Annotation of src/sys/kern/uipc_socket.c, Revision 1.93
1.93 ! yamt 1: /* $NetBSD: uipc_socket.c,v 1.92 2004/03/17 09:58:15 yamt Exp $ */
1.64 thorpej 2:
3: /*-
4: * Copyright (c) 2002 The NetBSD Foundation, Inc.
5: * All rights reserved.
6: *
7: * This code is derived from software contributed to The NetBSD Foundation
8: * by Jason R. Thorpe of Wasabi Systems, Inc.
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: * 3. All advertising materials mentioning features or use of this software
19: * must display the following acknowledgement:
20: * This product includes software developed by the NetBSD
21: * Foundation, Inc. and its contributors.
22: * 4. Neither the name of The NetBSD Foundation nor the names of its
23: * contributors may be used to endorse or promote products derived
24: * from this software without specific prior written permission.
25: *
26: * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27: * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28: * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29: * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30: * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31: * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32: * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33: * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34: * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35: * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36: * POSSIBILITY OF SUCH DAMAGE.
37: */
1.16 cgd 38:
1.1 cgd 39: /*
1.15 mycroft 40: * Copyright (c) 1982, 1986, 1988, 1990, 1993
41: * The Regents of the University of California. All rights reserved.
1.1 cgd 42: *
43: * Redistribution and use in source and binary forms, with or without
44: * modification, are permitted provided that the following conditions
45: * are met:
46: * 1. Redistributions of source code must retain the above copyright
47: * notice, this list of conditions and the following disclaimer.
48: * 2. Redistributions in binary form must reproduce the above copyright
49: * notice, this list of conditions and the following disclaimer in the
50: * documentation and/or other materials provided with the distribution.
1.85 agc 51: * 3. Neither the name of the University nor the names of its contributors
1.1 cgd 52: * may be used to endorse or promote products derived from this software
53: * without specific prior written permission.
54: *
55: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65: * SUCH DAMAGE.
66: *
1.32 fvdl 67: * @(#)uipc_socket.c 8.6 (Berkeley) 5/2/95
1.1 cgd 68: */
1.59 lukem 69:
70: #include <sys/cdefs.h>
1.93 ! yamt 71: __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.92 2004/03/17 09:58:15 yamt Exp $");
1.64 thorpej 72:
73: #include "opt_sock_counters.h"
74: #include "opt_sosend_loan.h"
1.81 martin 75: #include "opt_mbuftrace.h"
1.84 ragge 76: #include "opt_somaxkva.h"
1.1 cgd 77:
1.9 mycroft 78: #include <sys/param.h>
79: #include <sys/systm.h>
80: #include <sys/proc.h>
81: #include <sys/file.h>
82: #include <sys/malloc.h>
83: #include <sys/mbuf.h>
84: #include <sys/domain.h>
85: #include <sys/kernel.h>
86: #include <sys/protosw.h>
87: #include <sys/socket.h>
88: #include <sys/socketvar.h>
1.21 christos 89: #include <sys/signalvar.h>
1.9 mycroft 90: #include <sys/resourcevar.h>
1.37 thorpej 91: #include <sys/pool.h>
1.72 jdolecek 92: #include <sys/event.h>
1.89 christos 93: #include <sys/poll.h>
1.37 thorpej 94:
1.64 thorpej 95: #include <uvm/uvm.h>
96:
1.54 lukem 97: struct pool socket_pool;
1.77 thorpej 98:
99: MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
100: MALLOC_DEFINE(M_SONAME, "soname", "socket name");
1.37 thorpej 101:
1.54 lukem 102: extern int somaxconn; /* patchable (XXX sysctl) */
103: int somaxconn = SOMAXCONN;
1.49 jonathan 104:
1.64 thorpej 105: #ifdef SOSEND_COUNTERS
106: #include <sys/device.h>
107:
108: struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
109: NULL, "sosend", "loan big");
110: struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
111: NULL, "sosend", "copy big");
112: struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
113: NULL, "sosend", "copy small");
114: struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
115: NULL, "sosend", "kva limit");
116:
117: #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++
118:
119: #else
120:
121: #define SOSEND_COUNTER_INCR(ev) /* nothing */
122:
123: #endif /* SOSEND_COUNTERS */
124:
1.37 thorpej 125: void
1.54 lukem 126: soinit(void)
1.37 thorpej 127: {
1.91 thorpej 128:
129: /* Set the initial adjusted socket buffer size. */
130: if (sb_max_set(sb_max))
131: panic("bad initial sb_max value: %lu\n", sb_max);
1.37 thorpej 132:
133: pool_init(&socket_pool, sizeof(struct socket), 0, 0, 0,
1.62 thorpej 134: "sockpl", NULL);
1.64 thorpej 135:
136: #ifdef SOSEND_COUNTERS
137: evcnt_attach_static(&sosend_loan_big);
138: evcnt_attach_static(&sosend_copy_big);
139: evcnt_attach_static(&sosend_copy_small);
140: evcnt_attach_static(&sosend_kvalimit);
141: #endif /* SOSEND_COUNTERS */
1.37 thorpej 142: }
1.1 cgd 143:
1.71 thorpej 144: #ifdef SOSEND_NO_LOAN
145: int use_sosend_loan = 0;
146: #else
1.65 thorpej 147: int use_sosend_loan = 1;
148: #endif
1.64 thorpej 149:
1.93 ! yamt 150: struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER;
1.64 thorpej 151: struct mbuf *so_pendfree;
152:
1.84 ragge 153: #ifndef SOMAXKVA
154: #define SOMAXKVA (16 * 1024 * 1024)
155: #endif
156: int somaxkva = SOMAXKVA;
1.64 thorpej 157: int socurkva;
158: int sokvawaiters;
159:
160: #define SOCK_LOAN_THRESH 4096
161: #define SOCK_LOAN_CHUNK 65536
162:
1.80 yamt 163: static size_t sodopendfree(struct socket *);
1.93 ! yamt 164: static size_t sodopendfreel(struct socket *);
! 165:
! 166: /*
! 167: * sokvaalloc: allocate kva for loan.
! 168: */
1.80 yamt 169:
170: vaddr_t
171: sokvaalloc(vsize_t len, struct socket *so)
172: {
173: vaddr_t lva;
174: int s;
175:
1.93 ! yamt 176: /*
! 177: * reserve kva.
! 178: */
! 179:
! 180: s = splvm();
! 181: simple_lock(&so_pendfree_slock);
1.80 yamt 182: while (socurkva + len > somaxkva) {
1.93 ! yamt 183: size_t freed;
! 184:
! 185: /*
! 186: * try to do pendfree.
! 187: */
! 188:
! 189: freed = sodopendfreel(so);
! 190:
! 191: /*
! 192: * if some kva was freed, try again.
! 193: */
! 194:
! 195: if (freed)
1.80 yamt 196: continue;
1.93 ! yamt 197:
1.80 yamt 198: SOSEND_COUNTER_INCR(&sosend_kvalimit);
199: sokvawaiters++;
1.93 ! yamt 200: (void) ltsleep(&socurkva, PVM, "sokva", 0, &so_pendfree_slock);
1.80 yamt 201: sokvawaiters--;
202: }
1.93 ! yamt 203: socurkva += len;
! 204: simple_unlock(&so_pendfree_slock);
! 205: splx(s);
! 206:
! 207: /*
! 208: * allocate kva.
! 209: */
1.80 yamt 210:
211: lva = uvm_km_valloc_wait(kernel_map, len);
212: if (lva == 0)
213: return (0);
214:
215: return lva;
216: }
217:
1.93 ! yamt 218: /*
! 219: * sokvafree: free kva for loan.
! 220: */
! 221:
1.80 yamt 222: void
223: sokvafree(vaddr_t sva, vsize_t len)
224: {
1.93 ! yamt 225: int s;
! 226:
! 227: /*
! 228: * free kva.
! 229: */
1.80 yamt 230:
231: uvm_km_free(kernel_map, sva, len);
1.93 ! yamt 232:
! 233: /*
! 234: * unreserve kva.
! 235: */
! 236:
! 237: s = splvm();
! 238: simple_lock(&so_pendfree_slock);
1.80 yamt 239: socurkva -= len;
240: if (sokvawaiters)
241: wakeup(&socurkva);
1.93 ! yamt 242: simple_unlock(&so_pendfree_slock);
! 243: splx(s);
1.80 yamt 244: }
245:
1.64 thorpej 246: static void
1.79 thorpej 247: sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size)
1.64 thorpej 248: {
249: vaddr_t va, sva, eva;
250: vsize_t len;
251: paddr_t pa;
252: int i, npgs;
253:
254: eva = round_page((vaddr_t) buf + size);
255: sva = trunc_page((vaddr_t) buf);
256: len = eva - sva;
257: npgs = len >> PAGE_SHIFT;
258:
1.79 thorpej 259: if (__predict_false(pgs == NULL)) {
260: pgs = alloca(npgs * sizeof(*pgs));
1.64 thorpej 261:
1.79 thorpej 262: for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) {
263: if (pmap_extract(pmap_kernel(), va, &pa) == FALSE)
264: panic("sodoloanfree: va 0x%lx not mapped", va);
265: pgs[i] = PHYS_TO_VM_PAGE(pa);
266: }
1.64 thorpej 267: }
268:
269: pmap_kremove(sva, len);
270: pmap_update(pmap_kernel());
271: uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
1.80 yamt 272: sokvafree(sva, len);
1.64 thorpej 273: }
274:
275: static size_t
276: sodopendfree(struct socket *so)
277: {
278: int s;
1.93 ! yamt 279: size_t rv;
1.64 thorpej 280:
281: s = splvm();
1.93 ! yamt 282: simple_lock(&so_pendfree_slock);
! 283: rv = sodopendfreel(so);
! 284: simple_unlock(&so_pendfree_slock);
! 285: splx(s);
! 286:
! 287: return rv;
! 288: }
! 289:
! 290: /*
! 291: * sodopendfreel: free mbufs on "pendfree" list.
! 292: * unlock and relock so_pendfree_slock when freeing mbufs.
! 293: *
! 294: * => called with so_pendfree_slock held.
! 295: * => called at splvm.
! 296: */
! 297:
! 298: static size_t
! 299: sodopendfreel(struct socket *so)
! 300: {
! 301: size_t rv = 0;
! 302:
! 303: LOCK_ASSERT(simple_lock_held(&so_pendfree_slock));
1.64 thorpej 304:
305: for (;;) {
1.93 ! yamt 306: struct mbuf *m;
! 307: struct mbuf *next;
! 308:
1.64 thorpej 309: m = so_pendfree;
310: if (m == NULL)
311: break;
1.93 ! yamt 312: so_pendfree = NULL;
! 313: simple_unlock(&so_pendfree_slock);
! 314: /* XXX splx */
! 315:
! 316: for (; m != NULL; m = next) {
! 317: next = m->m_next;
! 318:
! 319: rv += m->m_ext.ext_size;
! 320: sodoloanfree((m->m_flags & M_EXT_PAGES) ?
! 321: m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf,
! 322: m->m_ext.ext_size);
! 323: pool_cache_put(&mbpool_cache, m);
! 324: }
1.64 thorpej 325:
1.93 ! yamt 326: /* XXX splvm */
! 327: simple_lock(&so_pendfree_slock);
1.64 thorpej 328: }
329:
330: return (rv);
331: }
332:
1.80 yamt 333: void
1.76 thorpej 334: soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1.64 thorpej 335: {
336: int s;
337:
338: if (m == NULL) {
1.93 ! yamt 339:
! 340: /*
! 341: * called from MEXTREMOVE.
! 342: */
! 343:
1.79 thorpej 344: sodoloanfree(NULL, buf, size);
1.64 thorpej 345: return;
346: }
347:
1.93 ! yamt 348: /*
! 349: * postpone freeing mbuf.
! 350: *
! 351: * we can't do it in interrupt context
! 352: * because we need to put kva back to kernel_map.
! 353: */
! 354:
1.64 thorpej 355: s = splvm();
1.93 ! yamt 356: simple_lock(&so_pendfree_slock);
1.92 yamt 357: m->m_next = so_pendfree;
358: so_pendfree = m;
1.64 thorpej 359: if (sokvawaiters)
360: wakeup(&socurkva);
1.93 ! yamt 361: simple_unlock(&so_pendfree_slock);
! 362: splx(s);
1.64 thorpej 363: }
364:
365: static long
366: sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
367: {
368: struct iovec *iov = uio->uio_iov;
369: vaddr_t sva, eva;
370: vsize_t len;
371: vaddr_t lva, va;
1.80 yamt 372: int npgs, i, error;
1.64 thorpej 373:
374: if (uio->uio_segflg != UIO_USERSPACE)
375: return (0);
376:
377: if (iov->iov_len < (size_t) space)
378: space = iov->iov_len;
379: if (space > SOCK_LOAN_CHUNK)
380: space = SOCK_LOAN_CHUNK;
381:
382: eva = round_page((vaddr_t) iov->iov_base + space);
383: sva = trunc_page((vaddr_t) iov->iov_base);
384: len = eva - sva;
385: npgs = len >> PAGE_SHIFT;
386:
1.79 thorpej 387: /* XXX KDASSERT */
388: KASSERT(npgs <= M_EXT_MAXPAGES);
389:
1.80 yamt 390: lva = sokvaalloc(len, so);
1.64 thorpej 391: if (lva == 0)
1.80 yamt 392: return 0;
1.64 thorpej 393:
1.83 fvdl 394: error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len,
1.79 thorpej 395: m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
1.64 thorpej 396: if (error) {
1.80 yamt 397: sokvafree(lva, len);
1.64 thorpej 398: return (0);
399: }
400:
401: for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
1.79 thorpej 402: pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
403: VM_PROT_READ);
1.64 thorpej 404: pmap_update(pmap_kernel());
405:
406: lva += (vaddr_t) iov->iov_base & PAGE_MASK;
407:
408: MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so);
1.79 thorpej 409: m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
1.64 thorpej 410:
411: uio->uio_resid -= space;
412: /* uio_offset not updated, not set/used for write(2) */
413: uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space;
414: uio->uio_iov->iov_len -= space;
415: if (uio->uio_iov->iov_len == 0) {
416: uio->uio_iov++;
417: uio->uio_iovcnt--;
418: }
419:
420: return (space);
421: }
422:
1.1 cgd 423: /*
424: * Socket operation routines.
425: * These routines are called by the routines in
426: * sys_socket.c or from a system process, and
427: * implement the semantics of socket operations by
428: * switching out to the protocol specific routines.
429: */
430: /*ARGSUSED*/
1.3 andrew 431: int
1.54 lukem 432: socreate(int dom, struct socket **aso, int type, int proto)
1.1 cgd 433: {
1.54 lukem 434: struct proc *p;
435: struct protosw *prp;
436: struct socket *so;
437: int error, s;
1.1 cgd 438:
1.54 lukem 439: p = curproc; /* XXX */
1.1 cgd 440: if (proto)
441: prp = pffindproto(dom, proto, type);
442: else
443: prp = pffindtype(dom, type);
1.15 mycroft 444: if (prp == 0 || prp->pr_usrreq == 0)
1.1 cgd 445: return (EPROTONOSUPPORT);
446: if (prp->pr_type != type)
447: return (EPROTOTYPE);
1.39 matt 448: s = splsoftnet();
1.37 thorpej 449: so = pool_get(&socket_pool, PR_WAITOK);
1.38 perry 450: memset((caddr_t)so, 0, sizeof(*so));
1.31 thorpej 451: TAILQ_INIT(&so->so_q0);
452: TAILQ_INIT(&so->so_q);
1.1 cgd 453: so->so_type = type;
454: so->so_proto = prp;
1.33 matt 455: so->so_send = sosend;
456: so->so_receive = soreceive;
1.78 matt 457: #ifdef MBUFTRACE
458: so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
459: so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
460: so->so_mowner = &prp->pr_domain->dom_mowner;
461: #endif
1.44 lukem 462: if (p != 0)
463: so->so_uid = p->p_ucred->cr_uid;
1.22 mycroft 464: error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0,
1.83 fvdl 465: (struct mbuf *)(long)proto, (struct mbuf *)0, p);
1.1 cgd 466: if (error) {
467: so->so_state |= SS_NOFDREF;
468: sofree(so);
1.39 matt 469: splx(s);
1.1 cgd 470: return (error);
471: }
1.39 matt 472: splx(s);
1.1 cgd 473: *aso = so;
474: return (0);
475: }
476:
1.3 andrew 477: int
1.83 fvdl 478: sobind(struct socket *so, struct mbuf *nam, struct proc *p)
1.1 cgd 479: {
1.54 lukem 480: int s, error;
1.1 cgd 481:
1.54 lukem 482: s = splsoftnet();
1.22 mycroft 483: error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0,
1.83 fvdl 484: nam, (struct mbuf *)0, p);
1.1 cgd 485: splx(s);
486: return (error);
487: }
488:
1.3 andrew 489: int
1.54 lukem 490: solisten(struct socket *so, int backlog)
1.1 cgd 491: {
1.54 lukem 492: int s, error;
1.1 cgd 493:
1.54 lukem 494: s = splsoftnet();
1.22 mycroft 495: error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0,
1.83 fvdl 496: (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1.1 cgd 497: if (error) {
498: splx(s);
499: return (error);
500: }
1.63 matt 501: if (TAILQ_EMPTY(&so->so_q))
1.1 cgd 502: so->so_options |= SO_ACCEPTCONN;
503: if (backlog < 0)
504: backlog = 0;
1.49 jonathan 505: so->so_qlimit = min(backlog, somaxconn);
1.1 cgd 506: splx(s);
507: return (0);
508: }
509:
1.21 christos 510: void
1.54 lukem 511: sofree(struct socket *so)
1.1 cgd 512: {
513:
1.43 mycroft 514: if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
1.1 cgd 515: return;
1.43 mycroft 516: if (so->so_head) {
517: /*
518: * We must not decommission a socket that's on the accept(2)
519: * queue. If we do, then accept(2) may hang after select(2)
520: * indicated that the listening socket was ready.
521: */
522: if (!soqremque(so, 0))
523: return;
524: }
1.1 cgd 525: sbrelease(&so->so_snd);
526: sorflush(so);
1.37 thorpej 527: pool_put(&socket_pool, so);
1.1 cgd 528: }
529:
530: /*
531: * Close a socket on last file table reference removal.
532: * Initiate disconnect if connected.
533: * Free socket when disconnect complete.
534: */
1.3 andrew 535: int
1.54 lukem 536: soclose(struct socket *so)
1.1 cgd 537: {
1.54 lukem 538: struct socket *so2;
539: int s, error;
1.1 cgd 540:
1.54 lukem 541: error = 0;
542: s = splsoftnet(); /* conservative */
1.1 cgd 543: if (so->so_options & SO_ACCEPTCONN) {
1.63 matt 544: while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
1.42 mycroft 545: (void) soqremque(so2, 0);
1.41 mycroft 546: (void) soabort(so2);
547: }
1.63 matt 548: while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
1.42 mycroft 549: (void) soqremque(so2, 1);
1.41 mycroft 550: (void) soabort(so2);
551: }
1.1 cgd 552: }
553: if (so->so_pcb == 0)
554: goto discard;
555: if (so->so_state & SS_ISCONNECTED) {
556: if ((so->so_state & SS_ISDISCONNECTING) == 0) {
557: error = sodisconnect(so);
558: if (error)
559: goto drop;
560: }
561: if (so->so_options & SO_LINGER) {
562: if ((so->so_state & SS_ISDISCONNECTING) &&
563: (so->so_state & SS_NBIO))
564: goto drop;
1.21 christos 565: while (so->so_state & SS_ISCONNECTED) {
566: error = tsleep((caddr_t)&so->so_timeo,
567: PSOCK | PCATCH, netcls,
1.30 thorpej 568: so->so_linger * hz);
1.21 christos 569: if (error)
1.1 cgd 570: break;
1.21 christos 571: }
1.1 cgd 572: }
573: }
1.54 lukem 574: drop:
1.1 cgd 575: if (so->so_pcb) {
1.22 mycroft 576: int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
577: (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
1.83 fvdl 578: (struct proc *)0);
1.1 cgd 579: if (error == 0)
580: error = error2;
581: }
1.54 lukem 582: discard:
1.1 cgd 583: if (so->so_state & SS_NOFDREF)
584: panic("soclose: NOFDREF");
585: so->so_state |= SS_NOFDREF;
586: sofree(so);
587: splx(s);
588: return (error);
589: }
590:
591: /*
1.20 mycroft 592: * Must be called at splsoftnet...
1.1 cgd 593: */
1.3 andrew 594: int
1.54 lukem 595: soabort(struct socket *so)
1.1 cgd 596: {
597:
1.22 mycroft 598: return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0,
1.83 fvdl 599: (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1.1 cgd 600: }
601:
1.3 andrew 602: int
1.54 lukem 603: soaccept(struct socket *so, struct mbuf *nam)
1.1 cgd 604: {
1.54 lukem 605: int s, error;
1.1 cgd 606:
1.54 lukem 607: error = 0;
608: s = splsoftnet();
1.1 cgd 609: if ((so->so_state & SS_NOFDREF) == 0)
610: panic("soaccept: !NOFDREF");
611: so->so_state &= ~SS_NOFDREF;
1.55 thorpej 612: if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
613: (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
1.41 mycroft 614: error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
1.83 fvdl 615: (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0);
1.41 mycroft 616: else
1.53 itojun 617: error = ECONNABORTED;
1.52 itojun 618:
1.1 cgd 619: splx(s);
620: return (error);
621: }
622:
1.3 andrew 623: int
1.54 lukem 624: soconnect(struct socket *so, struct mbuf *nam)
1.1 cgd 625: {
1.83 fvdl 626: struct proc *p;
1.54 lukem 627: int s, error;
1.1 cgd 628:
1.83 fvdl 629: p = curproc; /* XXX */
1.1 cgd 630: if (so->so_options & SO_ACCEPTCONN)
631: return (EOPNOTSUPP);
1.20 mycroft 632: s = splsoftnet();
1.1 cgd 633: /*
634: * If protocol is connection-based, can only connect once.
635: * Otherwise, if connected, try to disconnect first.
636: * This allows user to disconnect by connecting to, e.g.,
637: * a null address.
638: */
639: if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
640: ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
641: (error = sodisconnect(so))))
642: error = EISCONN;
643: else
644: error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
1.83 fvdl 645: (struct mbuf *)0, nam, (struct mbuf *)0, p);
1.1 cgd 646: splx(s);
647: return (error);
648: }
649:
1.3 andrew 650: int
1.54 lukem 651: soconnect2(struct socket *so1, struct socket *so2)
1.1 cgd 652: {
1.54 lukem 653: int s, error;
1.1 cgd 654:
1.54 lukem 655: s = splsoftnet();
1.22 mycroft 656: error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
657: (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0,
1.83 fvdl 658: (struct proc *)0);
1.1 cgd 659: splx(s);
660: return (error);
661: }
662:
1.3 andrew 663: int
1.54 lukem 664: sodisconnect(struct socket *so)
1.1 cgd 665: {
1.54 lukem 666: int s, error;
1.1 cgd 667:
1.54 lukem 668: s = splsoftnet();
1.1 cgd 669: if ((so->so_state & SS_ISCONNECTED) == 0) {
670: error = ENOTCONN;
671: goto bad;
672: }
673: if (so->so_state & SS_ISDISCONNECTING) {
674: error = EALREADY;
675: goto bad;
676: }
1.22 mycroft 677: error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
678: (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
1.83 fvdl 679: (struct proc *)0);
1.54 lukem 680: bad:
1.1 cgd 681: splx(s);
1.64 thorpej 682: sodopendfree(so);
1.1 cgd 683: return (error);
684: }
685:
1.15 mycroft 686: #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1.1 cgd 687: /*
688: * Send on a socket.
689: * If send must go all at once and message is larger than
690: * send buffering, then hard error.
691: * Lock against other senders.
692: * If must go all at once and not enough room now, then
693: * inform user that this would block and do nothing.
694: * Otherwise, if nonblocking, send as much as possible.
695: * The data to be sent is described by "uio" if nonzero,
696: * otherwise by the mbuf chain "top" (which must be null
697: * if uio is not). Data provided in mbuf chain must be small
698: * enough to send all at once.
699: *
700: * Returns nonzero on error, timeout or signal; callers
701: * must check for short counts if EINTR/ERESTART are returned.
702: * Data and control buffers are freed on return.
703: */
1.3 andrew 704: int
1.54 lukem 705: sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
706: struct mbuf *control, int flags)
1.1 cgd 707: {
1.54 lukem 708: struct proc *p;
709: struct mbuf **mp, *m;
1.58 jdolecek 710: long space, len, resid, clen, mlen;
711: int error, s, dontroute, atomic;
1.54 lukem 712:
1.64 thorpej 713: sodopendfree(so);
714:
1.83 fvdl 715: p = curproc; /* XXX */
1.54 lukem 716: clen = 0;
717: atomic = sosendallatonce(so) || top;
1.1 cgd 718: if (uio)
719: resid = uio->uio_resid;
720: else
721: resid = top->m_pkthdr.len;
1.7 cgd 722: /*
723: * In theory resid should be unsigned.
724: * However, space must be signed, as it might be less than 0
725: * if we over-committed, and we must use a signed comparison
726: * of space and resid. On the other hand, a negative resid
727: * causes us to loop sending 0-length segments to the protocol.
728: */
1.29 mycroft 729: if (resid < 0) {
730: error = EINVAL;
731: goto out;
732: }
1.1 cgd 733: dontroute =
734: (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
735: (so->so_proto->pr_flags & PR_ATOMIC);
1.12 mycroft 736: p->p_stats->p_ru.ru_msgsnd++;
1.1 cgd 737: if (control)
738: clen = control->m_len;
739: #define snderr(errno) { error = errno; splx(s); goto release; }
740:
1.54 lukem 741: restart:
1.21 christos 742: if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
1.1 cgd 743: goto out;
744: do {
1.20 mycroft 745: s = splsoftnet();
1.1 cgd 746: if (so->so_state & SS_CANTSENDMORE)
747: snderr(EPIPE);
1.48 thorpej 748: if (so->so_error) {
749: error = so->so_error;
750: so->so_error = 0;
751: splx(s);
752: goto release;
753: }
1.1 cgd 754: if ((so->so_state & SS_ISCONNECTED) == 0) {
755: if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
756: if ((so->so_state & SS_ISCONFIRMING) == 0 &&
757: !(resid == 0 && clen != 0))
758: snderr(ENOTCONN);
759: } else if (addr == 0)
760: snderr(EDESTADDRREQ);
761: }
762: space = sbspace(&so->so_snd);
763: if (flags & MSG_OOB)
764: space += 1024;
1.21 christos 765: if ((atomic && resid > so->so_snd.sb_hiwat) ||
1.11 mycroft 766: clen > so->so_snd.sb_hiwat)
767: snderr(EMSGSIZE);
768: if (space < resid + clen && uio &&
1.1 cgd 769: (atomic || space < so->so_snd.sb_lowat || space < clen)) {
770: if (so->so_state & SS_NBIO)
771: snderr(EWOULDBLOCK);
772: sbunlock(&so->so_snd);
773: error = sbwait(&so->so_snd);
774: splx(s);
775: if (error)
776: goto out;
777: goto restart;
778: }
779: splx(s);
780: mp = ⊤
781: space -= clen;
782: do {
1.45 tv 783: if (uio == NULL) {
784: /*
785: * Data is prepackaged in "top".
786: */
787: resid = 0;
788: if (flags & MSG_EOR)
789: top->m_flags |= M_EOR;
790: } else do {
791: if (top == 0) {
1.78 matt 792: m = m_gethdr(M_WAIT, MT_DATA);
1.45 tv 793: mlen = MHLEN;
794: m->m_pkthdr.len = 0;
795: m->m_pkthdr.rcvif = (struct ifnet *)0;
796: } else {
1.78 matt 797: m = m_get(M_WAIT, MT_DATA);
1.45 tv 798: mlen = MLEN;
799: }
1.78 matt 800: MCLAIM(m, so->so_snd.sb_mowner);
1.65 thorpej 801: if (use_sosend_loan &&
802: uio->uio_iov->iov_len >= SOCK_LOAN_THRESH &&
1.64 thorpej 803: space >= SOCK_LOAN_THRESH &&
804: (len = sosend_loan(so, uio, m,
805: space)) != 0) {
806: SOSEND_COUNTER_INCR(&sosend_loan_big);
807: space -= len;
808: goto have_data;
809: }
1.45 tv 810: if (resid >= MINCLSIZE && space >= MCLBYTES) {
1.64 thorpej 811: SOSEND_COUNTER_INCR(&sosend_copy_big);
1.78 matt 812: m_clget(m, M_WAIT);
1.45 tv 813: if ((m->m_flags & M_EXT) == 0)
814: goto nopages;
815: mlen = MCLBYTES;
816: if (atomic && top == 0) {
1.58 jdolecek 817: len = lmin(MCLBYTES - max_hdr,
1.54 lukem 818: resid);
1.45 tv 819: m->m_data += max_hdr;
820: } else
1.58 jdolecek 821: len = lmin(MCLBYTES, resid);
1.45 tv 822: space -= len;
823: } else {
1.64 thorpej 824: nopages:
825: SOSEND_COUNTER_INCR(&sosend_copy_small);
1.58 jdolecek 826: len = lmin(lmin(mlen, resid), space);
1.45 tv 827: space -= len;
828: /*
829: * For datagram protocols, leave room
830: * for protocol headers in first mbuf.
831: */
832: if (atomic && top == 0 && len < mlen)
833: MH_ALIGN(m, len);
834: }
1.54 lukem 835: error = uiomove(mtod(m, caddr_t), (int)len,
836: uio);
1.64 thorpej 837: have_data:
1.45 tv 838: resid = uio->uio_resid;
839: m->m_len = len;
840: *mp = m;
841: top->m_pkthdr.len += len;
842: if (error)
843: goto release;
844: mp = &m->m_next;
845: if (resid <= 0) {
846: if (flags & MSG_EOR)
847: top->m_flags |= M_EOR;
848: break;
849: }
850: } while (space > 0 && atomic);
1.46 sommerfe 851:
852: s = splsoftnet();
853:
854: if (so->so_state & SS_CANTSENDMORE)
855: snderr(EPIPE);
1.45 tv 856:
857: if (dontroute)
858: so->so_options |= SO_DONTROUTE;
859: if (resid > 0)
860: so->so_state |= SS_MORETOCOME;
1.46 sommerfe 861: error = (*so->so_proto->pr_usrreq)(so,
862: (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1.83 fvdl 863: top, addr, control, p);
1.45 tv 864: if (dontroute)
865: so->so_options &= ~SO_DONTROUTE;
866: if (resid > 0)
867: so->so_state &= ~SS_MORETOCOME;
1.46 sommerfe 868: splx(s);
869:
1.45 tv 870: clen = 0;
871: control = 0;
872: top = 0;
873: mp = ⊤
1.1 cgd 874: if (error)
875: goto release;
876: } while (resid && space > 0);
877: } while (resid);
878:
1.54 lukem 879: release:
1.1 cgd 880: sbunlock(&so->so_snd);
1.54 lukem 881: out:
1.1 cgd 882: if (top)
883: m_freem(top);
884: if (control)
885: m_freem(control);
886: return (error);
887: }
888:
889: /*
890: * Implement receive operations on a socket.
891: * We depend on the way that records are added to the sockbuf
892: * by sbappend*. In particular, each record (mbufs linked through m_next)
893: * must begin with an address if the protocol so specifies,
894: * followed by an optional mbuf or mbufs containing ancillary data,
895: * and then zero or more mbufs of data.
896: * In order to avoid blocking network interrupts for the entire time here,
897: * we splx() while doing the actual copy to user space.
898: * Although the sockbuf is locked, new data may still be appended,
899: * and thus we must maintain consistency of the sockbuf during that time.
900: *
901: * The caller may receive the data as a single mbuf chain by supplying
902: * an mbuf **mp0 for use in returning the chain. The uio is then used
903: * only for the count in uio_resid.
904: */
1.3 andrew 905: int
1.54 lukem 906: soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
907: struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1.1 cgd 908: {
1.54 lukem 909: struct mbuf *m, **mp;
910: int flags, len, error, s, offset, moff, type, orig_resid;
911: struct protosw *pr;
912: struct mbuf *nextrecord;
1.67 he 913: int mbuf_removed = 0;
1.64 thorpej 914:
1.54 lukem 915: pr = so->so_proto;
1.1 cgd 916: mp = mp0;
1.54 lukem 917: type = 0;
918: orig_resid = uio->uio_resid;
1.1 cgd 919: if (paddr)
920: *paddr = 0;
921: if (controlp)
922: *controlp = 0;
923: if (flagsp)
924: flags = *flagsp &~ MSG_EOR;
925: else
926: flags = 0;
1.66 enami 927:
928: if ((flags & MSG_DONTWAIT) == 0)
929: sodopendfree(so);
930:
1.1 cgd 931: if (flags & MSG_OOB) {
932: m = m_get(M_WAIT, MT_DATA);
1.17 cgd 933: error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1.22 mycroft 934: (struct mbuf *)(long)(flags & MSG_PEEK), (struct mbuf *)0,
1.83 fvdl 935: (struct proc *)0);
1.1 cgd 936: if (error)
937: goto bad;
938: do {
939: error = uiomove(mtod(m, caddr_t),
940: (int) min(uio->uio_resid, m->m_len), uio);
941: m = m_free(m);
942: } while (uio->uio_resid && error == 0 && m);
1.54 lukem 943: bad:
1.1 cgd 944: if (m)
945: m_freem(m);
946: return (error);
947: }
948: if (mp)
949: *mp = (struct mbuf *)0;
950: if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
1.22 mycroft 951: (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
1.83 fvdl 952: (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1.1 cgd 953:
1.54 lukem 954: restart:
1.21 christos 955: if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
1.1 cgd 956: return (error);
1.20 mycroft 957: s = splsoftnet();
1.1 cgd 958:
959: m = so->so_rcv.sb_mb;
960: /*
961: * If we have less data than requested, block awaiting more
962: * (subject to any timeout) if:
1.15 mycroft 963: * 1. the current count is less than the low water mark,
1.1 cgd 964: * 2. MSG_WAITALL is set, and it is possible to do the entire
1.15 mycroft 965: * receive operation at once if we block (resid <= hiwat), or
966: * 3. MSG_DONTWAIT is not set.
1.1 cgd 967: * If MSG_WAITALL is set but resid is larger than the receive buffer,
968: * we have to do the receive in sections, and thus risk returning
969: * a short count if a timeout or signal occurs after we start.
970: */
1.21 christos 971: if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
1.15 mycroft 972: so->so_rcv.sb_cc < uio->uio_resid) &&
1.1 cgd 973: (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
974: ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1.21 christos 975: m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1.1 cgd 976: #ifdef DIAGNOSTIC
977: if (m == 0 && so->so_rcv.sb_cc)
978: panic("receive 1");
979: #endif
980: if (so->so_error) {
981: if (m)
1.15 mycroft 982: goto dontblock;
1.1 cgd 983: error = so->so_error;
984: if ((flags & MSG_PEEK) == 0)
985: so->so_error = 0;
986: goto release;
987: }
988: if (so->so_state & SS_CANTRCVMORE) {
989: if (m)
1.15 mycroft 990: goto dontblock;
1.1 cgd 991: else
992: goto release;
993: }
994: for (; m; m = m->m_next)
995: if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
996: m = so->so_rcv.sb_mb;
997: goto dontblock;
998: }
999: if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1000: (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1001: error = ENOTCONN;
1002: goto release;
1003: }
1004: if (uio->uio_resid == 0)
1005: goto release;
1.15 mycroft 1006: if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
1.1 cgd 1007: error = EWOULDBLOCK;
1008: goto release;
1009: }
1.69 thorpej 1010: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1011: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1.1 cgd 1012: sbunlock(&so->so_rcv);
1013: error = sbwait(&so->so_rcv);
1014: splx(s);
1015: if (error)
1016: return (error);
1017: goto restart;
1018: }
1.54 lukem 1019: dontblock:
1.69 thorpej 1020: /*
1021: * On entry here, m points to the first record of the socket buffer.
1022: * While we process the initial mbufs containing address and control
1023: * info, we save a copy of m->m_nextpkt into nextrecord.
1024: */
1.15 mycroft 1025: #ifdef notyet /* XXXX */
1.83 fvdl 1026: if (uio->uio_procp)
1027: uio->uio_procp->p_stats->p_ru.ru_msgrcv++;
1.15 mycroft 1028: #endif
1.69 thorpej 1029: KASSERT(m == so->so_rcv.sb_mb);
1030: SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1031: SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1.1 cgd 1032: nextrecord = m->m_nextpkt;
1033: if (pr->pr_flags & PR_ADDR) {
1034: #ifdef DIAGNOSTIC
1035: if (m->m_type != MT_SONAME)
1036: panic("receive 1a");
1037: #endif
1.3 andrew 1038: orig_resid = 0;
1.1 cgd 1039: if (flags & MSG_PEEK) {
1040: if (paddr)
1041: *paddr = m_copy(m, 0, m->m_len);
1042: m = m->m_next;
1043: } else {
1044: sbfree(&so->so_rcv, m);
1.67 he 1045: mbuf_removed = 1;
1.1 cgd 1046: if (paddr) {
1047: *paddr = m;
1048: so->so_rcv.sb_mb = m->m_next;
1049: m->m_next = 0;
1050: m = so->so_rcv.sb_mb;
1051: } else {
1052: MFREE(m, so->so_rcv.sb_mb);
1053: m = so->so_rcv.sb_mb;
1054: }
1055: }
1056: }
1057: while (m && m->m_type == MT_CONTROL && error == 0) {
1058: if (flags & MSG_PEEK) {
1059: if (controlp)
1060: *controlp = m_copy(m, 0, m->m_len);
1061: m = m->m_next;
1062: } else {
1063: sbfree(&so->so_rcv, m);
1.67 he 1064: mbuf_removed = 1;
1.1 cgd 1065: if (controlp) {
1066: if (pr->pr_domain->dom_externalize &&
1067: mtod(m, struct cmsghdr *)->cmsg_type ==
1068: SCM_RIGHTS)
1.45 tv 1069: error = (*pr->pr_domain->dom_externalize)(m);
1.1 cgd 1070: *controlp = m;
1071: so->so_rcv.sb_mb = m->m_next;
1072: m->m_next = 0;
1073: m = so->so_rcv.sb_mb;
1074: } else {
1075: MFREE(m, so->so_rcv.sb_mb);
1076: m = so->so_rcv.sb_mb;
1077: }
1078: }
1.3 andrew 1079: if (controlp) {
1080: orig_resid = 0;
1.1 cgd 1081: controlp = &(*controlp)->m_next;
1.3 andrew 1082: }
1.1 cgd 1083: }
1.69 thorpej 1084:
1085: /*
1086: * If m is non-NULL, we have some data to read. From now on,
1087: * make sure to keep sb_lastrecord consistent when working on
1088: * the last packet on the chain (nextrecord == NULL) and we
1089: * change m->m_nextpkt.
1090: */
1.1 cgd 1091: if (m) {
1.69 thorpej 1092: if ((flags & MSG_PEEK) == 0) {
1.1 cgd 1093: m->m_nextpkt = nextrecord;
1.69 thorpej 1094: /*
1095: * If nextrecord == NULL (this is a single chain),
1096: * then sb_lastrecord may not be valid here if m
1097: * was changed earlier.
1098: */
1099: if (nextrecord == NULL) {
1100: KASSERT(so->so_rcv.sb_mb == m);
1101: so->so_rcv.sb_lastrecord = m;
1102: }
1103: }
1.1 cgd 1104: type = m->m_type;
1105: if (type == MT_OOBDATA)
1106: flags |= MSG_OOB;
1.69 thorpej 1107: } else {
1108: if ((flags & MSG_PEEK) == 0) {
1109: KASSERT(so->so_rcv.sb_mb == m);
1110: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1111: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1112: }
1.1 cgd 1113: }
1.69 thorpej 1114: SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1115: SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1116:
1.1 cgd 1117: moff = 0;
1118: offset = 0;
1119: while (m && uio->uio_resid > 0 && error == 0) {
1120: if (m->m_type == MT_OOBDATA) {
1121: if (type != MT_OOBDATA)
1122: break;
1123: } else if (type == MT_OOBDATA)
1124: break;
1125: #ifdef DIAGNOSTIC
1126: else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1127: panic("receive 3");
1128: #endif
1129: so->so_state &= ~SS_RCVATMARK;
1130: len = uio->uio_resid;
1131: if (so->so_oobmark && len > so->so_oobmark - offset)
1132: len = so->so_oobmark - offset;
1133: if (len > m->m_len - moff)
1134: len = m->m_len - moff;
1135: /*
1136: * If mp is set, just pass back the mbufs.
1137: * Otherwise copy them out via the uio, then free.
1138: * Sockbuf must be consistent here (points to current mbuf,
1139: * it points to next record) when we drop priority;
1140: * we must note any additions to the sockbuf when we
1141: * block interrupts again.
1142: */
1143: if (mp == 0) {
1.69 thorpej 1144: SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1145: SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1.1 cgd 1146: splx(s);
1147: error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
1.20 mycroft 1148: s = splsoftnet();
1.67 he 1149: if (error) {
1150: /*
1151: * If any part of the record has been removed
1152: * (such as the MT_SONAME mbuf, which will
1153: * happen when PR_ADDR, and thus also
1154: * PR_ATOMIC, is set), then drop the entire
1155: * record to maintain the atomicity of the
1156: * receive operation.
1157: *
1158: * This avoids a later panic("receive 1a")
1159: * when compiled with DIAGNOSTIC.
1160: */
1161: if (m && mbuf_removed
1162: && (pr->pr_flags & PR_ATOMIC))
1163: (void) sbdroprecord(&so->so_rcv);
1164:
1.57 jdolecek 1165: goto release;
1.67 he 1166: }
1.1 cgd 1167: } else
1168: uio->uio_resid -= len;
1169: if (len == m->m_len - moff) {
1170: if (m->m_flags & M_EOR)
1171: flags |= MSG_EOR;
1172: if (flags & MSG_PEEK) {
1173: m = m->m_next;
1174: moff = 0;
1175: } else {
1176: nextrecord = m->m_nextpkt;
1177: sbfree(&so->so_rcv, m);
1178: if (mp) {
1179: *mp = m;
1180: mp = &m->m_next;
1181: so->so_rcv.sb_mb = m = m->m_next;
1182: *mp = (struct mbuf *)0;
1183: } else {
1184: MFREE(m, so->so_rcv.sb_mb);
1185: m = so->so_rcv.sb_mb;
1186: }
1.69 thorpej 1187: /*
1188: * If m != NULL, we also know that
1189: * so->so_rcv.sb_mb != NULL.
1190: */
1191: KASSERT(so->so_rcv.sb_mb == m);
1192: if (m) {
1.1 cgd 1193: m->m_nextpkt = nextrecord;
1.69 thorpej 1194: if (nextrecord == NULL)
1195: so->so_rcv.sb_lastrecord = m;
1196: } else {
1197: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1198: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1199: }
1200: SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1201: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1.1 cgd 1202: }
1203: } else {
1204: if (flags & MSG_PEEK)
1205: moff += len;
1206: else {
1207: if (mp)
1208: *mp = m_copym(m, 0, len, M_WAIT);
1209: m->m_data += len;
1210: m->m_len -= len;
1211: so->so_rcv.sb_cc -= len;
1212: }
1213: }
1214: if (so->so_oobmark) {
1215: if ((flags & MSG_PEEK) == 0) {
1216: so->so_oobmark -= len;
1217: if (so->so_oobmark == 0) {
1218: so->so_state |= SS_RCVATMARK;
1219: break;
1220: }
1.7 cgd 1221: } else {
1.1 cgd 1222: offset += len;
1.7 cgd 1223: if (offset == so->so_oobmark)
1224: break;
1225: }
1.1 cgd 1226: }
1227: if (flags & MSG_EOR)
1228: break;
1229: /*
1230: * If the MSG_WAITALL flag is set (for non-atomic socket),
1231: * we must not quit until "uio->uio_resid == 0" or an error
1232: * termination. If a signal/timeout occurs, return
1233: * with a short count but without error.
1234: * Keep sockbuf locked against other readers.
1235: */
1236: while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 &&
1.3 andrew 1237: !sosendallatonce(so) && !nextrecord) {
1.1 cgd 1238: if (so->so_error || so->so_state & SS_CANTRCVMORE)
1239: break;
1.68 matt 1240: /*
1241: * If we are peeking and the socket receive buffer is
1242: * full, stop since we can't get more data to peek at.
1243: */
1244: if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1245: break;
1246: /*
1247: * If we've drained the socket buffer, tell the
1248: * protocol in case it needs to do something to
1249: * get it filled again.
1250: */
1251: if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1252: (*pr->pr_usrreq)(so, PRU_RCVD,
1253: (struct mbuf *)0,
1254: (struct mbuf *)(long)flags,
1255: (struct mbuf *)0,
1.83 fvdl 1256: (struct proc *)0);
1.69 thorpej 1257: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1258: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1.1 cgd 1259: error = sbwait(&so->so_rcv);
1260: if (error) {
1261: sbunlock(&so->so_rcv);
1262: splx(s);
1263: return (0);
1264: }
1.21 christos 1265: if ((m = so->so_rcv.sb_mb) != NULL)
1.1 cgd 1266: nextrecord = m->m_nextpkt;
1267: }
1268: }
1.3 andrew 1269:
1270: if (m && pr->pr_flags & PR_ATOMIC) {
1271: flags |= MSG_TRUNC;
1272: if ((flags & MSG_PEEK) == 0)
1273: (void) sbdroprecord(&so->so_rcv);
1274: }
1.1 cgd 1275: if ((flags & MSG_PEEK) == 0) {
1.69 thorpej 1276: if (m == 0) {
1277: /*
1.70 thorpej 1278: * First part is an inline SB_EMPTY_FIXUP(). Second
1.69 thorpej 1279: * part makes sure sb_lastrecord is up-to-date if
1280: * there is still data in the socket buffer.
1281: */
1.1 cgd 1282: so->so_rcv.sb_mb = nextrecord;
1.69 thorpej 1283: if (so->so_rcv.sb_mb == NULL) {
1284: so->so_rcv.sb_mbtail = NULL;
1285: so->so_rcv.sb_lastrecord = NULL;
1286: } else if (nextrecord->m_nextpkt == NULL)
1287: so->so_rcv.sb_lastrecord = nextrecord;
1288: }
1289: SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1290: SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1.1 cgd 1291: if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1.22 mycroft 1292: (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
1293: (struct mbuf *)(long)flags, (struct mbuf *)0,
1.83 fvdl 1294: (struct proc *)0);
1.1 cgd 1295: }
1.3 andrew 1296: if (orig_resid == uio->uio_resid && orig_resid &&
1297: (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1298: sbunlock(&so->so_rcv);
1299: splx(s);
1300: goto restart;
1301: }
1302:
1.1 cgd 1303: if (flagsp)
1304: *flagsp |= flags;
1.54 lukem 1305: release:
1.1 cgd 1306: sbunlock(&so->so_rcv);
1307: splx(s);
1308: return (error);
1309: }
1310:
1.14 mycroft 1311: int
1.54 lukem 1312: soshutdown(struct socket *so, int how)
1.1 cgd 1313: {
1.54 lukem 1314: struct protosw *pr;
1.34 kleink 1315:
1.54 lukem 1316: pr = so->so_proto;
1.34 kleink 1317: if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1318: return (EINVAL);
1.1 cgd 1319:
1.34 kleink 1320: if (how == SHUT_RD || how == SHUT_RDWR)
1.1 cgd 1321: sorflush(so);
1.34 kleink 1322: if (how == SHUT_WR || how == SHUT_RDWR)
1.22 mycroft 1323: return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0,
1.83 fvdl 1324: (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1.1 cgd 1325: return (0);
1326: }
1327:
1.14 mycroft 1328: void
1.54 lukem 1329: sorflush(struct socket *so)
1.1 cgd 1330: {
1.54 lukem 1331: struct sockbuf *sb, asb;
1332: struct protosw *pr;
1333: int s;
1.1 cgd 1334:
1.54 lukem 1335: sb = &so->so_rcv;
1336: pr = so->so_proto;
1.1 cgd 1337: sb->sb_flags |= SB_NOINTR;
1.15 mycroft 1338: (void) sblock(sb, M_WAITOK);
1.56 thorpej 1339: s = splnet();
1.1 cgd 1340: socantrcvmore(so);
1341: sbunlock(sb);
1342: asb = *sb;
1.86 wrstuden 1343: /*
1344: * Clear most of the sockbuf structure, but leave some of the
1345: * fields valid.
1346: */
1347: memset(&sb->sb_startzero, 0,
1348: sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1.1 cgd 1349: splx(s);
1350: if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1351: (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1352: sbrelease(&asb);
1353: }
1354:
1.14 mycroft 1355: int
1.54 lukem 1356: sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
1.1 cgd 1357: {
1.54 lukem 1358: int error;
1359: struct mbuf *m;
1.1 cgd 1360:
1.54 lukem 1361: error = 0;
1362: m = m0;
1.1 cgd 1363: if (level != SOL_SOCKET) {
1364: if (so->so_proto && so->so_proto->pr_ctloutput)
1365: return ((*so->so_proto->pr_ctloutput)
1366: (PRCO_SETOPT, so, level, optname, &m0));
1367: error = ENOPROTOOPT;
1368: } else {
1369: switch (optname) {
1370:
1371: case SO_LINGER:
1.36 perry 1372: if (m == NULL || m->m_len != sizeof(struct linger)) {
1.1 cgd 1373: error = EINVAL;
1374: goto bad;
1375: }
1376: so->so_linger = mtod(m, struct linger *)->l_linger;
1377: /* fall thru... */
1378:
1379: case SO_DEBUG:
1380: case SO_KEEPALIVE:
1381: case SO_DONTROUTE:
1382: case SO_USELOOPBACK:
1383: case SO_BROADCAST:
1384: case SO_REUSEADDR:
1.15 mycroft 1385: case SO_REUSEPORT:
1.1 cgd 1386: case SO_OOBINLINE:
1.26 thorpej 1387: case SO_TIMESTAMP:
1.36 perry 1388: if (m == NULL || m->m_len < sizeof(int)) {
1.1 cgd 1389: error = EINVAL;
1390: goto bad;
1391: }
1392: if (*mtod(m, int *))
1393: so->so_options |= optname;
1394: else
1395: so->so_options &= ~optname;
1396: break;
1397:
1398: case SO_SNDBUF:
1399: case SO_RCVBUF:
1400: case SO_SNDLOWAT:
1401: case SO_RCVLOWAT:
1.28 thorpej 1402: {
1403: int optval;
1404:
1.36 perry 1405: if (m == NULL || m->m_len < sizeof(int)) {
1.1 cgd 1406: error = EINVAL;
1407: goto bad;
1408: }
1.28 thorpej 1409:
1410: /*
1411: * Values < 1 make no sense for any of these
1412: * options, so disallow them.
1413: */
1414: optval = *mtod(m, int *);
1415: if (optval < 1) {
1416: error = EINVAL;
1417: goto bad;
1418: }
1419:
1.1 cgd 1420: switch (optname) {
1421:
1422: case SO_SNDBUF:
1423: case SO_RCVBUF:
1424: if (sbreserve(optname == SO_SNDBUF ?
1425: &so->so_snd : &so->so_rcv,
1.28 thorpej 1426: (u_long) optval) == 0) {
1.1 cgd 1427: error = ENOBUFS;
1428: goto bad;
1429: }
1430: break;
1431:
1.28 thorpej 1432: /*
1433: * Make sure the low-water is never greater than
1434: * the high-water.
1435: */
1.1 cgd 1436: case SO_SNDLOWAT:
1.28 thorpej 1437: so->so_snd.sb_lowat =
1438: (optval > so->so_snd.sb_hiwat) ?
1439: so->so_snd.sb_hiwat : optval;
1.1 cgd 1440: break;
1441: case SO_RCVLOWAT:
1.28 thorpej 1442: so->so_rcv.sb_lowat =
1443: (optval > so->so_rcv.sb_hiwat) ?
1444: so->so_rcv.sb_hiwat : optval;
1.1 cgd 1445: break;
1446: }
1447: break;
1.28 thorpej 1448: }
1.1 cgd 1449:
1450: case SO_SNDTIMEO:
1451: case SO_RCVTIMEO:
1452: {
1453: struct timeval *tv;
1454: short val;
1455:
1.36 perry 1456: if (m == NULL || m->m_len < sizeof(*tv)) {
1.1 cgd 1457: error = EINVAL;
1458: goto bad;
1459: }
1460: tv = mtod(m, struct timeval *);
1.75 itojun 1461: if (tv->tv_sec > (SHRT_MAX - tv->tv_usec / tick) / hz) {
1.1 cgd 1462: error = EDOM;
1463: goto bad;
1464: }
1465: val = tv->tv_sec * hz + tv->tv_usec / tick;
1.74 itojun 1466: if (val == 0 && tv->tv_usec != 0)
1467: val = 1;
1.1 cgd 1468:
1469: switch (optname) {
1470:
1471: case SO_SNDTIMEO:
1472: so->so_snd.sb_timeo = val;
1473: break;
1474: case SO_RCVTIMEO:
1475: so->so_rcv.sb_timeo = val;
1476: break;
1477: }
1478: break;
1479: }
1480:
1481: default:
1482: error = ENOPROTOOPT;
1483: break;
1484: }
1.15 mycroft 1485: if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1486: (void) ((*so->so_proto->pr_ctloutput)
1487: (PRCO_SETOPT, so, level, optname, &m0));
1488: m = NULL; /* freed by protocol */
1489: }
1.1 cgd 1490: }
1.54 lukem 1491: bad:
1.1 cgd 1492: if (m)
1493: (void) m_free(m);
1494: return (error);
1495: }
1496:
1.14 mycroft 1497: int
1.54 lukem 1498: sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1.1 cgd 1499: {
1.54 lukem 1500: struct mbuf *m;
1.1 cgd 1501:
1502: if (level != SOL_SOCKET) {
1503: if (so->so_proto && so->so_proto->pr_ctloutput) {
1504: return ((*so->so_proto->pr_ctloutput)
1505: (PRCO_GETOPT, so, level, optname, mp));
1506: } else
1507: return (ENOPROTOOPT);
1508: } else {
1509: m = m_get(M_WAIT, MT_SOOPTS);
1.36 perry 1510: m->m_len = sizeof(int);
1.1 cgd 1511:
1512: switch (optname) {
1513:
1514: case SO_LINGER:
1.36 perry 1515: m->m_len = sizeof(struct linger);
1.1 cgd 1516: mtod(m, struct linger *)->l_onoff =
1517: so->so_options & SO_LINGER;
1518: mtod(m, struct linger *)->l_linger = so->so_linger;
1519: break;
1520:
1521: case SO_USELOOPBACK:
1522: case SO_DONTROUTE:
1523: case SO_DEBUG:
1524: case SO_KEEPALIVE:
1525: case SO_REUSEADDR:
1.15 mycroft 1526: case SO_REUSEPORT:
1.1 cgd 1527: case SO_BROADCAST:
1528: case SO_OOBINLINE:
1.26 thorpej 1529: case SO_TIMESTAMP:
1.1 cgd 1530: *mtod(m, int *) = so->so_options & optname;
1531: break;
1532:
1533: case SO_TYPE:
1534: *mtod(m, int *) = so->so_type;
1535: break;
1536:
1537: case SO_ERROR:
1538: *mtod(m, int *) = so->so_error;
1539: so->so_error = 0;
1540: break;
1541:
1542: case SO_SNDBUF:
1543: *mtod(m, int *) = so->so_snd.sb_hiwat;
1544: break;
1545:
1546: case SO_RCVBUF:
1547: *mtod(m, int *) = so->so_rcv.sb_hiwat;
1548: break;
1549:
1550: case SO_SNDLOWAT:
1551: *mtod(m, int *) = so->so_snd.sb_lowat;
1552: break;
1553:
1554: case SO_RCVLOWAT:
1555: *mtod(m, int *) = so->so_rcv.sb_lowat;
1556: break;
1557:
1558: case SO_SNDTIMEO:
1559: case SO_RCVTIMEO:
1560: {
1561: int val = (optname == SO_SNDTIMEO ?
1562: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1563:
1564: m->m_len = sizeof(struct timeval);
1565: mtod(m, struct timeval *)->tv_sec = val / hz;
1566: mtod(m, struct timeval *)->tv_usec =
1.27 kleink 1567: (val % hz) * tick;
1.1 cgd 1568: break;
1569: }
1570:
1571: default:
1572: (void)m_free(m);
1573: return (ENOPROTOOPT);
1574: }
1575: *mp = m;
1576: return (0);
1577: }
1578: }
1579:
1.14 mycroft 1580: void
1.54 lukem 1581: sohasoutofband(struct socket *so)
1.1 cgd 1582: {
1.90 christos 1583: fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1.2 cgd 1584: selwakeup(&so->so_rcv.sb_sel);
1.1 cgd 1585: }
1.72 jdolecek 1586:
1587: static void
1588: filt_sordetach(struct knote *kn)
1589: {
1590: struct socket *so;
1591:
1592: so = (struct socket *)kn->kn_fp->f_data;
1.73 christos 1593: SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1594: if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1.72 jdolecek 1595: so->so_rcv.sb_flags &= ~SB_KNOTE;
1596: }
1597:
1598: /*ARGSUSED*/
1599: static int
1600: filt_soread(struct knote *kn, long hint)
1601: {
1602: struct socket *so;
1603:
1604: so = (struct socket *)kn->kn_fp->f_data;
1605: kn->kn_data = so->so_rcv.sb_cc;
1606: if (so->so_state & SS_CANTRCVMORE) {
1607: kn->kn_flags |= EV_EOF;
1608: kn->kn_fflags = so->so_error;
1609: return (1);
1610: }
1611: if (so->so_error) /* temporary udp error */
1612: return (1);
1613: if (kn->kn_sfflags & NOTE_LOWAT)
1614: return (kn->kn_data >= kn->kn_sdata);
1615: return (kn->kn_data >= so->so_rcv.sb_lowat);
1616: }
1617:
1618: static void
1619: filt_sowdetach(struct knote *kn)
1620: {
1621: struct socket *so;
1622:
1623: so = (struct socket *)kn->kn_fp->f_data;
1.73 christos 1624: SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1625: if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1.72 jdolecek 1626: so->so_snd.sb_flags &= ~SB_KNOTE;
1627: }
1628:
1629: /*ARGSUSED*/
1630: static int
1631: filt_sowrite(struct knote *kn, long hint)
1632: {
1633: struct socket *so;
1634:
1635: so = (struct socket *)kn->kn_fp->f_data;
1636: kn->kn_data = sbspace(&so->so_snd);
1637: if (so->so_state & SS_CANTSENDMORE) {
1638: kn->kn_flags |= EV_EOF;
1639: kn->kn_fflags = so->so_error;
1640: return (1);
1641: }
1642: if (so->so_error) /* temporary udp error */
1643: return (1);
1644: if (((so->so_state & SS_ISCONNECTED) == 0) &&
1645: (so->so_proto->pr_flags & PR_CONNREQUIRED))
1646: return (0);
1647: if (kn->kn_sfflags & NOTE_LOWAT)
1648: return (kn->kn_data >= kn->kn_sdata);
1649: return (kn->kn_data >= so->so_snd.sb_lowat);
1650: }
1651:
1652: /*ARGSUSED*/
1653: static int
1654: filt_solisten(struct knote *kn, long hint)
1655: {
1656: struct socket *so;
1657:
1658: so = (struct socket *)kn->kn_fp->f_data;
1659:
1660: /*
1661: * Set kn_data to number of incoming connections, not
1662: * counting partial (incomplete) connections.
1663: */
1664: kn->kn_data = so->so_qlen;
1665: return (kn->kn_data > 0);
1666: }
1667:
1668: static const struct filterops solisten_filtops =
1669: { 1, NULL, filt_sordetach, filt_solisten };
1670: static const struct filterops soread_filtops =
1671: { 1, NULL, filt_sordetach, filt_soread };
1672: static const struct filterops sowrite_filtops =
1673: { 1, NULL, filt_sowdetach, filt_sowrite };
1674:
1675: int
1676: soo_kqfilter(struct file *fp, struct knote *kn)
1677: {
1678: struct socket *so;
1679: struct sockbuf *sb;
1680:
1681: so = (struct socket *)kn->kn_fp->f_data;
1682: switch (kn->kn_filter) {
1683: case EVFILT_READ:
1684: if (so->so_options & SO_ACCEPTCONN)
1685: kn->kn_fop = &solisten_filtops;
1686: else
1687: kn->kn_fop = &soread_filtops;
1688: sb = &so->so_rcv;
1689: break;
1690: case EVFILT_WRITE:
1691: kn->kn_fop = &sowrite_filtops;
1692: sb = &so->so_snd;
1693: break;
1694: default:
1695: return (1);
1696: }
1.73 christos 1697: SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1.72 jdolecek 1698: sb->sb_flags |= SB_KNOTE;
1699: return (0);
1700: }
1701:
CVSweb <webmaster@jp.NetBSD.org>