1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * sendfile(2) and related extensions:
6 * Copyright (c) 1998, David Greenman. All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/10.4/sys/kern/uipc_syscalls.c 321021 2017-07-15 17:28:03Z dchagin $");
37
38 #include "opt_capsicum.h"
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 #include "opt_compat.h"
42 #include "opt_ktrace.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/capsicum.h>
47 #include <sys/condvar.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/sysproto.h>
52 #include <sys/malloc.h>
53 #include <sys/filedesc.h>
54 #include <sys/event.h>
55 #include <sys/proc.h>
56 #include <sys/fcntl.h>
57 #include <sys/file.h>
58 #include <sys/filio.h>
59 #include <sys/jail.h>
60 #include <sys/mman.h>
61 #include <sys/mount.h>
62 #include <sys/mbuf.h>
63 #include <sys/protosw.h>
64 #include <sys/rwlock.h>
65 #include <sys/sf_buf.h>
66 #include <sys/sysent.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/signalvar.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysctl.h>
72 #include <sys/uio.h>
73 #include <sys/vnode.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 #ifdef COMPAT_FREEBSD32
78 #include <compat/freebsd32/freebsd32_util.h>
79 #endif
80
81 #include <net/vnet.h>
82
83 #include <security/audit/audit.h>
84 #include <security/mac/mac_framework.h>
85
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93
94 /*
95 * Flags for accept1() and kern_accept4(), in addition to SOCK_CLOEXEC
96 * and SOCK_NONBLOCK.
97 */
98 #define ACCEPT4_INHERIT 0x1
99 #define ACCEPT4_COMPAT 0x2
100
101 static int sendit(struct thread *td, int s, struct msghdr *mp, int flags);
102 static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp);
103
104 static int accept1(struct thread *td, int s, struct sockaddr *uname,
105 socklen_t *anamelen, int flags);
106 static int do_sendfile(struct thread *td, struct sendfile_args *uap,
107 int compat);
108 static int getsockname1(struct thread *td, struct getsockname_args *uap,
109 int compat);
110 static int getpeername1(struct thread *td, struct getpeername_args *uap,
111 int compat);
112
113 counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)];
114
115 /*
116 * sendfile(2)-related variables and associated sysctls
117 */
118 static SYSCTL_NODE(_kern_ipc, OID_AUTO, sendfile, CTLFLAG_RW, 0,
119 "sendfile(2) tunables");
120 static int sfreadahead = 1;
121 SYSCTL_INT(_kern_ipc_sendfile, OID_AUTO, readahead, CTLFLAG_RW,
122 &sfreadahead, 0, "Number of sendfile(2) read-ahead MAXBSIZE blocks");
123
124
125 static void
126 sfstat_init(const void *unused)
127 {
128
129 COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t),
130 M_WAITOK);
131 }
132 SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL);
133
134 static int
135 sfstat_sysctl(SYSCTL_HANDLER_ARGS)
136 {
137 struct sfstat s;
138
139 COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t));
140 if (req->newptr)
141 COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t));
142 return (SYSCTL_OUT(req, &s, sizeof(s)));
143 }
144 SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW,
145 NULL, 0, sfstat_sysctl, "I", "sendfile statistics");
146
147 /*
148 * Convert a user file descriptor to a kernel file entry and check if required
149 * capability rights are present.
150 * A reference on the file entry is held upon returning.
151 */
152 int
153 getsock_cap(struct thread *td, int fd, cap_rights_t *rightsp,
154 struct file **fpp, u_int *fflagp)
155 {
156 struct file *fp;
157 int error;
158
159 error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, 0, &fp, NULL);
160 if (error != 0)
161 return (error);
162 if (fp->f_type != DTYPE_SOCKET) {
163 fdrop(fp, td);
164 return (ENOTSOCK);
165 }
166 if (fflagp != NULL)
167 *fflagp = fp->f_flag;
168 *fpp = fp;
169 return (0);
170 }
171
172 /*
173 * System call interface to the socket abstraction.
174 */
175 #if defined(COMPAT_43)
176 #define COMPAT_OLDSOCK
177 #endif
178
179 int
180 sys_socket(td, uap)
181 struct thread *td;
182 struct socket_args /* {
183 int domain;
184 int type;
185 int protocol;
186 } */ *uap;
187 {
188 struct socket *so;
189 struct file *fp;
190 int fd, error, type, oflag, fflag;
191
192 AUDIT_ARG_SOCKET(uap->domain, uap->type, uap->protocol);
193
194 type = uap->type;
195 oflag = 0;
196 fflag = 0;
197 if ((type & SOCK_CLOEXEC) != 0) {
198 type &= ~SOCK_CLOEXEC;
199 oflag |= O_CLOEXEC;
200 }
201 if ((type & SOCK_NONBLOCK) != 0) {
202 type &= ~SOCK_NONBLOCK;
203 fflag |= FNONBLOCK;
204 }
205
206 #ifdef MAC
207 error = mac_socket_check_create(td->td_ucred, uap->domain, type,
208 uap->protocol);
209 if (error != 0)
210 return (error);
211 #endif
212 error = falloc(td, &fp, &fd, oflag);
213 if (error != 0)
214 return (error);
215 /* An extra reference on `fp' has been held for us by falloc(). */
216 error = socreate(uap->domain, &so, type, uap->protocol,
217 td->td_ucred, td);
218 if (error != 0) {
219 fdclose(td, fp, fd);
220 } else {
221 finit(fp, FREAD | FWRITE | fflag, DTYPE_SOCKET, so, &socketops);
222 if ((fflag & FNONBLOCK) != 0)
223 (void) fo_ioctl(fp, FIONBIO, &fflag, td->td_ucred, td);
224 td->td_retval[0] = fd;
225 }
226 fdrop(fp, td);
227 return (error);
228 }
229
230 /* ARGSUSED */
231 int
232 sys_bind(td, uap)
233 struct thread *td;
234 struct bind_args /* {
235 int s;
236 caddr_t name;
237 int namelen;
238 } */ *uap;
239 {
240 struct sockaddr *sa;
241 int error;
242
243 error = getsockaddr(&sa, uap->name, uap->namelen);
244 if (error == 0) {
245 error = kern_bind(td, uap->s, sa);
246 free(sa, M_SONAME);
247 }
248 return (error);
249 }
250
251 static int
252 kern_bindat(struct thread *td, int dirfd, int fd, struct sockaddr *sa)
253 {
254 struct socket *so;
255 struct file *fp;
256 cap_rights_t rights;
257 int error;
258
259 AUDIT_ARG_FD(fd);
260 AUDIT_ARG_SOCKADDR(td, dirfd, sa);
261 error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_BIND),
262 &fp, NULL);
263 if (error != 0)
264 return (error);
265 so = fp->f_data;
266 #ifdef KTRACE
267 if (KTRPOINT(td, KTR_STRUCT))
268 ktrsockaddr(sa);
269 #endif
270 #ifdef MAC
271 error = mac_socket_check_bind(td->td_ucred, so, sa);
272 if (error == 0) {
273 #endif
274 if (dirfd == AT_FDCWD)
275 error = sobind(so, sa, td);
276 else
277 error = sobindat(dirfd, so, sa, td);
278 #ifdef MAC
279 }
280 #endif
281 fdrop(fp, td);
282 return (error);
283 }
284
285 int
286 kern_bind(struct thread *td, int fd, struct sockaddr *sa)
287 {
288
289 return (kern_bindat(td, AT_FDCWD, fd, sa));
290 }
291
292 /* ARGSUSED */
293 int
294 sys_bindat(td, uap)
295 struct thread *td;
296 struct bindat_args /* {
297 int fd;
298 int s;
299 caddr_t name;
300 int namelen;
301 } */ *uap;
302 {
303 struct sockaddr *sa;
304 int error;
305
306 error = getsockaddr(&sa, uap->name, uap->namelen);
307 if (error == 0) {
308 error = kern_bindat(td, uap->fd, uap->s, sa);
309 free(sa, M_SONAME);
310 }
311 return (error);
312 }
313
314 /* ARGSUSED */
315 int
316 sys_listen(td, uap)
317 struct thread *td;
318 struct listen_args /* {
319 int s;
320 int backlog;
321 } */ *uap;
322 {
323 struct socket *so;
324 struct file *fp;
325 cap_rights_t rights;
326 int error;
327
328 AUDIT_ARG_FD(uap->s);
329 error = getsock_cap(td, uap->s, cap_rights_init(&rights, CAP_LISTEN),
330 &fp, NULL);
331 if (error == 0) {
332 so = fp->f_data;
333 #ifdef MAC
334 error = mac_socket_check_listen(td->td_ucred, so);
335 if (error == 0)
336 #endif
337 error = solisten(so, uap->backlog, td);
338 fdrop(fp, td);
339 }
340 return(error);
341 }
342
343 /*
344 * accept1()
345 */
346 static int
347 accept1(td, s, uname, anamelen, flags)
348 struct thread *td;
349 int s;
350 struct sockaddr *uname;
351 socklen_t *anamelen;
352 int flags;
353 {
354 struct sockaddr *name;
355 socklen_t namelen;
356 struct file *fp;
357 int error;
358
359 if (uname == NULL)
360 return (kern_accept4(td, s, NULL, NULL, flags, NULL));
361
362 error = copyin(anamelen, &namelen, sizeof (namelen));
363 if (error != 0)
364 return (error);
365
366 error = kern_accept4(td, s, &name, &namelen, flags, &fp);
367
368 /*
369 * return a namelen of zero for older code which might
370 * ignore the return value from accept.
371 */
372 if (error != 0) {
373 (void) copyout(&namelen, anamelen, sizeof(*anamelen));
374 return (error);
375 }
376
377 if (error == 0 && uname != NULL) {
378 #ifdef COMPAT_OLDSOCK
379 if (flags & ACCEPT4_COMPAT)
380 ((struct osockaddr *)name)->sa_family =
381 name->sa_family;
382 #endif
383 error = copyout(name, uname, namelen);
384 }
385 if (error == 0)
386 error = copyout(&namelen, anamelen,
387 sizeof(namelen));
388 if (error != 0)
389 fdclose(td, fp, td->td_retval[0]);
390 fdrop(fp, td);
391 free(name, M_SONAME);
392 return (error);
393 }
394
395 int
396 kern_accept(struct thread *td, int s, struct sockaddr **name,
397 socklen_t *namelen, struct file **fp)
398 {
399 return (kern_accept4(td, s, name, namelen, ACCEPT4_INHERIT, fp));
400 }
401
402 int
403 kern_accept4(struct thread *td, int s, struct sockaddr **name,
404 socklen_t *namelen, int flags, struct file **fp)
405 {
406 struct file *headfp, *nfp = NULL;
407 struct sockaddr *sa = NULL;
408 struct socket *head, *so;
409 cap_rights_t rights;
410 u_int fflag;
411 pid_t pgid;
412 int error, fd, tmp;
413
414 if (name != NULL)
415 *name = NULL;
416
417 AUDIT_ARG_FD(s);
418 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_ACCEPT),
419 &headfp, &fflag);
420 if (error != 0)
421 return (error);
422 head = headfp->f_data;
423 if ((head->so_options & SO_ACCEPTCONN) == 0) {
424 error = EINVAL;
425 goto done;
426 }
427 #ifdef MAC
428 error = mac_socket_check_accept(td->td_ucred, head);
429 if (error != 0)
430 goto done;
431 #endif
432 error = falloc(td, &nfp, &fd, (flags & SOCK_CLOEXEC) ? O_CLOEXEC : 0);
433 if (error != 0)
434 goto done;
435 ACCEPT_LOCK();
436 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) {
437 ACCEPT_UNLOCK();
438 error = EWOULDBLOCK;
439 goto noconnection;
440 }
441 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
442 if (head->so_rcv.sb_state & SBS_CANTRCVMORE) {
443 head->so_error = ECONNABORTED;
444 break;
445 }
446 error = msleep(&head->so_timeo, &accept_mtx, PSOCK | PCATCH,
447 "accept", 0);
448 if (error != 0) {
449 ACCEPT_UNLOCK();
450 goto noconnection;
451 }
452 }
453 if (head->so_error) {
454 error = head->so_error;
455 head->so_error = 0;
456 ACCEPT_UNLOCK();
457 goto noconnection;
458 }
459 so = TAILQ_FIRST(&head->so_comp);
460 KASSERT(!(so->so_qstate & SQ_INCOMP), ("accept1: so SQ_INCOMP"));
461 KASSERT(so->so_qstate & SQ_COMP, ("accept1: so not SQ_COMP"));
462
463 /*
464 * Before changing the flags on the socket, we have to bump the
465 * reference count. Otherwise, if the protocol calls sofree(),
466 * the socket will be released due to a zero refcount.
467 */
468 SOCK_LOCK(so); /* soref() and so_state update */
469 soref(so); /* file descriptor reference */
470
471 TAILQ_REMOVE(&head->so_comp, so, so_list);
472 head->so_qlen--;
473 if (flags & ACCEPT4_INHERIT)
474 so->so_state |= (head->so_state & SS_NBIO);
475 else
476 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
477 so->so_qstate &= ~SQ_COMP;
478 so->so_head = NULL;
479
480 SOCK_UNLOCK(so);
481 ACCEPT_UNLOCK();
482
483 /* An extra reference on `nfp' has been held for us by falloc(). */
484 td->td_retval[0] = fd;
485
486 /* connection has been removed from the listen queue */
487 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
488
489 if (flags & ACCEPT4_INHERIT) {
490 pgid = fgetown(&head->so_sigio);
491 if (pgid != 0)
492 fsetown(pgid, &so->so_sigio);
493 } else {
494 fflag &= ~(FNONBLOCK | FASYNC);
495 if (flags & SOCK_NONBLOCK)
496 fflag |= FNONBLOCK;
497 }
498
499 finit(nfp, fflag, DTYPE_SOCKET, so, &socketops);
500 /* Sync socket nonblocking/async state with file flags */
501 tmp = fflag & FNONBLOCK;
502 (void) fo_ioctl(nfp, FIONBIO, &tmp, td->td_ucred, td);
503 tmp = fflag & FASYNC;
504 (void) fo_ioctl(nfp, FIOASYNC, &tmp, td->td_ucred, td);
505 sa = 0;
506 error = soaccept(so, &sa);
507 if (error != 0) {
508 /*
509 * return a namelen of zero for older code which might
510 * ignore the return value from accept.
511 */
512 if (name)
513 *namelen = 0;
514 goto noconnection;
515 }
516 if (sa == NULL) {
517 if (name)
518 *namelen = 0;
519 goto done;
520 }
521 AUDIT_ARG_SOCKADDR(td, AT_FDCWD, sa);
522 if (name) {
523 /* check sa_len before it is destroyed */
524 if (*namelen > sa->sa_len)
525 *namelen = sa->sa_len;
526 #ifdef KTRACE
527 if (KTRPOINT(td, KTR_STRUCT))
528 ktrsockaddr(sa);
529 #endif
530 *name = sa;
531 sa = NULL;
532 }
533 noconnection:
534 free(sa, M_SONAME);
535
536 /*
537 * close the new descriptor, assuming someone hasn't ripped it
538 * out from under us.
539 */
540 if (error != 0)
541 fdclose(td, nfp, fd);
542
543 /*
544 * Release explicitly held references before returning. We return
545 * a reference on nfp to the caller on success if they request it.
546 */
547 done:
548 if (fp != NULL) {
549 if (error == 0) {
550 *fp = nfp;
551 nfp = NULL;
552 } else
553 *fp = NULL;
554 }
555 if (nfp != NULL)
556 fdrop(nfp, td);
557 fdrop(headfp, td);
558 return (error);
559 }
560
561 int
562 sys_accept(td, uap)
563 struct thread *td;
564 struct accept_args *uap;
565 {
566
567 return (accept1(td, uap->s, uap->name, uap->anamelen, ACCEPT4_INHERIT));
568 }
569
570 int
571 sys_accept4(td, uap)
572 struct thread *td;
573 struct accept4_args *uap;
574 {
575
576 if (uap->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
577 return (EINVAL);
578
579 return (accept1(td, uap->s, uap->name, uap->anamelen, uap->flags));
580 }
581
582 #ifdef COMPAT_OLDSOCK
583 int
584 oaccept(td, uap)
585 struct thread *td;
586 struct accept_args *uap;
587 {
588
589 return (accept1(td, uap->s, uap->name, uap->anamelen,
590 ACCEPT4_INHERIT | ACCEPT4_COMPAT));
591 }
592 #endif /* COMPAT_OLDSOCK */
593
594 /* ARGSUSED */
595 int
596 sys_connect(td, uap)
597 struct thread *td;
598 struct connect_args /* {
599 int s;
600 caddr_t name;
601 int namelen;
602 } */ *uap;
603 {
604 struct sockaddr *sa;
605 int error;
606
607 error = getsockaddr(&sa, uap->name, uap->namelen);
608 if (error == 0) {
609 error = kern_connect(td, uap->s, sa);
610 free(sa, M_SONAME);
611 }
612 return (error);
613 }
614
615 static int
616 kern_connectat(struct thread *td, int dirfd, int fd, struct sockaddr *sa)
617 {
618 struct socket *so;
619 struct file *fp;
620 cap_rights_t rights;
621 int error, interrupted = 0;
622
623 AUDIT_ARG_FD(fd);
624 AUDIT_ARG_SOCKADDR(td, dirfd, sa);
625 error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_CONNECT),
626 &fp, NULL);
627 if (error != 0)
628 return (error);
629 so = fp->f_data;
630 if (so->so_state & SS_ISCONNECTING) {
631 error = EALREADY;
632 goto done1;
633 }
634 #ifdef KTRACE
635 if (KTRPOINT(td, KTR_STRUCT))
636 ktrsockaddr(sa);
637 #endif
638 #ifdef MAC
639 error = mac_socket_check_connect(td->td_ucred, so, sa);
640 if (error != 0)
641 goto bad;
642 #endif
643 if (dirfd == AT_FDCWD)
644 error = soconnect(so, sa, td);
645 else
646 error = soconnectat(dirfd, so, sa, td);
647 if (error != 0)
648 goto bad;
649 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
650 error = EINPROGRESS;
651 goto done1;
652 }
653 SOCK_LOCK(so);
654 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
655 error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH,
656 "connec", 0);
657 if (error != 0) {
658 if (error == EINTR || error == ERESTART)
659 interrupted = 1;
660 break;
661 }
662 }
663 if (error == 0) {
664 error = so->so_error;
665 so->so_error = 0;
666 }
667 SOCK_UNLOCK(so);
668 bad:
669 if (!interrupted)
670 so->so_state &= ~SS_ISCONNECTING;
671 if (error == ERESTART)
672 error = EINTR;
673 done1:
674 fdrop(fp, td);
675 return (error);
676 }
677
678 int
679 kern_connect(struct thread *td, int fd, struct sockaddr *sa)
680 {
681
682 return (kern_connectat(td, AT_FDCWD, fd, sa));
683 }
684
685 /* ARGSUSED */
686 int
687 sys_connectat(td, uap)
688 struct thread *td;
689 struct connectat_args /* {
690 int fd;
691 int s;
692 caddr_t name;
693 int namelen;
694 } */ *uap;
695 {
696 struct sockaddr *sa;
697 int error;
698
699 error = getsockaddr(&sa, uap->name, uap->namelen);
700 if (error == 0) {
701 error = kern_connectat(td, uap->fd, uap->s, sa);
702 free(sa, M_SONAME);
703 }
704 return (error);
705 }
706
707 int
708 kern_socketpair(struct thread *td, int domain, int type, int protocol,
709 int *rsv)
710 {
711 struct file *fp1, *fp2;
712 struct socket *so1, *so2;
713 int fd, error, oflag, fflag;
714
715 AUDIT_ARG_SOCKET(domain, type, protocol);
716
717 oflag = 0;
718 fflag = 0;
719 if ((type & SOCK_CLOEXEC) != 0) {
720 type &= ~SOCK_CLOEXEC;
721 oflag |= O_CLOEXEC;
722 }
723 if ((type & SOCK_NONBLOCK) != 0) {
724 type &= ~SOCK_NONBLOCK;
725 fflag |= FNONBLOCK;
726 }
727 #ifdef MAC
728 /* We might want to have a separate check for socket pairs. */
729 error = mac_socket_check_create(td->td_ucred, domain, type,
730 protocol);
731 if (error != 0)
732 return (error);
733 #endif
734 error = socreate(domain, &so1, type, protocol, td->td_ucred, td);
735 if (error != 0)
736 return (error);
737 error = socreate(domain, &so2, type, protocol, td->td_ucred, td);
738 if (error != 0)
739 goto free1;
740 /* On success extra reference to `fp1' and 'fp2' is set by falloc. */
741 error = falloc(td, &fp1, &fd, oflag);
742 if (error != 0)
743 goto free2;
744 rsv[0] = fd;
745 fp1->f_data = so1; /* so1 already has ref count */
746 error = falloc(td, &fp2, &fd, oflag);
747 if (error != 0)
748 goto free3;
749 fp2->f_data = so2; /* so2 already has ref count */
750 rsv[1] = fd;
751 error = soconnect2(so1, so2);
752 if (error != 0)
753 goto free4;
754 if (type == SOCK_DGRAM) {
755 /*
756 * Datagram socket connection is asymmetric.
757 */
758 error = soconnect2(so2, so1);
759 if (error != 0)
760 goto free4;
761 }
762 finit(fp1, FREAD | FWRITE | fflag, DTYPE_SOCKET, fp1->f_data,
763 &socketops);
764 finit(fp2, FREAD | FWRITE | fflag, DTYPE_SOCKET, fp2->f_data,
765 &socketops);
766 if ((fflag & FNONBLOCK) != 0) {
767 (void) fo_ioctl(fp1, FIONBIO, &fflag, td->td_ucred, td);
768 (void) fo_ioctl(fp2, FIONBIO, &fflag, td->td_ucred, td);
769 }
770 fdrop(fp1, td);
771 fdrop(fp2, td);
772 return (0);
773 free4:
774 fdclose(td, fp2, rsv[1]);
775 fdrop(fp2, td);
776 free3:
777 fdclose(td, fp1, rsv[0]);
778 fdrop(fp1, td);
779 free2:
780 if (so2 != NULL)
781 (void)soclose(so2);
782 free1:
783 if (so1 != NULL)
784 (void)soclose(so1);
785 return (error);
786 }
787
788 int
789 sys_socketpair(struct thread *td, struct socketpair_args *uap)
790 {
791 int error, sv[2];
792
793 error = kern_socketpair(td, uap->domain, uap->type,
794 uap->protocol, sv);
795 if (error != 0)
796 return (error);
797 error = copyout(sv, uap->rsv, 2 * sizeof(int));
798 if (error != 0) {
799 (void)kern_close(td, sv[0]);
800 (void)kern_close(td, sv[1]);
801 }
802 return (error);
803 }
804
805 static int
806 sendit(td, s, mp, flags)
807 struct thread *td;
808 int s;
809 struct msghdr *mp;
810 int flags;
811 {
812 struct mbuf *control;
813 struct sockaddr *to;
814 int error;
815
816 #ifdef CAPABILITY_MODE
817 if (IN_CAPABILITY_MODE(td) && (mp->msg_name != NULL))
818 return (ECAPMODE);
819 #endif
820
821 if (mp->msg_name != NULL) {
822 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
823 if (error != 0) {
824 to = NULL;
825 goto bad;
826 }
827 mp->msg_name = to;
828 } else {
829 to = NULL;
830 }
831
832 if (mp->msg_control) {
833 if (mp->msg_controllen < sizeof(struct cmsghdr)
834 #ifdef COMPAT_OLDSOCK
835 && mp->msg_flags != MSG_COMPAT
836 #endif
837 ) {
838 error = EINVAL;
839 goto bad;
840 }
841 error = sockargs(&control, mp->msg_control,
842 mp->msg_controllen, MT_CONTROL);
843 if (error != 0)
844 goto bad;
845 #ifdef COMPAT_OLDSOCK
846 if (mp->msg_flags == MSG_COMPAT) {
847 struct cmsghdr *cm;
848
849 M_PREPEND(control, sizeof(*cm), M_WAITOK);
850 cm = mtod(control, struct cmsghdr *);
851 cm->cmsg_len = control->m_len;
852 cm->cmsg_level = SOL_SOCKET;
853 cm->cmsg_type = SCM_RIGHTS;
854 }
855 #endif
856 } else {
857 control = NULL;
858 }
859
860 error = kern_sendit(td, s, mp, flags, control, UIO_USERSPACE);
861
862 bad:
863 free(to, M_SONAME);
864 return (error);
865 }
866
867 int
868 kern_sendit(td, s, mp, flags, control, segflg)
869 struct thread *td;
870 int s;
871 struct msghdr *mp;
872 int flags;
873 struct mbuf *control;
874 enum uio_seg segflg;
875 {
876 struct file *fp;
877 struct uio auio;
878 struct iovec *iov;
879 struct socket *so;
880 cap_rights_t rights;
881 #ifdef KTRACE
882 struct uio *ktruio = NULL;
883 #endif
884 ssize_t len;
885 int i, error;
886
887 AUDIT_ARG_FD(s);
888 cap_rights_init(&rights, CAP_SEND);
889 if (mp->msg_name != NULL) {
890 AUDIT_ARG_SOCKADDR(td, AT_FDCWD, mp->msg_name);
891 cap_rights_set(&rights, CAP_CONNECT);
892 }
893 error = getsock_cap(td, s, &rights, &fp, NULL);
894 if (error != 0)
895 return (error);
896 so = (struct socket *)fp->f_data;
897
898 #ifdef KTRACE
899 if (mp->msg_name != NULL && KTRPOINT(td, KTR_STRUCT))
900 ktrsockaddr(mp->msg_name);
901 #endif
902 #ifdef MAC
903 if (mp->msg_name != NULL) {
904 error = mac_socket_check_connect(td->td_ucred, so,
905 mp->msg_name);
906 if (error != 0)
907 goto bad;
908 }
909 error = mac_socket_check_send(td->td_ucred, so);
910 if (error != 0)
911 goto bad;
912 #endif
913
914 auio.uio_iov = mp->msg_iov;
915 auio.uio_iovcnt = mp->msg_iovlen;
916 auio.uio_segflg = segflg;
917 auio.uio_rw = UIO_WRITE;
918 auio.uio_td = td;
919 auio.uio_offset = 0; /* XXX */
920 auio.uio_resid = 0;
921 iov = mp->msg_iov;
922 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
923 if ((auio.uio_resid += iov->iov_len) < 0) {
924 error = EINVAL;
925 goto bad;
926 }
927 }
928 #ifdef KTRACE
929 if (KTRPOINT(td, KTR_GENIO))
930 ktruio = cloneuio(&auio);
931 #endif
932 len = auio.uio_resid;
933 error = sosend(so, mp->msg_name, &auio, 0, control, flags, td);
934 if (error != 0) {
935 if (auio.uio_resid != len && (error == ERESTART ||
936 error == EINTR || error == EWOULDBLOCK))
937 error = 0;
938 /* Generation of SIGPIPE can be controlled per socket */
939 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
940 !(flags & MSG_NOSIGNAL)) {
941 PROC_LOCK(td->td_proc);
942 tdsignal(td, SIGPIPE);
943 PROC_UNLOCK(td->td_proc);
944 }
945 }
946 if (error == 0)
947 td->td_retval[0] = len - auio.uio_resid;
948 #ifdef KTRACE
949 if (ktruio != NULL) {
950 ktruio->uio_resid = td->td_retval[0];
951 ktrgenio(s, UIO_WRITE, ktruio, error);
952 }
953 #endif
954 bad:
955 fdrop(fp, td);
956 return (error);
957 }
958
959 int
960 sys_sendto(td, uap)
961 struct thread *td;
962 struct sendto_args /* {
963 int s;
964 caddr_t buf;
965 size_t len;
966 int flags;
967 caddr_t to;
968 int tolen;
969 } */ *uap;
970 {
971 struct msghdr msg;
972 struct iovec aiov;
973
974 msg.msg_name = uap->to;
975 msg.msg_namelen = uap->tolen;
976 msg.msg_iov = &aiov;
977 msg.msg_iovlen = 1;
978 msg.msg_control = 0;
979 #ifdef COMPAT_OLDSOCK
980 msg.msg_flags = 0;
981 #endif
982 aiov.iov_base = uap->buf;
983 aiov.iov_len = uap->len;
984 return (sendit(td, uap->s, &msg, uap->flags));
985 }
986
987 #ifdef COMPAT_OLDSOCK
988 int
989 osend(td, uap)
990 struct thread *td;
991 struct osend_args /* {
992 int s;
993 caddr_t buf;
994 int len;
995 int flags;
996 } */ *uap;
997 {
998 struct msghdr msg;
999 struct iovec aiov;
1000
1001 msg.msg_name = 0;
1002 msg.msg_namelen = 0;
1003 msg.msg_iov = &aiov;
1004 msg.msg_iovlen = 1;
1005 aiov.iov_base = uap->buf;
1006 aiov.iov_len = uap->len;
1007 msg.msg_control = 0;
1008 msg.msg_flags = 0;
1009 return (sendit(td, uap->s, &msg, uap->flags));
1010 }
1011
1012 int
1013 osendmsg(td, uap)
1014 struct thread *td;
1015 struct osendmsg_args /* {
1016 int s;
1017 caddr_t msg;
1018 int flags;
1019 } */ *uap;
1020 {
1021 struct msghdr msg;
1022 struct iovec *iov;
1023 int error;
1024
1025 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
1026 if (error != 0)
1027 return (error);
1028 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1029 if (error != 0)
1030 return (error);
1031 msg.msg_iov = iov;
1032 msg.msg_flags = MSG_COMPAT;
1033 error = sendit(td, uap->s, &msg, uap->flags);
1034 free(iov, M_IOV);
1035 return (error);
1036 }
1037 #endif
1038
1039 int
1040 sys_sendmsg(td, uap)
1041 struct thread *td;
1042 struct sendmsg_args /* {
1043 int s;
1044 caddr_t msg;
1045 int flags;
1046 } */ *uap;
1047 {
1048 struct msghdr msg;
1049 struct iovec *iov;
1050 int error;
1051
1052 error = copyin(uap->msg, &msg, sizeof (msg));
1053 if (error != 0)
1054 return (error);
1055 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1056 if (error != 0)
1057 return (error);
1058 msg.msg_iov = iov;
1059 #ifdef COMPAT_OLDSOCK
1060 msg.msg_flags = 0;
1061 #endif
1062 error = sendit(td, uap->s, &msg, uap->flags);
1063 free(iov, M_IOV);
1064 return (error);
1065 }
1066
1067 int
1068 kern_recvit(td, s, mp, fromseg, controlp)
1069 struct thread *td;
1070 int s;
1071 struct msghdr *mp;
1072 enum uio_seg fromseg;
1073 struct mbuf **controlp;
1074 {
1075 struct uio auio;
1076 struct iovec *iov;
1077 struct mbuf *m, *control = NULL;
1078 caddr_t ctlbuf;
1079 struct file *fp;
1080 struct socket *so;
1081 struct sockaddr *fromsa = NULL;
1082 cap_rights_t rights;
1083 #ifdef KTRACE
1084 struct uio *ktruio = NULL;
1085 #endif
1086 ssize_t len;
1087 int error, i;
1088
1089 if (controlp != NULL)
1090 *controlp = NULL;
1091
1092 AUDIT_ARG_FD(s);
1093 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_RECV),
1094 &fp, NULL);
1095 if (error != 0)
1096 return (error);
1097 so = fp->f_data;
1098
1099 #ifdef MAC
1100 error = mac_socket_check_receive(td->td_ucred, so);
1101 if (error != 0) {
1102 fdrop(fp, td);
1103 return (error);
1104 }
1105 #endif
1106
1107 auio.uio_iov = mp->msg_iov;
1108 auio.uio_iovcnt = mp->msg_iovlen;
1109 auio.uio_segflg = UIO_USERSPACE;
1110 auio.uio_rw = UIO_READ;
1111 auio.uio_td = td;
1112 auio.uio_offset = 0; /* XXX */
1113 auio.uio_resid = 0;
1114 iov = mp->msg_iov;
1115 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
1116 if ((auio.uio_resid += iov->iov_len) < 0) {
1117 fdrop(fp, td);
1118 return (EINVAL);
1119 }
1120 }
1121 #ifdef KTRACE
1122 if (KTRPOINT(td, KTR_GENIO))
1123 ktruio = cloneuio(&auio);
1124 #endif
1125 len = auio.uio_resid;
1126 error = soreceive(so, &fromsa, &auio, NULL,
1127 (mp->msg_control || controlp) ? &control : NULL,
1128 &mp->msg_flags);
1129 if (error != 0) {
1130 if (auio.uio_resid != len && (error == ERESTART ||
1131 error == EINTR || error == EWOULDBLOCK))
1132 error = 0;
1133 }
1134 if (fromsa != NULL)
1135 AUDIT_ARG_SOCKADDR(td, AT_FDCWD, fromsa);
1136 #ifdef KTRACE
1137 if (ktruio != NULL) {
1138 ktruio->uio_resid = len - auio.uio_resid;
1139 ktrgenio(s, UIO_READ, ktruio, error);
1140 }
1141 #endif
1142 if (error != 0)
1143 goto out;
1144 td->td_retval[0] = len - auio.uio_resid;
1145 if (mp->msg_name) {
1146 len = mp->msg_namelen;
1147 if (len <= 0 || fromsa == NULL)
1148 len = 0;
1149 else {
1150 /* save sa_len before it is destroyed by MSG_COMPAT */
1151 len = MIN(len, fromsa->sa_len);
1152 #ifdef COMPAT_OLDSOCK
1153 if (mp->msg_flags & MSG_COMPAT)
1154 ((struct osockaddr *)fromsa)->sa_family =
1155 fromsa->sa_family;
1156 #endif
1157 if (fromseg == UIO_USERSPACE) {
1158 error = copyout(fromsa, mp->msg_name,
1159 (unsigned)len);
1160 if (error != 0)
1161 goto out;
1162 } else
1163 bcopy(fromsa, mp->msg_name, len);
1164 }
1165 mp->msg_namelen = len;
1166 }
1167 if (mp->msg_control && controlp == NULL) {
1168 #ifdef COMPAT_OLDSOCK
1169 /*
1170 * We assume that old recvmsg calls won't receive access
1171 * rights and other control info, esp. as control info
1172 * is always optional and those options didn't exist in 4.3.
1173 * If we receive rights, trim the cmsghdr; anything else
1174 * is tossed.
1175 */
1176 if (control && mp->msg_flags & MSG_COMPAT) {
1177 if (mtod(control, struct cmsghdr *)->cmsg_level !=
1178 SOL_SOCKET ||
1179 mtod(control, struct cmsghdr *)->cmsg_type !=
1180 SCM_RIGHTS) {
1181 mp->msg_controllen = 0;
1182 goto out;
1183 }
1184 control->m_len -= sizeof (struct cmsghdr);
1185 control->m_data += sizeof (struct cmsghdr);
1186 }
1187 #endif
1188 len = mp->msg_controllen;
1189 m = control;
1190 mp->msg_controllen = 0;
1191 ctlbuf = mp->msg_control;
1192
1193 while (m && len > 0) {
1194 unsigned int tocopy;
1195
1196 if (len >= m->m_len)
1197 tocopy = m->m_len;
1198 else {
1199 mp->msg_flags |= MSG_CTRUNC;
1200 tocopy = len;
1201 }
1202
1203 if ((error = copyout(mtod(m, caddr_t),
1204 ctlbuf, tocopy)) != 0)
1205 goto out;
1206
1207 ctlbuf += tocopy;
1208 len -= tocopy;
1209 m = m->m_next;
1210 }
1211 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
1212 }
1213 out:
1214 fdrop(fp, td);
1215 #ifdef KTRACE
1216 if (fromsa && KTRPOINT(td, KTR_STRUCT))
1217 ktrsockaddr(fromsa);
1218 #endif
1219 free(fromsa, M_SONAME);
1220
1221 if (error == 0 && controlp != NULL)
1222 *controlp = control;
1223 else if (control)
1224 m_freem(control);
1225
1226 return (error);
1227 }
1228
1229 static int
1230 recvit(td, s, mp, namelenp)
1231 struct thread *td;
1232 int s;
1233 struct msghdr *mp;
1234 void *namelenp;
1235 {
1236 int error;
1237
1238 error = kern_recvit(td, s, mp, UIO_USERSPACE, NULL);
1239 if (error != 0)
1240 return (error);
1241 if (namelenp != NULL) {
1242 error = copyout(&mp->msg_namelen, namelenp, sizeof (socklen_t));
1243 #ifdef COMPAT_OLDSOCK
1244 if (mp->msg_flags & MSG_COMPAT)
1245 error = 0; /* old recvfrom didn't check */
1246 #endif
1247 }
1248 return (error);
1249 }
1250
1251 int
1252 sys_recvfrom(td, uap)
1253 struct thread *td;
1254 struct recvfrom_args /* {
1255 int s;
1256 caddr_t buf;
1257 size_t len;
1258 int flags;
1259 struct sockaddr * __restrict from;
1260 socklen_t * __restrict fromlenaddr;
1261 } */ *uap;
1262 {
1263 struct msghdr msg;
1264 struct iovec aiov;
1265 int error;
1266
1267 if (uap->fromlenaddr) {
1268 error = copyin(uap->fromlenaddr,
1269 &msg.msg_namelen, sizeof (msg.msg_namelen));
1270 if (error != 0)
1271 goto done2;
1272 } else {
1273 msg.msg_namelen = 0;
1274 }
1275 msg.msg_name = uap->from;
1276 msg.msg_iov = &aiov;
1277 msg.msg_iovlen = 1;
1278 aiov.iov_base = uap->buf;
1279 aiov.iov_len = uap->len;
1280 msg.msg_control = 0;
1281 msg.msg_flags = uap->flags;
1282 error = recvit(td, uap->s, &msg, uap->fromlenaddr);
1283 done2:
1284 return (error);
1285 }
1286
1287 #ifdef COMPAT_OLDSOCK
1288 int
1289 orecvfrom(td, uap)
1290 struct thread *td;
1291 struct recvfrom_args *uap;
1292 {
1293
1294 uap->flags |= MSG_COMPAT;
1295 return (sys_recvfrom(td, uap));
1296 }
1297 #endif
1298
1299 #ifdef COMPAT_OLDSOCK
1300 int
1301 orecv(td, uap)
1302 struct thread *td;
1303 struct orecv_args /* {
1304 int s;
1305 caddr_t buf;
1306 int len;
1307 int flags;
1308 } */ *uap;
1309 {
1310 struct msghdr msg;
1311 struct iovec aiov;
1312
1313 msg.msg_name = 0;
1314 msg.msg_namelen = 0;
1315 msg.msg_iov = &aiov;
1316 msg.msg_iovlen = 1;
1317 aiov.iov_base = uap->buf;
1318 aiov.iov_len = uap->len;
1319 msg.msg_control = 0;
1320 msg.msg_flags = uap->flags;
1321 return (recvit(td, uap->s, &msg, NULL));
1322 }
1323
1324 /*
1325 * Old recvmsg. This code takes advantage of the fact that the old msghdr
1326 * overlays the new one, missing only the flags, and with the (old) access
1327 * rights where the control fields are now.
1328 */
1329 int
1330 orecvmsg(td, uap)
1331 struct thread *td;
1332 struct orecvmsg_args /* {
1333 int s;
1334 struct omsghdr *msg;
1335 int flags;
1336 } */ *uap;
1337 {
1338 struct msghdr msg;
1339 struct iovec *iov;
1340 int error;
1341
1342 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
1343 if (error != 0)
1344 return (error);
1345 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1346 if (error != 0)
1347 return (error);
1348 msg.msg_flags = uap->flags | MSG_COMPAT;
1349 msg.msg_iov = iov;
1350 error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen);
1351 if (msg.msg_controllen && error == 0)
1352 error = copyout(&msg.msg_controllen,
1353 &uap->msg->msg_accrightslen, sizeof (int));
1354 free(iov, M_IOV);
1355 return (error);
1356 }
1357 #endif
1358
1359 int
1360 sys_recvmsg(td, uap)
1361 struct thread *td;
1362 struct recvmsg_args /* {
1363 int s;
1364 struct msghdr *msg;
1365 int flags;
1366 } */ *uap;
1367 {
1368 struct msghdr msg;
1369 struct iovec *uiov, *iov;
1370 int error;
1371
1372 error = copyin(uap->msg, &msg, sizeof (msg));
1373 if (error != 0)
1374 return (error);
1375 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1376 if (error != 0)
1377 return (error);
1378 msg.msg_flags = uap->flags;
1379 #ifdef COMPAT_OLDSOCK
1380 msg.msg_flags &= ~MSG_COMPAT;
1381 #endif
1382 uiov = msg.msg_iov;
1383 msg.msg_iov = iov;
1384 error = recvit(td, uap->s, &msg, NULL);
1385 if (error == 0) {
1386 msg.msg_iov = uiov;
1387 error = copyout(&msg, uap->msg, sizeof(msg));
1388 }
1389 free(iov, M_IOV);
1390 return (error);
1391 }
1392
1393 /* ARGSUSED */
1394 int
1395 sys_shutdown(td, uap)
1396 struct thread *td;
1397 struct shutdown_args /* {
1398 int s;
1399 int how;
1400 } */ *uap;
1401 {
1402 struct socket *so;
1403 struct file *fp;
1404 cap_rights_t rights;
1405 int error;
1406
1407 AUDIT_ARG_FD(uap->s);
1408 error = getsock_cap(td, uap->s, cap_rights_init(&rights, CAP_SHUTDOWN),
1409 &fp, NULL);
1410 if (error == 0) {
1411 so = fp->f_data;
1412 error = soshutdown(so, uap->how);
1413 fdrop(fp, td);
1414 }
1415 return (error);
1416 }
1417
1418 /* ARGSUSED */
1419 int
1420 sys_setsockopt(td, uap)
1421 struct thread *td;
1422 struct setsockopt_args /* {
1423 int s;
1424 int level;
1425 int name;
1426 caddr_t val;
1427 int valsize;
1428 } */ *uap;
1429 {
1430
1431 return (kern_setsockopt(td, uap->s, uap->level, uap->name,
1432 uap->val, UIO_USERSPACE, uap->valsize));
1433 }
1434
1435 int
1436 kern_setsockopt(td, s, level, name, val, valseg, valsize)
1437 struct thread *td;
1438 int s;
1439 int level;
1440 int name;
1441 void *val;
1442 enum uio_seg valseg;
1443 socklen_t valsize;
1444 {
1445 struct socket *so;
1446 struct file *fp;
1447 struct sockopt sopt;
1448 cap_rights_t rights;
1449 int error;
1450
1451 if (val == NULL && valsize != 0)
1452 return (EFAULT);
1453 if ((int)valsize < 0)
1454 return (EINVAL);
1455
1456 sopt.sopt_dir = SOPT_SET;
1457 sopt.sopt_level = level;
1458 sopt.sopt_name = name;
1459 sopt.sopt_val = val;
1460 sopt.sopt_valsize = valsize;
1461 switch (valseg) {
1462 case UIO_USERSPACE:
1463 sopt.sopt_td = td;
1464 break;
1465 case UIO_SYSSPACE:
1466 sopt.sopt_td = NULL;
1467 break;
1468 default:
1469 panic("kern_setsockopt called with bad valseg");
1470 }
1471
1472 AUDIT_ARG_FD(s);
1473 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SETSOCKOPT),
1474 &fp, NULL);
1475 if (error == 0) {
1476 so = fp->f_data;
1477 error = sosetopt(so, &sopt);
1478 fdrop(fp, td);
1479 }
1480 return(error);
1481 }
1482
1483 /* ARGSUSED */
1484 int
1485 sys_getsockopt(td, uap)
1486 struct thread *td;
1487 struct getsockopt_args /* {
1488 int s;
1489 int level;
1490 int name;
1491 void * __restrict val;
1492 socklen_t * __restrict avalsize;
1493 } */ *uap;
1494 {
1495 socklen_t valsize;
1496 int error;
1497
1498 if (uap->val) {
1499 error = copyin(uap->avalsize, &valsize, sizeof (valsize));
1500 if (error != 0)
1501 return (error);
1502 }
1503
1504 error = kern_getsockopt(td, uap->s, uap->level, uap->name,
1505 uap->val, UIO_USERSPACE, &valsize);
1506
1507 if (error == 0)
1508 error = copyout(&valsize, uap->avalsize, sizeof (valsize));
1509 return (error);
1510 }
1511
1512 /*
1513 * Kernel version of getsockopt.
1514 * optval can be a userland or userspace. optlen is always a kernel pointer.
1515 */
1516 int
1517 kern_getsockopt(td, s, level, name, val, valseg, valsize)
1518 struct thread *td;
1519 int s;
1520 int level;
1521 int name;
1522 void *val;
1523 enum uio_seg valseg;
1524 socklen_t *valsize;
1525 {
1526 struct socket *so;
1527 struct file *fp;
1528 struct sockopt sopt;
1529 cap_rights_t rights;
1530 int error;
1531
1532 if (val == NULL)
1533 *valsize = 0;
1534 if ((int)*valsize < 0)
1535 return (EINVAL);
1536
1537 sopt.sopt_dir = SOPT_GET;
1538 sopt.sopt_level = level;
1539 sopt.sopt_name = name;
1540 sopt.sopt_val = val;
1541 sopt.sopt_valsize = (size_t)*valsize; /* checked non-negative above */
1542 switch (valseg) {
1543 case UIO_USERSPACE:
1544 sopt.sopt_td = td;
1545 break;
1546 case UIO_SYSSPACE:
1547 sopt.sopt_td = NULL;
1548 break;
1549 default:
1550 panic("kern_getsockopt called with bad valseg");
1551 }
1552
1553 AUDIT_ARG_FD(s);
1554 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_GETSOCKOPT),
1555 &fp, NULL);
1556 if (error == 0) {
1557 so = fp->f_data;
1558 error = sogetopt(so, &sopt);
1559 *valsize = sopt.sopt_valsize;
1560 fdrop(fp, td);
1561 }
1562 return (error);
1563 }
1564
1565 /*
1566 * getsockname1() - Get socket name.
1567 */
1568 /* ARGSUSED */
1569 static int
1570 getsockname1(td, uap, compat)
1571 struct thread *td;
1572 struct getsockname_args /* {
1573 int fdes;
1574 struct sockaddr * __restrict asa;
1575 socklen_t * __restrict alen;
1576 } */ *uap;
1577 int compat;
1578 {
1579 struct sockaddr *sa;
1580 socklen_t len;
1581 int error;
1582
1583 error = copyin(uap->alen, &len, sizeof(len));
1584 if (error != 0)
1585 return (error);
1586
1587 error = kern_getsockname(td, uap->fdes, &sa, &len);
1588 if (error != 0)
1589 return (error);
1590
1591 if (len != 0) {
1592 #ifdef COMPAT_OLDSOCK
1593 if (compat)
1594 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1595 #endif
1596 error = copyout(sa, uap->asa, (u_int)len);
1597 }
1598 free(sa, M_SONAME);
1599 if (error == 0)
1600 error = copyout(&len, uap->alen, sizeof(len));
1601 return (error);
1602 }
1603
1604 int
1605 kern_getsockname(struct thread *td, int fd, struct sockaddr **sa,
1606 socklen_t *alen)
1607 {
1608 struct socket *so;
1609 struct file *fp;
1610 cap_rights_t rights;
1611 socklen_t len;
1612 int error;
1613
1614 AUDIT_ARG_FD(fd);
1615 error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_GETSOCKNAME),
1616 &fp, NULL);
1617 if (error != 0)
1618 return (error);
1619 so = fp->f_data;
1620 *sa = NULL;
1621 CURVNET_SET(so->so_vnet);
1622 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, sa);
1623 CURVNET_RESTORE();
1624 if (error != 0)
1625 goto bad;
1626 if (*sa == NULL)
1627 len = 0;
1628 else
1629 len = MIN(*alen, (*sa)->sa_len);
1630 *alen = len;
1631 #ifdef KTRACE
1632 if (KTRPOINT(td, KTR_STRUCT))
1633 ktrsockaddr(*sa);
1634 #endif
1635 bad:
1636 fdrop(fp, td);
1637 if (error != 0 && *sa != NULL) {
1638 free(*sa, M_SONAME);
1639 *sa = NULL;
1640 }
1641 return (error);
1642 }
1643
1644 int
1645 sys_getsockname(td, uap)
1646 struct thread *td;
1647 struct getsockname_args *uap;
1648 {
1649
1650 return (getsockname1(td, uap, 0));
1651 }
1652
1653 #ifdef COMPAT_OLDSOCK
1654 int
1655 ogetsockname(td, uap)
1656 struct thread *td;
1657 struct getsockname_args *uap;
1658 {
1659
1660 return (getsockname1(td, uap, 1));
1661 }
1662 #endif /* COMPAT_OLDSOCK */
1663
1664 /*
1665 * getpeername1() - Get name of peer for connected socket.
1666 */
1667 /* ARGSUSED */
1668 static int
1669 getpeername1(td, uap, compat)
1670 struct thread *td;
1671 struct getpeername_args /* {
1672 int fdes;
1673 struct sockaddr * __restrict asa;
1674 socklen_t * __restrict alen;
1675 } */ *uap;
1676 int compat;
1677 {
1678 struct sockaddr *sa;
1679 socklen_t len;
1680 int error;
1681
1682 error = copyin(uap->alen, &len, sizeof (len));
1683 if (error != 0)
1684 return (error);
1685
1686 error = kern_getpeername(td, uap->fdes, &sa, &len);
1687 if (error != 0)
1688 return (error);
1689
1690 if (len != 0) {
1691 #ifdef COMPAT_OLDSOCK
1692 if (compat)
1693 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1694 #endif
1695 error = copyout(sa, uap->asa, (u_int)len);
1696 }
1697 free(sa, M_SONAME);
1698 if (error == 0)
1699 error = copyout(&len, uap->alen, sizeof(len));
1700 return (error);
1701 }
1702
1703 int
1704 kern_getpeername(struct thread *td, int fd, struct sockaddr **sa,
1705 socklen_t *alen)
1706 {
1707 struct socket *so;
1708 struct file *fp;
1709 cap_rights_t rights;
1710 socklen_t len;
1711 int error;
1712
1713 AUDIT_ARG_FD(fd);
1714 error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_GETPEERNAME),
1715 &fp, NULL);
1716 if (error != 0)
1717 return (error);
1718 so = fp->f_data;
1719 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1720 error = ENOTCONN;
1721 goto done;
1722 }
1723 *sa = NULL;
1724 CURVNET_SET(so->so_vnet);
1725 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, sa);
1726 CURVNET_RESTORE();
1727 if (error != 0)
1728 goto bad;
1729 if (*sa == NULL)
1730 len = 0;
1731 else
1732 len = MIN(*alen, (*sa)->sa_len);
1733 *alen = len;
1734 #ifdef KTRACE
1735 if (KTRPOINT(td, KTR_STRUCT))
1736 ktrsockaddr(*sa);
1737 #endif
1738 bad:
1739 if (error != 0 && *sa != NULL) {
1740 free(*sa, M_SONAME);
1741 *sa = NULL;
1742 }
1743 done:
1744 fdrop(fp, td);
1745 return (error);
1746 }
1747
1748 int
1749 sys_getpeername(td, uap)
1750 struct thread *td;
1751 struct getpeername_args *uap;
1752 {
1753
1754 return (getpeername1(td, uap, 0));
1755 }
1756
1757 #ifdef COMPAT_OLDSOCK
1758 int
1759 ogetpeername(td, uap)
1760 struct thread *td;
1761 struct ogetpeername_args *uap;
1762 {
1763
1764 /* XXX uap should have type `getpeername_args *' to begin with. */
1765 return (getpeername1(td, (struct getpeername_args *)uap, 1));
1766 }
1767 #endif /* COMPAT_OLDSOCK */
1768
1769 int
1770 sockargs(mp, buf, buflen, type)
1771 struct mbuf **mp;
1772 caddr_t buf;
1773 int buflen, type;
1774 {
1775 struct sockaddr *sa;
1776 struct mbuf *m;
1777 int error;
1778
1779 if (buflen < 0)
1780 return (EINVAL);
1781
1782 if (buflen > MLEN) {
1783 #ifdef COMPAT_OLDSOCK
1784 if (type == MT_SONAME && buflen <= 112)
1785 buflen = MLEN; /* unix domain compat. hack */
1786 else
1787 #endif
1788 if (buflen > MCLBYTES)
1789 return (EINVAL);
1790 }
1791 m = m_get2(buflen, M_WAITOK, type, 0);
1792 m->m_len = buflen;
1793 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1794 if (error != 0)
1795 (void) m_free(m);
1796 else {
1797 *mp = m;
1798 if (type == MT_SONAME) {
1799 sa = mtod(m, struct sockaddr *);
1800
1801 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1802 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1803 sa->sa_family = sa->sa_len;
1804 #endif
1805 sa->sa_len = buflen;
1806 }
1807 }
1808 return (error);
1809 }
1810
1811 int
1812 getsockaddr(namp, uaddr, len)
1813 struct sockaddr **namp;
1814 caddr_t uaddr;
1815 size_t len;
1816 {
1817 struct sockaddr *sa;
1818 int error;
1819
1820 if (len > SOCK_MAXADDRLEN)
1821 return (ENAMETOOLONG);
1822 if (len < offsetof(struct sockaddr, sa_data[0]))
1823 return (EINVAL);
1824 sa = malloc(len, M_SONAME, M_WAITOK);
1825 error = copyin(uaddr, sa, len);
1826 if (error != 0) {
1827 free(sa, M_SONAME);
1828 } else {
1829 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1830 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1831 sa->sa_family = sa->sa_len;
1832 #endif
1833 sa->sa_len = len;
1834 *namp = sa;
1835 }
1836 return (error);
1837 }
1838
1839 struct sendfile_sync {
1840 struct mtx mtx;
1841 struct cv cv;
1842 unsigned count;
1843 };
1844
1845 /*
1846 * Detach mapped page and release resources back to the system.
1847 */
1848 int
1849 sf_buf_mext(struct mbuf *mb, void *addr, void *args)
1850 {
1851 vm_page_t m;
1852 struct sendfile_sync *sfs;
1853
1854 m = sf_buf_page(args);
1855 sf_buf_free(args);
1856 vm_page_lock(m);
1857 vm_page_unwire(m, 0);
1858 /*
1859 * Check for the object going away on us. This can
1860 * happen since we don't hold a reference to it.
1861 * If so, we're responsible for freeing the page.
1862 */
1863 if (m->wire_count == 0 && m->object == NULL)
1864 vm_page_free(m);
1865 vm_page_unlock(m);
1866 if (addr == NULL)
1867 return (EXT_FREE_OK);
1868 sfs = addr;
1869 mtx_lock(&sfs->mtx);
1870 KASSERT(sfs->count> 0, ("Sendfile sync botchup count == 0"));
1871 if (--sfs->count == 0)
1872 cv_signal(&sfs->cv);
1873 mtx_unlock(&sfs->mtx);
1874 return (EXT_FREE_OK);
1875 }
1876
1877 /*
1878 * sendfile(2)
1879 *
1880 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1881 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1882 *
1883 * Send a file specified by 'fd' and starting at 'offset' to a socket
1884 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1885 * 0. Optionally add a header and/or trailer to the socket output. If
1886 * specified, write the total number of bytes sent into *sbytes.
1887 */
1888 int
1889 sys_sendfile(struct thread *td, struct sendfile_args *uap)
1890 {
1891
1892 return (do_sendfile(td, uap, 0));
1893 }
1894
1895 static int
1896 do_sendfile(struct thread *td, struct sendfile_args *uap, int compat)
1897 {
1898 struct sf_hdtr hdtr;
1899 struct uio *hdr_uio, *trl_uio;
1900 struct file *fp;
1901 cap_rights_t rights;
1902 int error;
1903
1904 /*
1905 * File offset must be positive. If it goes beyond EOF
1906 * we send only the header/trailer and no payload data.
1907 */
1908 if (uap->offset < 0)
1909 return (EINVAL);
1910
1911 hdr_uio = trl_uio = NULL;
1912
1913 if (uap->hdtr != NULL) {
1914 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1915 if (error != 0)
1916 goto out;
1917 if (hdtr.headers != NULL) {
1918 error = copyinuio(hdtr.headers, hdtr.hdr_cnt, &hdr_uio);
1919 if (error != 0)
1920 goto out;
1921 }
1922 if (hdtr.trailers != NULL) {
1923 error = copyinuio(hdtr.trailers, hdtr.trl_cnt, &trl_uio);
1924 if (error != 0)
1925 goto out;
1926
1927 }
1928 }
1929
1930 AUDIT_ARG_FD(uap->fd);
1931
1932 /*
1933 * sendfile(2) can start at any offset within a file so we require
1934 * CAP_READ+CAP_SEEK = CAP_PREAD.
1935 */
1936 if ((error = fget_read(td, uap->fd,
1937 cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) {
1938 goto out;
1939 }
1940
1941 error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset,
1942 uap->nbytes, uap->sbytes, uap->flags, compat ? SFK_COMPAT : 0, td);
1943 fdrop(fp, td);
1944
1945 out:
1946 free(hdr_uio, M_IOV);
1947 free(trl_uio, M_IOV);
1948 return (error);
1949 }
1950
1951 #ifdef COMPAT_FREEBSD4
1952 int
1953 freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1954 {
1955 struct sendfile_args args;
1956
1957 args.fd = uap->fd;
1958 args.s = uap->s;
1959 args.offset = uap->offset;
1960 args.nbytes = uap->nbytes;
1961 args.hdtr = uap->hdtr;
1962 args.sbytes = uap->sbytes;
1963 args.flags = uap->flags;
1964
1965 return (do_sendfile(td, &args, 1));
1966 }
1967 #endif /* COMPAT_FREEBSD4 */
1968
1969 static int
1970 sendfile_readpage(vm_object_t obj, struct vnode *vp, int nd,
1971 off_t off, int xfsize, int bsize, struct thread *td, vm_page_t *res)
1972 {
1973 vm_page_t m;
1974 vm_pindex_t pindex;
1975 ssize_t resid;
1976 int error, readahead, rv;
1977
1978 pindex = OFF_TO_IDX(off);
1979 VM_OBJECT_WLOCK(obj);
1980 m = vm_page_grab(obj, pindex, (vp != NULL ? VM_ALLOC_NOBUSY |
1981 VM_ALLOC_IGN_SBUSY : 0) | VM_ALLOC_WIRED | VM_ALLOC_NORMAL);
1982
1983 /*
1984 * Check if page is valid for what we need, otherwise initiate I/O.
1985 *
1986 * The non-zero nd argument prevents disk I/O, instead we
1987 * return the caller what he specified in nd. In particular,
1988 * if we already turned some pages into mbufs, nd == EAGAIN
1989 * and the main function send them the pages before we come
1990 * here again and block.
1991 */
1992 if (m->valid != 0 && vm_page_is_valid(m, off & PAGE_MASK, xfsize)) {
1993 if (vp == NULL)
1994 vm_page_xunbusy(m);
1995 VM_OBJECT_WUNLOCK(obj);
1996 *res = m;
1997 return (0);
1998 } else if (nd != 0) {
1999 if (vp == NULL)
2000 vm_page_xunbusy(m);
2001 error = nd;
2002 goto free_page;
2003 }
2004
2005 /*
2006 * Get the page from backing store.
2007 */
2008 error = 0;
2009 if (vp != NULL) {
2010 VM_OBJECT_WUNLOCK(obj);
2011 readahead = sfreadahead * MAXBSIZE;
2012
2013 /*
2014 * Use vn_rdwr() instead of the pager interface for
2015 * the vnode, to allow the read-ahead.
2016 *
2017 * XXXMAC: Because we don't have fp->f_cred here, we
2018 * pass in NOCRED. This is probably wrong, but is
2019 * consistent with our original implementation.
2020 */
2021 error = vn_rdwr(UIO_READ, vp, NULL, readahead, trunc_page(off),
2022 UIO_NOCOPY, IO_NODELOCKED | IO_VMIO | ((readahead /
2023 bsize) << IO_SEQSHIFT), td->td_ucred, NOCRED, &resid, td);
2024 SFSTAT_INC(sf_iocnt);
2025 VM_OBJECT_WLOCK(obj);
2026 } else {
2027 if (vm_pager_has_page(obj, pindex, NULL, NULL)) {
2028 rv = vm_pager_get_pages(obj, &m, 1, 0);
2029 SFSTAT_INC(sf_iocnt);
2030 m = vm_page_lookup(obj, pindex);
2031 if (m == NULL)
2032 error = EIO;
2033 else if (rv != VM_PAGER_OK) {
2034 vm_page_lock(m);
2035 vm_page_free(m);
2036 vm_page_unlock(m);
2037 m = NULL;
2038 error = EIO;
2039 }
2040 } else {
2041 pmap_zero_page(m);
2042 m->valid = VM_PAGE_BITS_ALL;
2043 m->dirty = 0;
2044 }
2045 if (m != NULL)
2046 vm_page_xunbusy(m);
2047 }
2048 if (error == 0) {
2049 *res = m;
2050 } else if (m != NULL) {
2051 free_page:
2052 vm_page_lock(m);
2053 vm_page_unwire(m, 0);
2054
2055 /*
2056 * See if anyone else might know about this page. If
2057 * not and it is not valid, then free it.
2058 */
2059 if (m->wire_count == 0 && m->valid == 0 && !vm_page_busied(m))
2060 vm_page_free(m);
2061 vm_page_unlock(m);
2062 }
2063 KASSERT(error != 0 || (m->wire_count > 0 &&
2064 vm_page_is_valid(m, off & PAGE_MASK, xfsize)),
2065 ("wrong page state m %p off %#jx xfsize %d", m, (uintmax_t)off,
2066 xfsize));
2067 VM_OBJECT_WUNLOCK(obj);
2068 return (error);
2069 }
2070
2071 static int
2072 sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res,
2073 struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size,
2074 int *bsize)
2075 {
2076 struct vattr va;
2077 vm_object_t obj;
2078 struct vnode *vp;
2079 struct shmfd *shmfd;
2080 int error;
2081
2082 vp = *vp_res = NULL;
2083 obj = NULL;
2084 shmfd = *shmfd_res = NULL;
2085 *bsize = 0;
2086
2087 /*
2088 * The file descriptor must be a regular file and have a
2089 * backing VM object.
2090 */
2091 if (fp->f_type == DTYPE_VNODE) {
2092 vp = fp->f_vnode;
2093 vn_lock(vp, LK_SHARED | LK_RETRY);
2094 if (vp->v_type != VREG) {
2095 error = EINVAL;
2096 goto out;
2097 }
2098 *bsize = vp->v_mount->mnt_stat.f_iosize;
2099 error = VOP_GETATTR(vp, &va, td->td_ucred);
2100 if (error != 0)
2101 goto out;
2102 *obj_size = va.va_size;
2103 obj = vp->v_object;
2104 if (obj == NULL) {
2105 error = EINVAL;
2106 goto out;
2107 }
2108 } else if (fp->f_type == DTYPE_SHM) {
2109 error = 0;
2110 shmfd = fp->f_data;
2111 obj = shmfd->shm_object;
2112 *obj_size = shmfd->shm_size;
2113 } else {
2114 error = EINVAL;
2115 goto out;
2116 }
2117
2118 VM_OBJECT_WLOCK(obj);
2119 if ((obj->flags & OBJ_DEAD) != 0) {
2120 VM_OBJECT_WUNLOCK(obj);
2121 error = EBADF;
2122 goto out;
2123 }
2124
2125 /*
2126 * Temporarily increase the backing VM object's reference
2127 * count so that a forced reclamation of its vnode does not
2128 * immediately destroy it.
2129 */
2130 vm_object_reference_locked(obj);
2131 VM_OBJECT_WUNLOCK(obj);
2132 *obj_res = obj;
2133 *vp_res = vp;
2134 *shmfd_res = shmfd;
2135
2136 out:
2137 if (vp != NULL)
2138 VOP_UNLOCK(vp, 0);
2139 return (error);
2140 }
2141
2142 static int
2143 kern_sendfile_getsock(struct thread *td, int s, struct file **sock_fp,
2144 struct socket **so)
2145 {
2146 cap_rights_t rights;
2147 int error;
2148
2149 *sock_fp = NULL;
2150 *so = NULL;
2151
2152 /*
2153 * The socket must be a stream socket and connected.
2154 */
2155 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SEND),
2156 sock_fp, NULL);
2157 if (error != 0)
2158 return (error);
2159 *so = (*sock_fp)->f_data;
2160 if ((*so)->so_type != SOCK_STREAM)
2161 return (EINVAL);
2162 if (((*so)->so_state & SS_ISCONNECTED) == 0)
2163 return (ENOTCONN);
2164 return (0);
2165 }
2166
2167 int
2168 vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
2169 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
2170 int kflags, struct thread *td)
2171 {
2172 struct file *sock_fp;
2173 struct vnode *vp;
2174 struct vm_object *obj;
2175 struct socket *so;
2176 struct mbuf *m;
2177 struct sf_buf *sf;
2178 struct vm_page *pg;
2179 struct shmfd *shmfd;
2180 struct sendfile_sync *sfs;
2181 struct vattr va;
2182 off_t off, xfsize, fsbytes, sbytes, rem, obj_size;
2183 int error, bsize, nd, hdrlen, mnw;
2184 bool inflight_called;
2185
2186 pg = NULL;
2187 obj = NULL;
2188 so = NULL;
2189 m = NULL;
2190 sfs = NULL;
2191 fsbytes = sbytes = 0;
2192 hdrlen = mnw = 0;
2193 rem = nbytes;
2194 obj_size = 0;
2195 inflight_called = false;
2196
2197 error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize);
2198 if (error != 0)
2199 return (error);
2200 if (rem == 0)
2201 rem = obj_size;
2202
2203 error = kern_sendfile_getsock(td, sockfd, &sock_fp, &so);
2204 if (error != 0)
2205 goto out;
2206
2207 /*
2208 * Do not wait on memory allocations but return ENOMEM for
2209 * caller to retry later.
2210 * XXX: Experimental.
2211 */
2212 if (flags & SF_MNOWAIT)
2213 mnw = 1;
2214
2215 if (flags & SF_SYNC) {
2216 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO);
2217 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
2218 cv_init(&sfs->cv, "sendfile");
2219 }
2220
2221 #ifdef MAC
2222 error = mac_socket_check_send(td->td_ucred, so);
2223 if (error != 0)
2224 goto out;
2225 #endif
2226
2227 /* If headers are specified copy them into mbufs. */
2228 if (hdr_uio != NULL) {
2229 hdr_uio->uio_td = td;
2230 hdr_uio->uio_rw = UIO_WRITE;
2231 if (hdr_uio->uio_resid > 0) {
2232 /*
2233 * In FBSD < 5.0 the nbytes to send also included
2234 * the header. If compat is specified subtract the
2235 * header size from nbytes.
2236 */
2237 if (kflags & SFK_COMPAT) {
2238 if (nbytes > hdr_uio->uio_resid)
2239 nbytes -= hdr_uio->uio_resid;
2240 else
2241 nbytes = 0;
2242 }
2243 m = m_uiotombuf(hdr_uio, (mnw ? M_NOWAIT : M_WAITOK),
2244 0, 0, 0);
2245 if (m == NULL) {
2246 error = mnw ? EAGAIN : ENOBUFS;
2247 goto out;
2248 }
2249 hdrlen = m_length(m, NULL);
2250 }
2251 }
2252
2253 /*
2254 * Protect against multiple writers to the socket.
2255 *
2256 * XXXRW: Historically this has assumed non-interruptibility, so now
2257 * we implement that, but possibly shouldn't.
2258 */
2259 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
2260
2261 /*
2262 * Loop through the pages of the file, starting with the requested
2263 * offset. Get a file page (do I/O if necessary), map the file page
2264 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
2265 * it on the socket.
2266 * This is done in two loops. The inner loop turns as many pages
2267 * as it can, up to available socket buffer space, without blocking
2268 * into mbufs to have it bulk delivered into the socket send buffer.
2269 * The outer loop checks the state and available space of the socket
2270 * and takes care of the overall progress.
2271 */
2272 for (off = offset; ; ) {
2273 struct mbuf *mtail;
2274 int loopbytes;
2275 int space;
2276 int done;
2277
2278 if ((nbytes != 0 && nbytes == fsbytes) ||
2279 (nbytes == 0 && obj_size == fsbytes))
2280 break;
2281
2282 mtail = NULL;
2283 loopbytes = 0;
2284 space = 0;
2285 done = 0;
2286
2287 /*
2288 * Check the socket state for ongoing connection,
2289 * no errors and space in socket buffer.
2290 * If space is low allow for the remainder of the
2291 * file to be processed if it fits the socket buffer.
2292 * Otherwise block in waiting for sufficient space
2293 * to proceed, or if the socket is nonblocking, return
2294 * to userland with EAGAIN while reporting how far
2295 * we've come.
2296 * We wait until the socket buffer has significant free
2297 * space to do bulk sends. This makes good use of file
2298 * system read ahead and allows packet segmentation
2299 * offloading hardware to take over lots of work. If
2300 * we were not careful here we would send off only one
2301 * sfbuf at a time.
2302 */
2303 SOCKBUF_LOCK(&so->so_snd);
2304 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
2305 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
2306 retry_space:
2307 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2308 error = EPIPE;
2309 SOCKBUF_UNLOCK(&so->so_snd);
2310 goto done;
2311 } else if (so->so_error) {
2312 error = so->so_error;
2313 so->so_error = 0;
2314 SOCKBUF_UNLOCK(&so->so_snd);
2315 goto done;
2316 }
2317 space = sbspace(&so->so_snd);
2318 if (space < rem &&
2319 (space <= 0 ||
2320 space < so->so_snd.sb_lowat)) {
2321 if (so->so_state & SS_NBIO) {
2322 SOCKBUF_UNLOCK(&so->so_snd);
2323 error = EAGAIN;
2324 goto done;
2325 }
2326 /*
2327 * sbwait drops the lock while sleeping.
2328 * When we loop back to retry_space the
2329 * state may have changed and we retest
2330 * for it.
2331 */
2332 error = sbwait(&so->so_snd);
2333 /*
2334 * An error from sbwait usually indicates that we've
2335 * been interrupted by a signal. If we've sent anything
2336 * then return bytes sent, otherwise return the error.
2337 */
2338 if (error != 0) {
2339 SOCKBUF_UNLOCK(&so->so_snd);
2340 goto done;
2341 }
2342 goto retry_space;
2343 }
2344 SOCKBUF_UNLOCK(&so->so_snd);
2345
2346 /*
2347 * Reduce space in the socket buffer by the size of
2348 * the header mbuf chain.
2349 * hdrlen is set to 0 after the first loop.
2350 */
2351 space -= hdrlen;
2352
2353 if (vp != NULL) {
2354 error = vn_lock(vp, LK_SHARED);
2355 if (error != 0)
2356 goto done;
2357 error = VOP_GETATTR(vp, &va, td->td_ucred);
2358 if (error != 0 || off >= va.va_size) {
2359 VOP_UNLOCK(vp, 0);
2360 goto done;
2361 }
2362 obj_size = va.va_size;
2363 }
2364
2365 /*
2366 * Loop and construct maximum sized mbuf chain to be bulk
2367 * dumped into socket buffer.
2368 */
2369 while (space > loopbytes) {
2370 vm_offset_t pgoff;
2371 struct mbuf *m0;
2372
2373 /*
2374 * Calculate the amount to transfer.
2375 * Not to exceed a page, the EOF,
2376 * or the passed in nbytes.
2377 */
2378 pgoff = (vm_offset_t)(off & PAGE_MASK);
2379 rem = obj_size - offset;
2380 if (nbytes != 0)
2381 rem = omin(rem, nbytes);
2382 rem -= fsbytes + loopbytes;
2383 xfsize = omin(PAGE_SIZE - pgoff, rem);
2384 xfsize = omin(space - loopbytes, xfsize);
2385 if (xfsize <= 0) {
2386 done = 1; /* all data sent */
2387 break;
2388 }
2389
2390 /*
2391 * Attempt to look up the page. Allocate
2392 * if not found or wait and loop if busy.
2393 */
2394 if (m != NULL)
2395 nd = EAGAIN; /* send what we already got */
2396 else if ((flags & SF_NODISKIO) != 0)
2397 nd = EBUSY;
2398 else
2399 nd = 0;
2400 error = sendfile_readpage(obj, vp, nd, off,
2401 xfsize, bsize, td, &pg);
2402 if (error != 0) {
2403 if (error == EAGAIN)
2404 error = 0; /* not a real error */
2405 break;
2406 }
2407
2408 /*
2409 * Get a sendfile buf. When allocating the
2410 * first buffer for mbuf chain, we usually
2411 * wait as long as necessary, but this wait
2412 * can be interrupted. For consequent
2413 * buffers, do not sleep, since several
2414 * threads might exhaust the buffers and then
2415 * deadlock.
2416 */
2417 sf = sf_buf_alloc(pg, (mnw || m != NULL) ? SFB_NOWAIT :
2418 SFB_CATCH);
2419 if (sf == NULL) {
2420 SFSTAT_INC(sf_allocfail);
2421 vm_page_lock(pg);
2422 vm_page_unwire(pg, 0);
2423 KASSERT(pg->object != NULL,
2424 ("%s: object disappeared", __func__));
2425 vm_page_unlock(pg);
2426 if (m == NULL)
2427 error = (mnw ? EAGAIN : EINTR);
2428 break;
2429 }
2430
2431 /*
2432 * Get an mbuf and set it up as having
2433 * external storage.
2434 */
2435 m0 = m_get((mnw ? M_NOWAIT : M_WAITOK), MT_DATA);
2436 if (m0 == NULL) {
2437 error = (mnw ? EAGAIN : ENOBUFS);
2438 (void)sf_buf_mext(NULL, NULL, sf);
2439 break;
2440 }
2441 if (m_extadd(m0, (caddr_t )sf_buf_kva(sf), PAGE_SIZE,
2442 sf_buf_mext, sfs, sf, M_RDONLY, EXT_SFBUF,
2443 (mnw ? M_NOWAIT : M_WAITOK)) != 0) {
2444 error = (mnw ? EAGAIN : ENOBUFS);
2445 (void)sf_buf_mext(NULL, NULL, sf);
2446 m_freem(m0);
2447 break;
2448 }
2449 m0->m_data = (char *)sf_buf_kva(sf) + pgoff;
2450 m0->m_len = xfsize;
2451
2452 /* Append to mbuf chain. */
2453 if (mtail != NULL)
2454 mtail->m_next = m0;
2455 else if (m != NULL)
2456 m_last(m)->m_next = m0;
2457 else
2458 m = m0;
2459 mtail = m0;
2460
2461 /* Keep track of bits processed. */
2462 loopbytes += xfsize;
2463 off += xfsize;
2464
2465 if (sfs != NULL) {
2466 mtx_lock(&sfs->mtx);
2467 sfs->count++;
2468 mtx_unlock(&sfs->mtx);
2469 }
2470 }
2471
2472 if (vp != NULL)
2473 VOP_UNLOCK(vp, 0);
2474
2475 /* Add the buffer chain to the socket buffer. */
2476 if (m != NULL) {
2477 int mlen, err;
2478
2479 mlen = m_length(m, NULL);
2480 SOCKBUF_LOCK(&so->so_snd);
2481 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2482 error = EPIPE;
2483 SOCKBUF_UNLOCK(&so->so_snd);
2484 goto done;
2485 }
2486 SOCKBUF_UNLOCK(&so->so_snd);
2487 CURVNET_SET(so->so_vnet);
2488 /* Avoid error aliasing. */
2489 err = (*so->so_proto->pr_usrreqs->pru_send)
2490 (so, 0, m, NULL, NULL, td);
2491 CURVNET_RESTORE();
2492 if (err == 0) {
2493 /*
2494 * We need two counters to get the
2495 * file offset and nbytes to send
2496 * right:
2497 * - sbytes contains the total amount
2498 * of bytes sent, including headers.
2499 * - fsbytes contains the total amount
2500 * of bytes sent from the file.
2501 */
2502 sbytes += mlen;
2503 fsbytes += mlen;
2504 if (hdrlen) {
2505 fsbytes -= hdrlen;
2506 hdrlen = 0;
2507 }
2508 } else if (error == 0)
2509 error = err;
2510 m = NULL; /* pru_send always consumes */
2511 }
2512
2513 /* Quit outer loop on error or when we're done. */
2514 if (done)
2515 break;
2516 if (error != 0)
2517 goto done;
2518 }
2519
2520 /*
2521 * Send trailers. Wimp out and use writev(2).
2522 */
2523 if (trl_uio != NULL) {
2524 sbunlock(&so->so_snd);
2525 error = kern_writev(td, sockfd, trl_uio);
2526 if (error == 0)
2527 sbytes += td->td_retval[0];
2528 goto out;
2529 }
2530
2531 done:
2532 sbunlock(&so->so_snd);
2533 out:
2534 /*
2535 * If there was no error we have to clear td->td_retval[0]
2536 * because it may have been set by writev.
2537 */
2538 if (error == 0) {
2539 td->td_retval[0] = 0;
2540 }
2541 if (sent != NULL) {
2542 copyout(&sbytes, sent, sizeof(off_t));
2543 }
2544 if (obj != NULL)
2545 vm_object_deallocate(obj);
2546 if (so)
2547 fdrop(sock_fp, td);
2548 if (m)
2549 m_freem(m);
2550
2551 if (sfs != NULL) {
2552 mtx_lock(&sfs->mtx);
2553 if (sfs->count != 0)
2554 cv_wait(&sfs->cv, &sfs->mtx);
2555 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
2556 cv_destroy(&sfs->cv);
2557 mtx_destroy(&sfs->mtx);
2558 free(sfs, M_TEMP);
2559 }
2560
2561 if (error == ERESTART)
2562 error = EINTR;
2563
2564 return (error);
2565 }
Cache object: d9be4be3c222c4c9f8f17759b297e665
|