt 227:
1.109 yamt 228: lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
1.95 yamt 229: if (lva == 0) {
230: sokvaunreserve(len);
1.80 yamt 231: return (0);
1.95 yamt 232: }
1.80 yamt 233:
234: return lva;
235: }
236:
1.93 yamt 237: /*
238: * sokvafree: free kva for loan.
239: */
240:
1.80 yamt 241: void
242: sokvafree(vaddr_t sva, vsize_t len)
243: {
1.93 yamt 244:
245: /*
246: * free kva.
247: */
1.80 yamt 248:
1.109 yamt 249: uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
1.93 yamt 250:
251: /*
252: * unreserve kva.
253: */
254:
1.95 yamt 255: sokvaunreserve(len);
1.80 yamt 256: }
257:
1.64 thorpej 258: static void
1.134 christos 259: sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
1.64 thorpej 260: {
1.156 yamt 261: vaddr_t sva, eva;
1.64 thorpej 262: vsize_t len;
1.156 yamt 263: int npgs;
264:
265: KASSERT(pgs != NULL);
1.64 thorpej 266:
267: eva = round_page((vaddr_t) buf + size);
268: sva = trunc_page((vaddr_t) buf);
269: len = eva - sva;
270: npgs = len >> PAGE_SHIFT;
271:
272: pmap_kremove(sva, len);
273: pmap_update(pmap_kernel());
274: uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
1.80 yamt 275: sokvafree(sva, len);
1.64 thorpej 276: }
277:
278: static size_t
1.152 matt 279: sodopendfree(void)
1.64 thorpej 280: {
1.93 yamt 281: size_t rv;
1.64 thorpej 282:
1.160 ad 283: if (__predict_true(so_pendfree == NULL))
284: return 0;
285:
1.136 ad 286: mutex_enter(&so_pendfree_lock);
1.117 yamt 287: rv = sodopendfreel();
1.136 ad 288: mutex_exit(&so_pendfree_lock);
1.93 yamt 289:
290: return rv;
291: }
292:
293: /*
294: * sodopendfreel: free mbufs on "pendfree" list.
1.136 ad 295: * unlock and relock so_pendfree_lock when freeing mbufs.
1.93 yamt 296: *
1.136 ad 297: * => called with so_pendfree_lock held.
1.93 yamt 298: */
299:
300: static size_t
1.152 matt 301: sodopendfreel(void)
1.93 yamt 302: {
1.137 ad 303: struct mbuf *m, *next;
1.93 yamt 304: size_t rv = 0;
305:
1.136 ad 306: KASSERT(mutex_owned(&so_pendfree_lock));
1.64 thorpej 307:
1.137 ad 308: while (so_pendfree != NULL) {
1.64 thorpej 309: m = so_pendfree;
1.93 yamt 310: so_pendfree = NULL;
1.136 ad 311: mutex_exit(&so_pendfree_lock);
1.93 yamt 312:
313: for (; m != NULL; m = next) {
314: next = m->m_next;
1.156 yamt 315: KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
316: KASSERT(m->m_ext.ext_refcnt == 0);
1.93 yamt 317:
318: rv += m->m_ext.ext_size;
1.156 yamt 319: sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
1.93 yamt 320: m->m_ext.ext_size);
1.145 ad 321: pool_cache_put(mb_cache, m);
1.93 yamt 322: }
1.64 thorpej 323:
1.136 ad 324: mutex_enter(&so_pendfree_lock);
1.64 thorpej 325: }
326:
327: return (rv);
328: }
329:
1.80 yamt 330: void
1.134 christos 331: soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
1.64 thorpej 332: {
333:
1.156 yamt 334: KASSERT(m != NULL);
1.64 thorpej 335:
1.93 yamt 336: /*
337: * postpone freeing mbuf.
338: *
339: * we can't do it in interrupt context
340: * because we need to put kva back to kernel_map.
341: */
342:
1.136 ad 343: mutex_enter(&so_pendfree_lock);
1.92 yamt 344: m->m_next = so_pendfree;
345: so_pendfree = m;
1.136 ad 346: cv_broadcast(&socurkva_cv);
347: mutex_exit(&so_pendfree_lock);
1.64 thorpej 348: }
349:
350: static long
351: sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
352: {
353: struct iovec *iov = uio->uio_iov;
354: vaddr_t sva, eva;
355: vsize_t len;
1.156 yamt 356: vaddr_t lva;
357: int npgs, error;
358: vaddr_t va;
359: int i;
1.64 thorpej 360:
1.116 yamt 361: if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
1.64 thorpej 362: return (0);
363:
364: if (iov->iov_len < (size_t) space)
365: space = iov->iov_len;
366: if (space > SOCK_LOAN_CHUNK)
367: space = SOCK_LOAN_CHUNK;
368:
369: eva = round_page((vaddr_t) iov->iov_base + space);
370: sva = trunc_page((vaddr_t) iov->iov_base);
371: len = eva - sva;
372: npgs = len >> PAGE_SHIFT;
373:
1.79 thorpej 374: KASSERT(npgs <= M_EXT_MAXPAGES);
375:
1.80 yamt 376: lva = sokvaalloc(len, so);
1.64 thorpej 377: if (lva == 0)
1.80 yamt 378: return 0;
1.64 thorpej 379:
1.116 yamt 380: error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
1.79 thorpej 381: m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
1.64 thorpej 382: if (error) {
1.80 yamt 383: sokvafree(lva, len);
1.64 thorpej 384: return (0);
385: }
386:
387: for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
1.79 thorpej 388: pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
1.194 ! cegger 389: VM_PROT_READ, 0);
1.64 thorpej 390: pmap_update(pmap_kernel());
391:
392: lva += (vaddr_t) iov->iov_base & PAGE_MASK;
393:
1.134 christos 394: MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
1.79 thorpej 395: m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
1.64 thorpej 396:
397: uio->uio_resid -= space;
398: /* uio_offset not updated, not set/used for write(2) */
1.134 christos 399: uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
1.64 thorpej 400: uio->uio_iov->iov_len -= space;
401: if (uio->uio_iov->iov_len == 0) {
402: uio->uio_iov++;
403: uio->uio_iovcnt--;
404: }
405:
406: return (space);
407: }
408:
1.119 yamt 409: static int
1.129 yamt 410: sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.119 yamt 411: {
412:
413: KASSERT(ce == &sokva_reclaimerentry);
414: KASSERT(obj == NULL);
415:
416: sodopendfree();
417: if (!vm_map_starved_p(kernel_map)) {
418: return CALLBACK_CHAIN_ABORT;
419: }
420: return CALLBACK_CHAIN_CONTINUE;
421: }
422:
1.142 dyoung 423: struct mbuf *
1.147 dyoung 424: getsombuf(struct socket *so, int type)
1.142 dyoung 425: {
426: struct mbuf *m;
427:
1.147 dyoung 428: m = m_get(M_WAIT, type);
1.142 dyoung 429: MCLAIM(m, so->so_mowner);
430: return m;
431: }
432:
1.191 elad 433: static int
434: socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
435: void *arg0, void *arg1, void *arg2, void *arg3)
436: {
437: int result;
438: enum kauth_network_req req;
439:
440: result = KAUTH_RESULT_DEFER;
441: req = (enum kauth_network_req)arg0;
442:
1.193 elad 443: if ((action != KAUTH_NETWORK_SOCKET) &&
444: (action != KAUTH_NETWORK_BIND))
1.191 elad 445: return result;
446:
447: switch (req) {
1.193 elad 448: case KAUTH_REQ_NETWORK_BIND_PORT:
449: result = KAUTH_RESULT_ALLOW;
450: break;
451:
1.191 elad 452: case KAUTH_REQ_NETWORK_SOCKET_DROP: {
453: /* Normal users can only drop their own connections. */
454: struct socket *so = (struct socket *)arg1;
455: uid_t sockuid = so->so_uidinfo->ui_uid;
456:
457: if (sockuid == kauth_cred_getuid(cred) ||
458: sockuid == kauth_cred_geteuid(cred))
459: result = KAUTH_RESULT_ALLOW;
460:
461: break;
462: }
463:
464: case KAUTH_REQ_NETWORK_SOCKET_OPEN:
465: /* We allow "raw" routing/bluetooth sockets to anyone. */
466: if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_BLUETOOTH)
467: result = KAUTH_RESULT_ALLOW;
468: else {
469: /* Privileged, let secmodel handle this. */
470: if ((u_long)arg2 == SOCK_RAW)
471: break;
472: }
473:
474: result = KAUTH_RESULT_ALLOW;
475:
476: break;
477:
1.192 elad 478: case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
479: result = KAUTH_RESULT_ALLOW;
480:
481: break;
482:
1.191 elad 483: default:
484: break;
485: }
486:
487: return result;
488: }
489:
1.119 yamt 490: void
491: soinit(void)
492: {
493:
1.178 pooka 494: sysctl_kern_somaxkva_setup();
495:
1.148 ad 496: mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
1.160 ad 497: softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
1.136 ad 498: cv_init(&socurkva_cv, "sokva");
1.166 ad 499: soinit2();
1.136 ad 500:
1.119 yamt 501: /* Set the initial adjusted socket buffer size. */
502: if (sb_max_set(sb_max))
503: panic("bad initial sb_max value: %lu", sb_max);
504:
505: callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback,
506: &sokva_reclaimerentry, NULL, sokva_reclaim_callback);
1.191 elad 507:
508: socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
509: socket_listener_cb, NULL);
1.119 yamt 510: }
511:
1.1 cgd 512: /*
513: * Socket operation routines.
514: * These routines are called by the routines in
515: * sys_socket.c or from a system process, and
516: * implement the semantics of socket operations by
517: * switching out to the protocol specific routines.
518: */
519: /*ARGSUSED*/
1.3 andrew 520: int
1.160 ad 521: socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
522: struct socket *lockso)
1.1 cgd 523: {
1.99 matt 524: const struct protosw *prp;
1.54 lukem 525: struct socket *so;
1.115 yamt 526: uid_t uid;
1.160 ad 527: int error;
528: kmutex_t *lock;
1.1 cgd 529:
1.132 elad 530: error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
531: KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
532: KAUTH_ARG(proto));
1.140 dyoung 533: if (error != 0)
534: return error;
1.127 elad 535:
1.1 cgd 536: if (proto)
537: prp = pffindproto(dom, proto, type);
538: else
539: prp = pffindtype(dom, type);
1.140 dyoung 540: if (prp == NULL) {
1.120 ginsbach 541: /* no support for domain */
542: if (pffinddomain(dom) == 0)
1.140 dyoung 543: return EAFNOSUPPORT;
1.120 ginsbach 544: /* no support for socket type */
545: if (proto == 0 && type != 0)
1.140 dyoung 546: return EPROTOTYPE;
547: return EPROTONOSUPPORT;
1.120 ginsbach 548: }
1.140 dyoung 549: if (prp->pr_usrreq == NULL)
550: return EPROTONOSUPPORT;
1.1 cgd 551: if (prp->pr_type != type)
1.140 dyoung 552: return EPROTOTYPE;
1.160 ad 553:
554: so = soget(true);
1.1 cgd 555: so->so_type = type;
556: so->so_proto = prp;
1.33 matt 557: so->so_send = sosend;
558: so->so_receive = soreceive;
1.78 matt 559: #ifdef MBUFTRACE
560: so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
561: so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
562: so->so_mowner = &prp->pr_domain->dom_mowner;
563: #endif
1.191 elad 564: /* so->so_cred = kauth_cred_dup(l->l_cred); */
1.138 rmind 565: uid = kauth_cred_geteuid(l->l_cred);
1.115 yamt 566: so->so_uidinfo = uid_find(uid);
1.168 yamt 567: so->so_egid = kauth_cred_getegid(l->l_cred);
568: so->so_cpid = l->l_proc->p_pid;
1.160 ad 569: if (lockso != NULL) {
570: /* Caller wants us to share a lock. */
571: lock = lockso->so_lock;
572: so->so_lock = lock;
573: mutex_obj_hold(lock);
574: mutex_enter(lock);
575: } else {
576: /* Lock assigned and taken during PRU_ATTACH. */
577: }
1.140 dyoung 578: error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
579: (struct mbuf *)(long)proto, NULL, l);
1.160 ad 580: KASSERT(solocked(so));
1.140 dyoung 581: if (error != 0) {
1.1 cgd 582: so->so_state |= SS_NOFDREF;
583: sofree(so);
1.140 dyoung 584: return error;
1.1 cgd 585: }
1.160 ad 586: sounlock(so);
1.1 cgd 587: *aso = so;
1.140 dyoung 588: return 0;
1.1 cgd 589: }
590:
1.142 dyoung 591: /* On success, write file descriptor to fdout and return zero. On
592: * failure, return non-zero; *fdout will be undefined.
593: */
594: int
595: fsocreate(int domain, struct socket **sop, int type, int protocol,
596: struct lwp *l, int *fdout)
597: {
598: struct socket *so;
599: struct file *fp;
600: int fd, error;
601:
1.155 ad 602: if ((error = fd_allocfile(&fp, &fd)) != 0)
1.142 dyoung 603: return (error);
604: fp->f_flag = FREAD|FWRITE;
605: fp->f_type = DTYPE_SOCKET;
606: fp->f_ops = &socketops;
1.160 ad 607: error = socreate(domain, &so, type, protocol, l, NULL);
1.142 dyoung 608: if (error != 0) {
1.155 ad 609: fd_abort(curproc, fp, fd);
1.142 dyoung 610: } else {
611: if (sop != NULL)
612: *sop = so;
613: fp->f_data = so;
1.155 ad 614: fd_affix(curproc, fp, fd);
1.142 dyoung 615: *fdout = fd;
616: }
617: return error;
618: }
619:
1.3 andrew 620: int
1.190 dyoung 621: sofamily(const struct socket *so)
622: {
623: const struct protosw *pr;
624: const struct domain *dom;
625:
626: if ((pr = so->so_proto) == NULL)
627: return AF_UNSPEC;
628: if ((dom = pr->pr_domain) == NULL)
629: return AF_UNSPEC;
630: return dom->dom_family;
631: }
632:
633: int
1.114 christos 634: sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 635: {
1.160 ad 636: int error;
1.1 cgd 637:
1.160 ad 638: solock(so);
1.140 dyoung 639: error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
1.160 ad 640: sounlock(so);
1.140 dyoung 641: return error;
1.1 cgd 642: }
643:
1.3 andrew 644: int
1.150 elad 645: solisten(struct socket *so, int backlog, struct lwp *l)
1.1 cgd 646: {
1.160 ad 647: int error;
1.1 cgd 648:
1.160 ad 649: solock(so);
1.158 ad 650: if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1.163 ad 651: SS_ISDISCONNECTING)) != 0) {
652: sounlock(so);
1.158 ad 653: return (EOPNOTSUPP);
1.163 ad 654: }
1.140 dyoung 655: error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
1.150 elad 656: NULL, NULL, l);
1.140 dyoung 657: if (error != 0) {
1.160 ad 658: sounlock(so);
1.140 dyoung 659: return error;
1.1 cgd 660: }
1.63 matt 661: if (TAILQ_EMPTY(&so->so_q))
1.1 cgd 662: so->so_options |= SO_ACCEPTCONN;
663: if (backlog < 0)
664: backlog = 0;
1.49 jonathan 665: so->so_qlimit = min(backlog, somaxconn);
1.160 ad 666: sounlock(so);
1.140 dyoung 667: return 0;
1.1 cgd 668: }
669:
1.21 christos 670: void
1.54 lukem 671: sofree(struct socket *so)
1.1 cgd 672: {
1.161 ad 673: u_int refs;
1.1 cgd 674:
1.160 ad 675: KASSERT(solocked(so));
676:
677: if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
678: sounlock(so);
1.1 cgd 679: return;
1.160 ad 680: }
1.43 mycroft 681: if (so->so_head) {
682: /*
683: * We must not decommission a socket that's on the accept(2)
684: * queue. If we do, then accept(2) may hang after select(2)
685: * indicated that the listening socket was ready.
686: */
1.160 ad 687: if (!soqremque(so, 0)) {
688: sounlock(so);
1.43 mycroft 689: return;
1.160 ad 690: }
1.43 mycroft 691: }
1.98 christos 692: if (so->so_rcv.sb_hiwat)
1.110 christos 693: (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
1.98 christos 694: RLIM_INFINITY);
695: if (so->so_snd.sb_hiwat)
1.110 christos 696: (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
1.98 christos 697: RLIM_INFINITY);
698: sbrelease(&so->so_snd, so);
1.160 ad 699: KASSERT(!cv_has_waiters(&so->so_cv));
700: KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
701: KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
1.1 cgd 702: sorflush(so);
1.161 ad 703: refs = so->so_aborting; /* XXX */
1.177 ad 704: /* Remove acccept filter if one is present. */
1.170 tls 705: if (so->so_accf != NULL)
1.177 ad 706: (void)accept_filt_clear(so);
1.191 elad 707: /* kauth_cred_free(so->so_cred); */
1.160 ad 708: sounlock(so);
1.161 ad 709: if (refs == 0) /* XXX */
710: soput(so);
1.1 cgd 711: }
712:
713: /*
714: * Close a socket on last file table reference removal.
715: * Initiate disconnect if connected.
716: * Free socket when disconnect complete.
717: */
1.3 andrew 718: int
1.54 lukem 719: soclose(struct socket *so)
1.1 cgd 720: {
1.54 lukem 721: struct socket *so2;
1.160 ad 722: int error;
723: int error2;
1.1 cgd 724:
1.54 lukem 725: error = 0;
1.160 ad 726: solock(so);
1.1 cgd 727: if (so->so_options & SO_ACCEPTCONN) {
1.172 ad 728: for (;;) {
729: if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
1.160 ad 730: KASSERT(solocked2(so, so2));
731: (void) soqremque(so2, 0);
732: /* soabort drops the lock. */
733: (void) soabort(so2);
734: solock(so);
1.172 ad 735: continue;
1.160 ad 736: }
1.172 ad 737: if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
1.160 ad 738: KASSERT(solocked2(so, so2));
739: (void) soqremque(so2, 1);
740: /* soabort drops the lock. */
741: (void) soabort(so2);
742: solock(so);
1.172 ad 743: continue;
1.160 ad 744: }
1.172 ad 745: break;
746: }
1.1 cgd 747: }
748: if (so->so_pcb == 0)
749: goto discard;
750: if (so->so_state & SS_ISCONNECTED) {
751: if ((so->so_state & SS_ISDISCONNECTING) == 0) {
752: error = sodisconnect(so);
753: if (error)
754: goto drop;
755: }
756: if (so->so_options & SO_LINGER) {
1.151 ad 757: if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio)
1.1 cgd 758: goto drop;
1.21 christos 759: while (so->so_state & SS_ISCONNECTED) {
1.185 yamt 760: error = sowait(so, true, so->so_linger * hz);
1.21 christos 761: if (error)
1.1 cgd 762: break;
1.21 christos 763: }
1.1 cgd 764: }
765: }
1.54 lukem 766: drop:
1.1 cgd 767: if (so->so_pcb) {
1.160 ad 768: error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
1.140 dyoung 769: NULL, NULL, NULL, NULL);
1.1 cgd 770: if (error == 0)
771: error = error2;
772: }
1.54 lukem 773: discard:
1.1 cgd 774: if (so->so_state & SS_NOFDREF)
775: panic("soclose: NOFDREF");
776: so->so_state |= SS_NOFDREF;
777: sofree(so);
778: return (error);
779: }
780:
781: /*
1.160 ad 782: * Must be called with the socket locked.. Will return with it unlocked.
1.1 cgd 783: */
1.3 andrew 784: int
1.54 lukem 785: soabort(struct socket *so)
1.1 cgd 786: {
1.161 ad 787: u_int refs;
1.139 yamt 788: int error;
1.160 ad 789:
790: KASSERT(solocked(so));
791: KASSERT(so->so_head == NULL);
1.1 cgd 792:
1.161 ad 793: so->so_aborting++; /* XXX */
1.140 dyoung 794: error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
795: NULL, NULL, NULL);
1.161 ad 796: refs = --so->so_aborting; /* XXX */
1.164 drochner 797: if (error || (refs == 0)) {
1.139 yamt 798: sofree(so);
1.160 ad 799: } else {
800: sounlock(so);
1.139 yamt 801: }
802: return error;
1.1 cgd 803: }
804:
1.3 andrew 805: int
1.54 lukem 806: soaccept(struct socket *so, struct mbuf *nam)
1.1 cgd 807: {
1.160 ad 808: int error;
809:
810: KASSERT(solocked(so));
1.1 cgd 811:
1.54 lukem 812: error = 0;
1.1 cgd 813: if ((so->so_state & SS_NOFDREF) == 0)
814: panic("soaccept: !NOFDREF");
815: so->so_state &= ~SS_NOFDREF;
1.55 thorpej 816: if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
817: (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
1.41 mycroft 818: error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
1.140 dyoung 819: NULL, nam, NULL, NULL);
1.41 mycroft 820: else
1.53 itojun 821: error = ECONNABORTED;
1.52 itojun 822:
1.1 cgd 823: return (error);
824: }
825:
1.3 andrew 826: int
1.114 christos 827: soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 828: {
1.160 ad 829: int error;
830:
831: KASSERT(solocked(so));
1.1 cgd 832:
833: if (so->so_options & SO_ACCEPTCONN)
834: return (EOPNOTSUPP);
835: /*
836: * If protocol is connection-based, can only connect once.
837: * Otherwise, if connected, try to disconnect first.
838: * This allows user to disconnect by connecting to, e.g.,
839: * a null address.
840: */
841: if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
842: ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
843: (error = sodisconnect(so))))
844: error = EISCONN;
845: else
846: error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
1.140 dyoung 847: NULL, nam, NULL, l);
1.1 cgd 848: return (error);
849: }
850:
1.3 andrew 851: int
1.54 lukem 852: soconnect2(struct socket *so1, struct socket *so2)
1.1 cgd 853: {
1.160 ad 854: int error;
855:
856: KASSERT(solocked2(so1, so2));
1.1 cgd 857:
1.22 mycroft 858: error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
1.140 dyoung 859: NULL, (struct mbuf *)so2, NULL, NULL);
1.1 cgd 860: return (error);
861: }
862:
1.3 andrew 863: int
1.54 lukem 864: sodisconnect(struct socket *so)
1.1 cgd 865: {
1.160 ad 866: int error;
867:
868: KASSERT(solocked(so));
1.1 cgd 869:
870: if ((so->so_state & SS_ISCONNECTED) == 0) {
871: error = ENOTCONN;
1.160 ad 872: } else if (so->so_state & SS_ISDISCONNECTING) {
1.1 cgd 873: error = EALREADY;
1.160 ad 874: } else {
875: error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
876: NULL, NULL, NULL, NULL);
1.1 cgd 877: }
1.117 yamt 878: sodopendfree();
1.1 cgd 879: return (error);
880: }
881:
1.15 mycroft 882: #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1.1 cgd 883: /*
884: * Send on a socket.
885: * If send must go all at once and message is larger than
886: * send buffering, then hard error.
887: * Lock against other senders.
888: * If must go all at once and not enough room now, then
889: * inform user that this would block and do nothing.
890: * Otherwise, if nonblocking, send as much as possible.
891: * The data to be sent is described by "uio" if nonzero,
892: * otherwise by the mbuf chain "top" (which must be null
893: * if uio is not). Data provided in mbuf chain must be small
894: * enough to send all at once.
895: *
896: * Returns nonzero on error, timeout or signal; callers
897: * must check for short counts if EINTR/ERESTART are returned.
898: * Data and control buffers are freed on return.
899: */
1.3 andrew 900: int
1.54 lukem 901: sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
1.114 christos 902: struct mbuf *control, int flags, struct lwp *l)
1.1 cgd 903: {
1.54 lukem 904: struct mbuf **mp, *m;
1.114 christos 905: struct proc *p;
1.58 jdolecek 906: long space, len, resid, clen, mlen;
907: int error, s, dontroute, atomic;
1.54 lukem 908:
1.114 christos 909: p = l->l_proc;
1.117 yamt 910: sodopendfree();
1.160 ad 911: clen = 0;
1.64 thorpej 912:
1.160 ad 913: /*
914: * solock() provides atomicity of access. splsoftnet() prevents
915: * protocol processing soft interrupts from interrupting us and
916: * blocking (expensive).
917: */
918: s = splsoftnet();
919: solock(so);
1.54 lukem 920: atomic = sosendallatonce(so) || top;
1.1 cgd 921: if (uio)
922: resid = uio->uio_resid;
923: else
924: resid = top->m_pkthdr.len;
1.7 cgd 925: /*
926: * In theory resid should be unsigned.
927: * However, space must be signed, as it might be less than 0
928: * if we over-committed, and we must use a signed comparison
929: * of space and resid. On the other hand, a negative resid
930: * causes us to loop sending 0-length segments to the protocol.
931: */
1.29 mycroft 932: if (resid < 0) {
933: error = EINVAL;
934: goto out;
935: }
1.1 cgd 936: dontroute =
937: (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
938: (so->so_proto->pr_flags & PR_ATOMIC);
1.165 christos 939: l->l_ru.ru_msgsnd++;
1.1 cgd 940: if (control)
941: clen = control->m_len;
1.54 lukem 942: restart:
1.21 christos 943: if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
1.1 cgd 944: goto out;
945: do {
1.160 ad 946: if (so->so_state & SS_CANTSENDMORE) {
947: error = EPIPE;
948: goto release;
949: }
1.48 thorpej 950: if (so->so_error) {
951: error = so->so_error;
952: so->so_error = 0;
953: goto release;
954: }
1.1 cgd 955: if ((so->so_state & SS_ISCONNECTED) == 0) {
956: if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
957: if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1.160 ad 958: !(resid == 0 && clen != 0)) {
959: error = ENOTCONN;
960: goto release;
961: }
962: } else if (addr == 0) {
963: error = EDESTADDRREQ;
964: goto release;
965: }
1.1 cgd 966: }
967: space = sbspace(&so->so_snd);
968: if (flags & MSG_OOB)
969: space += 1024;
1.21 christos 970: if ((atomic && resid > so->so_snd.sb_hiwat) ||
1.160 ad 971: clen > so->so_snd.sb_hiwat) {
972: error = EMSGSIZE;
973: goto release;
974: }
1.96 mycroft 975: if (space < resid + clen &&
1.1 cgd 976: (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1.160 ad 977: if (so->so_nbio) {
978: error = EWOULDBLOCK;
979: goto release;
980: }
1.1 cgd 981: sbunlock(&so->so_snd);
982: error = sbwait(&so->so_snd);
983: if (error)
984: goto out;
985: goto restart;
986: }
987: mp = ⊤
988: space -= clen;
989: do {
1.45 tv 990: if (uio == NULL) {
991: /*
992: * Data is prepackaged in "top".
993: */
994: resid = 0;
995: if (flags & MSG_EOR)
996: top->m_flags |= M_EOR;
997: } else do {
1.160 ad 998: sounlock(so);
999: splx(s);
1.144 dyoung 1000: if (top == NULL) {
1.78 matt 1001: m = m_gethdr(M_WAIT, MT_DATA);
1.45 tv 1002: mlen = MHLEN;
1003: m->m_pkthdr.len = 0;
1.140 dyoung 1004: m->m_pkthdr.rcvif = NULL;
1.45 tv 1005: } else {
1.78 matt 1006: m = m_get(M_WAIT, MT_DATA);
1.45 tv 1007: mlen = MLEN;
1008: }
1.78 matt 1009: MCLAIM(m, so->so_snd.sb_mowner);
1.121 yamt 1010: if (sock_loan_thresh >= 0 &&
1011: uio->uio_iov->iov_len >= sock_loan_thresh &&
1012: space >= sock_loan_thresh &&
1.64 thorpej 1013: (len = sosend_loan(so, uio, m,
1014: space)) != 0) {
1015: SOSEND_COUNTER_INCR(&sosend_loan_big);
1016: space -= len;
1017: goto have_data;
1018: }
1.45 tv 1019: if (resid >= MINCLSIZE && space >= MCLBYTES) {
1.64 thorpej 1020: SOSEND_COUNTER_INCR(&sosend_copy_big);
1.78 matt 1021: m_clget(m, M_WAIT);
1.45 tv 1022: if ((m->m_flags & M_EXT) == 0)
1023: goto nopages;
1024: mlen = MCLBYTES;
1025: if (atomic && top == 0) {
1.58 jdolecek 1026: len = lmin(MCLBYTES - max_hdr,
1.54 lukem 1027: resid);
1.45 tv 1028: m->m_data += max_hdr;
1029: } else
1.58 jdolecek 1030: len = lmin(MCLBYTES, resid);
1.45 tv 1031: space -= len;
1032: } else {
1.64 thorpej 1033: nopages:
1034: SOSEND_COUNTER_INCR(&sosend_copy_small);
1.58 jdolecek 1035: len = lmin(lmin(mlen, resid), space);
1.45 tv 1036: space -= len;
1037: /*
1038: * For datagram protocols, leave room
1039: * for protocol headers in first mbuf.
1040: */
1041: if (atomic && top == 0 && len < mlen)
1042: MH_ALIGN(m, len);
1043: }
1.144 dyoung 1044: error = uiomove(mtod(m, void *), (int)len, uio);
1.64 thorpej 1045: have_data:
1.45 tv 1046: resid = uio->uio_resid;
1047: m->m_len = len;
1048: *mp = m;
1049: top->m_pkthdr.len += len;
1.160 ad 1050: s = splsoftnet();
1051: solock(so);
1.144 dyoung 1052: if (error != 0)
1.45 tv 1053: goto release;
1054: mp = &m->m_next;
1055: if (resid <= 0) {
1056: if (flags & MSG_EOR)
1057: top->m_flags |= M_EOR;
1058: break;
1059: }
1060: } while (space > 0 && atomic);
1.108 perry 1061:
1.160 ad 1062: if (so->so_state & SS_CANTSENDMORE) {
1063: error = EPIPE;
1064: goto release;
1065: }
1.45 tv 1066: if (dontroute)
1067: so->so_options |= SO_DONTROUTE;
1068: if (resid > 0)
1069: so->so_state |= SS_MORETOCOME;
1.46 sommerfe 1070: error = (*so->so_proto->pr_usrreq)(so,
1071: (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1.160 ad 1072: top, addr, control, curlwp);
1.45 tv 1073: if (dontroute)
1074: so->so_options &= ~SO_DONTROUTE;
1075: if (resid > 0)
1076: so->so_state &= ~SS_MORETOCOME;
1077: clen = 0;
1.144 dyoung 1078: control = NULL;
1079: top = NULL;
1.45 tv 1080: mp = ⊤
1.144 dyoung 1081: if (error != 0)
1.1 cgd 1082: goto release;
1083: } while (resid && space > 0);
1084: } while (resid);
1085:
1.54 lukem 1086: release:
1.1 cgd 1087: sbunlock(&so->so_snd);
1.54 lukem 1088: out:
1.160 ad 1089: sounlock(so);
1090: splx(s);
1.1 cgd 1091: if (top)
1092: m_freem(top);
1093: if (control)
1094: m_freem(control);
1095: return (error);
1096: }
1097:
1098: /*
1.159 ad 1099: * Following replacement or removal of the first mbuf on the first
1100: * mbuf chain of a socket buffer, push necessary state changes back
1101: * into the socket buffer so that other consumers see the values
1102: * consistently. 'nextrecord' is the callers locally stored value of
1103: * the original value of sb->sb_mb->m_nextpkt which must be restored
1104: * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL.
1105: */
1106: static void
1107: sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1108: {
1109:
1.160 ad 1110: KASSERT(solocked(sb->sb_so));
1111:
1.159 ad 1112: /*
1113: * First, update for the new value of nextrecord. If necessary,
1114: * make it the first record.
1115: */
1116: if (sb->sb_mb != NULL)
1117: sb->sb_mb->m_nextpkt = nextrecord;
1118: else
1119: sb->sb_mb = nextrecord;
1120:
1121: /*
1122: * Now update any dependent socket buffer fields to reflect
1123: * the new state. This is an inline of SB_EMPTY_FIXUP, with
1124: * the addition of a second clause that takes care of the
1125: * case where sb_mb has been updated, but remains the last
1126: * record.
1127: */
1128: if (sb->sb_mb == NULL) {
1129: sb->sb_mbtail = NULL;
1130: sb->sb_lastrecord = NULL;
1131: } else if (sb->sb_mb->m_nextpkt == NULL)
1132: sb->sb_lastrecord = sb->sb_mb;
1133: }
1134:
1135: /*
1.1 cgd 1136: * Implement receive operations on a socket.
1137: * We depend on the way that records are added to the sockbuf
1138: * by sbappend*. In particular, each record (mbufs linked through m_next)
1139: * must begin with an address if the protocol so specifies,
1140: * followed by an optional mbuf or mbufs containing ancillary data,
1141: * and then zero or more mbufs of data.
1142: * In order to avoid blocking network interrupts for the entire time here,
1143: * we splx() while doing the actual copy to user space.
1144: * Although the sockbuf is locked, new data may still be appended,
1145: * and thus we must maintain consistency of the sockbuf during that time.
1146: *
1147: * The caller may receive the data as a single mbuf chain by supplying
1148: * an mbuf **mp0 for use in returning the chain. The uio is then used
1149: * only for the count in uio_resid.
1150: */
1.3 andrew 1151: int
1.54 lukem 1152: soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1153: struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1.1 cgd 1154: {
1.116 yamt 1155: struct lwp *l = curlwp;
1.160 ad 1156: struct mbuf *m, **mp, *mt;
1.146 dyoung 1157: int atomic, flags, len, error, s, offset, moff, type, orig_resid;
1.99 matt 1158: const struct protosw *pr;
1.54 lukem 1159: struct mbuf *nextrecord;
1.67 he 1160: int mbuf_removed = 0;
1.146 dyoung 1161: const struct domain *dom;
1.64 thorpej 1162:
1.54 lukem 1163: pr = so->so_proto;
1.146 dyoung 1164: atomic = pr->pr_flags & PR_ATOMIC;
1165: dom = pr->pr_domain;
1.1 cgd 1166: mp = mp0;
1.54 lukem 1167: type = 0;
1168: orig_resid = uio->uio_resid;
1.102 jonathan 1169:
1.144 dyoung 1170: if (paddr != NULL)
1171: *paddr = NULL;
1172: if (controlp != NULL)
1173: *controlp = NULL;
1174: if (flagsp != NULL)
1.1 cgd 1175: flags = *flagsp &~ MSG_EOR;
1176: else
1177: flags = 0;
1.66 enami 1178:
1179: if ((flags & MSG_DONTWAIT) == 0)
1.117 yamt 1180: sodopendfree();
1.66 enami 1181:
1.1 cgd 1182: if (flags & MSG_OOB) {
1183: m = m_get(M_WAIT, MT_DATA);
1.160 ad 1184: solock(so);
1.17 cgd 1185: error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1.140 dyoung 1186: (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1.160 ad 1187: sounlock(so);
1.1 cgd 1188: if (error)
1189: goto bad;
1190: do {
1.134 christos 1191: error = uiomove(mtod(m, void *),
1.1 cgd 1192: (int) min(uio->uio_resid, m->m_len), uio);
1193: m = m_free(m);
1.144 dyoung 1194: } while (uio->uio_resid > 0 && error == 0 && m);
1.54 lukem 1195: bad:
1.144 dyoung 1196: if (m != NULL)
1.1 cgd 1197: m_freem(m);
1.144 dyoung 1198: return error;
1.1 cgd 1199: }
1.144 dyoung 1200: if (mp != NULL)
1.140 dyoung 1201: *mp = NULL;
1.160 ad 1202:
1203: /*
1204: * solock() provides atomicity of access. splsoftnet() prevents
1205: * protocol processing soft interrupts from interrupting us and
1206: * blocking (expensive).
1207: */
1208: s = splsoftnet();
1209: solock(so);
1.1 cgd 1210: if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1.140 dyoung 1211: (*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
1.1 cgd 1212:
1.54 lukem 1213: restart:
1.160 ad 1214: if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1215: sounlock(so);
1216: splx(s);
1.144 dyoung 1217: return error;
1.160 ad 1218: }
1.1 cgd 1219:
1220: m = so->so_rcv.sb_mb;
1221: /*
1222: * If we have less data than requested, block awaiting more
1223: * (subject to any timeout) if:
1.15 mycroft 1224: * 1. the current count is less than the low water mark,
1.1 cgd 1225: * 2. MSG_WAITALL is set, and it is possible to do the entire
1.15 mycroft 1226: * receive operation at once if we block (resid <= hiwat), or
1227: * 3. MSG_DONTWAIT is not set.
1.1 cgd 1228: * If MSG_WAITALL is set but resid is larger than the receive buffer,
1229: * we have to do the receive in sections, and thus risk returning
1230: * a short count if a timeout or signal occurs after we start.
1231: */
1.144 dyoung 1232: if (m == NULL ||
1233: ((flags & MSG_DONTWAIT) == 0 &&
1234: so->so_rcv.sb_cc < uio->uio_resid &&
1235: (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1236: ((flags & MSG_WAITALL) &&
1237: uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1.146 dyoung 1238: m->m_nextpkt == NULL && !atomic)) {
1.1 cgd 1239: #ifdef DIAGNOSTIC
1.144 dyoung 1240: if (m == NULL && so->so_rcv.sb_cc)
1.1 cgd 1241: panic("receive 1");
1242: #endif
1243: if (so->so_error) {
1.144 dyoung 1244: if (m != NULL)
1.15 mycroft 1245: goto dontblock;
1.1 cgd 1246: error = so->so_error;
1247: if ((flags & MSG_PEEK) == 0)
1248: so->so_error = 0;
1249: goto release;
1250: }
1251: if (so->so_state & SS_CANTRCVMORE) {
1.144 dyoung 1252: if (m != NULL)
1.15 mycroft 1253: goto dontblock;
1.1 cgd 1254: else
1255: goto release;
1256: }
1.144 dyoung 1257: for (; m != NULL; m = m->m_next)
1.1 cgd 1258: if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1259: m = so->so_rcv.sb_mb;
1260: goto dontblock;
1261: }
1262: if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1263: (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1264: error = ENOTCONN;
1265: goto release;
1266: }
1267: if (uio->uio_resid == 0)
1268: goto release;
1.151 ad 1269: if (so->so_nbio || (flags & MSG_DONTWAIT)) {
1.1 cgd 1270: error = EWOULDBLOCK;
1271: goto release;
1272: }
1.69 thorpej 1273: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1274: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1.1 cgd 1275: sbunlock(&so->so_rcv);
1276: error = sbwait(&so->so_rcv);
1.160 ad 1277: if (error != 0) {
1278: sounlock(so);
1279: splx(s);
1.144 dyoung 1280: return error;
1.160 ad 1281: }
1.1 cgd 1282: goto restart;
1283: }
1.54 lukem 1284: dontblock:
1.69 thorpej 1285: /*
1286: * On entry here, m points to the first record of the socket buffer.
1.159 ad 1287: * From this point onward, we maintain 'nextrecord' as a cache of the
1288: * pointer to the next record in the socket buffer. We must keep the
1289: * various socket buffer pointers and local stack versions of the
1290: * pointers in sync, pushing out modifications before dropping the
1.160 ad 1291: * socket lock, and re-reading them when picking it up.
1.159 ad 1292: *
1293: * Otherwise, we will race with the network stack appending new data
1294: * or records onto the socket buffer by using inconsistent/stale
1295: * versions of the field, possibly resulting in socket buffer
1296: * corruption.
1297: *
1298: * By holding the high-level sblock(), we prevent simultaneous
1299: * readers from pulling off the front of the socket buffer.
1.69 thorpej 1300: */
1.144 dyoung 1301: if (l != NULL)
1.157 ad 1302: l->l_ru.ru_msgrcv++;
1.69 thorpej 1303: KASSERT(m == so->so_rcv.sb_mb);
1304: SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1305: SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1.1 cgd 1306: nextrecord = m->m_nextpkt;
1307: if (pr->pr_flags & PR_ADDR) {
1308: #ifdef DIAGNOSTIC
1309: if (m->m_type != MT_SONAME)
1310: panic("receive 1a");
1311: #endif
1.3 andrew 1312: orig_resid = 0;
1.1 cgd 1313: if (flags & MSG_PEEK) {
1314: if (paddr)
1315: *paddr = m_copy(m, 0, m->m_len);
1316: m = m->m_next;
1317: } else {
1318: sbfree(&so->so_rcv, m);
1.67 he 1319: mbuf_removed = 1;
1.144 dyoung 1320: if (paddr != NULL) {
1.1 cgd 1321: *paddr = m;
1322: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1323: m->m_next = NULL;
1.1 cgd 1324: m = so->so_rcv.sb_mb;
1325: } else {
1326: MFREE(m, so->so_rcv.sb_mb);
1327: m = so->so_rcv.sb_mb;
1328: }
1.159 ad 1329: sbsync(&so->so_rcv, nextrecord);
1.1 cgd 1330: }
1331: }
1.159 ad 1332:
1333: /*
1334: * Process one or more MT_CONTROL mbufs present before any data mbufs
1335: * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1336: * just copy the data; if !MSG_PEEK, we call into the protocol to
1337: * perform externalization (or freeing if controlp == NULL).
1338: */
1339: if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1340: struct mbuf *cm = NULL, *cmn;
1341: struct mbuf **cme = &cm;
1342:
1343: do {
1344: if (flags & MSG_PEEK) {
1345: if (controlp != NULL) {
1346: *controlp = m_copy(m, 0, m->m_len);
1347: controlp = &(*controlp)->m_next;
1348: }
1349: m = m->m_next;
1350: } else {
1351: sbfree(&so->so_rcv, m);
1.1 cgd 1352: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1353: m->m_next = NULL;
1.159 ad 1354: *cme = m;
1355: cme = &(*cme)->m_next;
1.1 cgd 1356: m = so->so_rcv.sb_mb;
1.159 ad 1357: }
1358: } while (m != NULL && m->m_type == MT_CONTROL);
1359: if ((flags & MSG_PEEK) == 0)
1360: sbsync(&so->so_rcv, nextrecord);
1361: for (; cm != NULL; cm = cmn) {
1362: cmn = cm->m_next;
1363: cm->m_next = NULL;
1364: type = mtod(cm, struct cmsghdr *)->cmsg_type;
1365: if (controlp != NULL) {
1366: if (dom->dom_externalize != NULL &&
1367: type == SCM_RIGHTS) {
1.160 ad 1368: sounlock(so);
1.159 ad 1369: splx(s);
1370: error = (*dom->dom_externalize)(cm, l);
1371: s = splsoftnet();
1.160 ad 1372: solock(so);
1.159 ad 1373: }
1374: *controlp = cm;
1375: while (*controlp != NULL)
1376: controlp = &(*controlp)->m_next;
1.1 cgd 1377: } else {
1.106 itojun 1378: /*
1379: * Dispose of any SCM_RIGHTS message that went
1380: * through the read path rather than recv.
1381: */
1.159 ad 1382: if (dom->dom_dispose != NULL &&
1383: type == SCM_RIGHTS) {
1.160 ad 1384: sounlock(so);
1.159 ad 1385: (*dom->dom_dispose)(cm);
1.160 ad 1386: solock(so);
1.159 ad 1387: }
1388: m_freem(cm);
1.1 cgd 1389: }
1390: }
1.159 ad 1391: if (m != NULL)
1392: nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1393: else
1394: nextrecord = so->so_rcv.sb_mb;
1395: orig_resid = 0;
1.1 cgd 1396: }
1.69 thorpej 1397:
1.159 ad 1398: /* If m is non-NULL, we have some data to read. */
1399: if (__predict_true(m != NULL)) {
1.1 cgd 1400: type = m->m_type;
1401: if (type == MT_OOBDATA)
1402: flags |= MSG_OOB;
1403: }
1.69 thorpej 1404: SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1405: SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1406:
1.1 cgd 1407: moff = 0;
1408: offset = 0;
1.144 dyoung 1409: while (m != NULL && uio->uio_resid > 0 && error == 0) {
1.1 cgd 1410: if (m->m_type == MT_OOBDATA) {
1411: if (type != MT_OOBDATA)
1412: break;
1413: } else if (type == MT_OOBDATA)
1414: break;
1415: #ifdef DIAGNOSTIC
1416: else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1417: panic("receive 3");
1418: #endif
1419: so->so_state &= ~SS_RCVATMARK;
1420: len = uio->uio_resid;
1421: if (so->so_oobmark && len > so->so_oobmark - offset)
1422: len = so->so_oobmark - offset;
1423: if (len > m->m_len - moff)
1424: len = m->m_len - moff;
1425: /*
1426: * If mp is set, just pass back the mbufs.
1427: * Otherwise copy them out via the uio, then free.
1428: * Sockbuf must be consistent here (points to current mbuf,
1429: * it points to next record) when we drop priority;
1430: * we must note any additions to the sockbuf when we
1431: * block interrupts again.
1432: */
1.144 dyoung 1433: if (mp == NULL) {
1.69 thorpej 1434: SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1435: SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1.160 ad 1436: sounlock(so);
1.1 cgd 1437: splx(s);
1.134 christos 1438: error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1.20 mycroft 1439: s = splsoftnet();
1.160 ad 1440: solock(so);
1.144 dyoung 1441: if (error != 0) {
1.67 he 1442: /*
1443: * If any part of the record has been removed
1444: * (such as the MT_SONAME mbuf, which will
1445: * happen when PR_ADDR, and thus also
1446: * PR_ATOMIC, is set), then drop the entire
1447: * record to maintain the atomicity of the
1448: * receive operation.
1449: *
1450: * This avoids a later panic("receive 1a")
1451: * when compiled with DIAGNOSTIC.
1452: */
1.146 dyoung 1453: if (m && mbuf_removed && atomic)
1.67 he 1454: (void) sbdroprecord(&so->so_rcv);
1455:
1.57 jdolecek 1456: goto release;
1.67 he 1457: }
1.1 cgd 1458: } else
1459: uio->uio_resid -= len;
1460: if (len == m->m_len - moff) {
1461: if (m->m_flags & M_EOR)
1462: flags |= MSG_EOR;
1463: if (flags & MSG_PEEK) {
1464: m = m->m_next;
1465: moff = 0;
1466: } else {
1467: nextrecord = m->m_nextpkt;
1468: sbfree(&so->so_rcv, m);
1469: if (mp) {
1470: *mp = m;
1471: mp = &m->m_next;
1472: so->so_rcv.sb_mb = m = m->m_next;
1.140 dyoung 1473: *mp = NULL;
1.1 cgd 1474: } else {
1475: MFREE(m, so->so_rcv.sb_mb);
1476: m = so->so_rcv.sb_mb;
1477: }
1.69 thorpej 1478: /*
1479: * If m != NULL, we also know that
1480: * so->so_rcv.sb_mb != NULL.
1481: */
1482: KASSERT(so->so_rcv.sb_mb == m);
1483: if (m) {
1.1 cgd 1484: m->m_nextpkt = nextrecord;
1.69 thorpej 1485: if (nextrecord == NULL)
1486: so->so_rcv.sb_lastrecord = m;
1487: } else {
1488: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1489: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1490: }
1491: SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1492: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1.1 cgd 1493: }
1.144 dyoung 1494: } else if (flags & MSG_PEEK)
1495: moff += len;
1496: else {
1.160 ad 1497: if (mp != NULL) {
1498: mt = m_copym(m, 0, len, M_NOWAIT);
1499: if (__predict_false(mt == NULL)) {
1500: sounlock(so);
1501: mt = m_copym(m, 0, len, M_WAIT);
1502: solock(so);
1503: }
1504: *mp = mt;
1505: }
1.144 dyoung 1506: m->m_data += len;
1507: m->m_len -= len;
1508: so->so_rcv.sb_cc -= len;
1.1 cgd 1509: }
1510: if (so->so_oobmark) {
1511: if ((flags & MSG_PEEK) == 0) {
1512: so->so_oobmark -= len;
1513: if (so->so_oobmark == 0) {
1514: so->so_state |= SS_RCVATMARK;
1515: break;
1516: }
1.7 cgd 1517: } else {
1.1 cgd 1518: offset += len;
1.7 cgd 1519: if (offset == so->so_oobmark)
1520: break;
1521: }
1.1 cgd 1522: }
1523: if (flags & MSG_EOR)
1524: break;
1525: /*
1526: * If the MSG_WAITALL flag is set (for non-atomic socket),
1527: * we must not quit until "uio->uio_resid == 0" or an error
1528: * termination. If a signal/timeout occurs, return
1529: * with a short count but without error.
1530: * Keep sockbuf locked against other readers.
1531: */
1.144 dyoung 1532: while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1.3 andrew 1533: !sosendallatonce(so) && !nextrecord) {
1.1 cgd 1534: if (so->so_error || so->so_state & SS_CANTRCVMORE)
1535: break;
1.68 matt 1536: /*
1537: * If we are peeking and the socket receive buffer is
1538: * full, stop since we can't get more data to peek at.
1539: */
1540: if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1541: break;
1542: /*
1543: * If we've drained the socket buffer, tell the
1544: * protocol in case it needs to do something to
1545: * get it filled again.
1546: */
1547: if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1548: (*pr->pr_usrreq)(so, PRU_RCVD,
1.140 dyoung 1549: NULL, (struct mbuf *)(long)flags, NULL, l);
1.69 thorpej 1550: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1551: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1.1 cgd 1552: error = sbwait(&so->so_rcv);
1.144 dyoung 1553: if (error != 0) {
1.1 cgd 1554: sbunlock(&so->so_rcv);
1.160 ad 1555: sounlock(so);
1.1 cgd 1556: splx(s);
1.144 dyoung 1557: return 0;
1.1 cgd 1558: }
1.21 christos 1559: if ((m = so->so_rcv.sb_mb) != NULL)
1.1 cgd 1560: nextrecord = m->m_nextpkt;
1561: }
1562: }
1.3 andrew 1563:
1.146 dyoung 1564: if (m && atomic) {
1.3 andrew 1565: flags |= MSG_TRUNC;
1566: if ((flags & MSG_PEEK) == 0)
1567: (void) sbdroprecord(&so->so_rcv);
1568: }
1.1 cgd 1569: if ((flags & MSG_PEEK) == 0) {
1.144 dyoung 1570: if (m == NULL) {
1.69 thorpej 1571: /*
1.70 thorpej 1572: * First part is an inline SB_EMPTY_FIXUP(). Second
1.69 thorpej 1573: * part makes sure sb_lastrecord is up-to-date if
1574: * there is still data in the socket buffer.
1575: */
1.1 cgd 1576: so->so_rcv.sb_mb = nextrecord;
1.69 thorpej 1577: if (so->so_rcv.sb_mb == NULL) {
1578: so->so_rcv.sb_mbtail = NULL;
1579: so->so_rcv.sb_lastrecord = NULL;
1580: } else if (nextrecord->m_nextpkt == NULL)
1581: so->so_rcv.sb_lastrecord = nextrecord;
1582: }
1583: SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1584: SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1.1 cgd 1585: if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1.140 dyoung 1586: (*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1587: (struct mbuf *)(long)flags, NULL, l);
1.1 cgd 1588: }
1.3 andrew 1589: if (orig_resid == uio->uio_resid && orig_resid &&
1590: (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1591: sbunlock(&so->so_rcv);
1592: goto restart;
1593: }
1.108 perry 1594:
1.144 dyoung 1595: if (flagsp != NULL)
1.1 cgd 1596: *flagsp |= flags;
1.54 lukem 1597: release:
1.1 cgd 1598: sbunlock(&so->so_rcv);
1.160 ad 1599: sounlock(so);
1.1 cgd 1600: splx(s);
1.144 dyoung 1601: return error;
1.1 cgd 1602: }
1603:
1.14 mycroft 1604: int
1.54 lukem 1605: soshutdown(struct socket *so, int how)
1.1 cgd 1606: {
1.99 matt 1607: const struct protosw *pr;
1.160 ad 1608: int error;
1609:
1610: KASSERT(solocked(so));
1.34 kleink 1611:
1.54 lukem 1612: pr = so->so_proto;
1.34 kleink 1613: if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1614: return (EINVAL);
1.1 cgd 1615:
1.160 ad 1616: if (how == SHUT_RD || how == SHUT_RDWR) {
1.1 cgd 1617: sorflush(so);
1.160 ad 1618: error = 0;
1619: }
1.34 kleink 1620: if (how == SHUT_WR || how == SHUT_RDWR)
1.160 ad 1621: error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1.140 dyoung 1622: NULL, NULL, NULL);
1.160 ad 1623:
1624: return error;
1.1 cgd 1625: }
1626:
1.188 ad 1627: int
1628: sodrain(struct socket *so)
1629: {
1630: int error;
1631:
1632: solock(so);
1633: so->so_state |= SS_ISDRAINING;
1634: cv_broadcast(&so->so_cv);
1635: error = soshutdown(so, SHUT_RDWR);
1636: sounlock(so);
1637:
1638: return error;
1639: }
1640:
1.14 mycroft 1641: void
1.54 lukem 1642: sorflush(struct socket *so)
1.1 cgd 1643: {
1.54 lukem 1644: struct sockbuf *sb, asb;
1.99 matt 1645: const struct protosw *pr;
1.160 ad 1646:
1647: KASSERT(solocked(so));
1.1 cgd 1648:
1.54 lukem 1649: sb = &so->so_rcv;
1650: pr = so->so_proto;
1.160 ad 1651: socantrcvmore(so);
1.1 cgd 1652: sb->sb_flags |= SB_NOINTR;
1.160 ad 1653: (void )sblock(sb, M_WAITOK);
1.1 cgd 1654: sbunlock(sb);
1655: asb = *sb;
1.86 wrstuden 1656: /*
1657: * Clear most of the sockbuf structure, but leave some of the
1658: * fields valid.
1659: */
1660: memset(&sb->sb_startzero, 0,
1661: sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1.160 ad 1662: if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1663: sounlock(so);
1.1 cgd 1664: (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1.160 ad 1665: solock(so);
1666: }
1.98 christos 1667: sbrelease(&asb, so);
1.1 cgd 1668: }
1669:
1.171 plunky 1670: /*
1671: * internal set SOL_SOCKET options
1672: */
1.142 dyoung 1673: static int
1.171 plunky 1674: sosetopt1(struct socket *so, const struct sockopt *sopt)
1.1 cgd 1675: {
1.182 christos 1676: int error = EINVAL, optval, opt;
1.171 plunky 1677: struct linger l;
1678: struct timeval tv;
1.142 dyoung 1679:
1.179 christos 1680: switch ((opt = sopt->sopt_name)) {
1.142 dyoung 1681:
1.170 tls 1682: case SO_ACCEPTFILTER:
1.177 ad 1683: error = accept_filt_setopt(so, sopt);
1684: KASSERT(solocked(so));
1.170 tls 1685: break;
1686:
1.171 plunky 1687: case SO_LINGER:
1688: error = sockopt_get(sopt, &l, sizeof(l));
1.177 ad 1689: solock(so);
1.171 plunky 1690: if (error)
1.177 ad 1691: break;
1.171 plunky 1692: if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1.177 ad 1693: l.l_linger > (INT_MAX / hz)) {
1694: error = EDOM;
1695: break;
1696: }
1.171 plunky 1697: so->so_linger = l.l_linger;
1698: if (l.l_onoff)
1699: so->so_options |= SO_LINGER;
1700: else
1701: so->so_options &= ~SO_LINGER;
1.177 ad 1702: break;
1.1 cgd 1703:
1.142 dyoung 1704: case SO_DEBUG:
1705: case SO_KEEPALIVE:
1706: case SO_DONTROUTE:
1707: case SO_USELOOPBACK:
1708: case SO_BROADCAST:
1709: case SO_REUSEADDR:
1710: case SO_REUSEPORT:
1711: case SO_OOBINLINE:
1712: case SO_TIMESTAMP:
1.184 christos 1713: #ifdef SO_OTIMESTAMP
1714: case SO_OTIMESTAMP:
1715: #endif
1.171 plunky 1716: error = sockopt_getint(sopt, &optval);
1.177 ad 1717: solock(so);
1.171 plunky 1718: if (error)
1.177 ad 1719: break;
1.171 plunky 1720: if (optval)
1.179 christos 1721: so->so_options |= opt;
1.142 dyoung 1722: else
1.179 christos 1723: so->so_options &= ~opt;
1.142 dyoung 1724: break;
1725:
1726: case SO_SNDBUF:
1727: case SO_RCVBUF:
1728: case SO_SNDLOWAT:
1729: case SO_RCVLOWAT:
1.171 plunky 1730: error = sockopt_getint(sopt, &optval);
1.177 ad 1731: solock(so);
1.171 plunky 1732: if (error)
1.177 ad 1733: break;
1.1 cgd 1734:
1.142 dyoung 1735: /*
1736: * Values < 1 make no sense for any of these
1737: * options, so disallow them.
1738: */
1.177 ad 1739: if (optval < 1) {
1740: error = EINVAL;
1741: break;
1742: }
1.1 cgd 1743:
1.179 christos 1744: switch (opt) {
1.171 plunky 1745: case SO_SNDBUF:
1.177 ad 1746: if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1747: error = ENOBUFS;
1748: break;
1749: }
1.171 plunky 1750: so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1751: break;
1.1 cgd 1752:
1753: case SO_RCVBUF:
1.177 ad 1754: if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1755: error = ENOBUFS;
1756: break;
1757: }
1.171 plunky 1758: so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1.142 dyoung 1759: break;
1760:
1761: /*
1762: * Make sure the low-water is never greater than
1763: * the high-water.
1764: */
1.1 cgd 1765: case SO_SNDLOWAT:
1.171 plunky 1766: if (optval > so->so_snd.sb_hiwat)
1767: optval = so->so_snd.sb_hiwat;
1768:
1769: so->so_snd.sb_lowat = optval;
1.142 dyoung 1770: break;
1.171 plunky 1771:
1.1 cgd 1772: case SO_RCVLOWAT:
1.171 plunky 1773: if (optval > so->so_rcv.sb_hiwat)
1774: optval = so->so_rcv.sb_hiwat;
1775:
1776: so->so_rcv.sb_lowat = optval;
1.142 dyoung 1777: break;
1778: }
1779: break;
1.28 thorpej 1780:
1.179 christos 1781: #ifdef COMPAT_50
1782: case SO_OSNDTIMEO:
1783: case SO_ORCVTIMEO: {
1784: struct timeval50 otv;
1785: error = sockopt_get(sopt, &otv, sizeof(otv));
1.186 pooka 1786: if (error) {
1787: solock(so);
1.183 christos 1788: break;
1.186 pooka 1789: }
1.179 christos 1790: timeval50_to_timeval(&otv, &tv);
1791: opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1.182 christos 1792: error = 0;
1.179 christos 1793: /*FALLTHROUGH*/
1794: }
1795: #endif /* COMPAT_50 */
1796:
1.142 dyoung 1797: case SO_SNDTIMEO:
1798: case SO_RCVTIMEO:
1.182 christos 1799: if (error)
1.179 christos 1800: error = sockopt_get(sopt, &tv, sizeof(tv));
1.177 ad 1801: solock(so);
1.171 plunky 1802: if (error)
1.177 ad 1803: break;
1.171 plunky 1804:
1.177 ad 1805: if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1806: error = EDOM;
1807: break;
1808: }
1.28 thorpej 1809:
1.171 plunky 1810: optval = tv.tv_sec * hz + tv.tv_usec / tick;
1811: if (optval == 0 && tv.tv_usec != 0)
1812: optval = 1;
1.28 thorpej 1813:
1.179 christos 1814: switch (opt) {
1.142 dyoung 1815: case SO_SNDTIMEO:
1.171 plunky 1816: so->so_snd.sb_timeo = optval;
1.1 cgd 1817: break;
1818: case SO_RCVTIMEO:
1.171 plunky 1819: so->so_rcv.sb_timeo = optval;
1.142 dyoung 1820: break;
1821: }
1822: break;
1.1 cgd 1823:
1.142 dyoung 1824: default:
1.177 ad 1825: solock(so);
1826: error = ENOPROTOOPT;
1827: break;
1.142 dyoung 1828: }
1.177 ad 1829: KASSERT(solocked(so));
1830: return error;
1.142 dyoung 1831: }
1.1 cgd 1832:
1.142 dyoung 1833: int
1.171 plunky 1834: sosetopt(struct socket *so, struct sockopt *sopt)
1.142 dyoung 1835: {
1836: int error, prerr;
1.1 cgd 1837:
1.177 ad 1838: if (sopt->sopt_level == SOL_SOCKET) {
1.171 plunky 1839: error = sosetopt1(so, sopt);
1.177 ad 1840: KASSERT(solocked(so));
1841: } else {
1.142 dyoung 1842: error = ENOPROTOOPT;
1.177 ad 1843: solock(so);
1844: }
1.1 cgd 1845:
1.142 dyoung 1846: if ((error == 0 || error == ENOPROTOOPT) &&
1847: so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1848: /* give the protocol stack a shot */
1.171 plunky 1849: prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1.142 dyoung 1850: if (prerr == 0)
1851: error = 0;
1852: else if (prerr != ENOPROTOOPT)
1853: error = prerr;
1.171 plunky 1854: }
1.160 ad 1855: sounlock(so);
1.142 dyoung 1856: return error;
1.1 cgd 1857: }
1858:
1.171 plunky 1859: /*
1860: * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1861: */
1862: int
1863: so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1864: const void *val, size_t valsize)
1865: {
1866: struct sockopt sopt;
1867: int error;
1868:
1869: KASSERT(valsize == 0 || val != NULL);
1870:
1871: sockopt_init(&sopt, level, name, valsize);
1872: sockopt_set(&sopt, val, valsize);
1873:
1874: error = sosetopt(so, &sopt);
1875:
1876: sockopt_destroy(&sopt);
1877:
1878: return error;
1879: }
1880:
1881: /*
1882: * internal get SOL_SOCKET options
1883: */
1884: static int
1885: sogetopt1(struct socket *so, struct sockopt *sopt)
1886: {
1.179 christos 1887: int error, optval, opt;
1.171 plunky 1888: struct linger l;
1889: struct timeval tv;
1890:
1.179 christos 1891: switch ((opt = sopt->sopt_name)) {
1.171 plunky 1892:
1893: case SO_ACCEPTFILTER:
1.177 ad 1894: error = accept_filt_getopt(so, sopt);
1.171 plunky 1895: break;
1896:
1897: case SO_LINGER:
1898: l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1899: l.l_linger = so->so_linger;
1900:
1901: error = sockopt_set(sopt, &l, sizeof(l));
1902: break;
1903:
1904: case SO_USELOOPBACK:
1905: case SO_DONTROUTE:
1906: case SO_DEBUG:
1907: case SO_KEEPALIVE:
1908: case SO_REUSEADDR:
1909: case SO_REUSEPORT:
1910: case SO_BROADCAST:
1911: case SO_OOBINLINE:
1912: case SO_TIMESTAMP:
1.184 christos 1913: #ifdef SO_OTIMESTAMP
1914: case SO_OTIMESTAMP:
1915: #endif
1.179 christos 1916: error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1.171 plunky 1917: break;
1918:
1919: case SO_TYPE:
1920: error = sockopt_setint(sopt, so->so_type);
1921: break;
1922:
1923: case SO_ERROR:
1924: error = sockopt_setint(sopt, so->so_error);
1925: so->so_error = 0;
1926: break;
1927:
1928: case SO_SNDBUF:
1929: error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1930: break;
1931:
1932: case SO_RCVBUF:
1933: error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1934: break;
1935:
1936: case SO_SNDLOWAT:
1937: error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1938: break;
1939:
1940: case SO_RCVLOWAT:
1941: error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1942: break;
1943:
1.179 christos 1944: #ifdef COMPAT_50
1945: case SO_OSNDTIMEO:
1946: case SO_ORCVTIMEO: {
1947: struct timeval50 otv;
1948:
1949: optval = (opt == SO_OSNDTIMEO ?
1950: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1951:
1952: otv.tv_sec = optval / hz;
1953: otv.tv_usec = (optval % hz) * tick;
1954:
1955: error = sockopt_set(sopt, &otv, sizeof(otv));
1956: break;
1957: }
1958: #endif /* COMPAT_50 */
1959:
1.171 plunky 1960: case SO_SNDTIMEO:
1961: case SO_RCVTIMEO:
1.179 christos 1962: optval = (opt == SO_SNDTIMEO ?
1.171 plunky 1963: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1964:
1965: tv.tv_sec = optval / hz;
1966: tv.tv_usec = (optval % hz) * tick;
1967:
1968: error = sockopt_set(sopt, &tv, sizeof(tv));
1969: break;
1970:
1971: case SO_OVERFLOWED:
1972: error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1973: break;
1974:
1975: default:
1976: error = ENOPROTOOPT;
1977: break;
1978: }
1979:
1980: return (error);
1981: }
1982:
1.14 mycroft 1983: int
1.171 plunky 1984: sogetopt(struct socket *so, struct sockopt *sopt)
1.1 cgd 1985: {
1.160 ad 1986: int error;
1.1 cgd 1987:
1.160 ad 1988: solock(so);
1.171 plunky 1989: if (sopt->sopt_level != SOL_SOCKET) {
1.1 cgd 1990: if (so->so_proto && so->so_proto->pr_ctloutput) {
1.160 ad 1991: error = ((*so->so_proto->pr_ctloutput)
1.171 plunky 1992: (PRCO_GETOPT, so, sopt));
1.1 cgd 1993: } else
1.160 ad 1994: error = (ENOPROTOOPT);
1.1 cgd 1995: } else {
1.171 plunky 1996: error = sogetopt1(so, sopt);
1997: }
1998: sounlock(so);
1999: return (error);
2000: }
2001:
2002: /*
2003: * alloc sockopt data buffer buffer
2004: * - will be released at destroy
2005: */
1.176 plunky 2006: static int
2007: sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
1.171 plunky 2008: {
2009:
2010: KASSERT(sopt->sopt_size == 0);
2011:
1.176 plunky 2012: if (len > sizeof(sopt->sopt_buf)) {
2013: sopt->sopt_data = kmem_zalloc(len, kmflag);
2014: if (sopt->sopt_data == NULL)
2015: return ENOMEM;
2016: } else
1.171 plunky 2017: sopt->sopt_data = sopt->sopt_buf;
2018:
2019: sopt->sopt_size = len;
1.176 plunky 2020: return 0;
1.171 plunky 2021: }
2022:
2023: /*
2024: * initialise sockopt storage
1.176 plunky 2025: * - MAY sleep during allocation
1.171 plunky 2026: */
2027: void
2028: sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2029: {
1.1 cgd 2030:
1.171 plunky 2031: memset(sopt, 0, sizeof(*sopt));
1.1 cgd 2032:
1.171 plunky 2033: sopt->sopt_level = level;
2034: sopt->sopt_name = name;
1.176 plunky 2035: (void)sockopt_alloc(sopt, size, KM_SLEEP);
1.171 plunky 2036: }
2037:
2038: /*
2039: * destroy sockopt storage
2040: * - will release any held memory references
2041: */
2042: void
2043: sockopt_destroy(struct sockopt *sopt)
2044: {
2045:
2046: if (sopt->sopt_data != sopt->sopt_buf)
1.173 plunky 2047: kmem_free(sopt->sopt_data, sopt->sopt_size);
1.171 plunky 2048:
2049: memset(sopt, 0, sizeof(*sopt));
2050: }
2051:
2052: /*
2053: * set sockopt value
2054: * - value is copied into sockopt
1.176 plunky 2055: * - memory is allocated when necessary, will not sleep
1.171 plunky 2056: */
2057: int
2058: sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2059: {
1.176 plunky 2060: int error;
1.171 plunky 2061:
1.176 plunky 2062: if (sopt->sopt_size == 0) {
2063: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2064: if (error)
2065: return error;
2066: }
1.171 plunky 2067:
2068: KASSERT(sopt->sopt_size == len);
2069: memcpy(sopt->sopt_data, buf, len);
2070: return 0;
2071: }
2072:
2073: /*
2074: * common case of set sockopt integer value
2075: */
2076: int
2077: sockopt_setint(struct sockopt *sopt, int val)
2078: {
2079:
2080: return sockopt_set(sopt, &val, sizeof(int));
2081: }
2082:
2083: /*
2084: * get sockopt value
2085: * - correct size must be given
2086: */
2087: int
2088: sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2089: {
1.170 tls 2090:
1.171 plunky 2091: if (sopt->sopt_size != len)
2092: return EINVAL;
1.1 cgd 2093:
1.171 plunky 2094: memcpy(buf, sopt->sopt_data, len);
2095: return 0;
2096: }
1.1 cgd 2097:
1.171 plunky 2098: /*
2099: * common case of get sockopt integer value
2100: */
2101: int
2102: sockopt_getint(const struct sockopt *sopt, int *valp)
2103: {
1.1 cgd 2104:
1.171 plunky 2105: return sockopt_get(sopt, valp, sizeof(int));
2106: }
1.1 cgd 2107:
1.171 plunky 2108: /*
2109: * set sockopt value from mbuf
2110: * - ONLY for legacy code
2111: * - mbuf is released by sockopt
1.176 plunky 2112: * - will not sleep
1.171 plunky 2113: */
2114: int
2115: sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2116: {
2117: size_t len;
1.176 plunky 2118: int error;
1.1 cgd 2119:
1.171 plunky 2120: len = m_length(m);
1.1 cgd 2121:
1.176 plunky 2122: if (sopt->sopt_size == 0) {
2123: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2124: if (error)
2125: return error;
2126: }
1.1 cgd 2127:
1.171 plunky 2128: KASSERT(sopt->sopt_size == len);
2129: m_copydata(m, 0, len, sopt->sopt_data);
2130: m_freem(m);
1.1 cgd 2131:
1.171 plunky 2132: return 0;
2133: }
1.1 cgd 2134:
1.171 plunky 2135: /*
2136: * get sockopt value into mbuf
2137: * - ONLY for legacy code
2138: * - mbuf to be released by the caller
1.176 plunky 2139: * - will not sleep
1.171 plunky 2140: */
2141: struct mbuf *
2142: sockopt_getmbuf(const struct sockopt *sopt)
2143: {
2144: struct mbuf *m;
1.107 darrenr 2145:
1.176 plunky 2146: if (sopt->sopt_size > MCLBYTES)
2147: return NULL;
2148:
2149: m = m_get(M_DONTWAIT, MT_SOOPTS);
1.171 plunky 2150: if (m == NULL)
2151: return NULL;
2152:
1.176 plunky 2153: if (sopt->sopt_size > MLEN) {
2154: MCLGET(m, M_DONTWAIT);
2155: if ((m->m_flags & M_EXT) == 0) {
2156: m_free(m);
2157: return NULL;
2158: }
1.1 cgd 2159: }
1.176 plunky 2160:
2161: memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2162: m->m_len = sopt->sopt_size;
1.160 ad 2163:
1.171 plunky 2164: return m;
1.1 cgd 2165: }
2166:
1.14 mycroft 2167: void
1.54 lukem 2168: sohasoutofband(struct socket *so)
1.1 cgd 2169: {
1.153 rmind 2170:
1.90 christos 2171: fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1.189 ad 2172: selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
1.1 cgd 2173: }
1.72 jdolecek 2174:
2175: static void
2176: filt_sordetach(struct knote *kn)
2177: {
2178: struct socket *so;
2179:
1.155 ad 2180: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2181: solock(so);
1.73 christos 2182: SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2183: if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1.72 jdolecek 2184: so->so_rcv.sb_flags &= ~SB_KNOTE;
1.160 ad 2185: sounlock(so);
1.72 jdolecek 2186: }
2187:
2188: /*ARGSUSED*/
2189: static int
1.129 yamt 2190: filt_soread(struct knote *kn, long hint)
1.72 jdolecek 2191: {
2192: struct socket *so;
1.160 ad 2193: int rv;
1.72 jdolecek 2194:
1.155 ad 2195: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2196: if (hint != NOTE_SUBMIT)
2197: solock(so);
1.72 jdolecek 2198: kn->kn_data = so->so_rcv.sb_cc;
2199: if (so->so_state & SS_CANTRCVMORE) {
1.108 perry 2200: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2201: kn->kn_fflags = so->so_error;
1.160 ad 2202: rv = 1;
2203: } else if (so->so_error) /* temporary udp error */
2204: rv = 1;
2205: else if (kn->kn_sfflags & NOTE_LOWAT)
2206: rv = (kn->kn_data >= kn->kn_sdata);
2207: else
2208: rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2209: if (hint != NOTE_SUBMIT)
2210: sounlock(so);
2211: return rv;
1.72 jdolecek 2212: }
2213:
2214: static void
2215: filt_sowdetach(struct knote *kn)
2216: {
2217: struct socket *so;
2218:
1.155 ad 2219: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2220: solock(so);
1.73 christos 2221: SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2222: if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1.72 jdolecek 2223: so->so_snd.sb_flags &= ~SB_KNOTE;
1.160 ad 2224: sounlock(so);
1.72 jdolecek 2225: }
2226:
2227: /*ARGSUSED*/
2228: static int
1.129 yamt 2229: filt_sowrite(struct knote *kn, long hint)
1.72 jdolecek 2230: {
2231: struct socket *so;
1.160 ad 2232: int rv;
1.72 jdolecek 2233:
1.155 ad 2234: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2235: if (hint != NOTE_SUBMIT)
2236: solock(so);
1.72 jdolecek 2237: kn->kn_data = sbspace(&so->so_snd);
2238: if (so->so_state & SS_CANTSENDMORE) {
1.108 perry 2239: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2240: kn->kn_fflags = so->so_error;
1.160 ad 2241: rv = 1;
2242: } else if (so->so_error) /* temporary udp error */
2243: rv = 1;
2244: else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1.72 jdolecek 2245: (so->so_proto->pr_flags & PR_CONNREQUIRED))
1.160 ad 2246: rv = 0;
2247: else if (kn->kn_sfflags & NOTE_LOWAT)
2248: rv = (kn->kn_data >= kn->kn_sdata);
2249: else
2250: rv = (kn->kn_data >= so->so_snd.sb_lowat);
2251: if (hint != NOTE_SUBMIT)
2252: sounlock(so);
2253: return rv;
1.72 jdolecek 2254: }
2255:
2256: /*ARGSUSED*/
2257: static int
1.129 yamt 2258: filt_solisten(struct knote *kn, long hint)
1.72 jdolecek 2259: {
2260: struct socket *so;
1.160 ad 2261: int rv;
1.72 jdolecek 2262:
1.155 ad 2263: so = ((file_t *)kn->kn_obj)->f_data;
1.72 jdolecek 2264:
2265: /*
2266: * Set kn_data to number of incoming connections, not
2267: * counting partial (incomplete) connections.
1.108 perry 2268: */
1.160 ad 2269: if (hint != NOTE_SUBMIT)
2270: solock(so);
1.72 jdolecek 2271: kn->kn_data = so->so_qlen;
1.160 ad 2272: rv = (kn->kn_data > 0);
2273: if (hint != NOTE_SUBMIT)
2274: sounlock(so);
2275: return rv;
1.72 jdolecek 2276: }
2277:
2278: static const struct filterops solisten_filtops =
2279: { 1, NULL, filt_sordetach, filt_solisten };
2280: static const struct filterops soread_filtops =
2281: { 1, NULL, filt_sordetach, filt_soread };
2282: static const struct filterops sowrite_filtops =
2283: { 1, NULL, filt_sowdetach, filt_sowrite };
2284:
2285: int
1.129 yamt 2286: soo_kqfilter(struct file *fp, struct knote *kn)
1.72 jdolecek 2287: {
2288: struct socket *so;
2289: struct sockbuf *sb;
2290:
1.155 ad 2291: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2292: solock(so);
1.72 jdolecek 2293: switch (kn->kn_filter) {
2294: case EVFILT_READ:
2295: if (so->so_options & SO_ACCEPTCONN)
2296: kn->kn_fop = &solisten_filtops;
2297: else
2298: kn->kn_fop = &soread_filtops;
2299: sb = &so->so_rcv;
2300: break;
2301: case EVFILT_WRITE:
2302: kn->kn_fop = &sowrite_filtops;
2303: sb = &so->so_snd;
2304: break;
2305: default:
1.160 ad 2306: sounlock(so);
1.149 pooka 2307: return (EINVAL);
1.72 jdolecek 2308: }
1.73 christos 2309: SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1.72 jdolecek 2310: sb->sb_flags |= SB_KNOTE;
1.160 ad 2311: sounlock(so);
1.72 jdolecek 2312: return (0);
2313: }
2314:
1.154 ad 2315: static int
2316: sodopoll(struct socket *so, int events)
2317: {
2318: int revents;
2319:
2320: revents = 0;
2321:
2322: if (events & (POLLIN | POLLRDNORM))
2323: if (soreadable(so))
2324: revents |= events & (POLLIN | POLLRDNORM);
2325:
2326: if (events & (POLLOUT | POLLWRNORM))
2327: if (sowritable(so))
2328: revents |= events & (POLLOUT | POLLWRNORM);
2329:
2330: if (events & (POLLPRI | POLLRDBAND))
2331: if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2332: revents |= events & (POLLPRI | POLLRDBAND);
2333:
2334: return revents;
2335: }
2336:
2337: int
2338: sopoll(struct socket *so, int events)
2339: {
2340: int revents = 0;
2341:
1.160 ad 2342: #ifndef DIAGNOSTIC
2343: /*
2344: * Do a quick, unlocked check in expectation that the socket
2345: * will be ready for I/O. Don't do this check if DIAGNOSTIC,
2346: * as the solocked() assertions will fail.
2347: */
1.154 ad 2348: if ((revents = sodopoll(so, events)) != 0)
2349: return revents;
1.160 ad 2350: #endif
1.154 ad 2351:
1.160 ad 2352: solock(so);
1.154 ad 2353: if ((revents = sodopoll(so, events)) == 0) {
2354: if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2355: selrecord(curlwp, &so->so_rcv.sb_sel);
1.160 ad 2356: so->so_rcv.sb_flags |= SB_NOTIFY;
1.154 ad 2357: }
2358:
2359: if (events & (POLLOUT | POLLWRNORM)) {
2360: selrecord(curlwp, &so->so_snd.sb_sel);
1.160 ad 2361: so->so_snd.sb_flags |= SB_NOTIFY;
1.154 ad 2362: }
2363: }
1.160 ad 2364: sounlock(so);
1.154 ad 2365:
2366: return revents;
2367: }
2368:
2369:
1.94 yamt 2370: #include <sys/sysctl.h>
2371:
2372: static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
2373:
2374: /*
2375: * sysctl helper routine for kern.somaxkva. ensures that the given
2376: * value is not too small.
2377: * (XXX should we maybe make sure it's not too large as well?)
2378: */
2379: static int
2380: sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2381: {
2382: int error, new_somaxkva;
2383: struct sysctlnode node;
2384:
2385: new_somaxkva = somaxkva;
2386: node = *rnode;
2387: node.sysctl_data = &new_somaxkva;
2388: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2389: if (error || newp == NULL)
2390: return (error);
2391:
2392: if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2393: return (EINVAL);
2394:
1.136 ad 2395: mutex_enter(&so_pendfree_lock);
1.94 yamt 2396: somaxkva = new_somaxkva;
1.136 ad 2397: cv_broadcast(&socurkva_cv);
2398: mutex_exit(&so_pendfree_lock);
1.94 yamt 2399:
2400: return (error);
2401: }
2402:
1.178 pooka 2403: static void
1.187 cegger 2404: sysctl_kern_somaxkva_setup(void)
1.94 yamt 2405: {
2406:
1.178 pooka 2407: KASSERT(socket_sysctllog == NULL);
2408: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2409: CTLFLAG_PERMANENT,
2410: CTLTYPE_NODE, "kern", NULL,
2411: NULL, 0, NULL, 0,
2412: CTL_KERN, CTL_EOL);
2413:
1.178 pooka 2414: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2415: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1.103 atatat 2416: CTLTYPE_INT, "somaxkva",
2417: SYSCTL_DESCR("Maximum amount of kernel memory to be "
2418: "used for socket buffers"),
1.94 yamt 2419: sysctl_kern_somaxkva, 0, NULL, 0,
2420: CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
2421: }
CVSweb <webmaster@jp.NetBSD.org>