Annotation of src/sys/kern/uipc_socket.c, Revision 1.222
1.222 ! rmind 1: /* $NetBSD: uipc_socket.c,v 1.221 2014/02/25 18:30:11 pooka 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:
1.222 ! rmind 65: /*
! 66: * Socket operation routines.
! 67: *
! 68: * These routines are called by the routines in sys_socket.c or from a
! 69: * system process, and implement the semantics of socket operations by
! 70: * switching out to the protocol specific routines.
! 71: */
! 72:
1.59 lukem 73: #include <sys/cdefs.h>
1.222 ! rmind 74: __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.221 2014/02/25 18:30:11 pooka Exp $");
1.64 thorpej 75:
1.179 christos 76: #include "opt_compat_netbsd.h"
1.64 thorpej 77: #include "opt_sock_counters.h"
78: #include "opt_sosend_loan.h"
1.81 martin 79: #include "opt_mbuftrace.h"
1.84 ragge 80: #include "opt_somaxkva.h"
1.167 ad 81: #include "opt_multiprocessor.h" /* XXX */
1.1 cgd 82:
1.9 mycroft 83: #include <sys/param.h>
84: #include <sys/systm.h>
85: #include <sys/proc.h>
86: #include <sys/file.h>
1.142 dyoung 87: #include <sys/filedesc.h>
1.173 plunky 88: #include <sys/kmem.h>
1.9 mycroft 89: #include <sys/mbuf.h>
90: #include <sys/domain.h>
91: #include <sys/kernel.h>
92: #include <sys/protosw.h>
93: #include <sys/socket.h>
94: #include <sys/socketvar.h>
1.21 christos 95: #include <sys/signalvar.h>
1.9 mycroft 96: #include <sys/resourcevar.h>
1.174 pooka 97: #include <sys/uidinfo.h>
1.72 jdolecek 98: #include <sys/event.h>
1.89 christos 99: #include <sys/poll.h>
1.118 elad 100: #include <sys/kauth.h>
1.136 ad 101: #include <sys/mutex.h>
102: #include <sys/condvar.h>
1.205 bouyer 103: #include <sys/kthread.h>
1.37 thorpej 104:
1.179 christos 105: #ifdef COMPAT_50
106: #include <compat/sys/time.h>
1.184 christos 107: #include <compat/sys/socket.h>
1.179 christos 108: #endif
109:
1.202 uebayasi 110: #include <uvm/uvm_extern.h>
111: #include <uvm/uvm_loan.h>
112: #include <uvm/uvm_page.h>
1.64 thorpej 113:
1.77 thorpej 114: MALLOC_DEFINE(M_SONAME, "soname", "socket name");
1.37 thorpej 115:
1.142 dyoung 116: extern const struct fileops socketops;
117:
1.54 lukem 118: extern int somaxconn; /* patchable (XXX sysctl) */
119: int somaxconn = SOMAXCONN;
1.160 ad 120: kmutex_t *softnet_lock;
1.49 jonathan 121:
1.64 thorpej 122: #ifdef SOSEND_COUNTERS
123: #include <sys/device.h>
124:
1.113 thorpej 125: static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 126: NULL, "sosend", "loan big");
1.113 thorpej 127: static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 128: NULL, "sosend", "copy big");
1.113 thorpej 129: static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 130: NULL, "sosend", "copy small");
1.113 thorpej 131: static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 132: NULL, "sosend", "kva limit");
133:
134: #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++
135:
1.101 matt 136: EVCNT_ATTACH_STATIC(sosend_loan_big);
137: EVCNT_ATTACH_STATIC(sosend_copy_big);
138: EVCNT_ATTACH_STATIC(sosend_copy_small);
139: EVCNT_ATTACH_STATIC(sosend_kvalimit);
1.64 thorpej 140: #else
141:
142: #define SOSEND_COUNTER_INCR(ev) /* nothing */
143:
144: #endif /* SOSEND_COUNTERS */
145:
1.167 ad 146: #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
1.121 yamt 147: int sock_loan_thresh = -1;
1.71 thorpej 148: #else
1.121 yamt 149: int sock_loan_thresh = 4096;
1.65 thorpej 150: #endif
1.64 thorpej 151:
1.136 ad 152: static kmutex_t so_pendfree_lock;
1.205 bouyer 153: static struct mbuf *so_pendfree = NULL;
1.64 thorpej 154:
1.84 ragge 155: #ifndef SOMAXKVA
156: #define SOMAXKVA (16 * 1024 * 1024)
157: #endif
158: int somaxkva = SOMAXKVA;
1.113 thorpej 159: static int socurkva;
1.136 ad 160: static kcondvar_t socurkva_cv;
1.64 thorpej 161:
1.191 elad 162: static kauth_listener_t socket_listener;
163:
1.64 thorpej 164: #define SOCK_LOAN_CHUNK 65536
165:
1.205 bouyer 166: static void sopendfree_thread(void *);
167: static kcondvar_t pendfree_thread_cv;
168: static lwp_t *sopendfree_lwp;
1.93 yamt 169:
1.212 pooka 170: static void sysctl_kern_socket_setup(void);
1.178 pooka 171: static struct sysctllog *socket_sysctllog;
172:
1.113 thorpej 173: static vsize_t
1.129 yamt 174: sokvareserve(struct socket *so, vsize_t len)
1.80 yamt 175: {
1.98 christos 176: int error;
1.80 yamt 177:
1.136 ad 178: mutex_enter(&so_pendfree_lock);
1.80 yamt 179: while (socurkva + len > somaxkva) {
180: SOSEND_COUNTER_INCR(&sosend_kvalimit);
1.136 ad 181: error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
1.98 christos 182: if (error) {
183: len = 0;
184: break;
185: }
1.80 yamt 186: }
1.93 yamt 187: socurkva += len;
1.136 ad 188: mutex_exit(&so_pendfree_lock);
1.98 christos 189: return len;
1.95 yamt 190: }
191:
1.113 thorpej 192: static void
1.95 yamt 193: sokvaunreserve(vsize_t len)
194: {
195:
1.136 ad 196: mutex_enter(&so_pendfree_lock);
1.95 yamt 197: socurkva -= len;
1.136 ad 198: cv_broadcast(&socurkva_cv);
199: mutex_exit(&so_pendfree_lock);
1.95 yamt 200: }
201:
202: /*
203: * sokvaalloc: allocate kva for loan.
204: */
205:
206: vaddr_t
1.209 matt 207: sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
1.95 yamt 208: {
209: vaddr_t lva;
210:
211: /*
212: * reserve kva.
213: */
214:
1.98 christos 215: if (sokvareserve(so, len) == 0)
216: return 0;
1.93 yamt 217:
218: /*
219: * allocate kva.
220: */
1.80 yamt 221:
1.209 matt 222: lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
223: UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
1.95 yamt 224: if (lva == 0) {
225: sokvaunreserve(len);
1.80 yamt 226: return (0);
1.95 yamt 227: }
1.80 yamt 228:
229: return lva;
230: }
231:
1.93 yamt 232: /*
233: * sokvafree: free kva for loan.
234: */
235:
1.80 yamt 236: void
237: sokvafree(vaddr_t sva, vsize_t len)
238: {
1.93 yamt 239:
240: /*
241: * free kva.
242: */
1.80 yamt 243:
1.109 yamt 244: uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
1.93 yamt 245:
246: /*
247: * unreserve kva.
248: */
249:
1.95 yamt 250: sokvaunreserve(len);
1.80 yamt 251: }
252:
1.64 thorpej 253: static void
1.134 christos 254: sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
1.64 thorpej 255: {
1.156 yamt 256: vaddr_t sva, eva;
1.64 thorpej 257: vsize_t len;
1.156 yamt 258: int npgs;
259:
260: KASSERT(pgs != NULL);
1.64 thorpej 261:
262: eva = round_page((vaddr_t) buf + size);
263: sva = trunc_page((vaddr_t) buf);
264: len = eva - sva;
265: npgs = len >> PAGE_SHIFT;
266:
267: pmap_kremove(sva, len);
268: pmap_update(pmap_kernel());
269: uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
1.80 yamt 270: sokvafree(sva, len);
1.64 thorpej 271: }
272:
1.93 yamt 273: /*
1.205 bouyer 274: * sopendfree_thread: free mbufs on "pendfree" list.
1.136 ad 275: * unlock and relock so_pendfree_lock when freeing mbufs.
1.93 yamt 276: */
277:
1.205 bouyer 278: static void
279: sopendfree_thread(void *v)
1.93 yamt 280: {
1.137 ad 281: struct mbuf *m, *next;
1.205 bouyer 282: size_t rv;
1.93 yamt 283:
1.205 bouyer 284: mutex_enter(&so_pendfree_lock);
1.64 thorpej 285:
1.205 bouyer 286: for (;;) {
287: rv = 0;
288: while (so_pendfree != NULL) {
289: m = so_pendfree;
290: so_pendfree = NULL;
291: mutex_exit(&so_pendfree_lock);
292:
293: for (; m != NULL; m = next) {
294: next = m->m_next;
295: KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0);
296: KASSERT(m->m_ext.ext_refcnt == 0);
297:
298: rv += m->m_ext.ext_size;
299: sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
300: m->m_ext.ext_size);
301: pool_cache_put(mb_cache, m);
302: }
1.93 yamt 303:
1.205 bouyer 304: mutex_enter(&so_pendfree_lock);
1.93 yamt 305: }
1.205 bouyer 306: if (rv)
307: cv_broadcast(&socurkva_cv);
308: cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
1.64 thorpej 309: }
1.205 bouyer 310: panic("sopendfree_thread");
311: /* NOTREACHED */
1.64 thorpej 312: }
313:
1.80 yamt 314: void
1.134 christos 315: soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
1.64 thorpej 316: {
317:
1.156 yamt 318: KASSERT(m != NULL);
1.64 thorpej 319:
1.93 yamt 320: /*
321: * postpone freeing mbuf.
322: *
323: * we can't do it in interrupt context
324: * because we need to put kva back to kernel_map.
325: */
326:
1.136 ad 327: mutex_enter(&so_pendfree_lock);
1.92 yamt 328: m->m_next = so_pendfree;
329: so_pendfree = m;
1.205 bouyer 330: cv_signal(&pendfree_thread_cv);
1.136 ad 331: mutex_exit(&so_pendfree_lock);
1.64 thorpej 332: }
333:
334: static long
335: sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
336: {
337: struct iovec *iov = uio->uio_iov;
338: vaddr_t sva, eva;
339: vsize_t len;
1.156 yamt 340: vaddr_t lva;
341: int npgs, error;
342: vaddr_t va;
343: int i;
1.64 thorpej 344:
1.116 yamt 345: if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
1.64 thorpej 346: return (0);
347:
348: if (iov->iov_len < (size_t) space)
349: space = iov->iov_len;
350: if (space > SOCK_LOAN_CHUNK)
351: space = SOCK_LOAN_CHUNK;
352:
353: eva = round_page((vaddr_t) iov->iov_base + space);
354: sva = trunc_page((vaddr_t) iov->iov_base);
355: len = eva - sva;
356: npgs = len >> PAGE_SHIFT;
357:
1.79 thorpej 358: KASSERT(npgs <= M_EXT_MAXPAGES);
359:
1.209 matt 360: lva = sokvaalloc(sva, len, so);
1.64 thorpej 361: if (lva == 0)
1.80 yamt 362: return 0;
1.64 thorpej 363:
1.116 yamt 364: error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
1.79 thorpej 365: m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
1.64 thorpej 366: if (error) {
1.80 yamt 367: sokvafree(lva, len);
1.64 thorpej 368: return (0);
369: }
370:
371: for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
1.79 thorpej 372: pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
1.194 cegger 373: VM_PROT_READ, 0);
1.64 thorpej 374: pmap_update(pmap_kernel());
375:
376: lva += (vaddr_t) iov->iov_base & PAGE_MASK;
377:
1.134 christos 378: MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
1.79 thorpej 379: m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
1.64 thorpej 380:
381: uio->uio_resid -= space;
382: /* uio_offset not updated, not set/used for write(2) */
1.134 christos 383: uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
1.64 thorpej 384: uio->uio_iov->iov_len -= space;
385: if (uio->uio_iov->iov_len == 0) {
386: uio->uio_iov++;
387: uio->uio_iovcnt--;
388: }
389:
390: return (space);
391: }
392:
1.142 dyoung 393: struct mbuf *
1.147 dyoung 394: getsombuf(struct socket *so, int type)
1.142 dyoung 395: {
396: struct mbuf *m;
397:
1.147 dyoung 398: m = m_get(M_WAIT, type);
1.142 dyoung 399: MCLAIM(m, so->so_mowner);
400: return m;
401: }
402:
1.191 elad 403: static int
404: socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
405: void *arg0, void *arg1, void *arg2, void *arg3)
406: {
407: int result;
408: enum kauth_network_req req;
409:
410: result = KAUTH_RESULT_DEFER;
411: req = (enum kauth_network_req)arg0;
412:
1.193 elad 413: if ((action != KAUTH_NETWORK_SOCKET) &&
414: (action != KAUTH_NETWORK_BIND))
1.191 elad 415: return result;
416:
417: switch (req) {
1.193 elad 418: case KAUTH_REQ_NETWORK_BIND_PORT:
419: result = KAUTH_RESULT_ALLOW;
420: break;
421:
1.191 elad 422: case KAUTH_REQ_NETWORK_SOCKET_DROP: {
423: /* Normal users can only drop their own connections. */
424: struct socket *so = (struct socket *)arg1;
425:
1.220 christos 426: if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
1.191 elad 427: result = KAUTH_RESULT_ALLOW;
428:
429: break;
430: }
431:
432: case KAUTH_REQ_NETWORK_SOCKET_OPEN:
433: /* We allow "raw" routing/bluetooth sockets to anyone. */
1.203 matt 434: if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE
435: || (u_long)arg1 == PF_BLUETOOTH) {
1.191 elad 436: result = KAUTH_RESULT_ALLOW;
1.203 matt 437: } else {
1.191 elad 438: /* Privileged, let secmodel handle this. */
439: if ((u_long)arg2 == SOCK_RAW)
440: break;
441: }
442:
443: result = KAUTH_RESULT_ALLOW;
444:
445: break;
446:
1.192 elad 447: case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
448: result = KAUTH_RESULT_ALLOW;
449:
450: break;
451:
1.191 elad 452: default:
453: break;
454: }
455:
456: return result;
457: }
458:
1.119 yamt 459: void
460: soinit(void)
461: {
462:
1.212 pooka 463: sysctl_kern_socket_setup();
1.178 pooka 464:
1.148 ad 465: mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
1.160 ad 466: softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
1.136 ad 467: cv_init(&socurkva_cv, "sokva");
1.205 bouyer 468: cv_init(&pendfree_thread_cv, "sopendfr");
1.166 ad 469: soinit2();
1.136 ad 470:
1.119 yamt 471: /* Set the initial adjusted socket buffer size. */
472: if (sb_max_set(sb_max))
473: panic("bad initial sb_max value: %lu", sb_max);
474:
1.191 elad 475: socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
476: socket_listener_cb, NULL);
1.119 yamt 477: }
478:
1.205 bouyer 479: void
480: soinit1(void)
481: {
482: int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
483: sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
484: if (error)
485: panic("soinit1 %d", error);
486: }
487:
1.1 cgd 488: /*
1.222 ! rmind 489: * socreate: create a new socket of the specified type and the protocol.
! 490: *
! 491: * => Caller may specify another socket for lock sharing (must not be held).
! 492: * => Returns the new socket without lock held.
! 493: */
1.3 andrew 494: int
1.160 ad 495: socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
496: struct socket *lockso)
1.1 cgd 497: {
1.99 matt 498: const struct protosw *prp;
1.54 lukem 499: struct socket *so;
1.115 yamt 500: uid_t uid;
1.160 ad 501: int error;
502: kmutex_t *lock;
1.1 cgd 503:
1.132 elad 504: error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
505: KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
506: KAUTH_ARG(proto));
1.140 dyoung 507: if (error != 0)
508: return error;
1.127 elad 509:
1.1 cgd 510: if (proto)
511: prp = pffindproto(dom, proto, type);
512: else
513: prp = pffindtype(dom, type);
1.140 dyoung 514: if (prp == NULL) {
1.120 ginsbach 515: /* no support for domain */
516: if (pffinddomain(dom) == 0)
1.140 dyoung 517: return EAFNOSUPPORT;
1.120 ginsbach 518: /* no support for socket type */
519: if (proto == 0 && type != 0)
1.140 dyoung 520: return EPROTOTYPE;
521: return EPROTONOSUPPORT;
1.120 ginsbach 522: }
1.140 dyoung 523: if (prp->pr_usrreq == NULL)
524: return EPROTONOSUPPORT;
1.1 cgd 525: if (prp->pr_type != type)
1.140 dyoung 526: return EPROTOTYPE;
1.160 ad 527:
528: so = soget(true);
1.1 cgd 529: so->so_type = type;
530: so->so_proto = prp;
1.33 matt 531: so->so_send = sosend;
532: so->so_receive = soreceive;
1.78 matt 533: #ifdef MBUFTRACE
534: so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
535: so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
536: so->so_mowner = &prp->pr_domain->dom_mowner;
537: #endif
1.138 rmind 538: uid = kauth_cred_geteuid(l->l_cred);
1.115 yamt 539: so->so_uidinfo = uid_find(uid);
1.168 yamt 540: so->so_cpid = l->l_proc->p_pid;
1.160 ad 541: if (lockso != NULL) {
542: /* Caller wants us to share a lock. */
543: lock = lockso->so_lock;
544: so->so_lock = lock;
545: mutex_obj_hold(lock);
1.214 gdt 546: /* XXX Why is this not solock, to match sounlock? */
1.160 ad 547: mutex_enter(lock);
548: } else {
549: /* Lock assigned and taken during PRU_ATTACH. */
550: }
1.140 dyoung 551: error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
552: (struct mbuf *)(long)proto, NULL, l);
1.160 ad 553: KASSERT(solocked(so));
1.140 dyoung 554: if (error != 0) {
1.222 ! rmind 555: KASSERT(so->so_pcb == NULL);
1.1 cgd 556: so->so_state |= SS_NOFDREF;
557: sofree(so);
1.140 dyoung 558: return error;
1.1 cgd 559: }
1.198 elad 560: so->so_cred = kauth_cred_dup(l->l_cred);
1.160 ad 561: sounlock(so);
1.1 cgd 562: *aso = so;
1.140 dyoung 563: return 0;
1.1 cgd 564: }
565:
1.222 ! rmind 566: /*
! 567: * fsocreate: create a socket and a file descriptor associated with it.
! 568: *
! 569: * => On success, write file descriptor to fdout and return zero.
! 570: * => On failure, return non-zero; *fdout will be undefined.
1.142 dyoung 571: */
572: int
1.222 ! rmind 573: fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
1.142 dyoung 574: {
1.222 ! rmind 575: lwp_t *l = curlwp;
! 576: int error, fd, flags;
! 577: struct socket *so;
! 578: struct file *fp;
1.142 dyoung 579:
1.222 ! rmind 580: if ((error = fd_allocfile(&fp, &fd)) != 0) {
1.204 christos 581: return error;
1.222 ! rmind 582: }
! 583: flags = type & SOCK_FLAGS_MASK;
1.204 christos 584: fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
1.207 christos 585: fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
586: ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
1.142 dyoung 587: fp->f_type = DTYPE_SOCKET;
588: fp->f_ops = &socketops;
1.222 ! rmind 589:
! 590: type &= ~SOCK_FLAGS_MASK;
! 591: error = socreate(domain, &so, type, proto, l, NULL);
! 592: if (error) {
1.155 ad 593: fd_abort(curproc, fp, fd);
1.222 ! rmind 594: return error;
! 595: }
! 596: if (flags & SOCK_NONBLOCK) {
! 597: so->so_state |= SS_NBIO;
! 598: }
! 599: fp->f_data = so;
! 600: fd_affix(curproc, fp, fd);
! 601:
! 602: if (sop != NULL) {
! 603: *sop = so;
1.142 dyoung 604: }
1.222 ! rmind 605: *fdout = fd;
1.142 dyoung 606: return error;
607: }
608:
1.3 andrew 609: int
1.190 dyoung 610: sofamily(const struct socket *so)
611: {
612: const struct protosw *pr;
613: const struct domain *dom;
614:
615: if ((pr = so->so_proto) == NULL)
616: return AF_UNSPEC;
617: if ((dom = pr->pr_domain) == NULL)
618: return AF_UNSPEC;
619: return dom->dom_family;
620: }
621:
622: int
1.114 christos 623: sobind(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 624: {
1.160 ad 625: int error;
1.1 cgd 626:
1.160 ad 627: solock(so);
1.140 dyoung 628: error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
1.160 ad 629: sounlock(so);
1.140 dyoung 630: return error;
1.1 cgd 631: }
632:
1.3 andrew 633: int
1.150 elad 634: solisten(struct socket *so, int backlog, struct lwp *l)
1.1 cgd 635: {
1.160 ad 636: int error;
1.1 cgd 637:
1.160 ad 638: solock(so);
1.158 ad 639: if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1.163 ad 640: SS_ISDISCONNECTING)) != 0) {
1.222 ! rmind 641: sounlock(so);
! 642: return EINVAL;
1.163 ad 643: }
1.140 dyoung 644: error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
1.150 elad 645: NULL, NULL, l);
1.140 dyoung 646: if (error != 0) {
1.160 ad 647: sounlock(so);
1.140 dyoung 648: return error;
1.1 cgd 649: }
1.63 matt 650: if (TAILQ_EMPTY(&so->so_q))
1.1 cgd 651: so->so_options |= SO_ACCEPTCONN;
652: if (backlog < 0)
653: backlog = 0;
1.49 jonathan 654: so->so_qlimit = min(backlog, somaxconn);
1.160 ad 655: sounlock(so);
1.140 dyoung 656: return 0;
1.1 cgd 657: }
658:
1.21 christos 659: void
1.54 lukem 660: sofree(struct socket *so)
1.1 cgd 661: {
1.161 ad 662: u_int refs;
1.1 cgd 663:
1.160 ad 664: KASSERT(solocked(so));
665:
666: if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
667: sounlock(so);
1.1 cgd 668: return;
1.160 ad 669: }
1.43 mycroft 670: if (so->so_head) {
671: /*
672: * We must not decommission a socket that's on the accept(2)
673: * queue. If we do, then accept(2) may hang after select(2)
674: * indicated that the listening socket was ready.
675: */
1.160 ad 676: if (!soqremque(so, 0)) {
677: sounlock(so);
1.43 mycroft 678: return;
1.160 ad 679: }
1.43 mycroft 680: }
1.98 christos 681: if (so->so_rcv.sb_hiwat)
1.110 christos 682: (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
1.98 christos 683: RLIM_INFINITY);
684: if (so->so_snd.sb_hiwat)
1.110 christos 685: (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
1.98 christos 686: RLIM_INFINITY);
687: sbrelease(&so->so_snd, so);
1.160 ad 688: KASSERT(!cv_has_waiters(&so->so_cv));
689: KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
690: KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
1.1 cgd 691: sorflush(so);
1.161 ad 692: refs = so->so_aborting; /* XXX */
1.177 ad 693: /* Remove acccept filter if one is present. */
1.170 tls 694: if (so->so_accf != NULL)
1.177 ad 695: (void)accept_filt_clear(so);
1.160 ad 696: sounlock(so);
1.161 ad 697: if (refs == 0) /* XXX */
698: soput(so);
1.1 cgd 699: }
700:
701: /*
1.222 ! rmind 702: * soclose: close a socket on last file table reference removal.
! 703: * Initiate disconnect if connected. Free socket when disconnect complete.
1.1 cgd 704: */
1.3 andrew 705: int
1.54 lukem 706: soclose(struct socket *so)
1.1 cgd 707: {
1.222 ! rmind 708: struct socket *so2;
! 709: int error = 0;
1.1 cgd 710:
1.160 ad 711: solock(so);
1.1 cgd 712: if (so->so_options & SO_ACCEPTCONN) {
1.172 ad 713: for (;;) {
714: if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
1.160 ad 715: KASSERT(solocked2(so, so2));
716: (void) soqremque(so2, 0);
717: /* soabort drops the lock. */
718: (void) soabort(so2);
719: solock(so);
1.172 ad 720: continue;
1.160 ad 721: }
1.172 ad 722: if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
1.160 ad 723: KASSERT(solocked2(so, so2));
724: (void) soqremque(so2, 1);
725: /* soabort drops the lock. */
726: (void) soabort(so2);
727: solock(so);
1.172 ad 728: continue;
1.160 ad 729: }
1.172 ad 730: break;
731: }
1.1 cgd 732: }
1.222 ! rmind 733: if (so->so_pcb == NULL)
1.1 cgd 734: goto discard;
735: if (so->so_state & SS_ISCONNECTED) {
736: if ((so->so_state & SS_ISDISCONNECTING) == 0) {
737: error = sodisconnect(so);
738: if (error)
739: goto drop;
740: }
741: if (so->so_options & SO_LINGER) {
1.206 christos 742: if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
743: (SS_ISDISCONNECTING|SS_NBIO))
1.1 cgd 744: goto drop;
1.21 christos 745: while (so->so_state & SS_ISCONNECTED) {
1.185 yamt 746: error = sowait(so, true, so->so_linger * hz);
1.21 christos 747: if (error)
1.1 cgd 748: break;
1.21 christos 749: }
1.1 cgd 750: }
751: }
1.54 lukem 752: drop:
1.1 cgd 753: if (so->so_pcb) {
1.222 ! rmind 754: int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
1.140 dyoung 755: NULL, NULL, NULL, NULL);
1.1 cgd 756: if (error == 0)
757: error = error2;
758: }
1.54 lukem 759: discard:
1.222 ! rmind 760: KASSERT((so->so_state & SS_NOFDREF) == 0);
1.198 elad 761: kauth_cred_free(so->so_cred);
1.1 cgd 762: so->so_state |= SS_NOFDREF;
763: sofree(so);
1.222 ! rmind 764: return error;
1.1 cgd 765: }
766:
767: /*
1.160 ad 768: * Must be called with the socket locked.. Will return with it unlocked.
1.1 cgd 769: */
1.3 andrew 770: int
1.54 lukem 771: soabort(struct socket *so)
1.1 cgd 772: {
1.161 ad 773: u_int refs;
1.139 yamt 774: int error;
1.160 ad 775:
776: KASSERT(solocked(so));
777: KASSERT(so->so_head == NULL);
1.1 cgd 778:
1.161 ad 779: so->so_aborting++; /* XXX */
1.140 dyoung 780: error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
781: NULL, NULL, NULL);
1.161 ad 782: refs = --so->so_aborting; /* XXX */
1.164 drochner 783: if (error || (refs == 0)) {
1.139 yamt 784: sofree(so);
1.160 ad 785: } else {
786: sounlock(so);
1.139 yamt 787: }
788: return error;
1.1 cgd 789: }
790:
1.3 andrew 791: int
1.54 lukem 792: soaccept(struct socket *so, struct mbuf *nam)
1.1 cgd 793: {
1.222 ! rmind 794: int error;
1.160 ad 795:
796: KASSERT(solocked(so));
1.222 ! rmind 797: KASSERT((so->so_state & SS_NOFDREF) != 0);
1.1 cgd 798:
799: so->so_state &= ~SS_NOFDREF;
1.55 thorpej 800: if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
801: (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
1.41 mycroft 802: error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
1.140 dyoung 803: NULL, nam, NULL, NULL);
1.41 mycroft 804: else
1.53 itojun 805: error = ECONNABORTED;
1.52 itojun 806:
1.222 ! rmind 807: return error;
1.1 cgd 808: }
809:
1.3 andrew 810: int
1.114 christos 811: soconnect(struct socket *so, struct mbuf *nam, struct lwp *l)
1.1 cgd 812: {
1.222 ! rmind 813: int error;
1.160 ad 814:
815: KASSERT(solocked(so));
1.1 cgd 816:
817: if (so->so_options & SO_ACCEPTCONN)
1.222 ! rmind 818: return EOPNOTSUPP;
1.1 cgd 819: /*
820: * If protocol is connection-based, can only connect once.
821: * Otherwise, if connected, try to disconnect first.
822: * This allows user to disconnect by connecting to, e.g.,
823: * a null address.
824: */
825: if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
826: ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
827: (error = sodisconnect(so))))
828: error = EISCONN;
829: else
830: error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
1.140 dyoung 831: NULL, nam, NULL, l);
1.222 ! rmind 832:
! 833: return error;
1.1 cgd 834: }
835:
1.3 andrew 836: int
1.54 lukem 837: soconnect2(struct socket *so1, struct socket *so2)
1.1 cgd 838: {
1.160 ad 839: KASSERT(solocked2(so1, so2));
1.1 cgd 840:
1.222 ! rmind 841: return (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
1.140 dyoung 842: NULL, (struct mbuf *)so2, NULL, NULL);
1.1 cgd 843: }
844:
1.3 andrew 845: int
1.54 lukem 846: sodisconnect(struct socket *so)
1.1 cgd 847: {
1.160 ad 848: int error;
849:
850: KASSERT(solocked(so));
1.1 cgd 851:
852: if ((so->so_state & SS_ISCONNECTED) == 0) {
853: error = ENOTCONN;
1.160 ad 854: } else if (so->so_state & SS_ISDISCONNECTING) {
1.1 cgd 855: error = EALREADY;
1.160 ad 856: } else {
857: error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
858: NULL, NULL, NULL, NULL);
1.1 cgd 859: }
860: return (error);
861: }
862:
1.15 mycroft 863: #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1.1 cgd 864: /*
865: * Send on a socket.
866: * If send must go all at once and message is larger than
867: * send buffering, then hard error.
868: * Lock against other senders.
869: * If must go all at once and not enough room now, then
870: * inform user that this would block and do nothing.
871: * Otherwise, if nonblocking, send as much as possible.
872: * The data to be sent is described by "uio" if nonzero,
873: * otherwise by the mbuf chain "top" (which must be null
874: * if uio is not). Data provided in mbuf chain must be small
875: * enough to send all at once.
876: *
877: * Returns nonzero on error, timeout or signal; callers
878: * must check for short counts if EINTR/ERESTART are returned.
879: * Data and control buffers are freed on return.
880: */
1.3 andrew 881: int
1.54 lukem 882: sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
1.114 christos 883: struct mbuf *control, int flags, struct lwp *l)
1.1 cgd 884: {
1.54 lukem 885: struct mbuf **mp, *m;
1.58 jdolecek 886: long space, len, resid, clen, mlen;
887: int error, s, dontroute, atomic;
1.196 dsl 888: short wakeup_state = 0;
1.54 lukem 889:
1.160 ad 890: clen = 0;
1.64 thorpej 891:
1.160 ad 892: /*
893: * solock() provides atomicity of access. splsoftnet() prevents
894: * protocol processing soft interrupts from interrupting us and
895: * blocking (expensive).
896: */
897: s = splsoftnet();
898: solock(so);
1.54 lukem 899: atomic = sosendallatonce(so) || top;
1.1 cgd 900: if (uio)
901: resid = uio->uio_resid;
902: else
903: resid = top->m_pkthdr.len;
1.7 cgd 904: /*
905: * In theory resid should be unsigned.
906: * However, space must be signed, as it might be less than 0
907: * if we over-committed, and we must use a signed comparison
908: * of space and resid. On the other hand, a negative resid
909: * causes us to loop sending 0-length segments to the protocol.
910: */
1.29 mycroft 911: if (resid < 0) {
912: error = EINVAL;
913: goto out;
914: }
1.1 cgd 915: dontroute =
916: (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
917: (so->so_proto->pr_flags & PR_ATOMIC);
1.165 christos 918: l->l_ru.ru_msgsnd++;
1.1 cgd 919: if (control)
920: clen = control->m_len;
1.54 lukem 921: restart:
1.21 christos 922: if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
1.1 cgd 923: goto out;
924: do {
1.160 ad 925: if (so->so_state & SS_CANTSENDMORE) {
926: error = EPIPE;
927: goto release;
928: }
1.48 thorpej 929: if (so->so_error) {
930: error = so->so_error;
931: so->so_error = 0;
932: goto release;
933: }
1.1 cgd 934: if ((so->so_state & SS_ISCONNECTED) == 0) {
935: if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1.217 rmind 936: if (resid || clen == 0) {
1.160 ad 937: error = ENOTCONN;
938: goto release;
939: }
940: } else if (addr == 0) {
941: error = EDESTADDRREQ;
942: goto release;
943: }
1.1 cgd 944: }
945: space = sbspace(&so->so_snd);
946: if (flags & MSG_OOB)
947: space += 1024;
1.21 christos 948: if ((atomic && resid > so->so_snd.sb_hiwat) ||
1.160 ad 949: clen > so->so_snd.sb_hiwat) {
950: error = EMSGSIZE;
951: goto release;
952: }
1.96 mycroft 953: if (space < resid + clen &&
1.1 cgd 954: (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1.206 christos 955: if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1.160 ad 956: error = EWOULDBLOCK;
957: goto release;
958: }
1.1 cgd 959: sbunlock(&so->so_snd);
1.196 dsl 960: if (wakeup_state & SS_RESTARTSYS) {
961: error = ERESTART;
962: goto out;
963: }
1.1 cgd 964: error = sbwait(&so->so_snd);
965: if (error)
966: goto out;
1.196 dsl 967: wakeup_state = so->so_state;
1.1 cgd 968: goto restart;
969: }
1.196 dsl 970: wakeup_state = 0;
1.1 cgd 971: mp = ⊤
972: space -= clen;
973: do {
1.45 tv 974: if (uio == NULL) {
975: /*
976: * Data is prepackaged in "top".
977: */
978: resid = 0;
979: if (flags & MSG_EOR)
980: top->m_flags |= M_EOR;
981: } else do {
1.160 ad 982: sounlock(so);
983: splx(s);
1.144 dyoung 984: if (top == NULL) {
1.78 matt 985: m = m_gethdr(M_WAIT, MT_DATA);
1.45 tv 986: mlen = MHLEN;
987: m->m_pkthdr.len = 0;
1.140 dyoung 988: m->m_pkthdr.rcvif = NULL;
1.45 tv 989: } else {
1.78 matt 990: m = m_get(M_WAIT, MT_DATA);
1.45 tv 991: mlen = MLEN;
992: }
1.78 matt 993: MCLAIM(m, so->so_snd.sb_mowner);
1.121 yamt 994: if (sock_loan_thresh >= 0 &&
995: uio->uio_iov->iov_len >= sock_loan_thresh &&
996: space >= sock_loan_thresh &&
1.64 thorpej 997: (len = sosend_loan(so, uio, m,
998: space)) != 0) {
999: SOSEND_COUNTER_INCR(&sosend_loan_big);
1000: space -= len;
1001: goto have_data;
1002: }
1.45 tv 1003: if (resid >= MINCLSIZE && space >= MCLBYTES) {
1.64 thorpej 1004: SOSEND_COUNTER_INCR(&sosend_copy_big);
1.201 oki 1005: m_clget(m, M_DONTWAIT);
1.45 tv 1006: if ((m->m_flags & M_EXT) == 0)
1007: goto nopages;
1008: mlen = MCLBYTES;
1009: if (atomic && top == 0) {
1.58 jdolecek 1010: len = lmin(MCLBYTES - max_hdr,
1.54 lukem 1011: resid);
1.45 tv 1012: m->m_data += max_hdr;
1013: } else
1.58 jdolecek 1014: len = lmin(MCLBYTES, resid);
1.45 tv 1015: space -= len;
1016: } else {
1.64 thorpej 1017: nopages:
1018: SOSEND_COUNTER_INCR(&sosend_copy_small);
1.58 jdolecek 1019: len = lmin(lmin(mlen, resid), space);
1.45 tv 1020: space -= len;
1021: /*
1022: * For datagram protocols, leave room
1023: * for protocol headers in first mbuf.
1024: */
1025: if (atomic && top == 0 && len < mlen)
1026: MH_ALIGN(m, len);
1027: }
1.144 dyoung 1028: error = uiomove(mtod(m, void *), (int)len, uio);
1.64 thorpej 1029: have_data:
1.45 tv 1030: resid = uio->uio_resid;
1031: m->m_len = len;
1032: *mp = m;
1033: top->m_pkthdr.len += len;
1.160 ad 1034: s = splsoftnet();
1035: solock(so);
1.144 dyoung 1036: if (error != 0)
1.45 tv 1037: goto release;
1038: mp = &m->m_next;
1039: if (resid <= 0) {
1040: if (flags & MSG_EOR)
1041: top->m_flags |= M_EOR;
1042: break;
1043: }
1044: } while (space > 0 && atomic);
1.108 perry 1045:
1.160 ad 1046: if (so->so_state & SS_CANTSENDMORE) {
1047: error = EPIPE;
1048: goto release;
1049: }
1.45 tv 1050: if (dontroute)
1051: so->so_options |= SO_DONTROUTE;
1052: if (resid > 0)
1053: so->so_state |= SS_MORETOCOME;
1.46 sommerfe 1054: error = (*so->so_proto->pr_usrreq)(so,
1055: (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
1.160 ad 1056: top, addr, control, curlwp);
1.45 tv 1057: if (dontroute)
1058: so->so_options &= ~SO_DONTROUTE;
1059: if (resid > 0)
1060: so->so_state &= ~SS_MORETOCOME;
1061: clen = 0;
1.144 dyoung 1062: control = NULL;
1063: top = NULL;
1.45 tv 1064: mp = ⊤
1.144 dyoung 1065: if (error != 0)
1.1 cgd 1066: goto release;
1067: } while (resid && space > 0);
1068: } while (resid);
1069:
1.54 lukem 1070: release:
1.1 cgd 1071: sbunlock(&so->so_snd);
1.54 lukem 1072: out:
1.160 ad 1073: sounlock(so);
1074: splx(s);
1.1 cgd 1075: if (top)
1076: m_freem(top);
1077: if (control)
1078: m_freem(control);
1079: return (error);
1080: }
1081:
1082: /*
1.159 ad 1083: * Following replacement or removal of the first mbuf on the first
1084: * mbuf chain of a socket buffer, push necessary state changes back
1085: * into the socket buffer so that other consumers see the values
1086: * consistently. 'nextrecord' is the callers locally stored value of
1087: * the original value of sb->sb_mb->m_nextpkt which must be restored
1088: * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL.
1089: */
1090: static void
1091: sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1092: {
1093:
1.160 ad 1094: KASSERT(solocked(sb->sb_so));
1095:
1.159 ad 1096: /*
1097: * First, update for the new value of nextrecord. If necessary,
1098: * make it the first record.
1099: */
1100: if (sb->sb_mb != NULL)
1101: sb->sb_mb->m_nextpkt = nextrecord;
1102: else
1103: sb->sb_mb = nextrecord;
1104:
1105: /*
1106: * Now update any dependent socket buffer fields to reflect
1107: * the new state. This is an inline of SB_EMPTY_FIXUP, with
1108: * the addition of a second clause that takes care of the
1109: * case where sb_mb has been updated, but remains the last
1110: * record.
1111: */
1112: if (sb->sb_mb == NULL) {
1113: sb->sb_mbtail = NULL;
1114: sb->sb_lastrecord = NULL;
1115: } else if (sb->sb_mb->m_nextpkt == NULL)
1116: sb->sb_lastrecord = sb->sb_mb;
1117: }
1118:
1119: /*
1.1 cgd 1120: * Implement receive operations on a socket.
1121: * We depend on the way that records are added to the sockbuf
1122: * by sbappend*. In particular, each record (mbufs linked through m_next)
1123: * must begin with an address if the protocol so specifies,
1124: * followed by an optional mbuf or mbufs containing ancillary data,
1125: * and then zero or more mbufs of data.
1126: * In order to avoid blocking network interrupts for the entire time here,
1127: * we splx() while doing the actual copy to user space.
1128: * Although the sockbuf is locked, new data may still be appended,
1129: * and thus we must maintain consistency of the sockbuf during that time.
1130: *
1131: * The caller may receive the data as a single mbuf chain by supplying
1132: * an mbuf **mp0 for use in returning the chain. The uio is then used
1133: * only for the count in uio_resid.
1134: */
1.3 andrew 1135: int
1.54 lukem 1136: soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1137: struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1.1 cgd 1138: {
1.116 yamt 1139: struct lwp *l = curlwp;
1.160 ad 1140: struct mbuf *m, **mp, *mt;
1.211 chs 1141: size_t len, offset, moff, orig_resid;
1142: int atomic, flags, error, s, type;
1.99 matt 1143: const struct protosw *pr;
1.54 lukem 1144: struct mbuf *nextrecord;
1.67 he 1145: int mbuf_removed = 0;
1.146 dyoung 1146: const struct domain *dom;
1.196 dsl 1147: short wakeup_state = 0;
1.64 thorpej 1148:
1.54 lukem 1149: pr = so->so_proto;
1.146 dyoung 1150: atomic = pr->pr_flags & PR_ATOMIC;
1151: dom = pr->pr_domain;
1.1 cgd 1152: mp = mp0;
1.54 lukem 1153: type = 0;
1154: orig_resid = uio->uio_resid;
1.102 jonathan 1155:
1.144 dyoung 1156: if (paddr != NULL)
1157: *paddr = NULL;
1158: if (controlp != NULL)
1159: *controlp = NULL;
1160: if (flagsp != NULL)
1.1 cgd 1161: flags = *flagsp &~ MSG_EOR;
1162: else
1163: flags = 0;
1.66 enami 1164:
1.1 cgd 1165: if (flags & MSG_OOB) {
1166: m = m_get(M_WAIT, MT_DATA);
1.160 ad 1167: solock(so);
1.17 cgd 1168: error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
1.140 dyoung 1169: (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
1.160 ad 1170: sounlock(so);
1.1 cgd 1171: if (error)
1172: goto bad;
1173: do {
1.134 christos 1174: error = uiomove(mtod(m, void *),
1.211 chs 1175: MIN(uio->uio_resid, m->m_len), uio);
1.1 cgd 1176: m = m_free(m);
1.144 dyoung 1177: } while (uio->uio_resid > 0 && error == 0 && m);
1.54 lukem 1178: bad:
1.144 dyoung 1179: if (m != NULL)
1.1 cgd 1180: m_freem(m);
1.144 dyoung 1181: return error;
1.1 cgd 1182: }
1.144 dyoung 1183: if (mp != NULL)
1.140 dyoung 1184: *mp = NULL;
1.160 ad 1185:
1186: /*
1187: * solock() provides atomicity of access. splsoftnet() prevents
1188: * protocol processing soft interrupts from interrupting us and
1189: * blocking (expensive).
1190: */
1191: s = splsoftnet();
1192: solock(so);
1.54 lukem 1193: restart:
1.160 ad 1194: if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1195: sounlock(so);
1196: splx(s);
1.144 dyoung 1197: return error;
1.160 ad 1198: }
1.1 cgd 1199:
1200: m = so->so_rcv.sb_mb;
1201: /*
1202: * If we have less data than requested, block awaiting more
1203: * (subject to any timeout) if:
1.15 mycroft 1204: * 1. the current count is less than the low water mark,
1.1 cgd 1205: * 2. MSG_WAITALL is set, and it is possible to do the entire
1.15 mycroft 1206: * receive operation at once if we block (resid <= hiwat), or
1207: * 3. MSG_DONTWAIT is not set.
1.1 cgd 1208: * If MSG_WAITALL is set but resid is larger than the receive buffer,
1209: * we have to do the receive in sections, and thus risk returning
1210: * a short count if a timeout or signal occurs after we start.
1211: */
1.144 dyoung 1212: if (m == NULL ||
1213: ((flags & MSG_DONTWAIT) == 0 &&
1214: so->so_rcv.sb_cc < uio->uio_resid &&
1215: (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1216: ((flags & MSG_WAITALL) &&
1217: uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1.146 dyoung 1218: m->m_nextpkt == NULL && !atomic)) {
1.1 cgd 1219: #ifdef DIAGNOSTIC
1.144 dyoung 1220: if (m == NULL && so->so_rcv.sb_cc)
1.1 cgd 1221: panic("receive 1");
1222: #endif
1223: if (so->so_error) {
1.144 dyoung 1224: if (m != NULL)
1.15 mycroft 1225: goto dontblock;
1.1 cgd 1226: error = so->so_error;
1227: if ((flags & MSG_PEEK) == 0)
1228: so->so_error = 0;
1229: goto release;
1230: }
1231: if (so->so_state & SS_CANTRCVMORE) {
1.144 dyoung 1232: if (m != NULL)
1.15 mycroft 1233: goto dontblock;
1.1 cgd 1234: else
1235: goto release;
1236: }
1.144 dyoung 1237: for (; m != NULL; m = m->m_next)
1.1 cgd 1238: if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1239: m = so->so_rcv.sb_mb;
1240: goto dontblock;
1241: }
1242: if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1243: (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1244: error = ENOTCONN;
1245: goto release;
1246: }
1247: if (uio->uio_resid == 0)
1248: goto release;
1.206 christos 1249: if ((so->so_state & SS_NBIO) ||
1250: (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1.1 cgd 1251: error = EWOULDBLOCK;
1252: goto release;
1253: }
1.69 thorpej 1254: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1255: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1.1 cgd 1256: sbunlock(&so->so_rcv);
1.196 dsl 1257: if (wakeup_state & SS_RESTARTSYS)
1258: error = ERESTART;
1259: else
1260: error = sbwait(&so->so_rcv);
1.160 ad 1261: if (error != 0) {
1262: sounlock(so);
1263: splx(s);
1.144 dyoung 1264: return error;
1.160 ad 1265: }
1.196 dsl 1266: wakeup_state = so->so_state;
1.1 cgd 1267: goto restart;
1268: }
1.54 lukem 1269: dontblock:
1.69 thorpej 1270: /*
1271: * On entry here, m points to the first record of the socket buffer.
1.159 ad 1272: * From this point onward, we maintain 'nextrecord' as a cache of the
1273: * pointer to the next record in the socket buffer. We must keep the
1274: * various socket buffer pointers and local stack versions of the
1275: * pointers in sync, pushing out modifications before dropping the
1.160 ad 1276: * socket lock, and re-reading them when picking it up.
1.159 ad 1277: *
1278: * Otherwise, we will race with the network stack appending new data
1279: * or records onto the socket buffer by using inconsistent/stale
1280: * versions of the field, possibly resulting in socket buffer
1281: * corruption.
1282: *
1283: * By holding the high-level sblock(), we prevent simultaneous
1284: * readers from pulling off the front of the socket buffer.
1.69 thorpej 1285: */
1.144 dyoung 1286: if (l != NULL)
1.157 ad 1287: l->l_ru.ru_msgrcv++;
1.69 thorpej 1288: KASSERT(m == so->so_rcv.sb_mb);
1289: SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1290: SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1.1 cgd 1291: nextrecord = m->m_nextpkt;
1292: if (pr->pr_flags & PR_ADDR) {
1293: #ifdef DIAGNOSTIC
1294: if (m->m_type != MT_SONAME)
1295: panic("receive 1a");
1296: #endif
1.3 andrew 1297: orig_resid = 0;
1.1 cgd 1298: if (flags & MSG_PEEK) {
1299: if (paddr)
1300: *paddr = m_copy(m, 0, m->m_len);
1301: m = m->m_next;
1302: } else {
1303: sbfree(&so->so_rcv, m);
1.67 he 1304: mbuf_removed = 1;
1.144 dyoung 1305: if (paddr != NULL) {
1.1 cgd 1306: *paddr = m;
1307: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1308: m->m_next = NULL;
1.1 cgd 1309: m = so->so_rcv.sb_mb;
1310: } else {
1311: MFREE(m, so->so_rcv.sb_mb);
1312: m = so->so_rcv.sb_mb;
1313: }
1.159 ad 1314: sbsync(&so->so_rcv, nextrecord);
1.1 cgd 1315: }
1316: }
1.159 ad 1317:
1318: /*
1319: * Process one or more MT_CONTROL mbufs present before any data mbufs
1320: * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1321: * just copy the data; if !MSG_PEEK, we call into the protocol to
1322: * perform externalization (or freeing if controlp == NULL).
1323: */
1324: if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1325: struct mbuf *cm = NULL, *cmn;
1326: struct mbuf **cme = &cm;
1327:
1328: do {
1329: if (flags & MSG_PEEK) {
1330: if (controlp != NULL) {
1331: *controlp = m_copy(m, 0, m->m_len);
1332: controlp = &(*controlp)->m_next;
1333: }
1334: m = m->m_next;
1335: } else {
1336: sbfree(&so->so_rcv, m);
1.1 cgd 1337: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1338: m->m_next = NULL;
1.159 ad 1339: *cme = m;
1340: cme = &(*cme)->m_next;
1.1 cgd 1341: m = so->so_rcv.sb_mb;
1.159 ad 1342: }
1343: } while (m != NULL && m->m_type == MT_CONTROL);
1344: if ((flags & MSG_PEEK) == 0)
1345: sbsync(&so->so_rcv, nextrecord);
1346: for (; cm != NULL; cm = cmn) {
1347: cmn = cm->m_next;
1348: cm->m_next = NULL;
1349: type = mtod(cm, struct cmsghdr *)->cmsg_type;
1350: if (controlp != NULL) {
1351: if (dom->dom_externalize != NULL &&
1352: type == SCM_RIGHTS) {
1.160 ad 1353: sounlock(so);
1.159 ad 1354: splx(s);
1.204 christos 1355: error = (*dom->dom_externalize)(cm, l,
1356: (flags & MSG_CMSG_CLOEXEC) ?
1357: O_CLOEXEC : 0);
1.159 ad 1358: s = splsoftnet();
1.160 ad 1359: solock(so);
1.159 ad 1360: }
1361: *controlp = cm;
1362: while (*controlp != NULL)
1363: controlp = &(*controlp)->m_next;
1.1 cgd 1364: } else {
1.106 itojun 1365: /*
1366: * Dispose of any SCM_RIGHTS message that went
1367: * through the read path rather than recv.
1368: */
1.159 ad 1369: if (dom->dom_dispose != NULL &&
1370: type == SCM_RIGHTS) {
1.160 ad 1371: sounlock(so);
1.159 ad 1372: (*dom->dom_dispose)(cm);
1.160 ad 1373: solock(so);
1.159 ad 1374: }
1375: m_freem(cm);
1.1 cgd 1376: }
1377: }
1.159 ad 1378: if (m != NULL)
1379: nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1380: else
1381: nextrecord = so->so_rcv.sb_mb;
1382: orig_resid = 0;
1.1 cgd 1383: }
1.69 thorpej 1384:
1.159 ad 1385: /* If m is non-NULL, we have some data to read. */
1386: if (__predict_true(m != NULL)) {
1.1 cgd 1387: type = m->m_type;
1388: if (type == MT_OOBDATA)
1389: flags |= MSG_OOB;
1390: }
1.69 thorpej 1391: SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1392: SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1393:
1.1 cgd 1394: moff = 0;
1395: offset = 0;
1.144 dyoung 1396: while (m != NULL && uio->uio_resid > 0 && error == 0) {
1.1 cgd 1397: if (m->m_type == MT_OOBDATA) {
1398: if (type != MT_OOBDATA)
1399: break;
1400: } else if (type == MT_OOBDATA)
1401: break;
1402: #ifdef DIAGNOSTIC
1403: else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1404: panic("receive 3");
1405: #endif
1406: so->so_state &= ~SS_RCVATMARK;
1.196 dsl 1407: wakeup_state = 0;
1.1 cgd 1408: len = uio->uio_resid;
1409: if (so->so_oobmark && len > so->so_oobmark - offset)
1410: len = so->so_oobmark - offset;
1411: if (len > m->m_len - moff)
1412: len = m->m_len - moff;
1413: /*
1414: * If mp is set, just pass back the mbufs.
1415: * Otherwise copy them out via the uio, then free.
1416: * Sockbuf must be consistent here (points to current mbuf,
1417: * it points to next record) when we drop priority;
1418: * we must note any additions to the sockbuf when we
1419: * block interrupts again.
1420: */
1.144 dyoung 1421: if (mp == NULL) {
1.69 thorpej 1422: SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1423: SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1.160 ad 1424: sounlock(so);
1.1 cgd 1425: splx(s);
1.211 chs 1426: error = uiomove(mtod(m, char *) + moff, len, uio);
1.20 mycroft 1427: s = splsoftnet();
1.160 ad 1428: solock(so);
1.144 dyoung 1429: if (error != 0) {
1.67 he 1430: /*
1431: * If any part of the record has been removed
1432: * (such as the MT_SONAME mbuf, which will
1433: * happen when PR_ADDR, and thus also
1434: * PR_ATOMIC, is set), then drop the entire
1435: * record to maintain the atomicity of the
1436: * receive operation.
1437: *
1438: * This avoids a later panic("receive 1a")
1439: * when compiled with DIAGNOSTIC.
1440: */
1.146 dyoung 1441: if (m && mbuf_removed && atomic)
1.67 he 1442: (void) sbdroprecord(&so->so_rcv);
1443:
1.57 jdolecek 1444: goto release;
1.67 he 1445: }
1.1 cgd 1446: } else
1447: uio->uio_resid -= len;
1448: if (len == m->m_len - moff) {
1449: if (m->m_flags & M_EOR)
1450: flags |= MSG_EOR;
1451: if (flags & MSG_PEEK) {
1452: m = m->m_next;
1453: moff = 0;
1454: } else {
1455: nextrecord = m->m_nextpkt;
1456: sbfree(&so->so_rcv, m);
1457: if (mp) {
1458: *mp = m;
1459: mp = &m->m_next;
1460: so->so_rcv.sb_mb = m = m->m_next;
1.140 dyoung 1461: *mp = NULL;
1.1 cgd 1462: } else {
1463: MFREE(m, so->so_rcv.sb_mb);
1464: m = so->so_rcv.sb_mb;
1465: }
1.69 thorpej 1466: /*
1467: * If m != NULL, we also know that
1468: * so->so_rcv.sb_mb != NULL.
1469: */
1470: KASSERT(so->so_rcv.sb_mb == m);
1471: if (m) {
1.1 cgd 1472: m->m_nextpkt = nextrecord;
1.69 thorpej 1473: if (nextrecord == NULL)
1474: so->so_rcv.sb_lastrecord = m;
1475: } else {
1476: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1477: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1478: }
1479: SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1480: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1.1 cgd 1481: }
1.144 dyoung 1482: } else if (flags & MSG_PEEK)
1483: moff += len;
1484: else {
1.160 ad 1485: if (mp != NULL) {
1486: mt = m_copym(m, 0, len, M_NOWAIT);
1487: if (__predict_false(mt == NULL)) {
1488: sounlock(so);
1489: mt = m_copym(m, 0, len, M_WAIT);
1490: solock(so);
1491: }
1492: *mp = mt;
1493: }
1.144 dyoung 1494: m->m_data += len;
1495: m->m_len -= len;
1496: so->so_rcv.sb_cc -= len;
1.1 cgd 1497: }
1498: if (so->so_oobmark) {
1499: if ((flags & MSG_PEEK) == 0) {
1500: so->so_oobmark -= len;
1501: if (so->so_oobmark == 0) {
1502: so->so_state |= SS_RCVATMARK;
1503: break;
1504: }
1.7 cgd 1505: } else {
1.1 cgd 1506: offset += len;
1.7 cgd 1507: if (offset == so->so_oobmark)
1508: break;
1509: }
1.1 cgd 1510: }
1511: if (flags & MSG_EOR)
1512: break;
1513: /*
1514: * If the MSG_WAITALL flag is set (for non-atomic socket),
1515: * we must not quit until "uio->uio_resid == 0" or an error
1516: * termination. If a signal/timeout occurs, return
1517: * with a short count but without error.
1518: * Keep sockbuf locked against other readers.
1519: */
1.144 dyoung 1520: while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1.3 andrew 1521: !sosendallatonce(so) && !nextrecord) {
1.1 cgd 1522: if (so->so_error || so->so_state & SS_CANTRCVMORE)
1523: break;
1.68 matt 1524: /*
1525: * If we are peeking and the socket receive buffer is
1526: * full, stop since we can't get more data to peek at.
1527: */
1528: if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1529: break;
1530: /*
1531: * If we've drained the socket buffer, tell the
1532: * protocol in case it needs to do something to
1533: * get it filled again.
1534: */
1535: if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1536: (*pr->pr_usrreq)(so, PRU_RCVD,
1.140 dyoung 1537: NULL, (struct mbuf *)(long)flags, NULL, l);
1.69 thorpej 1538: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1539: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1.196 dsl 1540: if (wakeup_state & SS_RESTARTSYS)
1541: error = ERESTART;
1542: else
1543: error = sbwait(&so->so_rcv);
1.144 dyoung 1544: if (error != 0) {
1.1 cgd 1545: sbunlock(&so->so_rcv);
1.160 ad 1546: sounlock(so);
1.1 cgd 1547: splx(s);
1.144 dyoung 1548: return 0;
1.1 cgd 1549: }
1.21 christos 1550: if ((m = so->so_rcv.sb_mb) != NULL)
1.1 cgd 1551: nextrecord = m->m_nextpkt;
1.196 dsl 1552: wakeup_state = so->so_state;
1.1 cgd 1553: }
1554: }
1.3 andrew 1555:
1.146 dyoung 1556: if (m && atomic) {
1.3 andrew 1557: flags |= MSG_TRUNC;
1558: if ((flags & MSG_PEEK) == 0)
1559: (void) sbdroprecord(&so->so_rcv);
1560: }
1.1 cgd 1561: if ((flags & MSG_PEEK) == 0) {
1.144 dyoung 1562: if (m == NULL) {
1.69 thorpej 1563: /*
1.70 thorpej 1564: * First part is an inline SB_EMPTY_FIXUP(). Second
1.69 thorpej 1565: * part makes sure sb_lastrecord is up-to-date if
1566: * there is still data in the socket buffer.
1567: */
1.1 cgd 1568: so->so_rcv.sb_mb = nextrecord;
1.69 thorpej 1569: if (so->so_rcv.sb_mb == NULL) {
1570: so->so_rcv.sb_mbtail = NULL;
1571: so->so_rcv.sb_lastrecord = NULL;
1572: } else if (nextrecord->m_nextpkt == NULL)
1573: so->so_rcv.sb_lastrecord = nextrecord;
1574: }
1575: SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1576: SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1.1 cgd 1577: if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1.140 dyoung 1578: (*pr->pr_usrreq)(so, PRU_RCVD, NULL,
1579: (struct mbuf *)(long)flags, NULL, l);
1.1 cgd 1580: }
1.3 andrew 1581: if (orig_resid == uio->uio_resid && orig_resid &&
1582: (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1583: sbunlock(&so->so_rcv);
1584: goto restart;
1585: }
1.108 perry 1586:
1.144 dyoung 1587: if (flagsp != NULL)
1.1 cgd 1588: *flagsp |= flags;
1.54 lukem 1589: release:
1.1 cgd 1590: sbunlock(&so->so_rcv);
1.160 ad 1591: sounlock(so);
1.1 cgd 1592: splx(s);
1.144 dyoung 1593: return error;
1.1 cgd 1594: }
1595:
1.14 mycroft 1596: int
1.54 lukem 1597: soshutdown(struct socket *so, int how)
1.1 cgd 1598: {
1.99 matt 1599: const struct protosw *pr;
1.160 ad 1600: int error;
1601:
1602: KASSERT(solocked(so));
1.34 kleink 1603:
1.54 lukem 1604: pr = so->so_proto;
1.34 kleink 1605: if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1606: return (EINVAL);
1.1 cgd 1607:
1.160 ad 1608: if (how == SHUT_RD || how == SHUT_RDWR) {
1.1 cgd 1609: sorflush(so);
1.160 ad 1610: error = 0;
1611: }
1.34 kleink 1612: if (how == SHUT_WR || how == SHUT_RDWR)
1.160 ad 1613: error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
1.140 dyoung 1614: NULL, NULL, NULL);
1.160 ad 1615:
1616: return error;
1.1 cgd 1617: }
1618:
1.195 dsl 1619: void
1.196 dsl 1620: sorestart(struct socket *so)
1.188 ad 1621: {
1.196 dsl 1622: /*
1623: * An application has called close() on an fd on which another
1624: * of its threads has called a socket system call.
1625: * Mark this and wake everyone up, and code that would block again
1626: * instead returns ERESTART.
1627: * On system call re-entry the fd is validated and EBADF returned.
1628: * Any other fd will block again on the 2nd syscall.
1629: */
1.188 ad 1630: solock(so);
1.196 dsl 1631: so->so_state |= SS_RESTARTSYS;
1.188 ad 1632: cv_broadcast(&so->so_cv);
1.196 dsl 1633: cv_broadcast(&so->so_snd.sb_cv);
1634: cv_broadcast(&so->so_rcv.sb_cv);
1.188 ad 1635: sounlock(so);
1636: }
1637:
1.14 mycroft 1638: void
1.54 lukem 1639: sorflush(struct socket *so)
1.1 cgd 1640: {
1.54 lukem 1641: struct sockbuf *sb, asb;
1.99 matt 1642: const struct protosw *pr;
1.160 ad 1643:
1644: KASSERT(solocked(so));
1.1 cgd 1645:
1.54 lukem 1646: sb = &so->so_rcv;
1647: pr = so->so_proto;
1.160 ad 1648: socantrcvmore(so);
1.1 cgd 1649: sb->sb_flags |= SB_NOINTR;
1.160 ad 1650: (void )sblock(sb, M_WAITOK);
1.1 cgd 1651: sbunlock(sb);
1652: asb = *sb;
1.86 wrstuden 1653: /*
1654: * Clear most of the sockbuf structure, but leave some of the
1655: * fields valid.
1656: */
1657: memset(&sb->sb_startzero, 0,
1658: sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1.160 ad 1659: if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1660: sounlock(so);
1.1 cgd 1661: (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1.160 ad 1662: solock(so);
1663: }
1.98 christos 1664: sbrelease(&asb, so);
1.1 cgd 1665: }
1666:
1.171 plunky 1667: /*
1668: * internal set SOL_SOCKET options
1669: */
1.142 dyoung 1670: static int
1.171 plunky 1671: sosetopt1(struct socket *so, const struct sockopt *sopt)
1.1 cgd 1672: {
1.219 christos 1673: int error = EINVAL, opt;
1674: int optval = 0; /* XXX: gcc */
1.171 plunky 1675: struct linger l;
1676: struct timeval tv;
1.142 dyoung 1677:
1.179 christos 1678: switch ((opt = sopt->sopt_name)) {
1.142 dyoung 1679:
1.170 tls 1680: case SO_ACCEPTFILTER:
1.177 ad 1681: error = accept_filt_setopt(so, sopt);
1682: KASSERT(solocked(so));
1.170 tls 1683: break;
1684:
1.171 plunky 1685: case SO_LINGER:
1686: error = sockopt_get(sopt, &l, sizeof(l));
1.177 ad 1687: solock(so);
1.171 plunky 1688: if (error)
1.177 ad 1689: break;
1.171 plunky 1690: if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1.177 ad 1691: l.l_linger > (INT_MAX / hz)) {
1692: error = EDOM;
1693: break;
1694: }
1.171 plunky 1695: so->so_linger = l.l_linger;
1696: if (l.l_onoff)
1697: so->so_options |= SO_LINGER;
1698: else
1699: so->so_options &= ~SO_LINGER;
1.177 ad 1700: break;
1.1 cgd 1701:
1.142 dyoung 1702: case SO_DEBUG:
1703: case SO_KEEPALIVE:
1704: case SO_DONTROUTE:
1705: case SO_USELOOPBACK:
1706: case SO_BROADCAST:
1707: case SO_REUSEADDR:
1708: case SO_REUSEPORT:
1709: case SO_OOBINLINE:
1710: case SO_TIMESTAMP:
1.207 christos 1711: case SO_NOSIGPIPE:
1.184 christos 1712: #ifdef SO_OTIMESTAMP
1713: case SO_OTIMESTAMP:
1714: #endif
1.171 plunky 1715: error = sockopt_getint(sopt, &optval);
1.177 ad 1716: solock(so);
1.171 plunky 1717: if (error)
1.177 ad 1718: break;
1.171 plunky 1719: if (optval)
1.179 christos 1720: so->so_options |= opt;
1.142 dyoung 1721: else
1.179 christos 1722: so->so_options &= ~opt;
1.142 dyoung 1723: break;
1724:
1725: case SO_SNDBUF:
1726: case SO_RCVBUF:
1727: case SO_SNDLOWAT:
1728: case SO_RCVLOWAT:
1.171 plunky 1729: error = sockopt_getint(sopt, &optval);
1.177 ad 1730: solock(so);
1.171 plunky 1731: if (error)
1.177 ad 1732: break;
1.1 cgd 1733:
1.142 dyoung 1734: /*
1735: * Values < 1 make no sense for any of these
1736: * options, so disallow them.
1737: */
1.177 ad 1738: if (optval < 1) {
1739: error = EINVAL;
1740: break;
1741: }
1.1 cgd 1742:
1.179 christos 1743: switch (opt) {
1.171 plunky 1744: case SO_SNDBUF:
1.177 ad 1745: if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1746: error = ENOBUFS;
1747: break;
1748: }
1.171 plunky 1749: so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1750: break;
1.1 cgd 1751:
1752: case SO_RCVBUF:
1.177 ad 1753: if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1754: error = ENOBUFS;
1755: break;
1756: }
1.171 plunky 1757: so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1.142 dyoung 1758: break;
1759:
1760: /*
1761: * Make sure the low-water is never greater than
1762: * the high-water.
1763: */
1.1 cgd 1764: case SO_SNDLOWAT:
1.171 plunky 1765: if (optval > so->so_snd.sb_hiwat)
1766: optval = so->so_snd.sb_hiwat;
1767:
1768: so->so_snd.sb_lowat = optval;
1.142 dyoung 1769: break;
1.171 plunky 1770:
1.1 cgd 1771: case SO_RCVLOWAT:
1.171 plunky 1772: if (optval > so->so_rcv.sb_hiwat)
1773: optval = so->so_rcv.sb_hiwat;
1774:
1775: so->so_rcv.sb_lowat = optval;
1.142 dyoung 1776: break;
1777: }
1778: break;
1.28 thorpej 1779:
1.179 christos 1780: #ifdef COMPAT_50
1781: case SO_OSNDTIMEO:
1782: case SO_ORCVTIMEO: {
1783: struct timeval50 otv;
1784: error = sockopt_get(sopt, &otv, sizeof(otv));
1.186 pooka 1785: if (error) {
1786: solock(so);
1.183 christos 1787: break;
1.186 pooka 1788: }
1.179 christos 1789: timeval50_to_timeval(&otv, &tv);
1790: opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1.182 christos 1791: error = 0;
1.179 christos 1792: /*FALLTHROUGH*/
1793: }
1794: #endif /* COMPAT_50 */
1795:
1.142 dyoung 1796: case SO_SNDTIMEO:
1797: case SO_RCVTIMEO:
1.182 christos 1798: if (error)
1.179 christos 1799: error = sockopt_get(sopt, &tv, sizeof(tv));
1.177 ad 1800: solock(so);
1.171 plunky 1801: if (error)
1.177 ad 1802: break;
1.171 plunky 1803:
1.177 ad 1804: if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1805: error = EDOM;
1806: break;
1807: }
1.28 thorpej 1808:
1.171 plunky 1809: optval = tv.tv_sec * hz + tv.tv_usec / tick;
1810: if (optval == 0 && tv.tv_usec != 0)
1811: optval = 1;
1.28 thorpej 1812:
1.179 christos 1813: switch (opt) {
1.142 dyoung 1814: case SO_SNDTIMEO:
1.171 plunky 1815: so->so_snd.sb_timeo = optval;
1.1 cgd 1816: break;
1817: case SO_RCVTIMEO:
1.171 plunky 1818: so->so_rcv.sb_timeo = optval;
1.142 dyoung 1819: break;
1820: }
1821: break;
1.1 cgd 1822:
1.142 dyoung 1823: default:
1.177 ad 1824: solock(so);
1825: error = ENOPROTOOPT;
1826: break;
1.142 dyoung 1827: }
1.177 ad 1828: KASSERT(solocked(so));
1829: return error;
1.142 dyoung 1830: }
1.1 cgd 1831:
1.142 dyoung 1832: int
1.171 plunky 1833: sosetopt(struct socket *so, struct sockopt *sopt)
1.142 dyoung 1834: {
1835: int error, prerr;
1.1 cgd 1836:
1.177 ad 1837: if (sopt->sopt_level == SOL_SOCKET) {
1.171 plunky 1838: error = sosetopt1(so, sopt);
1.177 ad 1839: KASSERT(solocked(so));
1840: } else {
1.142 dyoung 1841: error = ENOPROTOOPT;
1.177 ad 1842: solock(so);
1843: }
1.1 cgd 1844:
1.142 dyoung 1845: if ((error == 0 || error == ENOPROTOOPT) &&
1846: so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1847: /* give the protocol stack a shot */
1.171 plunky 1848: prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1.142 dyoung 1849: if (prerr == 0)
1850: error = 0;
1851: else if (prerr != ENOPROTOOPT)
1852: error = prerr;
1.171 plunky 1853: }
1.160 ad 1854: sounlock(so);
1.142 dyoung 1855: return error;
1.1 cgd 1856: }
1857:
1.171 plunky 1858: /*
1859: * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1860: */
1861: int
1862: so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1863: const void *val, size_t valsize)
1864: {
1865: struct sockopt sopt;
1866: int error;
1867:
1868: KASSERT(valsize == 0 || val != NULL);
1869:
1870: sockopt_init(&sopt, level, name, valsize);
1871: sockopt_set(&sopt, val, valsize);
1872:
1873: error = sosetopt(so, &sopt);
1874:
1875: sockopt_destroy(&sopt);
1876:
1877: return error;
1878: }
1879:
1880: /*
1881: * internal get SOL_SOCKET options
1882: */
1883: static int
1884: sogetopt1(struct socket *so, struct sockopt *sopt)
1885: {
1.179 christos 1886: int error, optval, opt;
1.171 plunky 1887: struct linger l;
1888: struct timeval tv;
1889:
1.179 christos 1890: switch ((opt = sopt->sopt_name)) {
1.171 plunky 1891:
1892: case SO_ACCEPTFILTER:
1.177 ad 1893: error = accept_filt_getopt(so, sopt);
1.171 plunky 1894: break;
1895:
1896: case SO_LINGER:
1897: l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1898: l.l_linger = so->so_linger;
1899:
1900: error = sockopt_set(sopt, &l, sizeof(l));
1901: break;
1902:
1903: case SO_USELOOPBACK:
1904: case SO_DONTROUTE:
1905: case SO_DEBUG:
1906: case SO_KEEPALIVE:
1907: case SO_REUSEADDR:
1908: case SO_REUSEPORT:
1909: case SO_BROADCAST:
1910: case SO_OOBINLINE:
1911: case SO_TIMESTAMP:
1.207 christos 1912: case SO_NOSIGPIPE:
1.184 christos 1913: #ifdef SO_OTIMESTAMP
1914: case SO_OTIMESTAMP:
1915: #endif
1.218 seanb 1916: case SO_ACCEPTCONN:
1.179 christos 1917: error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1.171 plunky 1918: break;
1919:
1920: case SO_TYPE:
1921: error = sockopt_setint(sopt, so->so_type);
1922: break;
1923:
1924: case SO_ERROR:
1925: error = sockopt_setint(sopt, so->so_error);
1926: so->so_error = 0;
1927: break;
1928:
1929: case SO_SNDBUF:
1930: error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1931: break;
1932:
1933: case SO_RCVBUF:
1934: error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1935: break;
1936:
1937: case SO_SNDLOWAT:
1938: error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1939: break;
1940:
1941: case SO_RCVLOWAT:
1942: error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1943: break;
1944:
1.179 christos 1945: #ifdef COMPAT_50
1946: case SO_OSNDTIMEO:
1947: case SO_ORCVTIMEO: {
1948: struct timeval50 otv;
1949:
1950: optval = (opt == SO_OSNDTIMEO ?
1951: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1952:
1953: otv.tv_sec = optval / hz;
1954: otv.tv_usec = (optval % hz) * tick;
1955:
1956: error = sockopt_set(sopt, &otv, sizeof(otv));
1957: break;
1958: }
1959: #endif /* COMPAT_50 */
1960:
1.171 plunky 1961: case SO_SNDTIMEO:
1962: case SO_RCVTIMEO:
1.179 christos 1963: optval = (opt == SO_SNDTIMEO ?
1.171 plunky 1964: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1965:
1966: tv.tv_sec = optval / hz;
1967: tv.tv_usec = (optval % hz) * tick;
1968:
1969: error = sockopt_set(sopt, &tv, sizeof(tv));
1970: break;
1971:
1972: case SO_OVERFLOWED:
1973: error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
1974: break;
1975:
1976: default:
1977: error = ENOPROTOOPT;
1978: break;
1979: }
1980:
1981: return (error);
1982: }
1983:
1.14 mycroft 1984: int
1.171 plunky 1985: sogetopt(struct socket *so, struct sockopt *sopt)
1.1 cgd 1986: {
1.160 ad 1987: int error;
1.1 cgd 1988:
1.160 ad 1989: solock(so);
1.171 plunky 1990: if (sopt->sopt_level != SOL_SOCKET) {
1.1 cgd 1991: if (so->so_proto && so->so_proto->pr_ctloutput) {
1.160 ad 1992: error = ((*so->so_proto->pr_ctloutput)
1.171 plunky 1993: (PRCO_GETOPT, so, sopt));
1.1 cgd 1994: } else
1.160 ad 1995: error = (ENOPROTOOPT);
1.1 cgd 1996: } else {
1.171 plunky 1997: error = sogetopt1(so, sopt);
1998: }
1999: sounlock(so);
2000: return (error);
2001: }
2002:
2003: /*
2004: * alloc sockopt data buffer buffer
2005: * - will be released at destroy
2006: */
1.176 plunky 2007: static int
2008: sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
1.171 plunky 2009: {
2010:
2011: KASSERT(sopt->sopt_size == 0);
2012:
1.176 plunky 2013: if (len > sizeof(sopt->sopt_buf)) {
2014: sopt->sopt_data = kmem_zalloc(len, kmflag);
2015: if (sopt->sopt_data == NULL)
2016: return ENOMEM;
2017: } else
1.171 plunky 2018: sopt->sopt_data = sopt->sopt_buf;
2019:
2020: sopt->sopt_size = len;
1.176 plunky 2021: return 0;
1.171 plunky 2022: }
2023:
2024: /*
2025: * initialise sockopt storage
1.176 plunky 2026: * - MAY sleep during allocation
1.171 plunky 2027: */
2028: void
2029: sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2030: {
1.1 cgd 2031:
1.171 plunky 2032: memset(sopt, 0, sizeof(*sopt));
1.1 cgd 2033:
1.171 plunky 2034: sopt->sopt_level = level;
2035: sopt->sopt_name = name;
1.176 plunky 2036: (void)sockopt_alloc(sopt, size, KM_SLEEP);
1.171 plunky 2037: }
2038:
2039: /*
2040: * destroy sockopt storage
2041: * - will release any held memory references
2042: */
2043: void
2044: sockopt_destroy(struct sockopt *sopt)
2045: {
2046:
2047: if (sopt->sopt_data != sopt->sopt_buf)
1.173 plunky 2048: kmem_free(sopt->sopt_data, sopt->sopt_size);
1.171 plunky 2049:
2050: memset(sopt, 0, sizeof(*sopt));
2051: }
2052:
2053: /*
2054: * set sockopt value
2055: * - value is copied into sockopt
1.176 plunky 2056: * - memory is allocated when necessary, will not sleep
1.171 plunky 2057: */
2058: int
2059: sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2060: {
1.176 plunky 2061: int error;
1.171 plunky 2062:
1.176 plunky 2063: if (sopt->sopt_size == 0) {
2064: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2065: if (error)
2066: return error;
2067: }
1.171 plunky 2068:
2069: KASSERT(sopt->sopt_size == len);
2070: memcpy(sopt->sopt_data, buf, len);
2071: return 0;
2072: }
2073:
2074: /*
2075: * common case of set sockopt integer value
2076: */
2077: int
2078: sockopt_setint(struct sockopt *sopt, int val)
2079: {
2080:
2081: return sockopt_set(sopt, &val, sizeof(int));
2082: }
2083:
2084: /*
2085: * get sockopt value
2086: * - correct size must be given
2087: */
2088: int
2089: sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2090: {
1.170 tls 2091:
1.171 plunky 2092: if (sopt->sopt_size != len)
2093: return EINVAL;
1.1 cgd 2094:
1.171 plunky 2095: memcpy(buf, sopt->sopt_data, len);
2096: return 0;
2097: }
1.1 cgd 2098:
1.171 plunky 2099: /*
2100: * common case of get sockopt integer value
2101: */
2102: int
2103: sockopt_getint(const struct sockopt *sopt, int *valp)
2104: {
1.1 cgd 2105:
1.171 plunky 2106: return sockopt_get(sopt, valp, sizeof(int));
2107: }
1.1 cgd 2108:
1.171 plunky 2109: /*
2110: * set sockopt value from mbuf
2111: * - ONLY for legacy code
2112: * - mbuf is released by sockopt
1.176 plunky 2113: * - will not sleep
1.171 plunky 2114: */
2115: int
2116: sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2117: {
2118: size_t len;
1.176 plunky 2119: int error;
1.1 cgd 2120:
1.171 plunky 2121: len = m_length(m);
1.1 cgd 2122:
1.176 plunky 2123: if (sopt->sopt_size == 0) {
2124: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2125: if (error)
2126: return error;
2127: }
1.1 cgd 2128:
1.171 plunky 2129: KASSERT(sopt->sopt_size == len);
2130: m_copydata(m, 0, len, sopt->sopt_data);
2131: m_freem(m);
1.1 cgd 2132:
1.171 plunky 2133: return 0;
2134: }
1.1 cgd 2135:
1.171 plunky 2136: /*
2137: * get sockopt value into mbuf
2138: * - ONLY for legacy code
2139: * - mbuf to be released by the caller
1.176 plunky 2140: * - will not sleep
1.171 plunky 2141: */
2142: struct mbuf *
2143: sockopt_getmbuf(const struct sockopt *sopt)
2144: {
2145: struct mbuf *m;
1.107 darrenr 2146:
1.176 plunky 2147: if (sopt->sopt_size > MCLBYTES)
2148: return NULL;
2149:
2150: m = m_get(M_DONTWAIT, MT_SOOPTS);
1.171 plunky 2151: if (m == NULL)
2152: return NULL;
2153:
1.176 plunky 2154: if (sopt->sopt_size > MLEN) {
2155: MCLGET(m, M_DONTWAIT);
2156: if ((m->m_flags & M_EXT) == 0) {
2157: m_free(m);
2158: return NULL;
2159: }
1.1 cgd 2160: }
1.176 plunky 2161:
2162: memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2163: m->m_len = sopt->sopt_size;
1.160 ad 2164:
1.171 plunky 2165: return m;
1.1 cgd 2166: }
2167:
1.14 mycroft 2168: void
1.54 lukem 2169: sohasoutofband(struct socket *so)
1.1 cgd 2170: {
1.153 rmind 2171:
1.90 christos 2172: fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1.189 ad 2173: selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
1.1 cgd 2174: }
1.72 jdolecek 2175:
2176: static void
2177: filt_sordetach(struct knote *kn)
2178: {
2179: struct socket *so;
2180:
1.155 ad 2181: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2182: solock(so);
1.73 christos 2183: SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2184: if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1.72 jdolecek 2185: so->so_rcv.sb_flags &= ~SB_KNOTE;
1.160 ad 2186: sounlock(so);
1.72 jdolecek 2187: }
2188:
2189: /*ARGSUSED*/
2190: static int
1.129 yamt 2191: filt_soread(struct knote *kn, long hint)
1.72 jdolecek 2192: {
2193: struct socket *so;
1.160 ad 2194: int rv;
1.72 jdolecek 2195:
1.155 ad 2196: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2197: if (hint != NOTE_SUBMIT)
2198: solock(so);
1.72 jdolecek 2199: kn->kn_data = so->so_rcv.sb_cc;
2200: if (so->so_state & SS_CANTRCVMORE) {
1.108 perry 2201: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2202: kn->kn_fflags = so->so_error;
1.160 ad 2203: rv = 1;
2204: } else if (so->so_error) /* temporary udp error */
2205: rv = 1;
2206: else if (kn->kn_sfflags & NOTE_LOWAT)
2207: rv = (kn->kn_data >= kn->kn_sdata);
2208: else
2209: rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2210: if (hint != NOTE_SUBMIT)
2211: sounlock(so);
2212: return rv;
1.72 jdolecek 2213: }
2214:
2215: static void
2216: filt_sowdetach(struct knote *kn)
2217: {
2218: struct socket *so;
2219:
1.155 ad 2220: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2221: solock(so);
1.73 christos 2222: SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2223: if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1.72 jdolecek 2224: so->so_snd.sb_flags &= ~SB_KNOTE;
1.160 ad 2225: sounlock(so);
1.72 jdolecek 2226: }
2227:
2228: /*ARGSUSED*/
2229: static int
1.129 yamt 2230: filt_sowrite(struct knote *kn, long hint)
1.72 jdolecek 2231: {
2232: struct socket *so;
1.160 ad 2233: int rv;
1.72 jdolecek 2234:
1.155 ad 2235: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2236: if (hint != NOTE_SUBMIT)
2237: solock(so);
1.72 jdolecek 2238: kn->kn_data = sbspace(&so->so_snd);
2239: if (so->so_state & SS_CANTSENDMORE) {
1.108 perry 2240: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2241: kn->kn_fflags = so->so_error;
1.160 ad 2242: rv = 1;
2243: } else if (so->so_error) /* temporary udp error */
2244: rv = 1;
2245: else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1.72 jdolecek 2246: (so->so_proto->pr_flags & PR_CONNREQUIRED))
1.160 ad 2247: rv = 0;
2248: else if (kn->kn_sfflags & NOTE_LOWAT)
2249: rv = (kn->kn_data >= kn->kn_sdata);
2250: else
2251: rv = (kn->kn_data >= so->so_snd.sb_lowat);
2252: if (hint != NOTE_SUBMIT)
2253: sounlock(so);
2254: return rv;
1.72 jdolecek 2255: }
2256:
2257: /*ARGSUSED*/
2258: static int
1.129 yamt 2259: filt_solisten(struct knote *kn, long hint)
1.72 jdolecek 2260: {
2261: struct socket *so;
1.160 ad 2262: int rv;
1.72 jdolecek 2263:
1.155 ad 2264: so = ((file_t *)kn->kn_obj)->f_data;
1.72 jdolecek 2265:
2266: /*
2267: * Set kn_data to number of incoming connections, not
2268: * counting partial (incomplete) connections.
1.108 perry 2269: */
1.160 ad 2270: if (hint != NOTE_SUBMIT)
2271: solock(so);
1.72 jdolecek 2272: kn->kn_data = so->so_qlen;
1.160 ad 2273: rv = (kn->kn_data > 0);
2274: if (hint != NOTE_SUBMIT)
2275: sounlock(so);
2276: return rv;
1.72 jdolecek 2277: }
2278:
2279: static const struct filterops solisten_filtops =
2280: { 1, NULL, filt_sordetach, filt_solisten };
2281: static const struct filterops soread_filtops =
2282: { 1, NULL, filt_sordetach, filt_soread };
2283: static const struct filterops sowrite_filtops =
2284: { 1, NULL, filt_sowdetach, filt_sowrite };
2285:
2286: int
1.129 yamt 2287: soo_kqfilter(struct file *fp, struct knote *kn)
1.72 jdolecek 2288: {
2289: struct socket *so;
2290: struct sockbuf *sb;
2291:
1.155 ad 2292: so = ((file_t *)kn->kn_obj)->f_data;
1.160 ad 2293: solock(so);
1.72 jdolecek 2294: switch (kn->kn_filter) {
2295: case EVFILT_READ:
2296: if (so->so_options & SO_ACCEPTCONN)
2297: kn->kn_fop = &solisten_filtops;
2298: else
2299: kn->kn_fop = &soread_filtops;
2300: sb = &so->so_rcv;
2301: break;
2302: case EVFILT_WRITE:
2303: kn->kn_fop = &sowrite_filtops;
2304: sb = &so->so_snd;
2305: break;
2306: default:
1.160 ad 2307: sounlock(so);
1.149 pooka 2308: return (EINVAL);
1.72 jdolecek 2309: }
1.73 christos 2310: SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1.72 jdolecek 2311: sb->sb_flags |= SB_KNOTE;
1.160 ad 2312: sounlock(so);
1.72 jdolecek 2313: return (0);
2314: }
2315:
1.154 ad 2316: static int
2317: sodopoll(struct socket *so, int events)
2318: {
2319: int revents;
2320:
2321: revents = 0;
2322:
2323: if (events & (POLLIN | POLLRDNORM))
2324: if (soreadable(so))
2325: revents |= events & (POLLIN | POLLRDNORM);
2326:
2327: if (events & (POLLOUT | POLLWRNORM))
2328: if (sowritable(so))
2329: revents |= events & (POLLOUT | POLLWRNORM);
2330:
2331: if (events & (POLLPRI | POLLRDBAND))
2332: if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2333: revents |= events & (POLLPRI | POLLRDBAND);
2334:
2335: return revents;
2336: }
2337:
2338: int
2339: sopoll(struct socket *so, int events)
2340: {
2341: int revents = 0;
2342:
1.160 ad 2343: #ifndef DIAGNOSTIC
2344: /*
2345: * Do a quick, unlocked check in expectation that the socket
2346: * will be ready for I/O. Don't do this check if DIAGNOSTIC,
2347: * as the solocked() assertions will fail.
2348: */
1.154 ad 2349: if ((revents = sodopoll(so, events)) != 0)
2350: return revents;
1.160 ad 2351: #endif
1.154 ad 2352:
1.160 ad 2353: solock(so);
1.154 ad 2354: if ((revents = sodopoll(so, events)) == 0) {
2355: if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2356: selrecord(curlwp, &so->so_rcv.sb_sel);
1.160 ad 2357: so->so_rcv.sb_flags |= SB_NOTIFY;
1.154 ad 2358: }
2359:
2360: if (events & (POLLOUT | POLLWRNORM)) {
2361: selrecord(curlwp, &so->so_snd.sb_sel);
1.160 ad 2362: so->so_snd.sb_flags |= SB_NOTIFY;
1.154 ad 2363: }
2364: }
1.160 ad 2365: sounlock(so);
1.154 ad 2366:
2367: return revents;
2368: }
2369:
2370:
1.94 yamt 2371: #include <sys/sysctl.h>
2372:
2373: static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
1.212 pooka 2374: static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
1.94 yamt 2375:
2376: /*
2377: * sysctl helper routine for kern.somaxkva. ensures that the given
2378: * value is not too small.
2379: * (XXX should we maybe make sure it's not too large as well?)
2380: */
2381: static int
2382: sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2383: {
2384: int error, new_somaxkva;
2385: struct sysctlnode node;
2386:
2387: new_somaxkva = somaxkva;
2388: node = *rnode;
2389: node.sysctl_data = &new_somaxkva;
2390: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2391: if (error || newp == NULL)
2392: return (error);
2393:
2394: if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
2395: return (EINVAL);
2396:
1.136 ad 2397: mutex_enter(&so_pendfree_lock);
1.94 yamt 2398: somaxkva = new_somaxkva;
1.136 ad 2399: cv_broadcast(&socurkva_cv);
2400: mutex_exit(&so_pendfree_lock);
1.94 yamt 2401:
2402: return (error);
2403: }
2404:
1.212 pooka 2405: /*
2406: * sysctl helper routine for kern.sbmax. Basically just ensures that
2407: * any new value is not too small.
2408: */
2409: static int
2410: sysctl_kern_sbmax(SYSCTLFN_ARGS)
2411: {
2412: int error, new_sbmax;
2413: struct sysctlnode node;
2414:
2415: new_sbmax = sb_max;
2416: node = *rnode;
2417: node.sysctl_data = &new_sbmax;
2418: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2419: if (error || newp == NULL)
2420: return (error);
2421:
2422: KERNEL_LOCK(1, NULL);
2423: error = sb_max_set(new_sbmax);
2424: KERNEL_UNLOCK_ONE(NULL);
2425:
2426: return (error);
2427: }
2428:
1.178 pooka 2429: static void
1.212 pooka 2430: sysctl_kern_socket_setup(void)
1.94 yamt 2431: {
2432:
1.178 pooka 2433: KASSERT(socket_sysctllog == NULL);
1.97 atatat 2434:
1.178 pooka 2435: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2436: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1.103 atatat 2437: CTLTYPE_INT, "somaxkva",
2438: SYSCTL_DESCR("Maximum amount of kernel memory to be "
2439: "used for socket buffers"),
1.94 yamt 2440: sysctl_kern_somaxkva, 0, NULL, 0,
2441: CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
1.212 pooka 2442:
2443: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2444: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2445: CTLTYPE_INT, "sbmax",
2446: SYSCTL_DESCR("Maximum socket buffer size"),
2447: sysctl_kern_sbmax, 0, NULL, 0,
2448: CTL_KERN, KERN_SBMAX, CTL_EOL);
1.94 yamt 2449: }
CVSweb <webmaster@jp.NetBSD.org>