Annotation of src/sys/kern/uipc_socket.c, Revision 1.273
1.273 ! maxv 1: /* $NetBSD: uipc_socket.c,v 1.272 2019/03/31 19:54:36 maxv Exp $ */
1.64 thorpej 2:
1.270 maxv 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>
1.273 ! maxv 74: __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.272 2019/03/31 19:54:36 maxv Exp $");
1.64 thorpej 75:
1.246 pooka 76: #ifdef _KERNEL_OPT
1.179 christos 77: #include "opt_compat_netbsd.h"
1.64 thorpej 78: #include "opt_sock_counters.h"
79: #include "opt_sosend_loan.h"
1.81 martin 80: #include "opt_mbuftrace.h"
1.84 ragge 81: #include "opt_somaxkva.h"
1.167 ad 82: #include "opt_multiprocessor.h" /* XXX */
1.247 rjs 83: #include "opt_sctp.h"
1.246 pooka 84: #endif
1.1 cgd 85:
1.9 mycroft 86: #include <sys/param.h>
87: #include <sys/systm.h>
88: #include <sys/proc.h>
89: #include <sys/file.h>
1.142 dyoung 90: #include <sys/filedesc.h>
1.173 plunky 91: #include <sys/kmem.h>
1.9 mycroft 92: #include <sys/mbuf.h>
93: #include <sys/domain.h>
94: #include <sys/kernel.h>
95: #include <sys/protosw.h>
96: #include <sys/socket.h>
97: #include <sys/socketvar.h>
1.21 christos 98: #include <sys/signalvar.h>
1.9 mycroft 99: #include <sys/resourcevar.h>
1.174 pooka 100: #include <sys/uidinfo.h>
1.72 jdolecek 101: #include <sys/event.h>
1.89 christos 102: #include <sys/poll.h>
1.118 elad 103: #include <sys/kauth.h>
1.136 ad 104: #include <sys/mutex.h>
105: #include <sys/condvar.h>
1.205 bouyer 106: #include <sys/kthread.h>
1.37 thorpej 107:
1.179 christos 108: #ifdef COMPAT_50
109: #include <compat/sys/time.h>
1.184 christos 110: #include <compat/sys/socket.h>
1.179 christos 111: #endif
112:
1.202 uebayasi 113: #include <uvm/uvm_extern.h>
114: #include <uvm/uvm_loan.h>
115: #include <uvm/uvm_page.h>
1.64 thorpej 116:
1.77 thorpej 117: MALLOC_DEFINE(M_SONAME, "soname", "socket name");
1.37 thorpej 118:
1.142 dyoung 119: extern const struct fileops socketops;
120:
1.266 christos 121: static int sooptions;
1.54 lukem 122: extern int somaxconn; /* patchable (XXX sysctl) */
123: int somaxconn = SOMAXCONN;
1.160 ad 124: kmutex_t *softnet_lock;
1.49 jonathan 125:
1.64 thorpej 126: #ifdef SOSEND_COUNTERS
127: #include <sys/device.h>
128:
1.113 thorpej 129: static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 130: NULL, "sosend", "loan big");
1.113 thorpej 131: static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 132: NULL, "sosend", "copy big");
1.113 thorpej 133: static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 134: NULL, "sosend", "copy small");
1.113 thorpej 135: static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
1.64 thorpej 136: NULL, "sosend", "kva limit");
137:
138: #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++
139:
1.101 matt 140: EVCNT_ATTACH_STATIC(sosend_loan_big);
141: EVCNT_ATTACH_STATIC(sosend_copy_big);
142: EVCNT_ATTACH_STATIC(sosend_copy_small);
143: EVCNT_ATTACH_STATIC(sosend_kvalimit);
1.64 thorpej 144: #else
145:
146: #define SOSEND_COUNTER_INCR(ev) /* nothing */
147:
148: #endif /* SOSEND_COUNTERS */
149:
1.167 ad 150: #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
1.121 yamt 151: int sock_loan_thresh = -1;
1.71 thorpej 152: #else
1.121 yamt 153: int sock_loan_thresh = 4096;
1.65 thorpej 154: #endif
1.64 thorpej 155:
1.136 ad 156: static kmutex_t so_pendfree_lock;
1.205 bouyer 157: static struct mbuf *so_pendfree = NULL;
1.64 thorpej 158:
1.84 ragge 159: #ifndef SOMAXKVA
160: #define SOMAXKVA (16 * 1024 * 1024)
161: #endif
162: int somaxkva = SOMAXKVA;
1.113 thorpej 163: static int socurkva;
1.136 ad 164: static kcondvar_t socurkva_cv;
1.64 thorpej 165:
1.191 elad 166: static kauth_listener_t socket_listener;
167:
1.64 thorpej 168: #define SOCK_LOAN_CHUNK 65536
169:
1.205 bouyer 170: static void sopendfree_thread(void *);
171: static kcondvar_t pendfree_thread_cv;
172: static lwp_t *sopendfree_lwp;
1.93 yamt 173:
1.212 pooka 174: static void sysctl_kern_socket_setup(void);
1.178 pooka 175: static struct sysctllog *socket_sysctllog;
176:
1.113 thorpej 177: static vsize_t
1.129 yamt 178: sokvareserve(struct socket *so, vsize_t len)
1.80 yamt 179: {
1.98 christos 180: int error;
1.80 yamt 181:
1.136 ad 182: mutex_enter(&so_pendfree_lock);
1.80 yamt 183: while (socurkva + len > somaxkva) {
184: SOSEND_COUNTER_INCR(&sosend_kvalimit);
1.136 ad 185: error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
1.98 christos 186: if (error) {
187: len = 0;
188: break;
189: }
1.80 yamt 190: }
1.93 yamt 191: socurkva += len;
1.136 ad 192: mutex_exit(&so_pendfree_lock);
1.98 christos 193: return len;
1.95 yamt 194: }
195:
1.113 thorpej 196: static void
1.95 yamt 197: sokvaunreserve(vsize_t len)
198: {
199:
1.136 ad 200: mutex_enter(&so_pendfree_lock);
1.95 yamt 201: socurkva -= len;
1.136 ad 202: cv_broadcast(&socurkva_cv);
203: mutex_exit(&so_pendfree_lock);
1.95 yamt 204: }
205:
206: /*
207: * sokvaalloc: allocate kva for loan.
208: */
209: vaddr_t
1.209 matt 210: sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
1.95 yamt 211: {
212: vaddr_t lva;
213:
1.98 christos 214: if (sokvareserve(so, len) == 0)
215: return 0;
1.93 yamt 216:
1.209 matt 217: lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
218: UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
1.95 yamt 219: if (lva == 0) {
220: sokvaunreserve(len);
1.270 maxv 221: return 0;
1.95 yamt 222: }
1.80 yamt 223:
224: return lva;
225: }
226:
1.93 yamt 227: /*
228: * sokvafree: free kva for loan.
229: */
1.80 yamt 230: void
231: sokvafree(vaddr_t sva, vsize_t len)
232: {
1.93 yamt 233:
1.109 yamt 234: uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
1.95 yamt 235: sokvaunreserve(len);
1.80 yamt 236: }
237:
1.64 thorpej 238: static void
1.134 christos 239: sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
1.64 thorpej 240: {
1.156 yamt 241: vaddr_t sva, eva;
1.64 thorpej 242: vsize_t len;
1.156 yamt 243: int npgs;
244:
245: KASSERT(pgs != NULL);
1.64 thorpej 246:
247: eva = round_page((vaddr_t) buf + size);
248: sva = trunc_page((vaddr_t) buf);
249: len = eva - sva;
250: npgs = len >> PAGE_SHIFT;
251:
252: pmap_kremove(sva, len);
253: pmap_update(pmap_kernel());
254: uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
1.80 yamt 255: sokvafree(sva, len);
1.64 thorpej 256: }
257:
1.93 yamt 258: /*
1.270 maxv 259: * sopendfree_thread: free mbufs on "pendfree" list. Unlock and relock
260: * so_pendfree_lock when freeing mbufs.
1.93 yamt 261: */
1.205 bouyer 262: static void
263: sopendfree_thread(void *v)
1.93 yamt 264: {
1.137 ad 265: struct mbuf *m, *next;
1.205 bouyer 266: size_t rv;
1.93 yamt 267:
1.205 bouyer 268: mutex_enter(&so_pendfree_lock);
1.64 thorpej 269:
1.205 bouyer 270: for (;;) {
271: rv = 0;
272: while (so_pendfree != NULL) {
273: m = so_pendfree;
274: so_pendfree = NULL;
275: mutex_exit(&so_pendfree_lock);
276:
277: for (; m != NULL; m = next) {
278: next = m->m_next;
1.253 ryo 279: KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) ==
280: 0);
1.205 bouyer 281: KASSERT(m->m_ext.ext_refcnt == 0);
282:
283: rv += m->m_ext.ext_size;
284: sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
285: m->m_ext.ext_size);
286: pool_cache_put(mb_cache, m);
287: }
1.93 yamt 288:
1.205 bouyer 289: mutex_enter(&so_pendfree_lock);
1.93 yamt 290: }
1.205 bouyer 291: if (rv)
292: cv_broadcast(&socurkva_cv);
293: cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
1.64 thorpej 294: }
1.205 bouyer 295: panic("sopendfree_thread");
296: /* NOTREACHED */
1.64 thorpej 297: }
298:
1.80 yamt 299: void
1.134 christos 300: soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
1.64 thorpej 301: {
302:
1.156 yamt 303: KASSERT(m != NULL);
1.64 thorpej 304:
1.93 yamt 305: /*
306: * postpone freeing mbuf.
307: *
308: * we can't do it in interrupt context
309: * because we need to put kva back to kernel_map.
310: */
311:
1.136 ad 312: mutex_enter(&so_pendfree_lock);
1.92 yamt 313: m->m_next = so_pendfree;
314: so_pendfree = m;
1.205 bouyer 315: cv_signal(&pendfree_thread_cv);
1.136 ad 316: mutex_exit(&so_pendfree_lock);
1.64 thorpej 317: }
318:
319: static long
320: sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
321: {
322: struct iovec *iov = uio->uio_iov;
323: vaddr_t sva, eva;
324: vsize_t len;
1.156 yamt 325: vaddr_t lva;
326: int npgs, error;
327: vaddr_t va;
328: int i;
1.64 thorpej 329:
1.116 yamt 330: if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
1.270 maxv 331: return 0;
1.64 thorpej 332:
333: if (iov->iov_len < (size_t) space)
334: space = iov->iov_len;
335: if (space > SOCK_LOAN_CHUNK)
336: space = SOCK_LOAN_CHUNK;
337:
338: eva = round_page((vaddr_t) iov->iov_base + space);
339: sva = trunc_page((vaddr_t) iov->iov_base);
340: len = eva - sva;
341: npgs = len >> PAGE_SHIFT;
342:
1.79 thorpej 343: KASSERT(npgs <= M_EXT_MAXPAGES);
344:
1.209 matt 345: lva = sokvaalloc(sva, len, so);
1.64 thorpej 346: if (lva == 0)
1.252 uwe 347: return 0;
1.64 thorpej 348:
1.116 yamt 349: error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
1.79 thorpej 350: m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
1.64 thorpej 351: if (error) {
1.80 yamt 352: sokvafree(lva, len);
1.270 maxv 353: return 0;
1.64 thorpej 354: }
355:
356: for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
1.79 thorpej 357: pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
1.194 cegger 358: VM_PROT_READ, 0);
1.64 thorpej 359: pmap_update(pmap_kernel());
360:
361: lva += (vaddr_t) iov->iov_base & PAGE_MASK;
362:
1.134 christos 363: MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
1.79 thorpej 364: m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
1.64 thorpej 365:
366: uio->uio_resid -= space;
367: /* uio_offset not updated, not set/used for write(2) */
1.134 christos 368: uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
1.64 thorpej 369: uio->uio_iov->iov_len -= space;
370: if (uio->uio_iov->iov_len == 0) {
371: uio->uio_iov++;
372: uio->uio_iovcnt--;
373: }
374:
1.270 maxv 375: return space;
1.64 thorpej 376: }
377:
1.191 elad 378: static int
379: socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
380: void *arg0, void *arg1, void *arg2, void *arg3)
381: {
382: int result;
383: enum kauth_network_req req;
384:
385: result = KAUTH_RESULT_DEFER;
386: req = (enum kauth_network_req)arg0;
387:
1.193 elad 388: if ((action != KAUTH_NETWORK_SOCKET) &&
389: (action != KAUTH_NETWORK_BIND))
1.191 elad 390: return result;
391:
392: switch (req) {
1.193 elad 393: case KAUTH_REQ_NETWORK_BIND_PORT:
394: result = KAUTH_RESULT_ALLOW;
395: break;
396:
1.191 elad 397: case KAUTH_REQ_NETWORK_SOCKET_DROP: {
398: /* Normal users can only drop their own connections. */
399: struct socket *so = (struct socket *)arg1;
400:
1.220 christos 401: if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
1.191 elad 402: result = KAUTH_RESULT_ALLOW;
403:
404: break;
405: }
406:
407: case KAUTH_REQ_NETWORK_SOCKET_OPEN:
408: /* We allow "raw" routing/bluetooth sockets to anyone. */
1.254 christos 409: switch ((u_long)arg1) {
410: case PF_ROUTE:
411: case PF_OROUTE:
412: case PF_BLUETOOTH:
1.255 bouyer 413: case PF_CAN:
1.191 elad 414: result = KAUTH_RESULT_ALLOW;
1.254 christos 415: break;
416: default:
1.191 elad 417: /* Privileged, let secmodel handle this. */
418: if ((u_long)arg2 == SOCK_RAW)
419: break;
1.254 christos 420: result = KAUTH_RESULT_ALLOW;
421: break;
1.191 elad 422: }
423: break;
424:
1.192 elad 425: case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
426: result = KAUTH_RESULT_ALLOW;
427:
428: break;
429:
1.191 elad 430: default:
431: break;
432: }
433:
434: return result;
435: }
436:
1.119 yamt 437: void
438: soinit(void)
439: {
440:
1.212 pooka 441: sysctl_kern_socket_setup();
1.178 pooka 442:
1.148 ad 443: mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
1.160 ad 444: softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
1.136 ad 445: cv_init(&socurkva_cv, "sokva");
1.205 bouyer 446: cv_init(&pendfree_thread_cv, "sopendfr");
1.166 ad 447: soinit2();
1.136 ad 448:
1.119 yamt 449: /* Set the initial adjusted socket buffer size. */
450: if (sb_max_set(sb_max))
451: panic("bad initial sb_max value: %lu", sb_max);
452:
1.191 elad 453: socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
454: socket_listener_cb, NULL);
1.119 yamt 455: }
456:
1.205 bouyer 457: void
458: soinit1(void)
459: {
460: int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
461: sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
462: if (error)
463: panic("soinit1 %d", error);
464: }
465:
1.1 cgd 466: /*
1.222 rmind 467: * socreate: create a new socket of the specified type and the protocol.
468: *
469: * => Caller may specify another socket for lock sharing (must not be held).
470: * => Returns the new socket without lock held.
1.224 rmind 471: */
1.3 andrew 472: int
1.160 ad 473: socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
1.270 maxv 474: struct socket *lockso)
1.1 cgd 475: {
1.270 maxv 476: const struct protosw *prp;
477: struct socket *so;
478: uid_t uid;
479: int error;
480: kmutex_t *lock;
1.1 cgd 481:
1.132 elad 482: error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
483: KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
484: KAUTH_ARG(proto));
1.140 dyoung 485: if (error != 0)
486: return error;
1.127 elad 487:
1.1 cgd 488: if (proto)
489: prp = pffindproto(dom, proto, type);
490: else
491: prp = pffindtype(dom, type);
1.140 dyoung 492: if (prp == NULL) {
1.120 ginsbach 493: /* no support for domain */
494: if (pffinddomain(dom) == 0)
1.140 dyoung 495: return EAFNOSUPPORT;
1.120 ginsbach 496: /* no support for socket type */
497: if (proto == 0 && type != 0)
1.140 dyoung 498: return EPROTOTYPE;
499: return EPROTONOSUPPORT;
1.120 ginsbach 500: }
1.223 rmind 501: if (prp->pr_usrreqs == NULL)
1.140 dyoung 502: return EPROTONOSUPPORT;
1.1 cgd 503: if (prp->pr_type != type)
1.140 dyoung 504: return EPROTOTYPE;
1.160 ad 505:
506: so = soget(true);
1.1 cgd 507: so->so_type = type;
508: so->so_proto = prp;
1.33 matt 509: so->so_send = sosend;
510: so->so_receive = soreceive;
1.266 christos 511: so->so_options = sooptions;
1.78 matt 512: #ifdef MBUFTRACE
513: so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
514: so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
515: so->so_mowner = &prp->pr_domain->dom_mowner;
516: #endif
1.138 rmind 517: uid = kauth_cred_geteuid(l->l_cred);
1.115 yamt 518: so->so_uidinfo = uid_find(uid);
1.168 yamt 519: so->so_cpid = l->l_proc->p_pid;
1.224 rmind 520:
521: /*
522: * Lock assigned and taken during PCB attach, unless we share
523: * the lock with another socket, e.g. socketpair(2) case.
524: */
525: if (lockso) {
1.160 ad 526: lock = lockso->so_lock;
527: so->so_lock = lock;
528: mutex_obj_hold(lock);
529: mutex_enter(lock);
530: }
1.224 rmind 531:
532: /* Attach the PCB (returns with the socket lock held). */
533: error = (*prp->pr_usrreqs->pr_attach)(so, proto);
1.160 ad 534: KASSERT(solocked(so));
1.224 rmind 535:
536: if (error) {
1.222 rmind 537: KASSERT(so->so_pcb == NULL);
1.1 cgd 538: so->so_state |= SS_NOFDREF;
539: sofree(so);
1.140 dyoung 540: return error;
1.1 cgd 541: }
1.198 elad 542: so->so_cred = kauth_cred_dup(l->l_cred);
1.160 ad 543: sounlock(so);
1.224 rmind 544:
1.1 cgd 545: *aso = so;
1.140 dyoung 546: return 0;
1.1 cgd 547: }
548:
1.222 rmind 549: /*
550: * fsocreate: create a socket and a file descriptor associated with it.
551: *
552: * => On success, write file descriptor to fdout and return zero.
553: * => On failure, return non-zero; *fdout will be undefined.
1.142 dyoung 554: */
555: int
1.222 rmind 556: fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout)
1.142 dyoung 557: {
1.222 rmind 558: lwp_t *l = curlwp;
559: int error, fd, flags;
560: struct socket *so;
561: struct file *fp;
1.142 dyoung 562:
1.222 rmind 563: if ((error = fd_allocfile(&fp, &fd)) != 0) {
1.204 christos 564: return error;
1.222 rmind 565: }
566: flags = type & SOCK_FLAGS_MASK;
1.204 christos 567: fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
1.207 christos 568: fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
569: ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
1.142 dyoung 570: fp->f_type = DTYPE_SOCKET;
571: fp->f_ops = &socketops;
1.222 rmind 572:
573: type &= ~SOCK_FLAGS_MASK;
574: error = socreate(domain, &so, type, proto, l, NULL);
575: if (error) {
1.155 ad 576: fd_abort(curproc, fp, fd);
1.222 rmind 577: return error;
578: }
579: if (flags & SOCK_NONBLOCK) {
580: so->so_state |= SS_NBIO;
581: }
1.235 matt 582: fp->f_socket = so;
1.222 rmind 583: fd_affix(curproc, fp, fd);
584:
585: if (sop != NULL) {
586: *sop = so;
1.142 dyoung 587: }
1.222 rmind 588: *fdout = fd;
1.142 dyoung 589: return error;
590: }
591:
1.3 andrew 592: int
1.190 dyoung 593: sofamily(const struct socket *so)
594: {
595: const struct protosw *pr;
596: const struct domain *dom;
597:
598: if ((pr = so->so_proto) == NULL)
599: return AF_UNSPEC;
600: if ((dom = pr->pr_domain) == NULL)
601: return AF_UNSPEC;
602: return dom->dom_family;
603: }
604:
605: int
1.236 rtr 606: sobind(struct socket *so, struct sockaddr *nam, struct lwp *l)
1.1 cgd 607: {
1.270 maxv 608: int error;
1.1 cgd 609:
1.160 ad 610: solock(so);
1.237 rtr 611: if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
612: sounlock(so);
1.238 rtr 613: return EAFNOSUPPORT;
1.237 rtr 614: }
1.231 rtr 615: error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
1.160 ad 616: sounlock(so);
1.140 dyoung 617: return error;
1.1 cgd 618: }
619:
1.3 andrew 620: int
1.150 elad 621: solisten(struct socket *so, int backlog, struct lwp *l)
1.1 cgd 622: {
1.270 maxv 623: int error;
624: short oldopt, oldqlimit;
1.1 cgd 625:
1.160 ad 626: solock(so);
1.253 ryo 627: if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1.163 ad 628: SS_ISDISCONNECTING)) != 0) {
1.222 rmind 629: sounlock(so);
630: return EINVAL;
1.163 ad 631: }
1.247 rjs 632: oldopt = so->so_options;
633: oldqlimit = so->so_qlimit;
634: if (TAILQ_EMPTY(&so->so_q))
635: so->so_options |= SO_ACCEPTCONN;
636: if (backlog < 0)
637: backlog = 0;
1.265 riastrad 638: so->so_qlimit = uimin(backlog, somaxconn);
1.247 rjs 639:
1.231 rtr 640: error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
1.140 dyoung 641: if (error != 0) {
1.247 rjs 642: so->so_options = oldopt;
643: so->so_qlimit = oldqlimit;
1.160 ad 644: sounlock(so);
1.140 dyoung 645: return error;
1.1 cgd 646: }
1.160 ad 647: sounlock(so);
1.140 dyoung 648: return 0;
1.1 cgd 649: }
650:
1.21 christos 651: void
1.54 lukem 652: sofree(struct socket *so)
1.1 cgd 653: {
1.161 ad 654: u_int refs;
1.1 cgd 655:
1.160 ad 656: KASSERT(solocked(so));
657:
658: if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
659: sounlock(so);
1.1 cgd 660: return;
1.160 ad 661: }
1.43 mycroft 662: if (so->so_head) {
663: /*
664: * We must not decommission a socket that's on the accept(2)
665: * queue. If we do, then accept(2) may hang after select(2)
666: * indicated that the listening socket was ready.
667: */
1.160 ad 668: if (!soqremque(so, 0)) {
669: sounlock(so);
1.43 mycroft 670: return;
1.160 ad 671: }
1.43 mycroft 672: }
1.98 christos 673: if (so->so_rcv.sb_hiwat)
1.110 christos 674: (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
1.98 christos 675: RLIM_INFINITY);
676: if (so->so_snd.sb_hiwat)
1.110 christos 677: (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
1.98 christos 678: RLIM_INFINITY);
679: sbrelease(&so->so_snd, so);
1.160 ad 680: KASSERT(!cv_has_waiters(&so->so_cv));
681: KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
682: KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
1.1 cgd 683: sorflush(so);
1.161 ad 684: refs = so->so_aborting; /* XXX */
1.177 ad 685: /* Remove acccept filter if one is present. */
1.170 tls 686: if (so->so_accf != NULL)
1.177 ad 687: (void)accept_filt_clear(so);
1.160 ad 688: sounlock(so);
1.161 ad 689: if (refs == 0) /* XXX */
690: soput(so);
1.1 cgd 691: }
692:
693: /*
1.222 rmind 694: * soclose: close a socket on last file table reference removal.
695: * Initiate disconnect if connected. Free socket when disconnect complete.
1.1 cgd 696: */
1.3 andrew 697: int
1.54 lukem 698: soclose(struct socket *so)
1.1 cgd 699: {
1.222 rmind 700: struct socket *so2;
701: int error = 0;
1.1 cgd 702:
1.160 ad 703: solock(so);
1.1 cgd 704: if (so->so_options & SO_ACCEPTCONN) {
1.172 ad 705: for (;;) {
706: if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
1.160 ad 707: KASSERT(solocked2(so, so2));
708: (void) soqremque(so2, 0);
709: /* soabort drops the lock. */
710: (void) soabort(so2);
711: solock(so);
1.172 ad 712: continue;
1.160 ad 713: }
1.172 ad 714: if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
1.160 ad 715: KASSERT(solocked2(so, so2));
716: (void) soqremque(so2, 1);
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: break;
723: }
1.1 cgd 724: }
1.222 rmind 725: if (so->so_pcb == NULL)
1.1 cgd 726: goto discard;
727: if (so->so_state & SS_ISCONNECTED) {
728: if ((so->so_state & SS_ISDISCONNECTING) == 0) {
729: error = sodisconnect(so);
730: if (error)
731: goto drop;
732: }
733: if (so->so_options & SO_LINGER) {
1.206 christos 734: if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
735: (SS_ISDISCONNECTING|SS_NBIO))
1.1 cgd 736: goto drop;
1.21 christos 737: while (so->so_state & SS_ISCONNECTED) {
1.185 yamt 738: error = sowait(so, true, so->so_linger * hz);
1.21 christos 739: if (error)
1.1 cgd 740: break;
1.21 christos 741: }
1.1 cgd 742: }
743: }
1.54 lukem 744: drop:
1.1 cgd 745: if (so->so_pcb) {
1.224 rmind 746: KASSERT(solocked(so));
747: (*so->so_proto->pr_usrreqs->pr_detach)(so);
1.1 cgd 748: }
1.54 lukem 749: discard:
1.222 rmind 750: KASSERT((so->so_state & SS_NOFDREF) == 0);
1.198 elad 751: kauth_cred_free(so->so_cred);
1.273 ! maxv 752: so->so_cred = NULL;
1.1 cgd 753: so->so_state |= SS_NOFDREF;
754: sofree(so);
1.222 rmind 755: return error;
1.1 cgd 756: }
757:
758: /*
1.160 ad 759: * Must be called with the socket locked.. Will return with it unlocked.
1.1 cgd 760: */
1.3 andrew 761: int
1.54 lukem 762: soabort(struct socket *so)
1.1 cgd 763: {
1.161 ad 764: u_int refs;
1.139 yamt 765: int error;
1.253 ryo 766:
1.160 ad 767: KASSERT(solocked(so));
768: KASSERT(so->so_head == NULL);
1.1 cgd 769:
1.161 ad 770: so->so_aborting++; /* XXX */
1.230 mrg 771: error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
1.161 ad 772: refs = --so->so_aborting; /* XXX */
1.164 drochner 773: if (error || (refs == 0)) {
1.139 yamt 774: sofree(so);
1.160 ad 775: } else {
776: sounlock(so);
1.139 yamt 777: }
778: return error;
1.1 cgd 779: }
780:
1.3 andrew 781: int
1.239 rtr 782: soaccept(struct socket *so, struct sockaddr *nam)
1.1 cgd 783: {
1.222 rmind 784: int error;
1.160 ad 785:
786: KASSERT(solocked(so));
1.222 rmind 787: KASSERT((so->so_state & SS_NOFDREF) != 0);
1.1 cgd 788:
789: so->so_state &= ~SS_NOFDREF;
1.55 thorpej 790: if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
791: (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
1.225 rtr 792: error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
1.41 mycroft 793: else
1.53 itojun 794: error = ECONNABORTED;
1.52 itojun 795:
1.222 rmind 796: return error;
1.1 cgd 797: }
798:
1.3 andrew 799: int
1.240 rtr 800: soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l)
1.1 cgd 801: {
1.222 rmind 802: int error;
1.160 ad 803:
804: KASSERT(solocked(so));
1.1 cgd 805:
806: if (so->so_options & SO_ACCEPTCONN)
1.222 rmind 807: return EOPNOTSUPP;
1.1 cgd 808: /*
809: * If protocol is connection-based, can only connect once.
810: * Otherwise, if connected, try to disconnect first.
811: * This allows user to disconnect by connecting to, e.g.,
812: * a null address.
813: */
814: if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
815: ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1.241 rtr 816: (error = sodisconnect(so)))) {
1.1 cgd 817: error = EISCONN;
1.241 rtr 818: } else {
1.242 rtr 819: if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
1.241 rtr 820: return EAFNOSUPPORT;
821: }
1.231 rtr 822: error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
1.241 rtr 823: }
1.222 rmind 824:
825: return error;
1.1 cgd 826: }
827:
1.3 andrew 828: int
1.54 lukem 829: soconnect2(struct socket *so1, struct socket *so2)
1.1 cgd 830: {
1.160 ad 831: KASSERT(solocked2(so1, so2));
1.1 cgd 832:
1.234 rtr 833: return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
1.1 cgd 834: }
835:
1.3 andrew 836: int
1.54 lukem 837: sodisconnect(struct socket *so)
1.1 cgd 838: {
1.270 maxv 839: int error;
1.160 ad 840:
841: KASSERT(solocked(so));
1.1 cgd 842:
843: if ((so->so_state & SS_ISCONNECTED) == 0) {
844: error = ENOTCONN;
1.160 ad 845: } else if (so->so_state & SS_ISDISCONNECTING) {
1.1 cgd 846: error = EALREADY;
1.160 ad 847: } else {
1.229 rtr 848: error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
1.1 cgd 849: }
1.270 maxv 850: return error;
1.1 cgd 851: }
852:
1.15 mycroft 853: #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1.1 cgd 854: /*
855: * Send on a socket.
856: * If send must go all at once and message is larger than
857: * send buffering, then hard error.
858: * Lock against other senders.
859: * If must go all at once and not enough room now, then
860: * inform user that this would block and do nothing.
861: * Otherwise, if nonblocking, send as much as possible.
862: * The data to be sent is described by "uio" if nonzero,
863: * otherwise by the mbuf chain "top" (which must be null
864: * if uio is not). Data provided in mbuf chain must be small
865: * enough to send all at once.
866: *
867: * Returns nonzero on error, timeout or signal; callers
868: * must check for short counts if EINTR/ERESTART are returned.
869: * Data and control buffers are freed on return.
870: */
1.3 andrew 871: int
1.245 rtr 872: sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
873: struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
1.1 cgd 874: {
1.270 maxv 875: struct mbuf **mp, *m;
876: long space, len, resid, clen, mlen;
877: int error, s, dontroute, atomic;
878: short wakeup_state = 0;
1.54 lukem 879:
1.160 ad 880: clen = 0;
1.64 thorpej 881:
1.160 ad 882: /*
883: * solock() provides atomicity of access. splsoftnet() prevents
884: * protocol processing soft interrupts from interrupting us and
885: * blocking (expensive).
886: */
887: s = splsoftnet();
888: solock(so);
1.54 lukem 889: atomic = sosendallatonce(so) || top;
1.1 cgd 890: if (uio)
891: resid = uio->uio_resid;
892: else
893: resid = top->m_pkthdr.len;
1.7 cgd 894: /*
895: * In theory resid should be unsigned.
896: * However, space must be signed, as it might be less than 0
897: * if we over-committed, and we must use a signed comparison
898: * of space and resid. On the other hand, a negative resid
899: * causes us to loop sending 0-length segments to the protocol.
900: */
1.29 mycroft 901: if (resid < 0) {
902: error = EINVAL;
903: goto out;
904: }
1.1 cgd 905: dontroute =
906: (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
907: (so->so_proto->pr_flags & PR_ATOMIC);
1.165 christos 908: l->l_ru.ru_msgsnd++;
1.1 cgd 909: if (control)
910: clen = control->m_len;
1.54 lukem 911: restart:
1.21 christos 912: if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
1.1 cgd 913: goto out;
914: do {
1.160 ad 915: if (so->so_state & SS_CANTSENDMORE) {
916: error = EPIPE;
917: goto release;
918: }
1.48 thorpej 919: if (so->so_error) {
920: error = so->so_error;
921: so->so_error = 0;
922: goto release;
923: }
1.1 cgd 924: if ((so->so_state & SS_ISCONNECTED) == 0) {
925: if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1.217 rmind 926: if (resid || clen == 0) {
1.160 ad 927: error = ENOTCONN;
928: goto release;
929: }
1.244 rtr 930: } else if (addr == NULL) {
1.160 ad 931: error = EDESTADDRREQ;
932: goto release;
933: }
1.1 cgd 934: }
935: space = sbspace(&so->so_snd);
936: if (flags & MSG_OOB)
937: space += 1024;
1.21 christos 938: if ((atomic && resid > so->so_snd.sb_hiwat) ||
1.160 ad 939: clen > so->so_snd.sb_hiwat) {
940: error = EMSGSIZE;
941: goto release;
942: }
1.96 mycroft 943: if (space < resid + clen &&
1.1 cgd 944: (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1.206 christos 945: if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1.160 ad 946: error = EWOULDBLOCK;
947: goto release;
948: }
1.1 cgd 949: sbunlock(&so->so_snd);
1.196 dsl 950: if (wakeup_state & SS_RESTARTSYS) {
951: error = ERESTART;
952: goto out;
953: }
1.1 cgd 954: error = sbwait(&so->so_snd);
955: if (error)
956: goto out;
1.196 dsl 957: wakeup_state = so->so_state;
1.1 cgd 958: goto restart;
959: }
1.196 dsl 960: wakeup_state = 0;
1.1 cgd 961: mp = ⊤
962: space -= clen;
963: do {
1.45 tv 964: if (uio == NULL) {
965: /*
966: * Data is prepackaged in "top".
967: */
968: resid = 0;
969: if (flags & MSG_EOR)
970: top->m_flags |= M_EOR;
971: } else do {
1.160 ad 972: sounlock(so);
973: splx(s);
1.144 dyoung 974: if (top == NULL) {
1.78 matt 975: m = m_gethdr(M_WAIT, MT_DATA);
1.45 tv 976: mlen = MHLEN;
977: m->m_pkthdr.len = 0;
1.248 ozaki-r 978: m_reset_rcvif(m);
1.45 tv 979: } else {
1.78 matt 980: m = m_get(M_WAIT, MT_DATA);
1.45 tv 981: mlen = MLEN;
982: }
1.78 matt 983: MCLAIM(m, so->so_snd.sb_mowner);
1.121 yamt 984: if (sock_loan_thresh >= 0 &&
985: uio->uio_iov->iov_len >= sock_loan_thresh &&
986: space >= sock_loan_thresh &&
1.64 thorpej 987: (len = sosend_loan(so, uio, m,
1.252 uwe 988: space)) != 0) {
1.64 thorpej 989: SOSEND_COUNTER_INCR(&sosend_loan_big);
990: space -= len;
991: goto have_data;
992: }
1.45 tv 993: if (resid >= MINCLSIZE && space >= MCLBYTES) {
1.64 thorpej 994: SOSEND_COUNTER_INCR(&sosend_copy_big);
1.201 oki 995: m_clget(m, M_DONTWAIT);
1.45 tv 996: if ((m->m_flags & M_EXT) == 0)
997: goto nopages;
998: mlen = MCLBYTES;
999: if (atomic && top == 0) {
1.58 jdolecek 1000: len = lmin(MCLBYTES - max_hdr,
1.54 lukem 1001: resid);
1.45 tv 1002: m->m_data += max_hdr;
1003: } else
1.58 jdolecek 1004: len = lmin(MCLBYTES, resid);
1.45 tv 1005: space -= len;
1006: } else {
1.64 thorpej 1007: nopages:
1008: SOSEND_COUNTER_INCR(&sosend_copy_small);
1.58 jdolecek 1009: len = lmin(lmin(mlen, resid), space);
1.45 tv 1010: space -= len;
1011: /*
1012: * For datagram protocols, leave room
1013: * for protocol headers in first mbuf.
1014: */
1015: if (atomic && top == 0 && len < mlen)
1.268 maxv 1016: m_align(m, len);
1.45 tv 1017: }
1.144 dyoung 1018: error = uiomove(mtod(m, void *), (int)len, uio);
1.64 thorpej 1019: have_data:
1.45 tv 1020: resid = uio->uio_resid;
1021: m->m_len = len;
1022: *mp = m;
1023: top->m_pkthdr.len += len;
1.160 ad 1024: s = splsoftnet();
1025: solock(so);
1.144 dyoung 1026: if (error != 0)
1.45 tv 1027: goto release;
1028: mp = &m->m_next;
1029: if (resid <= 0) {
1030: if (flags & MSG_EOR)
1031: top->m_flags |= M_EOR;
1032: break;
1033: }
1034: } while (space > 0 && atomic);
1.108 perry 1035:
1.160 ad 1036: if (so->so_state & SS_CANTSENDMORE) {
1037: error = EPIPE;
1038: goto release;
1039: }
1.45 tv 1040: if (dontroute)
1041: so->so_options |= SO_DONTROUTE;
1042: if (resid > 0)
1043: so->so_state |= SS_MORETOCOME;
1.240 rtr 1044: if (flags & MSG_OOB) {
1.253 ryo 1045: error = (*so->so_proto->pr_usrreqs->pr_sendoob)(
1046: so, top, control);
1.240 rtr 1047: } else {
1.232 rtr 1048: error = (*so->so_proto->pr_usrreqs->pr_send)(so,
1.245 rtr 1049: top, addr, control, l);
1.240 rtr 1050: }
1.45 tv 1051: if (dontroute)
1052: so->so_options &= ~SO_DONTROUTE;
1053: if (resid > 0)
1054: so->so_state &= ~SS_MORETOCOME;
1055: clen = 0;
1.144 dyoung 1056: control = NULL;
1057: top = NULL;
1.45 tv 1058: mp = ⊤
1.144 dyoung 1059: if (error != 0)
1.1 cgd 1060: goto release;
1061: } while (resid && space > 0);
1062: } while (resid);
1063:
1.54 lukem 1064: release:
1.1 cgd 1065: sbunlock(&so->so_snd);
1.54 lukem 1066: out:
1.160 ad 1067: sounlock(so);
1068: splx(s);
1.1 cgd 1069: if (top)
1070: m_freem(top);
1071: if (control)
1072: m_freem(control);
1.270 maxv 1073: return error;
1.1 cgd 1074: }
1075:
1076: /*
1.159 ad 1077: * Following replacement or removal of the first mbuf on the first
1078: * mbuf chain of a socket buffer, push necessary state changes back
1079: * into the socket buffer so that other consumers see the values
1.270 maxv 1080: * consistently. 'nextrecord' is the caller's locally stored value of
1.159 ad 1081: * the original value of sb->sb_mb->m_nextpkt which must be restored
1082: * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL.
1083: */
1084: static void
1085: sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
1086: {
1087:
1.160 ad 1088: KASSERT(solocked(sb->sb_so));
1089:
1.159 ad 1090: /*
1091: * First, update for the new value of nextrecord. If necessary,
1092: * make it the first record.
1093: */
1094: if (sb->sb_mb != NULL)
1095: sb->sb_mb->m_nextpkt = nextrecord;
1096: else
1097: sb->sb_mb = nextrecord;
1098:
1099: /*
1100: * Now update any dependent socket buffer fields to reflect
1101: * the new state. This is an inline of SB_EMPTY_FIXUP, with
1102: * the addition of a second clause that takes care of the
1103: * case where sb_mb has been updated, but remains the last
1104: * record.
1105: */
1106: if (sb->sb_mb == NULL) {
1107: sb->sb_mbtail = NULL;
1108: sb->sb_lastrecord = NULL;
1109: } else if (sb->sb_mb->m_nextpkt == NULL)
1110: sb->sb_lastrecord = sb->sb_mb;
1111: }
1112:
1113: /*
1.1 cgd 1114: * Implement receive operations on a socket.
1115: *
1.270 maxv 1116: * We depend on the way that records are added to the sockbuf by sbappend*. In
1117: * particular, each record (mbufs linked through m_next) must begin with an
1118: * address if the protocol so specifies, followed by an optional mbuf or mbufs
1119: * containing ancillary data, and then zero or more mbufs of data.
1120: *
1121: * In order to avoid blocking network interrupts for the entire time here, we
1122: * splx() while doing the actual copy to user space. Although the sockbuf is
1123: * locked, new data may still be appended, and thus we must maintain
1124: * consistency of the sockbuf during that time.
1125: *
1126: * The caller may receive the data as a single mbuf chain by supplying an mbuf
1127: * **mp0 for use in returning the chain. The uio is then used only for the
1128: * count in uio_resid.
1.1 cgd 1129: */
1.3 andrew 1130: int
1.54 lukem 1131: soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
1.270 maxv 1132: struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1.1 cgd 1133: {
1.116 yamt 1134: struct lwp *l = curlwp;
1.270 maxv 1135: struct mbuf *m, **mp, *mt;
1.211 chs 1136: size_t len, offset, moff, orig_resid;
1137: int atomic, flags, error, s, type;
1.270 maxv 1138: const struct protosw *pr;
1139: struct mbuf *nextrecord;
1140: int mbuf_removed = 0;
1.146 dyoung 1141: const struct domain *dom;
1.270 maxv 1142: short wakeup_state = 0;
1.64 thorpej 1143:
1.54 lukem 1144: pr = so->so_proto;
1.146 dyoung 1145: atomic = pr->pr_flags & PR_ATOMIC;
1146: dom = pr->pr_domain;
1.1 cgd 1147: mp = mp0;
1.54 lukem 1148: type = 0;
1149: orig_resid = uio->uio_resid;
1.102 jonathan 1150:
1.144 dyoung 1151: if (paddr != NULL)
1152: *paddr = NULL;
1153: if (controlp != NULL)
1154: *controlp = NULL;
1155: if (flagsp != NULL)
1.252 uwe 1156: flags = *flagsp &~ MSG_EOR;
1.1 cgd 1157: else
1158: flags = 0;
1.66 enami 1159:
1.1 cgd 1160: if (flags & MSG_OOB) {
1161: m = m_get(M_WAIT, MT_DATA);
1.160 ad 1162: solock(so);
1.226 rtr 1163: error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
1.160 ad 1164: sounlock(so);
1.1 cgd 1165: if (error)
1166: goto bad;
1167: do {
1.134 christos 1168: error = uiomove(mtod(m, void *),
1.211 chs 1169: MIN(uio->uio_resid, m->m_len), uio);
1.1 cgd 1170: m = m_free(m);
1.144 dyoung 1171: } while (uio->uio_resid > 0 && error == 0 && m);
1.270 maxv 1172: bad:
1.144 dyoung 1173: if (m != NULL)
1.1 cgd 1174: m_freem(m);
1.144 dyoung 1175: return error;
1.1 cgd 1176: }
1.144 dyoung 1177: if (mp != NULL)
1.140 dyoung 1178: *mp = NULL;
1.160 ad 1179:
1180: /*
1181: * solock() provides atomicity of access. splsoftnet() prevents
1182: * protocol processing soft interrupts from interrupting us and
1183: * blocking (expensive).
1184: */
1185: s = splsoftnet();
1186: solock(so);
1.270 maxv 1187: restart:
1.160 ad 1188: if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
1189: sounlock(so);
1190: splx(s);
1.144 dyoung 1191: return error;
1.160 ad 1192: }
1.270 maxv 1193: m = so->so_rcv.sb_mb;
1.1 cgd 1194:
1195: /*
1196: * If we have less data than requested, block awaiting more
1197: * (subject to any timeout) if:
1.15 mycroft 1198: * 1. the current count is less than the low water mark,
1.1 cgd 1199: * 2. MSG_WAITALL is set, and it is possible to do the entire
1.15 mycroft 1200: * receive operation at once if we block (resid <= hiwat), or
1201: * 3. MSG_DONTWAIT is not set.
1.1 cgd 1202: * If MSG_WAITALL is set but resid is larger than the receive buffer,
1203: * we have to do the receive in sections, and thus risk returning
1204: * a short count if a timeout or signal occurs after we start.
1205: */
1.144 dyoung 1206: if (m == NULL ||
1207: ((flags & MSG_DONTWAIT) == 0 &&
1208: so->so_rcv.sb_cc < uio->uio_resid &&
1209: (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1210: ((flags & MSG_WAITALL) &&
1211: uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1.146 dyoung 1212: m->m_nextpkt == NULL && !atomic)) {
1.1 cgd 1213: #ifdef DIAGNOSTIC
1.144 dyoung 1214: if (m == NULL && so->so_rcv.sb_cc)
1.1 cgd 1215: panic("receive 1");
1216: #endif
1.264 roy 1217: if (so->so_error || so->so_rerror) {
1.144 dyoung 1218: if (m != NULL)
1.15 mycroft 1219: goto dontblock;
1.264 roy 1220: if (so->so_error) {
1221: error = so->so_error;
1222: so->so_error = 0;
1223: } else {
1224: error = so->so_rerror;
1225: so->so_rerror = 0;
1226: }
1.1 cgd 1227: goto release;
1228: }
1229: if (so->so_state & SS_CANTRCVMORE) {
1.144 dyoung 1230: if (m != NULL)
1.15 mycroft 1231: goto dontblock;
1.1 cgd 1232: else
1233: goto release;
1234: }
1.144 dyoung 1235: for (; m != NULL; m = m->m_next)
1.1 cgd 1236: if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1237: m = so->so_rcv.sb_mb;
1238: goto dontblock;
1239: }
1240: if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1241: (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1242: error = ENOTCONN;
1243: goto release;
1244: }
1245: if (uio->uio_resid == 0)
1246: goto release;
1.206 christos 1247: if ((so->so_state & SS_NBIO) ||
1248: (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1.1 cgd 1249: error = EWOULDBLOCK;
1250: goto release;
1251: }
1.69 thorpej 1252: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
1253: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
1.1 cgd 1254: sbunlock(&so->so_rcv);
1.196 dsl 1255: if (wakeup_state & SS_RESTARTSYS)
1256: error = ERESTART;
1257: else
1258: error = sbwait(&so->so_rcv);
1.160 ad 1259: if (error != 0) {
1260: sounlock(so);
1261: splx(s);
1.144 dyoung 1262: return error;
1.160 ad 1263: }
1.196 dsl 1264: wakeup_state = so->so_state;
1.1 cgd 1265: goto restart;
1266: }
1.270 maxv 1267:
1268: dontblock:
1.69 thorpej 1269: /*
1270: * On entry here, m points to the first record of the socket buffer.
1.159 ad 1271: * From this point onward, we maintain 'nextrecord' as a cache of the
1272: * pointer to the next record in the socket buffer. We must keep the
1273: * various socket buffer pointers and local stack versions of the
1274: * pointers in sync, pushing out modifications before dropping the
1.160 ad 1275: * socket lock, and re-reading them when picking it up.
1.159 ad 1276: *
1277: * Otherwise, we will race with the network stack appending new data
1278: * or records onto the socket buffer by using inconsistent/stale
1279: * versions of the field, possibly resulting in socket buffer
1280: * corruption.
1281: *
1282: * By holding the high-level sblock(), we prevent simultaneous
1283: * readers from pulling off the front of the socket buffer.
1.69 thorpej 1284: */
1.144 dyoung 1285: if (l != NULL)
1.157 ad 1286: l->l_ru.ru_msgrcv++;
1.69 thorpej 1287: KASSERT(m == so->so_rcv.sb_mb);
1288: SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
1289: SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
1.1 cgd 1290: nextrecord = m->m_nextpkt;
1.270 maxv 1291:
1.1 cgd 1292: if (pr->pr_flags & PR_ADDR) {
1.270 maxv 1293: KASSERT(m->m_type == MT_SONAME);
1.3 andrew 1294: orig_resid = 0;
1.1 cgd 1295: if (flags & MSG_PEEK) {
1296: if (paddr)
1.263 maxv 1297: *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
1.1 cgd 1298: m = m->m_next;
1299: } else {
1300: sbfree(&so->so_rcv, m);
1.67 he 1301: mbuf_removed = 1;
1.144 dyoung 1302: if (paddr != NULL) {
1.1 cgd 1303: *paddr = m;
1304: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1305: m->m_next = NULL;
1.1 cgd 1306: m = so->so_rcv.sb_mb;
1307: } else {
1.249 christos 1308: m = so->so_rcv.sb_mb = m_free(m);
1.1 cgd 1309: }
1.159 ad 1310: sbsync(&so->so_rcv, nextrecord);
1.1 cgd 1311: }
1312: }
1.270 maxv 1313:
1.247 rjs 1314: if (pr->pr_flags & PR_ADDR_OPT) {
1315: /*
1.270 maxv 1316: * For SCTP we may be getting a whole message OR a partial
1317: * delivery.
1.247 rjs 1318: */
1319: if (m->m_type == MT_SONAME) {
1320: orig_resid = 0;
1321: if (flags & MSG_PEEK) {
1322: if (paddr)
1.263 maxv 1323: *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
1.247 rjs 1324: m = m->m_next;
1325: } else {
1326: sbfree(&so->so_rcv, m);
1327: if (paddr) {
1328: *paddr = m;
1329: so->so_rcv.sb_mb = m->m_next;
1330: m->m_next = 0;
1331: m = so->so_rcv.sb_mb;
1332: } else {
1.249 christos 1333: m = so->so_rcv.sb_mb = m_free(m);
1.247 rjs 1334: }
1335: }
1336: }
1337: }
1.159 ad 1338:
1339: /*
1340: * Process one or more MT_CONTROL mbufs present before any data mbufs
1341: * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1342: * just copy the data; if !MSG_PEEK, we call into the protocol to
1343: * perform externalization (or freeing if controlp == NULL).
1344: */
1345: if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
1346: struct mbuf *cm = NULL, *cmn;
1347: struct mbuf **cme = &cm;
1348:
1349: do {
1350: if (flags & MSG_PEEK) {
1351: if (controlp != NULL) {
1.263 maxv 1352: *controlp = m_copym(m, 0, m->m_len, M_DONTWAIT);
1.159 ad 1353: controlp = &(*controlp)->m_next;
1354: }
1355: m = m->m_next;
1356: } else {
1357: sbfree(&so->so_rcv, m);
1.1 cgd 1358: so->so_rcv.sb_mb = m->m_next;
1.144 dyoung 1359: m->m_next = NULL;
1.159 ad 1360: *cme = m;
1361: cme = &(*cme)->m_next;
1.1 cgd 1362: m = so->so_rcv.sb_mb;
1.159 ad 1363: }
1364: } while (m != NULL && m->m_type == MT_CONTROL);
1365: if ((flags & MSG_PEEK) == 0)
1366: sbsync(&so->so_rcv, nextrecord);
1.270 maxv 1367:
1.159 ad 1368: for (; cm != NULL; cm = cmn) {
1369: cmn = cm->m_next;
1370: cm->m_next = NULL;
1371: type = mtod(cm, struct cmsghdr *)->cmsg_type;
1372: if (controlp != NULL) {
1373: if (dom->dom_externalize != NULL &&
1374: type == SCM_RIGHTS) {
1.160 ad 1375: sounlock(so);
1.159 ad 1376: splx(s);
1.204 christos 1377: error = (*dom->dom_externalize)(cm, l,
1378: (flags & MSG_CMSG_CLOEXEC) ?
1379: O_CLOEXEC : 0);
1.159 ad 1380: s = splsoftnet();
1.160 ad 1381: solock(so);
1.159 ad 1382: }
1383: *controlp = cm;
1384: while (*controlp != NULL)
1385: controlp = &(*controlp)->m_next;
1.1 cgd 1386: } else {
1.106 itojun 1387: /*
1388: * Dispose of any SCM_RIGHTS message that went
1389: * through the read path rather than recv.
1390: */
1.159 ad 1391: if (dom->dom_dispose != NULL &&
1392: type == SCM_RIGHTS) {
1.253 ryo 1393: sounlock(so);
1.159 ad 1394: (*dom->dom_dispose)(cm);
1.160 ad 1395: solock(so);
1.159 ad 1396: }
1397: m_freem(cm);
1.1 cgd 1398: }
1399: }
1.159 ad 1400: if (m != NULL)
1401: nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1402: else
1403: nextrecord = so->so_rcv.sb_mb;
1404: orig_resid = 0;
1.1 cgd 1405: }
1.69 thorpej 1406:
1.159 ad 1407: /* If m is non-NULL, we have some data to read. */
1408: if (__predict_true(m != NULL)) {
1.1 cgd 1409: type = m->m_type;
1410: if (type == MT_OOBDATA)
1411: flags |= MSG_OOB;
1412: }
1.69 thorpej 1413: SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1414: SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1415:
1.1 cgd 1416: moff = 0;
1417: offset = 0;
1.144 dyoung 1418: while (m != NULL && uio->uio_resid > 0 && error == 0) {
1.272 maxv 1419: /*
1420: * If the type of mbuf has changed, end the receive
1421: * operation and do a short read.
1422: */
1.1 cgd 1423: if (m->m_type == MT_OOBDATA) {
1424: if (type != MT_OOBDATA)
1425: break;
1.270 maxv 1426: } else if (type == MT_OOBDATA) {
1.1 cgd 1427: break;
1.272 maxv 1428: } else if (m->m_type == MT_CONTROL) {
1429: break;
1.270 maxv 1430: }
1.1 cgd 1431: #ifdef DIAGNOSTIC
1.270 maxv 1432: else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
1.272 maxv 1433: panic("%s: m_type=%d", __func__, m->m_type);
1.270 maxv 1434: }
1.1 cgd 1435: #endif
1.270 maxv 1436:
1.1 cgd 1437: so->so_state &= ~SS_RCVATMARK;
1.196 dsl 1438: wakeup_state = 0;
1.1 cgd 1439: len = uio->uio_resid;
1440: if (so->so_oobmark && len > so->so_oobmark - offset)
1441: len = so->so_oobmark - offset;
1442: if (len > m->m_len - moff)
1443: len = m->m_len - moff;
1.270 maxv 1444:
1.1 cgd 1445: /*
1446: * If mp is set, just pass back the mbufs.
1447: * Otherwise copy them out via the uio, then free.
1448: * Sockbuf must be consistent here (points to current mbuf,
1449: * it points to next record) when we drop priority;
1450: * we must note any additions to the sockbuf when we
1451: * block interrupts again.
1452: */
1.144 dyoung 1453: if (mp == NULL) {
1.69 thorpej 1454: SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1455: SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1.160 ad 1456: sounlock(so);
1.1 cgd 1457: splx(s);
1.211 chs 1458: error = uiomove(mtod(m, char *) + moff, len, uio);
1.20 mycroft 1459: s = splsoftnet();
1.160 ad 1460: solock(so);
1.144 dyoung 1461: if (error != 0) {
1.67 he 1462: /*
1463: * If any part of the record has been removed
1464: * (such as the MT_SONAME mbuf, which will
1465: * happen when PR_ADDR, and thus also
1466: * PR_ATOMIC, is set), then drop the entire
1467: * record to maintain the atomicity of the
1468: * receive operation.
1469: *
1470: * This avoids a later panic("receive 1a")
1471: * when compiled with DIAGNOSTIC.
1472: */
1.146 dyoung 1473: if (m && mbuf_removed && atomic)
1.67 he 1474: (void) sbdroprecord(&so->so_rcv);
1475:
1.57 jdolecek 1476: goto release;
1.67 he 1477: }
1.270 maxv 1478: } else {
1.1 cgd 1479: uio->uio_resid -= len;
1.270 maxv 1480: }
1481:
1.1 cgd 1482: if (len == m->m_len - moff) {
1483: if (m->m_flags & M_EOR)
1484: flags |= MSG_EOR;
1.247 rjs 1485: #ifdef SCTP
1486: if (m->m_flags & M_NOTIFICATION)
1487: flags |= MSG_NOTIFICATION;
1.270 maxv 1488: #endif
1.1 cgd 1489: if (flags & MSG_PEEK) {
1490: m = m->m_next;
1491: moff = 0;
1492: } else {
1493: nextrecord = m->m_nextpkt;
1494: sbfree(&so->so_rcv, m);
1495: if (mp) {
1496: *mp = m;
1497: mp = &m->m_next;
1498: so->so_rcv.sb_mb = m = m->m_next;
1.140 dyoung 1499: *mp = NULL;
1.1 cgd 1500: } else {
1.249 christos 1501: m = so->so_rcv.sb_mb = m_free(m);
1.1 cgd 1502: }
1.69 thorpej 1503: /*
1504: * If m != NULL, we also know that
1505: * so->so_rcv.sb_mb != NULL.
1506: */
1507: KASSERT(so->so_rcv.sb_mb == m);
1508: if (m) {
1.1 cgd 1509: m->m_nextpkt = nextrecord;
1.69 thorpej 1510: if (nextrecord == NULL)
1511: so->so_rcv.sb_lastrecord = m;
1512: } else {
1513: so->so_rcv.sb_mb = nextrecord;
1.70 thorpej 1514: SB_EMPTY_FIXUP(&so->so_rcv);
1.69 thorpej 1515: }
1516: SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1517: SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1.1 cgd 1518: }
1.270 maxv 1519: } else if (flags & MSG_PEEK) {
1.144 dyoung 1520: moff += len;
1.270 maxv 1521: } else {
1.160 ad 1522: if (mp != NULL) {
1523: mt = m_copym(m, 0, len, M_NOWAIT);
1524: if (__predict_false(mt == NULL)) {
1525: sounlock(so);
1526: mt = m_copym(m, 0, len, M_WAIT);
1527: solock(so);
1528: }
1529: *mp = mt;
1530: }
1.144 dyoung 1531: m->m_data += len;
1532: m->m_len -= len;
1533: so->so_rcv.sb_cc -= len;
1.1 cgd 1534: }
1.270 maxv 1535:
1.1 cgd 1536: if (so->so_oobmark) {
1537: if ((flags & MSG_PEEK) == 0) {
1538: so->so_oobmark -= len;
1539: if (so->so_oobmark == 0) {
1540: so->so_state |= SS_RCVATMARK;
1541: break;
1542: }
1.7 cgd 1543: } else {
1.1 cgd 1544: offset += len;
1.7 cgd 1545: if (offset == so->so_oobmark)
1546: break;
1547: }
1.1 cgd 1548: }
1549: if (flags & MSG_EOR)
1550: break;
1.270 maxv 1551:
1.1 cgd 1552: /*
1553: * If the MSG_WAITALL flag is set (for non-atomic socket),
1554: * we must not quit until "uio->uio_resid == 0" or an error
1555: * termination. If a signal/timeout occurs, return
1556: * with a short count but without error.
1557: * Keep sockbuf locked against other readers.
1558: */
1.144 dyoung 1559: while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1.3 andrew 1560: !sosendallatonce(so) && !nextrecord) {
1.264 roy 1561: if (so->so_error || so->so_rerror ||
1562: so->so_state & SS_CANTRCVMORE)
1.1 cgd 1563: break;
1.68 matt 1564: /*
1565: * If we are peeking and the socket receive buffer is
1566: * full, stop since we can't get more data to peek at.
1567: */
1568: if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1569: break;
1570: /*
1571: * If we've drained the socket buffer, tell the
1572: * protocol in case it needs to do something to
1573: * get it filled again.
1574: */
1575: if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1.233 rtr 1576: (*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1.69 thorpej 1577: SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1578: SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1.196 dsl 1579: if (wakeup_state & SS_RESTARTSYS)
1580: error = ERESTART;
1581: else
1582: error = sbwait(&so->so_rcv);
1.144 dyoung 1583: if (error != 0) {
1.1 cgd 1584: sbunlock(&so->so_rcv);
1.160 ad 1585: sounlock(so);
1.1 cgd 1586: splx(s);
1.144 dyoung 1587: return 0;
1.1 cgd 1588: }
1.21 christos 1589: if ((m = so->so_rcv.sb_mb) != NULL)
1.1 cgd 1590: nextrecord = m->m_nextpkt;
1.196 dsl 1591: wakeup_state = so->so_state;
1.1 cgd 1592: }
1593: }
1.3 andrew 1594:
1.146 dyoung 1595: if (m && atomic) {
1.3 andrew 1596: flags |= MSG_TRUNC;
1597: if ((flags & MSG_PEEK) == 0)
1598: (void) sbdroprecord(&so->so_rcv);
1599: }
1.1 cgd 1600: if ((flags & MSG_PEEK) == 0) {
1.144 dyoung 1601: if (m == NULL) {
1.69 thorpej 1602: /*
1.70 thorpej 1603: * First part is an inline SB_EMPTY_FIXUP(). Second
1.69 thorpej 1604: * part makes sure sb_lastrecord is up-to-date if
1605: * there is still data in the socket buffer.
1606: */
1.1 cgd 1607: so->so_rcv.sb_mb = nextrecord;
1.69 thorpej 1608: if (so->so_rcv.sb_mb == NULL) {
1609: so->so_rcv.sb_mbtail = NULL;
1610: so->so_rcv.sb_lastrecord = NULL;
1611: } else if (nextrecord->m_nextpkt == NULL)
1612: so->so_rcv.sb_lastrecord = nextrecord;
1613: }
1614: SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1615: SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1.1 cgd 1616: if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1.233 rtr 1617: (*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
1.1 cgd 1618: }
1.3 andrew 1619: if (orig_resid == uio->uio_resid && orig_resid &&
1620: (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1621: sbunlock(&so->so_rcv);
1622: goto restart;
1623: }
1.108 perry 1624:
1.144 dyoung 1625: if (flagsp != NULL)
1.1 cgd 1626: *flagsp |= flags;
1.270 maxv 1627: release:
1.1 cgd 1628: sbunlock(&so->so_rcv);
1.160 ad 1629: sounlock(so);
1.1 cgd 1630: splx(s);
1.144 dyoung 1631: return error;
1.1 cgd 1632: }
1633:
1.14 mycroft 1634: int
1.54 lukem 1635: soshutdown(struct socket *so, int how)
1.1 cgd 1636: {
1.270 maxv 1637: const struct protosw *pr;
1638: int error;
1.160 ad 1639:
1640: KASSERT(solocked(so));
1.34 kleink 1641:
1.54 lukem 1642: pr = so->so_proto;
1.34 kleink 1643: if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1.270 maxv 1644: return EINVAL;
1.1 cgd 1645:
1.160 ad 1646: if (how == SHUT_RD || how == SHUT_RDWR) {
1.1 cgd 1647: sorflush(so);
1.160 ad 1648: error = 0;
1649: }
1.34 kleink 1650: if (how == SHUT_WR || how == SHUT_RDWR)
1.229 rtr 1651: error = (*pr->pr_usrreqs->pr_shutdown)(so);
1.160 ad 1652:
1653: return error;
1.1 cgd 1654: }
1655:
1.195 dsl 1656: void
1.196 dsl 1657: sorestart(struct socket *so)
1.188 ad 1658: {
1.196 dsl 1659: /*
1660: * An application has called close() on an fd on which another
1661: * of its threads has called a socket system call.
1662: * Mark this and wake everyone up, and code that would block again
1663: * instead returns ERESTART.
1664: * On system call re-entry the fd is validated and EBADF returned.
1665: * Any other fd will block again on the 2nd syscall.
1666: */
1.188 ad 1667: solock(so);
1.196 dsl 1668: so->so_state |= SS_RESTARTSYS;
1.188 ad 1669: cv_broadcast(&so->so_cv);
1.196 dsl 1670: cv_broadcast(&so->so_snd.sb_cv);
1671: cv_broadcast(&so->so_rcv.sb_cv);
1.188 ad 1672: sounlock(so);
1673: }
1674:
1.14 mycroft 1675: void
1.54 lukem 1676: sorflush(struct socket *so)
1.1 cgd 1677: {
1.270 maxv 1678: struct sockbuf *sb, asb;
1679: const struct protosw *pr;
1.160 ad 1680:
1681: KASSERT(solocked(so));
1.1 cgd 1682:
1.54 lukem 1683: sb = &so->so_rcv;
1684: pr = so->so_proto;
1.160 ad 1685: socantrcvmore(so);
1.1 cgd 1686: sb->sb_flags |= SB_NOINTR;
1.160 ad 1687: (void )sblock(sb, M_WAITOK);
1.1 cgd 1688: sbunlock(sb);
1689: asb = *sb;
1.86 wrstuden 1690: /*
1691: * Clear most of the sockbuf structure, but leave some of the
1692: * fields valid.
1693: */
1694: memset(&sb->sb_startzero, 0,
1695: sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1.160 ad 1696: if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
1697: sounlock(so);
1.1 cgd 1698: (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1.160 ad 1699: solock(so);
1700: }
1.98 christos 1701: sbrelease(&asb, so);
1.1 cgd 1702: }
1703:
1.171 plunky 1704: /*
1705: * internal set SOL_SOCKET options
1706: */
1.142 dyoung 1707: static int
1.171 plunky 1708: sosetopt1(struct socket *so, const struct sockopt *sopt)
1.1 cgd 1709: {
1.219 christos 1710: int error = EINVAL, opt;
1711: int optval = 0; /* XXX: gcc */
1.171 plunky 1712: struct linger l;
1713: struct timeval tv;
1.142 dyoung 1714:
1.179 christos 1715: switch ((opt = sopt->sopt_name)) {
1.142 dyoung 1716:
1.170 tls 1717: case SO_ACCEPTFILTER:
1.177 ad 1718: error = accept_filt_setopt(so, sopt);
1719: KASSERT(solocked(so));
1.170 tls 1720: break;
1721:
1.253 ryo 1722: case SO_LINGER:
1723: error = sockopt_get(sopt, &l, sizeof(l));
1.177 ad 1724: solock(so);
1.253 ryo 1725: if (error)
1726: break;
1727: if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
1728: l.l_linger > (INT_MAX / hz)) {
1.177 ad 1729: error = EDOM;
1730: break;
1731: }
1.253 ryo 1732: so->so_linger = l.l_linger;
1733: if (l.l_onoff)
1734: so->so_options |= SO_LINGER;
1735: else
1736: so->so_options &= ~SO_LINGER;
1737: break;
1.1 cgd 1738:
1.142 dyoung 1739: case SO_DEBUG:
1740: case SO_KEEPALIVE:
1741: case SO_DONTROUTE:
1742: case SO_USELOOPBACK:
1743: case SO_BROADCAST:
1744: case SO_REUSEADDR:
1745: case SO_REUSEPORT:
1746: case SO_OOBINLINE:
1747: case SO_TIMESTAMP:
1.207 christos 1748: case SO_NOSIGPIPE:
1.266 christos 1749: case SO_RERROR:
1.184 christos 1750: #ifdef SO_OTIMESTAMP
1751: case SO_OTIMESTAMP:
1752: #endif
1.171 plunky 1753: error = sockopt_getint(sopt, &optval);
1.177 ad 1754: solock(so);
1.171 plunky 1755: if (error)
1.177 ad 1756: break;
1.171 plunky 1757: if (optval)
1.179 christos 1758: so->so_options |= opt;
1.142 dyoung 1759: else
1.179 christos 1760: so->so_options &= ~opt;
1.142 dyoung 1761: break;
1762:
1763: case SO_SNDBUF:
1764: case SO_RCVBUF:
1765: case SO_SNDLOWAT:
1766: case SO_RCVLOWAT:
1.171 plunky 1767: error = sockopt_getint(sopt, &optval);
1.177 ad 1768: solock(so);
1.171 plunky 1769: if (error)
1.177 ad 1770: break;
1.1 cgd 1771:
1.142 dyoung 1772: /*
1773: * Values < 1 make no sense for any of these
1774: * options, so disallow them.
1775: */
1.177 ad 1776: if (optval < 1) {
1777: error = EINVAL;
1778: break;
1779: }
1.1 cgd 1780:
1.179 christos 1781: switch (opt) {
1.171 plunky 1782: case SO_SNDBUF:
1.177 ad 1783: if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
1784: error = ENOBUFS;
1785: break;
1786: }
1.171 plunky 1787: so->so_snd.sb_flags &= ~SB_AUTOSIZE;
1788: break;
1.1 cgd 1789:
1790: case SO_RCVBUF:
1.177 ad 1791: if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
1792: error = ENOBUFS;
1793: break;
1794: }
1.171 plunky 1795: so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1.142 dyoung 1796: break;
1797:
1798: /*
1799: * Make sure the low-water is never greater than
1800: * the high-water.
1801: */
1.1 cgd 1802: case SO_SNDLOWAT:
1.171 plunky 1803: if (optval > so->so_snd.sb_hiwat)
1804: optval = so->so_snd.sb_hiwat;
1805:
1806: so->so_snd.sb_lowat = optval;
1.142 dyoung 1807: break;
1.171 plunky 1808:
1.1 cgd 1809: case SO_RCVLOWAT:
1.171 plunky 1810: if (optval > so->so_rcv.sb_hiwat)
1811: optval = so->so_rcv.sb_hiwat;
1812:
1813: so->so_rcv.sb_lowat = optval;
1.142 dyoung 1814: break;
1815: }
1816: break;
1.28 thorpej 1817:
1.179 christos 1818: #ifdef COMPAT_50
1819: case SO_OSNDTIMEO:
1820: case SO_ORCVTIMEO: {
1821: struct timeval50 otv;
1822: error = sockopt_get(sopt, &otv, sizeof(otv));
1.186 pooka 1823: if (error) {
1824: solock(so);
1.183 christos 1825: break;
1.186 pooka 1826: }
1.179 christos 1827: timeval50_to_timeval(&otv, &tv);
1828: opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO;
1.182 christos 1829: error = 0;
1.179 christos 1830: /*FALLTHROUGH*/
1831: }
1832: #endif /* COMPAT_50 */
1833:
1.269 mrg 1834: /*FALLTHROUGH*/
1.142 dyoung 1835: case SO_SNDTIMEO:
1.269 mrg 1836: /*FALLTHROUGH*/
1.142 dyoung 1837: case SO_RCVTIMEO:
1.182 christos 1838: if (error)
1.179 christos 1839: error = sockopt_get(sopt, &tv, sizeof(tv));
1.177 ad 1840: solock(so);
1.171 plunky 1841: if (error)
1.177 ad 1842: break;
1.171 plunky 1843:
1.177 ad 1844: if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
1845: error = EDOM;
1846: break;
1847: }
1.28 thorpej 1848:
1.171 plunky 1849: optval = tv.tv_sec * hz + tv.tv_usec / tick;
1850: if (optval == 0 && tv.tv_usec != 0)
1851: optval = 1;
1.28 thorpej 1852:
1.179 christos 1853: switch (opt) {
1.142 dyoung 1854: case SO_SNDTIMEO:
1.171 plunky 1855: so->so_snd.sb_timeo = optval;
1.1 cgd 1856: break;
1857: case SO_RCVTIMEO:
1.171 plunky 1858: so->so_rcv.sb_timeo = optval;
1.142 dyoung 1859: break;
1860: }
1861: break;
1.1 cgd 1862:
1.142 dyoung 1863: default:
1.177 ad 1864: solock(so);
1865: error = ENOPROTOOPT;
1866: break;
1.142 dyoung 1867: }
1.177 ad 1868: KASSERT(solocked(so));
1869: return error;
1.142 dyoung 1870: }
1.1 cgd 1871:
1.142 dyoung 1872: int
1.171 plunky 1873: sosetopt(struct socket *so, struct sockopt *sopt)
1.142 dyoung 1874: {
1875: int error, prerr;
1.1 cgd 1876:
1.177 ad 1877: if (sopt->sopt_level == SOL_SOCKET) {
1.171 plunky 1878: error = sosetopt1(so, sopt);
1.177 ad 1879: KASSERT(solocked(so));
1880: } else {
1.142 dyoung 1881: error = ENOPROTOOPT;
1.177 ad 1882: solock(so);
1883: }
1.1 cgd 1884:
1.142 dyoung 1885: if ((error == 0 || error == ENOPROTOOPT) &&
1886: so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
1887: /* give the protocol stack a shot */
1.171 plunky 1888: prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
1.142 dyoung 1889: if (prerr == 0)
1890: error = 0;
1891: else if (prerr != ENOPROTOOPT)
1892: error = prerr;
1.171 plunky 1893: }
1.160 ad 1894: sounlock(so);
1.142 dyoung 1895: return error;
1.1 cgd 1896: }
1897:
1.171 plunky 1898: /*
1899: * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
1900: */
1901: int
1902: so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
1903: const void *val, size_t valsize)
1904: {
1905: struct sockopt sopt;
1906: int error;
1907:
1908: KASSERT(valsize == 0 || val != NULL);
1909:
1910: sockopt_init(&sopt, level, name, valsize);
1911: sockopt_set(&sopt, val, valsize);
1912:
1913: error = sosetopt(so, &sopt);
1914:
1915: sockopt_destroy(&sopt);
1916:
1917: return error;
1918: }
1.253 ryo 1919:
1.171 plunky 1920: /*
1921: * internal get SOL_SOCKET options
1922: */
1923: static int
1924: sogetopt1(struct socket *so, struct sockopt *sopt)
1925: {
1.179 christos 1926: int error, optval, opt;
1.171 plunky 1927: struct linger l;
1928: struct timeval tv;
1929:
1.179 christos 1930: switch ((opt = sopt->sopt_name)) {
1.171 plunky 1931:
1932: case SO_ACCEPTFILTER:
1.177 ad 1933: error = accept_filt_getopt(so, sopt);
1.171 plunky 1934: break;
1935:
1936: case SO_LINGER:
1937: l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
1938: l.l_linger = so->so_linger;
1939:
1940: error = sockopt_set(sopt, &l, sizeof(l));
1941: break;
1942:
1943: case SO_USELOOPBACK:
1944: case SO_DONTROUTE:
1945: case SO_DEBUG:
1946: case SO_KEEPALIVE:
1947: case SO_REUSEADDR:
1948: case SO_REUSEPORT:
1949: case SO_BROADCAST:
1950: case SO_OOBINLINE:
1951: case SO_TIMESTAMP:
1.207 christos 1952: case SO_NOSIGPIPE:
1.266 christos 1953: case SO_RERROR:
1.184 christos 1954: #ifdef SO_OTIMESTAMP
1955: case SO_OTIMESTAMP:
1956: #endif
1.218 seanb 1957: case SO_ACCEPTCONN:
1.179 christos 1958: error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
1.171 plunky 1959: break;
1960:
1961: case SO_TYPE:
1962: error = sockopt_setint(sopt, so->so_type);
1963: break;
1964:
1965: case SO_ERROR:
1.267 hannken 1966: if (so->so_error == 0) {
1967: so->so_error = so->so_rerror;
1968: so->so_rerror = 0;
1969: }
1.171 plunky 1970: error = sockopt_setint(sopt, so->so_error);
1971: so->so_error = 0;
1972: break;
1973:
1974: case SO_SNDBUF:
1975: error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
1976: break;
1977:
1978: case SO_RCVBUF:
1979: error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
1980: break;
1981:
1982: case SO_SNDLOWAT:
1983: error = sockopt_setint(sopt, so->so_snd.sb_lowat);
1984: break;
1985:
1986: case SO_RCVLOWAT:
1987: error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
1988: break;
1989:
1.179 christos 1990: #ifdef COMPAT_50
1991: case SO_OSNDTIMEO:
1992: case SO_ORCVTIMEO: {
1993: struct timeval50 otv;
1994:
1995: optval = (opt == SO_OSNDTIMEO ?
1996: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1997:
1998: otv.tv_sec = optval / hz;
1999: otv.tv_usec = (optval % hz) * tick;
2000:
2001: error = sockopt_set(sopt, &otv, sizeof(otv));
2002: break;
2003: }
2004: #endif /* COMPAT_50 */
2005:
1.171 plunky 2006: case SO_SNDTIMEO:
2007: case SO_RCVTIMEO:
1.179 christos 2008: optval = (opt == SO_SNDTIMEO ?
1.171 plunky 2009: so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2010:
2011: tv.tv_sec = optval / hz;
2012: tv.tv_usec = (optval % hz) * tick;
2013:
2014: error = sockopt_set(sopt, &tv, sizeof(tv));
2015: break;
2016:
2017: case SO_OVERFLOWED:
2018: error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
2019: break;
2020:
2021: default:
2022: error = ENOPROTOOPT;
2023: break;
2024: }
2025:
1.270 maxv 2026: return error;
1.171 plunky 2027: }
2028:
1.14 mycroft 2029: int
1.171 plunky 2030: sogetopt(struct socket *so, struct sockopt *sopt)
1.1 cgd 2031: {
1.270 maxv 2032: int error;
1.1 cgd 2033:
1.160 ad 2034: solock(so);
1.171 plunky 2035: if (sopt->sopt_level != SOL_SOCKET) {
1.1 cgd 2036: if (so->so_proto && so->so_proto->pr_ctloutput) {
1.160 ad 2037: error = ((*so->so_proto->pr_ctloutput)
1.171 plunky 2038: (PRCO_GETOPT, so, sopt));
1.1 cgd 2039: } else
1.160 ad 2040: error = (ENOPROTOOPT);
1.1 cgd 2041: } else {
1.171 plunky 2042: error = sogetopt1(so, sopt);
2043: }
2044: sounlock(so);
1.270 maxv 2045: return error;
1.171 plunky 2046: }
2047:
2048: /*
2049: * alloc sockopt data buffer buffer
2050: * - will be released at destroy
2051: */
1.176 plunky 2052: static int
2053: sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
1.171 plunky 2054: {
2055:
2056: KASSERT(sopt->sopt_size == 0);
2057:
1.176 plunky 2058: if (len > sizeof(sopt->sopt_buf)) {
2059: sopt->sopt_data = kmem_zalloc(len, kmflag);
2060: if (sopt->sopt_data == NULL)
2061: return ENOMEM;
2062: } else
1.171 plunky 2063: sopt->sopt_data = sopt->sopt_buf;
2064:
2065: sopt->sopt_size = len;
1.176 plunky 2066: return 0;
1.171 plunky 2067: }
2068:
2069: /*
2070: * initialise sockopt storage
1.176 plunky 2071: * - MAY sleep during allocation
1.171 plunky 2072: */
2073: void
2074: sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
2075: {
1.1 cgd 2076:
1.171 plunky 2077: memset(sopt, 0, sizeof(*sopt));
1.1 cgd 2078:
1.171 plunky 2079: sopt->sopt_level = level;
2080: sopt->sopt_name = name;
1.176 plunky 2081: (void)sockopt_alloc(sopt, size, KM_SLEEP);
1.171 plunky 2082: }
2083:
2084: /*
2085: * destroy sockopt storage
2086: * - will release any held memory references
2087: */
2088: void
2089: sockopt_destroy(struct sockopt *sopt)
2090: {
2091:
2092: if (sopt->sopt_data != sopt->sopt_buf)
1.173 plunky 2093: kmem_free(sopt->sopt_data, sopt->sopt_size);
1.171 plunky 2094:
2095: memset(sopt, 0, sizeof(*sopt));
2096: }
2097:
2098: /*
2099: * set sockopt value
2100: * - value is copied into sockopt
1.253 ryo 2101: * - memory is allocated when necessary, will not sleep
1.171 plunky 2102: */
2103: int
2104: sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
2105: {
1.176 plunky 2106: int error;
1.171 plunky 2107:
1.176 plunky 2108: if (sopt->sopt_size == 0) {
2109: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2110: if (error)
2111: return error;
2112: }
1.171 plunky 2113:
1.258 christos 2114: if (sopt->sopt_size < len)
2115: return EINVAL;
1.270 maxv 2116:
1.171 plunky 2117: memcpy(sopt->sopt_data, buf, len);
1.259 christos 2118: sopt->sopt_retsize = len;
2119:
1.171 plunky 2120: return 0;
2121: }
2122:
2123: /*
2124: * common case of set sockopt integer value
2125: */
2126: int
2127: sockopt_setint(struct sockopt *sopt, int val)
2128: {
2129:
2130: return sockopt_set(sopt, &val, sizeof(int));
2131: }
2132:
2133: /*
2134: * get sockopt value
2135: * - correct size must be given
2136: */
2137: int
2138: sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
2139: {
1.170 tls 2140:
1.171 plunky 2141: if (sopt->sopt_size != len)
2142: return EINVAL;
1.1 cgd 2143:
1.171 plunky 2144: memcpy(buf, sopt->sopt_data, len);
2145: return 0;
2146: }
1.1 cgd 2147:
1.171 plunky 2148: /*
2149: * common case of get sockopt integer value
2150: */
2151: int
2152: sockopt_getint(const struct sockopt *sopt, int *valp)
2153: {
1.1 cgd 2154:
1.171 plunky 2155: return sockopt_get(sopt, valp, sizeof(int));
2156: }
1.1 cgd 2157:
1.171 plunky 2158: /*
2159: * set sockopt value from mbuf
2160: * - ONLY for legacy code
2161: * - mbuf is released by sockopt
1.176 plunky 2162: * - will not sleep
1.171 plunky 2163: */
2164: int
2165: sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
2166: {
2167: size_t len;
1.176 plunky 2168: int error;
1.1 cgd 2169:
1.171 plunky 2170: len = m_length(m);
1.1 cgd 2171:
1.176 plunky 2172: if (sopt->sopt_size == 0) {
2173: error = sockopt_alloc(sopt, len, KM_NOSLEEP);
2174: if (error)
2175: return error;
2176: }
1.1 cgd 2177:
1.258 christos 2178: if (sopt->sopt_size < len)
2179: return EINVAL;
1.270 maxv 2180:
1.171 plunky 2181: m_copydata(m, 0, len, sopt->sopt_data);
2182: m_freem(m);
1.259 christos 2183: sopt->sopt_retsize = len;
1.1 cgd 2184:
1.171 plunky 2185: return 0;
2186: }
1.1 cgd 2187:
1.171 plunky 2188: /*
2189: * get sockopt value into mbuf
2190: * - ONLY for legacy code
2191: * - mbuf to be released by the caller
1.176 plunky 2192: * - will not sleep
1.171 plunky 2193: */
2194: struct mbuf *
2195: sockopt_getmbuf(const struct sockopt *sopt)
2196: {
2197: struct mbuf *m;
1.107 darrenr 2198:
1.176 plunky 2199: if (sopt->sopt_size > MCLBYTES)
2200: return NULL;
2201:
2202: m = m_get(M_DONTWAIT, MT_SOOPTS);
1.171 plunky 2203: if (m == NULL)
2204: return NULL;
2205:
1.176 plunky 2206: if (sopt->sopt_size > MLEN) {
2207: MCLGET(m, M_DONTWAIT);
2208: if ((m->m_flags & M_EXT) == 0) {
2209: m_free(m);
2210: return NULL;
2211: }
1.1 cgd 2212: }
1.176 plunky 2213:
2214: memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
2215: m->m_len = sopt->sopt_size;
1.160 ad 2216:
1.171 plunky 2217: return m;
1.1 cgd 2218: }
2219:
1.14 mycroft 2220: void
1.54 lukem 2221: sohasoutofband(struct socket *so)
1.1 cgd 2222: {
1.153 rmind 2223:
1.90 christos 2224: fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
1.189 ad 2225: selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
1.1 cgd 2226: }
1.72 jdolecek 2227:
2228: static void
2229: filt_sordetach(struct knote *kn)
2230: {
1.270 maxv 2231: struct socket *so;
1.72 jdolecek 2232:
1.235 matt 2233: so = ((file_t *)kn->kn_obj)->f_socket;
1.160 ad 2234: solock(so);
1.73 christos 2235: SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
2236: if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1.72 jdolecek 2237: so->so_rcv.sb_flags &= ~SB_KNOTE;
1.160 ad 2238: sounlock(so);
1.72 jdolecek 2239: }
2240:
2241: /*ARGSUSED*/
2242: static int
1.129 yamt 2243: filt_soread(struct knote *kn, long hint)
1.72 jdolecek 2244: {
1.270 maxv 2245: struct socket *so;
1.160 ad 2246: int rv;
1.72 jdolecek 2247:
1.235 matt 2248: so = ((file_t *)kn->kn_obj)->f_socket;
1.160 ad 2249: if (hint != NOTE_SUBMIT)
2250: solock(so);
1.72 jdolecek 2251: kn->kn_data = so->so_rcv.sb_cc;
2252: if (so->so_state & SS_CANTRCVMORE) {
1.108 perry 2253: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2254: kn->kn_fflags = so->so_error;
1.160 ad 2255: rv = 1;
1.264 roy 2256: } else if (so->so_error || so->so_rerror)
1.160 ad 2257: rv = 1;
2258: else if (kn->kn_sfflags & NOTE_LOWAT)
2259: rv = (kn->kn_data >= kn->kn_sdata);
1.253 ryo 2260: else
1.160 ad 2261: rv = (kn->kn_data >= so->so_rcv.sb_lowat);
2262: if (hint != NOTE_SUBMIT)
2263: sounlock(so);
2264: return rv;
1.72 jdolecek 2265: }
2266:
2267: static void
2268: filt_sowdetach(struct knote *kn)
2269: {
1.270 maxv 2270: struct socket *so;
1.72 jdolecek 2271:
1.235 matt 2272: so = ((file_t *)kn->kn_obj)->f_socket;
1.160 ad 2273: solock(so);
1.73 christos 2274: SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
2275: if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1.72 jdolecek 2276: so->so_snd.sb_flags &= ~SB_KNOTE;
1.160 ad 2277: sounlock(so);
1.72 jdolecek 2278: }
2279:
2280: /*ARGSUSED*/
2281: static int
1.129 yamt 2282: filt_sowrite(struct knote *kn, long hint)
1.72 jdolecek 2283: {
1.270 maxv 2284: struct socket *so;
1.160 ad 2285: int rv;
1.72 jdolecek 2286:
1.235 matt 2287: so = ((file_t *)kn->kn_obj)->f_socket;
1.160 ad 2288: if (hint != NOTE_SUBMIT)
2289: solock(so);
1.72 jdolecek 2290: kn->kn_data = sbspace(&so->so_snd);
2291: if (so->so_state & SS_CANTSENDMORE) {
1.108 perry 2292: kn->kn_flags |= EV_EOF;
1.72 jdolecek 2293: kn->kn_fflags = so->so_error;
1.160 ad 2294: rv = 1;
1.261 roy 2295: } else if (so->so_error)
1.160 ad 2296: rv = 1;
2297: else if (((so->so_state & SS_ISCONNECTED) == 0) &&
1.72 jdolecek 2298: (so->so_proto->pr_flags & PR_CONNREQUIRED))
1.160 ad 2299: rv = 0;
2300: else if (kn->kn_sfflags & NOTE_LOWAT)
2301: rv = (kn->kn_data >= kn->kn_sdata);
2302: else
2303: rv = (kn->kn_data >= so->so_snd.sb_lowat);
2304: if (hint != NOTE_SUBMIT)
2305: sounlock(so);
2306: return rv;
1.72 jdolecek 2307: }
2308:
2309: /*ARGSUSED*/
2310: static int
1.129 yamt 2311: filt_solisten(struct knote *kn, long hint)
1.72 jdolecek 2312: {
1.270 maxv 2313: struct socket *so;
1.160 ad 2314: int rv;
1.72 jdolecek 2315:
1.235 matt 2316: so = ((file_t *)kn->kn_obj)->f_socket;
1.72 jdolecek 2317:
2318: /*
2319: * Set kn_data to number of incoming connections, not
2320: * counting partial (incomplete) connections.
1.108 perry 2321: */
1.160 ad 2322: if (hint != NOTE_SUBMIT)
2323: solock(so);
1.72 jdolecek 2324: kn->kn_data = so->so_qlen;
1.160 ad 2325: rv = (kn->kn_data > 0);
2326: if (hint != NOTE_SUBMIT)
2327: sounlock(so);
2328: return rv;
1.72 jdolecek 2329: }
2330:
1.257 maya 2331: static const struct filterops solisten_filtops = {
2332: .f_isfd = 1,
2333: .f_attach = NULL,
2334: .f_detach = filt_sordetach,
2335: .f_event = filt_solisten,
2336: };
2337:
2338: static const struct filterops soread_filtops = {
2339: .f_isfd = 1,
2340: .f_attach = NULL,
2341: .f_detach = filt_sordetach,
2342: .f_event = filt_soread,
2343: };
2344:
2345: static const struct filterops sowrite_filtops = {
2346: .f_isfd = 1,
2347: .f_attach = NULL,
2348: .f_detach = filt_sowdetach,
2349: .f_event = filt_sowrite,
2350: };
1.72 jdolecek 2351:
2352: int
1.129 yamt 2353: soo_kqfilter(struct file *fp, struct knote *kn)
1.72 jdolecek 2354: {
1.270 maxv 2355: struct socket *so;
2356: struct sockbuf *sb;
1.72 jdolecek 2357:
1.235 matt 2358: so = ((file_t *)kn->kn_obj)->f_socket;
1.160 ad 2359: solock(so);
1.72 jdolecek 2360: switch (kn->kn_filter) {
2361: case EVFILT_READ:
2362: if (so->so_options & SO_ACCEPTCONN)
2363: kn->kn_fop = &solisten_filtops;
2364: else
2365: kn->kn_fop = &soread_filtops;
2366: sb = &so->so_rcv;
2367: break;
2368: case EVFILT_WRITE:
2369: kn->kn_fop = &sowrite_filtops;
2370: sb = &so->so_snd;
2371: break;
2372: default:
1.160 ad 2373: sounlock(so);
1.270 maxv 2374: return EINVAL;
1.72 jdolecek 2375: }
1.73 christos 2376: SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1.72 jdolecek 2377: sb->sb_flags |= SB_KNOTE;
1.160 ad 2378: sounlock(so);
1.270 maxv 2379: return 0;
1.72 jdolecek 2380: }
2381:
1.154 ad 2382: static int
2383: sodopoll(struct socket *so, int events)
2384: {
2385: int revents;
2386:
2387: revents = 0;
2388:
2389: if (events & (POLLIN | POLLRDNORM))
2390: if (soreadable(so))
2391: revents |= events & (POLLIN | POLLRDNORM);
2392:
2393: if (events & (POLLOUT | POLLWRNORM))
2394: if (sowritable(so))
2395: revents |= events & (POLLOUT | POLLWRNORM);
2396:
2397: if (events & (POLLPRI | POLLRDBAND))
2398: if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
2399: revents |= events & (POLLPRI | POLLRDBAND);
2400:
2401: return revents;
2402: }
2403:
2404: int
2405: sopoll(struct socket *so, int events)
2406: {
2407: int revents = 0;
2408:
1.160 ad 2409: #ifndef DIAGNOSTIC
2410: /*
2411: * Do a quick, unlocked check in expectation that the socket
2412: * will be ready for I/O. Don't do this check if DIAGNOSTIC,
2413: * as the solocked() assertions will fail.
2414: */
1.154 ad 2415: if ((revents = sodopoll(so, events)) != 0)
2416: return revents;
1.160 ad 2417: #endif
1.154 ad 2418:
1.160 ad 2419: solock(so);
1.154 ad 2420: if ((revents = sodopoll(so, events)) == 0) {
2421: if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2422: selrecord(curlwp, &so->so_rcv.sb_sel);
1.160 ad 2423: so->so_rcv.sb_flags |= SB_NOTIFY;
1.154 ad 2424: }
2425:
2426: if (events & (POLLOUT | POLLWRNORM)) {
2427: selrecord(curlwp, &so->so_snd.sb_sel);
1.160 ad 2428: so->so_snd.sb_flags |= SB_NOTIFY;
1.154 ad 2429: }
2430: }
1.160 ad 2431: sounlock(so);
1.154 ad 2432:
2433: return revents;
2434: }
2435:
1.256 christos 2436: struct mbuf **
1.262 maxv 2437: sbsavetimestamp(int opt, struct mbuf **mp)
1.256 christos 2438: {
2439: struct timeval tv;
2440: microtime(&tv);
2441:
2442: #ifdef SO_OTIMESTAMP
2443: if (opt & SO_OTIMESTAMP) {
2444: struct timeval50 tv50;
2445:
2446: timeval_to_timeval50(&tv, &tv50);
2447: *mp = sbcreatecontrol(&tv50, sizeof(tv50),
2448: SCM_OTIMESTAMP, SOL_SOCKET);
2449: if (*mp)
2450: mp = &(*mp)->m_next;
2451: } else
2452: #endif
2453:
2454: if (opt & SO_TIMESTAMP) {
2455: *mp = sbcreatecontrol(&tv, sizeof(tv),
2456: SCM_TIMESTAMP, SOL_SOCKET);
2457: if (*mp)
2458: mp = &(*mp)->m_next;
2459: }
2460: return mp;
2461: }
2462:
1.154 ad 2463:
1.94 yamt 2464: #include <sys/sysctl.h>
2465:
2466: static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
1.212 pooka 2467: static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
1.94 yamt 2468:
2469: /*
2470: * sysctl helper routine for kern.somaxkva. ensures that the given
2471: * value is not too small.
2472: * (XXX should we maybe make sure it's not too large as well?)
2473: */
2474: static int
2475: sysctl_kern_somaxkva(SYSCTLFN_ARGS)
2476: {
2477: int error, new_somaxkva;
2478: struct sysctlnode node;
2479:
2480: new_somaxkva = somaxkva;
2481: node = *rnode;
2482: node.sysctl_data = &new_somaxkva;
2483: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2484: if (error || newp == NULL)
1.270 maxv 2485: return error;
1.94 yamt 2486:
2487: if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
1.270 maxv 2488: return EINVAL;
1.94 yamt 2489:
1.136 ad 2490: mutex_enter(&so_pendfree_lock);
1.94 yamt 2491: somaxkva = new_somaxkva;
1.136 ad 2492: cv_broadcast(&socurkva_cv);
2493: mutex_exit(&so_pendfree_lock);
1.94 yamt 2494:
1.270 maxv 2495: return error;
1.94 yamt 2496: }
2497:
1.212 pooka 2498: /*
2499: * sysctl helper routine for kern.sbmax. Basically just ensures that
2500: * any new value is not too small.
2501: */
2502: static int
2503: sysctl_kern_sbmax(SYSCTLFN_ARGS)
2504: {
2505: int error, new_sbmax;
2506: struct sysctlnode node;
2507:
2508: new_sbmax = sb_max;
2509: node = *rnode;
2510: node.sysctl_data = &new_sbmax;
2511: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2512: if (error || newp == NULL)
1.270 maxv 2513: return error;
1.212 pooka 2514:
2515: KERNEL_LOCK(1, NULL);
2516: error = sb_max_set(new_sbmax);
2517: KERNEL_UNLOCK_ONE(NULL);
2518:
1.270 maxv 2519: return error;
1.212 pooka 2520: }
2521:
1.266 christos 2522: /*
2523: * sysctl helper routine for kern.sooptions. Ensures that only allowed
2524: * options can be set.
2525: */
2526: static int
2527: sysctl_kern_sooptions(SYSCTLFN_ARGS)
2528: {
2529: int error, new_options;
2530: struct sysctlnode node;
2531:
2532: new_options = sooptions;
2533: node = *rnode;
2534: node.sysctl_data = &new_options;
2535: error = sysctl_lookup(SYSCTLFN_CALL(&node));
2536: if (error || newp == NULL)
2537: return error;
2538:
2539: if (new_options & ~SO_DEFOPTS)
2540: return EINVAL;
2541:
2542: sooptions = new_options;
2543:
2544: return 0;
2545: }
2546:
1.178 pooka 2547: static void
1.212 pooka 2548: sysctl_kern_socket_setup(void)
1.94 yamt 2549: {
2550:
1.178 pooka 2551: KASSERT(socket_sysctllog == NULL);
1.97 atatat 2552:
1.178 pooka 2553: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
1.97 atatat 2554: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1.103 atatat 2555: CTLTYPE_INT, "somaxkva",
2556: SYSCTL_DESCR("Maximum amount of kernel memory to be "
1.270 maxv 2557: "used for socket buffers"),
1.94 yamt 2558: sysctl_kern_somaxkva, 0, NULL, 0,
2559: CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
1.212 pooka 2560:
2561: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2562: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2563: CTLTYPE_INT, "sbmax",
2564: SYSCTL_DESCR("Maximum socket buffer size"),
2565: sysctl_kern_sbmax, 0, NULL, 0,
2566: CTL_KERN, KERN_SBMAX, CTL_EOL);
1.266 christos 2567:
2568: sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
2569: CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2570: CTLTYPE_INT, "sooptions",
2571: SYSCTL_DESCR("Default socket options"),
2572: sysctl_kern_sooptions, 0, NULL, 0,
2573: CTL_KERN, CTL_CREATE, CTL_EOL);
1.94 yamt 2574: }
CVSweb <webmaster@jp.NetBSD.org>