37
38#include "opt_inet.h"
39#include "opt_inet6.h"
40#include "opt_sctp.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/kernel.h>
47#include <sys/lock.h>
48#include <sys/mutex.h>
49#include <sys/sysproto.h>
50#include <sys/malloc.h>
51#include <sys/filedesc.h>
52#include <sys/event.h>
53#include <sys/proc.h>
54#include <sys/fcntl.h>
55#include <sys/file.h>
56#include <sys/filio.h>
57#include <sys/jail.h>
58#include <sys/mount.h>
59#include <sys/mbuf.h>
60#include <sys/protosw.h>
61#include <sys/sf_buf.h>
62#include <sys/sysent.h>
63#include <sys/socket.h>
64#include <sys/socketvar.h>
65#include <sys/signalvar.h>
66#include <sys/syscallsubr.h>
67#include <sys/sysctl.h>
68#include <sys/uio.h>
69#include <sys/vnode.h>
70#ifdef KTRACE
71#include <sys/ktrace.h>
72#endif
73#ifdef COMPAT_FREEBSD32
74#include <compat/freebsd32/freebsd32_util.h>
75#endif
76
77#include <net/vnet.h>
78
79#include <security/audit/audit.h>
80#include <security/mac/mac_framework.h>
81
82#include <vm/vm.h>
83#include <vm/vm_object.h>
84#include <vm/vm_page.h>
85#include <vm/vm_pageout.h>
86#include <vm/vm_kern.h>
87#include <vm/vm_extern.h>
88
89#if defined(INET) || defined(INET6)
90#ifdef SCTP
91#include <netinet/sctp.h>
92#include <netinet/sctp_peeloff.h>
93#endif /* SCTP */
94#endif /* INET || INET6 */
95
96static int sendit(struct thread *td, int s, struct msghdr *mp, int flags);
97static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp);
98
99static int accept1(struct thread *td, struct accept_args *uap, int compat);
100static int do_sendfile(struct thread *td, struct sendfile_args *uap, int compat);
101static int getsockname1(struct thread *td, struct getsockname_args *uap,
102 int compat);
103static int getpeername1(struct thread *td, struct getpeername_args *uap,
104 int compat);
105
106/*
107 * NSFBUFS-related variables and associated sysctls
108 */
109int nsfbufs;
110int nsfbufspeak;
111int nsfbufsused;
112
113SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
114 "Maximum number of sendfile(2) sf_bufs available");
115SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
116 "Number of sendfile(2) sf_bufs at peak usage");
117SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
118 "Number of sendfile(2) sf_bufs in use");
119
120/*
121 * Convert a user file descriptor to a kernel file entry. A reference on the
122 * file entry is held upon returning. This is lighter weight than
123 * fgetsock(), which bumps the socket reference drops the file reference
124 * count instead, as this approach avoids several additional mutex operations
125 * associated with the additional reference count. If requested, return the
126 * open file flags.
127 */
128static int
129getsock(struct filedesc *fdp, int fd, struct file **fpp, u_int *fflagp)
130{
131 struct file *fp;
132 int error;
133
134 fp = NULL;
135 if (fdp == NULL || (fp = fget_unlocked(fdp, fd)) == NULL) {
136 error = EBADF;
137 } else if (fp->f_type != DTYPE_SOCKET) {
138 fdrop(fp, curthread);
139 fp = NULL;
140 error = ENOTSOCK;
141 } else {
142 if (fflagp != NULL)
143 *fflagp = fp->f_flag;
144 error = 0;
145 }
146 *fpp = fp;
147 return (error);
148}
149
150/*
151 * System call interface to the socket abstraction.
152 */
153#if defined(COMPAT_43)
154#define COMPAT_OLDSOCK
155#endif
156
157int
158socket(td, uap)
159 struct thread *td;
160 struct socket_args /* {
161 int domain;
162 int type;
163 int protocol;
164 } */ *uap;
165{
166 struct filedesc *fdp;
167 struct socket *so;
168 struct file *fp;
169 int fd, error;
170
171 AUDIT_ARG_SOCKET(uap->domain, uap->type, uap->protocol);
172#ifdef MAC
173 error = mac_socket_check_create(td->td_ucred, uap->domain, uap->type,
174 uap->protocol);
175 if (error)
176 return (error);
177#endif
178 fdp = td->td_proc->p_fd;
179 error = falloc(td, &fp, &fd);
180 if (error)
181 return (error);
182 /* An extra reference on `fp' has been held for us by falloc(). */
183 error = socreate(uap->domain, &so, uap->type, uap->protocol,
184 td->td_ucred, td);
185 if (error) {
186 fdclose(fdp, fp, fd, td);
187 } else {
188 finit(fp, FREAD | FWRITE, DTYPE_SOCKET, so, &socketops);
189 td->td_retval[0] = fd;
190 }
191 fdrop(fp, td);
192 return (error);
193}
194
195/* ARGSUSED */
196int
197bind(td, uap)
198 struct thread *td;
199 struct bind_args /* {
200 int s;
201 caddr_t name;
202 int namelen;
203 } */ *uap;
204{
205 struct sockaddr *sa;
206 int error;
207
208 if ((error = getsockaddr(&sa, uap->name, uap->namelen)) != 0)
209 return (error);
210
211 error = kern_bind(td, uap->s, sa);
212 free(sa, M_SONAME);
213 return (error);
214}
215
216int
217kern_bind(td, fd, sa)
218 struct thread *td;
219 int fd;
220 struct sockaddr *sa;
221{
222 struct socket *so;
223 struct file *fp;
224 int error;
225
226 AUDIT_ARG_FD(fd);
227 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
228 if (error)
229 return (error);
230 so = fp->f_data;
231#ifdef KTRACE
232 if (KTRPOINT(td, KTR_STRUCT))
233 ktrsockaddr(sa);
234#endif
235#ifdef MAC
236 error = mac_socket_check_bind(td->td_ucred, so, sa);
237 if (error == 0)
238#endif
239 error = sobind(so, sa, td);
240 fdrop(fp, td);
241 return (error);
242}
243
244/* ARGSUSED */
245int
246listen(td, uap)
247 struct thread *td;
248 struct listen_args /* {
249 int s;
250 int backlog;
251 } */ *uap;
252{
253 struct socket *so;
254 struct file *fp;
255 int error;
256
257 AUDIT_ARG_FD(uap->s);
258 error = getsock(td->td_proc->p_fd, uap->s, &fp, NULL);
259 if (error == 0) {
260 so = fp->f_data;
261#ifdef MAC
262 error = mac_socket_check_listen(td->td_ucred, so);
263 if (error == 0) {
264#endif
265 CURVNET_SET(so->so_vnet);
266 error = solisten(so, uap->backlog, td);
267 CURVNET_RESTORE();
268#ifdef MAC
269 }
270#endif
271 fdrop(fp, td);
272 }
273 return(error);
274}
275
276/*
277 * accept1()
278 */
279static int
280accept1(td, uap, compat)
281 struct thread *td;
282 struct accept_args /* {
283 int s;
284 struct sockaddr * __restrict name;
285 socklen_t * __restrict anamelen;
286 } */ *uap;
287 int compat;
288{
289 struct sockaddr *name;
290 socklen_t namelen;
291 struct file *fp;
292 int error;
293
294 if (uap->name == NULL)
295 return (kern_accept(td, uap->s, NULL, NULL, NULL));
296
297 error = copyin(uap->anamelen, &namelen, sizeof (namelen));
298 if (error)
299 return (error);
300
301 error = kern_accept(td, uap->s, &name, &namelen, &fp);
302
303 /*
304 * return a namelen of zero for older code which might
305 * ignore the return value from accept.
306 */
307 if (error) {
308 (void) copyout(&namelen,
309 uap->anamelen, sizeof(*uap->anamelen));
310 return (error);
311 }
312
313 if (error == 0 && name != NULL) {
314#ifdef COMPAT_OLDSOCK
315 if (compat)
316 ((struct osockaddr *)name)->sa_family =
317 name->sa_family;
318#endif
319 error = copyout(name, uap->name, namelen);
320 }
321 if (error == 0)
322 error = copyout(&namelen, uap->anamelen,
323 sizeof(namelen));
324 if (error)
325 fdclose(td->td_proc->p_fd, fp, td->td_retval[0], td);
326 fdrop(fp, td);
327 free(name, M_SONAME);
328 return (error);
329}
330
331int
332kern_accept(struct thread *td, int s, struct sockaddr **name,
333 socklen_t *namelen, struct file **fp)
334{
335 struct filedesc *fdp;
336 struct file *headfp, *nfp = NULL;
337 struct sockaddr *sa = NULL;
338 int error;
339 struct socket *head, *so;
340 int fd;
341 u_int fflag;
342 pid_t pgid;
343 int tmp;
344
345 if (name) {
346 *name = NULL;
347 if (*namelen < 0)
348 return (EINVAL);
349 }
350
351 AUDIT_ARG_FD(s);
352 fdp = td->td_proc->p_fd;
353 error = getsock(fdp, s, &headfp, &fflag);
354 if (error)
355 return (error);
356 head = headfp->f_data;
357 if ((head->so_options & SO_ACCEPTCONN) == 0) {
358 error = EINVAL;
359 goto done;
360 }
361#ifdef MAC
362 error = mac_socket_check_accept(td->td_ucred, head);
363 if (error != 0)
364 goto done;
365#endif
366 error = falloc(td, &nfp, &fd);
367 if (error)
368 goto done;
369 ACCEPT_LOCK();
370 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) {
371 ACCEPT_UNLOCK();
372 error = EWOULDBLOCK;
373 goto noconnection;
374 }
375 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
376 if (head->so_rcv.sb_state & SBS_CANTRCVMORE) {
377 head->so_error = ECONNABORTED;
378 break;
379 }
380 error = msleep(&head->so_timeo, &accept_mtx, PSOCK | PCATCH,
381 "accept", 0);
382 if (error) {
383 ACCEPT_UNLOCK();
384 goto noconnection;
385 }
386 }
387 if (head->so_error) {
388 error = head->so_error;
389 head->so_error = 0;
390 ACCEPT_UNLOCK();
391 goto noconnection;
392 }
393 so = TAILQ_FIRST(&head->so_comp);
394 KASSERT(!(so->so_qstate & SQ_INCOMP), ("accept1: so SQ_INCOMP"));
395 KASSERT(so->so_qstate & SQ_COMP, ("accept1: so not SQ_COMP"));
396
397 /*
398 * Before changing the flags on the socket, we have to bump the
399 * reference count. Otherwise, if the protocol calls sofree(),
400 * the socket will be released due to a zero refcount.
401 */
402 SOCK_LOCK(so); /* soref() and so_state update */
403 soref(so); /* file descriptor reference */
404
405 TAILQ_REMOVE(&head->so_comp, so, so_list);
406 head->so_qlen--;
407 so->so_state |= (head->so_state & SS_NBIO);
408 so->so_qstate &= ~SQ_COMP;
409 so->so_head = NULL;
410
411 SOCK_UNLOCK(so);
412 ACCEPT_UNLOCK();
413
414 /* An extra reference on `nfp' has been held for us by falloc(). */
415 td->td_retval[0] = fd;
416
417 /* connection has been removed from the listen queue */
418 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
419
420 pgid = fgetown(&head->so_sigio);
421 if (pgid != 0)
422 fsetown(pgid, &so->so_sigio);
423
424 finit(nfp, fflag, DTYPE_SOCKET, so, &socketops);
425 /* Sync socket nonblocking/async state with file flags */
426 tmp = fflag & FNONBLOCK;
427 (void) fo_ioctl(nfp, FIONBIO, &tmp, td->td_ucred, td);
428 tmp = fflag & FASYNC;
429 (void) fo_ioctl(nfp, FIOASYNC, &tmp, td->td_ucred, td);
430 sa = 0;
431 CURVNET_SET(so->so_vnet);
432 error = soaccept(so, &sa);
433 CURVNET_RESTORE();
434 if (error) {
435 /*
436 * return a namelen of zero for older code which might
437 * ignore the return value from accept.
438 */
439 if (name)
440 *namelen = 0;
441 goto noconnection;
442 }
443 if (sa == NULL) {
444 if (name)
445 *namelen = 0;
446 goto done;
447 }
448 if (name) {
449 /* check sa_len before it is destroyed */
450 if (*namelen > sa->sa_len)
451 *namelen = sa->sa_len;
452#ifdef KTRACE
453 if (KTRPOINT(td, KTR_STRUCT))
454 ktrsockaddr(sa);
455#endif
456 *name = sa;
457 sa = NULL;
458 }
459noconnection:
460 if (sa)
461 free(sa, M_SONAME);
462
463 /*
464 * close the new descriptor, assuming someone hasn't ripped it
465 * out from under us.
466 */
467 if (error)
468 fdclose(fdp, nfp, fd, td);
469
470 /*
471 * Release explicitly held references before returning. We return
472 * a reference on nfp to the caller on success if they request it.
473 */
474done:
475 if (fp != NULL) {
476 if (error == 0) {
477 *fp = nfp;
478 nfp = NULL;
479 } else
480 *fp = NULL;
481 }
482 if (nfp != NULL)
483 fdrop(nfp, td);
484 fdrop(headfp, td);
485 return (error);
486}
487
488int
489accept(td, uap)
490 struct thread *td;
491 struct accept_args *uap;
492{
493
494 return (accept1(td, uap, 0));
495}
496
497#ifdef COMPAT_OLDSOCK
498int
499oaccept(td, uap)
500 struct thread *td;
501 struct accept_args *uap;
502{
503
504 return (accept1(td, uap, 1));
505}
506#endif /* COMPAT_OLDSOCK */
507
508/* ARGSUSED */
509int
510connect(td, uap)
511 struct thread *td;
512 struct connect_args /* {
513 int s;
514 caddr_t name;
515 int namelen;
516 } */ *uap;
517{
518 struct sockaddr *sa;
519 int error;
520
521 error = getsockaddr(&sa, uap->name, uap->namelen);
522 if (error)
523 return (error);
524
525 error = kern_connect(td, uap->s, sa);
526 free(sa, M_SONAME);
527 return (error);
528}
529
530
531int
532kern_connect(td, fd, sa)
533 struct thread *td;
534 int fd;
535 struct sockaddr *sa;
536{
537 struct socket *so;
538 struct file *fp;
539 int error;
540 int interrupted = 0;
541
542 AUDIT_ARG_FD(fd);
543 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
544 if (error)
545 return (error);
546 so = fp->f_data;
547 if (so->so_state & SS_ISCONNECTING) {
548 error = EALREADY;
549 goto done1;
550 }
551#ifdef KTRACE
552 if (KTRPOINT(td, KTR_STRUCT))
553 ktrsockaddr(sa);
554#endif
555#ifdef MAC
556 error = mac_socket_check_connect(td->td_ucred, so, sa);
557 if (error)
558 goto bad;
559#endif
560 error = soconnect(so, sa, td);
561 if (error)
562 goto bad;
563 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
564 error = EINPROGRESS;
565 goto done1;
566 }
567 SOCK_LOCK(so);
568 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
569 error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH,
570 "connec", 0);
571 if (error) {
572 if (error == EINTR || error == ERESTART)
573 interrupted = 1;
574 break;
575 }
576 }
577 if (error == 0) {
578 error = so->so_error;
579 so->so_error = 0;
580 }
581 SOCK_UNLOCK(so);
582bad:
583 if (!interrupted)
584 so->so_state &= ~SS_ISCONNECTING;
585 if (error == ERESTART)
586 error = EINTR;
587done1:
588 fdrop(fp, td);
589 return (error);
590}
591
592int
593kern_socketpair(struct thread *td, int domain, int type, int protocol,
594 int *rsv)
595{
596 struct filedesc *fdp = td->td_proc->p_fd;
597 struct file *fp1, *fp2;
598 struct socket *so1, *so2;
599 int fd, error;
600
601 AUDIT_ARG_SOCKET(domain, type, protocol);
602#ifdef MAC
603 /* We might want to have a separate check for socket pairs. */
604 error = mac_socket_check_create(td->td_ucred, domain, type,
605 protocol);
606 if (error)
607 return (error);
608#endif
609 error = socreate(domain, &so1, type, protocol, td->td_ucred, td);
610 if (error)
611 return (error);
612 error = socreate(domain, &so2, type, protocol, td->td_ucred, td);
613 if (error)
614 goto free1;
615 /* On success extra reference to `fp1' and 'fp2' is set by falloc. */
616 error = falloc(td, &fp1, &fd);
617 if (error)
618 goto free2;
619 rsv[0] = fd;
620 fp1->f_data = so1; /* so1 already has ref count */
621 error = falloc(td, &fp2, &fd);
622 if (error)
623 goto free3;
624 fp2->f_data = so2; /* so2 already has ref count */
625 rsv[1] = fd;
626 error = soconnect2(so1, so2);
627 if (error)
628 goto free4;
629 if (type == SOCK_DGRAM) {
630 /*
631 * Datagram socket connection is asymmetric.
632 */
633 error = soconnect2(so2, so1);
634 if (error)
635 goto free4;
636 }
637 finit(fp1, FREAD | FWRITE, DTYPE_SOCKET, fp1->f_data, &socketops);
638 finit(fp2, FREAD | FWRITE, DTYPE_SOCKET, fp2->f_data, &socketops);
639 fdrop(fp1, td);
640 fdrop(fp2, td);
641 return (0);
642free4:
643 fdclose(fdp, fp2, rsv[1], td);
644 fdrop(fp2, td);
645free3:
646 fdclose(fdp, fp1, rsv[0], td);
647 fdrop(fp1, td);
648free2:
649 if (so2 != NULL)
650 (void)soclose(so2);
651free1:
652 if (so1 != NULL)
653 (void)soclose(so1);
654 return (error);
655}
656
657int
658socketpair(struct thread *td, struct socketpair_args *uap)
659{
660 int error, sv[2];
661
662 error = kern_socketpair(td, uap->domain, uap->type,
663 uap->protocol, sv);
664 if (error)
665 return (error);
666 error = copyout(sv, uap->rsv, 2 * sizeof(int));
667 if (error) {
668 (void)kern_close(td, sv[0]);
669 (void)kern_close(td, sv[1]);
670 }
671 return (error);
672}
673
674static int
675sendit(td, s, mp, flags)
676 struct thread *td;
677 int s;
678 struct msghdr *mp;
679 int flags;
680{
681 struct mbuf *control;
682 struct sockaddr *to;
683 int error;
684
685 if (mp->msg_name != NULL) {
686 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
687 if (error) {
688 to = NULL;
689 goto bad;
690 }
691 mp->msg_name = to;
692 } else {
693 to = NULL;
694 }
695
696 if (mp->msg_control) {
697 if (mp->msg_controllen < sizeof(struct cmsghdr)
698#ifdef COMPAT_OLDSOCK
699 && mp->msg_flags != MSG_COMPAT
700#endif
701 ) {
702 error = EINVAL;
703 goto bad;
704 }
705 error = sockargs(&control, mp->msg_control,
706 mp->msg_controllen, MT_CONTROL);
707 if (error)
708 goto bad;
709#ifdef COMPAT_OLDSOCK
710 if (mp->msg_flags == MSG_COMPAT) {
711 struct cmsghdr *cm;
712
713 M_PREPEND(control, sizeof(*cm), M_WAIT);
714 cm = mtod(control, struct cmsghdr *);
715 cm->cmsg_len = control->m_len;
716 cm->cmsg_level = SOL_SOCKET;
717 cm->cmsg_type = SCM_RIGHTS;
718 }
719#endif
720 } else {
721 control = NULL;
722 }
723
724 error = kern_sendit(td, s, mp, flags, control, UIO_USERSPACE);
725
726bad:
727 if (to)
728 free(to, M_SONAME);
729 return (error);
730}
731
732int
733kern_sendit(td, s, mp, flags, control, segflg)
734 struct thread *td;
735 int s;
736 struct msghdr *mp;
737 int flags;
738 struct mbuf *control;
739 enum uio_seg segflg;
740{
741 struct file *fp;
742 struct uio auio;
743 struct iovec *iov;
744 struct socket *so;
745 int i;
746 int len, error;
747#ifdef KTRACE
748 struct uio *ktruio = NULL;
749#endif
750
751 AUDIT_ARG_FD(s);
752 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
753 if (error)
754 return (error);
755 so = (struct socket *)fp->f_data;
756
757#ifdef MAC
758 if (mp->msg_name != NULL) {
759 error = mac_socket_check_connect(td->td_ucred, so,
760 mp->msg_name);
761 if (error)
762 goto bad;
763 }
764 error = mac_socket_check_send(td->td_ucred, so);
765 if (error)
766 goto bad;
767#endif
768
769 auio.uio_iov = mp->msg_iov;
770 auio.uio_iovcnt = mp->msg_iovlen;
771 auio.uio_segflg = segflg;
772 auio.uio_rw = UIO_WRITE;
773 auio.uio_td = td;
774 auio.uio_offset = 0; /* XXX */
775 auio.uio_resid = 0;
776 iov = mp->msg_iov;
777 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
778 if ((auio.uio_resid += iov->iov_len) < 0) {
779 error = EINVAL;
780 goto bad;
781 }
782 }
783#ifdef KTRACE
784 if (KTRPOINT(td, KTR_GENIO))
785 ktruio = cloneuio(&auio);
786#endif
787 len = auio.uio_resid;
788 error = sosend(so, mp->msg_name, &auio, 0, control, flags, td);
789 if (error) {
790 if (auio.uio_resid != len && (error == ERESTART ||
791 error == EINTR || error == EWOULDBLOCK))
792 error = 0;
793 /* Generation of SIGPIPE can be controlled per socket */
794 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
795 !(flags & MSG_NOSIGNAL)) {
796 PROC_LOCK(td->td_proc);
797 psignal(td->td_proc, SIGPIPE);
798 PROC_UNLOCK(td->td_proc);
799 }
800 }
801 if (error == 0)
802 td->td_retval[0] = len - auio.uio_resid;
803#ifdef KTRACE
804 if (ktruio != NULL) {
805 ktruio->uio_resid = td->td_retval[0];
806 ktrgenio(s, UIO_WRITE, ktruio, error);
807 }
808#endif
809bad:
810 fdrop(fp, td);
811 return (error);
812}
813
814int
815sendto(td, uap)
816 struct thread *td;
817 struct sendto_args /* {
818 int s;
819 caddr_t buf;
820 size_t len;
821 int flags;
822 caddr_t to;
823 int tolen;
824 } */ *uap;
825{
826 struct msghdr msg;
827 struct iovec aiov;
828 int error;
829
830 msg.msg_name = uap->to;
831 msg.msg_namelen = uap->tolen;
832 msg.msg_iov = &aiov;
833 msg.msg_iovlen = 1;
834 msg.msg_control = 0;
835#ifdef COMPAT_OLDSOCK
836 msg.msg_flags = 0;
837#endif
838 aiov.iov_base = uap->buf;
839 aiov.iov_len = uap->len;
840 error = sendit(td, uap->s, &msg, uap->flags);
841 return (error);
842}
843
844#ifdef COMPAT_OLDSOCK
845int
846osend(td, uap)
847 struct thread *td;
848 struct osend_args /* {
849 int s;
850 caddr_t buf;
851 int len;
852 int flags;
853 } */ *uap;
854{
855 struct msghdr msg;
856 struct iovec aiov;
857 int error;
858
859 msg.msg_name = 0;
860 msg.msg_namelen = 0;
861 msg.msg_iov = &aiov;
862 msg.msg_iovlen = 1;
863 aiov.iov_base = uap->buf;
864 aiov.iov_len = uap->len;
865 msg.msg_control = 0;
866 msg.msg_flags = 0;
867 error = sendit(td, uap->s, &msg, uap->flags);
868 return (error);
869}
870
871int
872osendmsg(td, uap)
873 struct thread *td;
874 struct osendmsg_args /* {
875 int s;
876 caddr_t msg;
877 int flags;
878 } */ *uap;
879{
880 struct msghdr msg;
881 struct iovec *iov;
882 int error;
883
884 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
885 if (error)
886 return (error);
887 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
888 if (error)
889 return (error);
890 msg.msg_iov = iov;
891 msg.msg_flags = MSG_COMPAT;
892 error = sendit(td, uap->s, &msg, uap->flags);
893 free(iov, M_IOV);
894 return (error);
895}
896#endif
897
898int
899sendmsg(td, uap)
900 struct thread *td;
901 struct sendmsg_args /* {
902 int s;
903 caddr_t msg;
904 int flags;
905 } */ *uap;
906{
907 struct msghdr msg;
908 struct iovec *iov;
909 int error;
910
911 error = copyin(uap->msg, &msg, sizeof (msg));
912 if (error)
913 return (error);
914 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
915 if (error)
916 return (error);
917 msg.msg_iov = iov;
918#ifdef COMPAT_OLDSOCK
919 msg.msg_flags = 0;
920#endif
921 error = sendit(td, uap->s, &msg, uap->flags);
922 free(iov, M_IOV);
923 return (error);
924}
925
926int
927kern_recvit(td, s, mp, fromseg, controlp)
928 struct thread *td;
929 int s;
930 struct msghdr *mp;
931 enum uio_seg fromseg;
932 struct mbuf **controlp;
933{
934 struct uio auio;
935 struct iovec *iov;
936 int i;
937 socklen_t len;
938 int error;
939 struct mbuf *m, *control = 0;
940 caddr_t ctlbuf;
941 struct file *fp;
942 struct socket *so;
943 struct sockaddr *fromsa = 0;
944#ifdef KTRACE
945 struct uio *ktruio = NULL;
946#endif
947
948 if (controlp != NULL)
949 *controlp = NULL;
950
951 AUDIT_ARG_FD(s);
952 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
953 if (error)
954 return (error);
955 so = fp->f_data;
956
957#ifdef MAC
958 error = mac_socket_check_receive(td->td_ucred, so);
959 if (error) {
960 fdrop(fp, td);
961 return (error);
962 }
963#endif
964
965 auio.uio_iov = mp->msg_iov;
966 auio.uio_iovcnt = mp->msg_iovlen;
967 auio.uio_segflg = UIO_USERSPACE;
968 auio.uio_rw = UIO_READ;
969 auio.uio_td = td;
970 auio.uio_offset = 0; /* XXX */
971 auio.uio_resid = 0;
972 iov = mp->msg_iov;
973 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
974 if ((auio.uio_resid += iov->iov_len) < 0) {
975 fdrop(fp, td);
976 return (EINVAL);
977 }
978 }
979#ifdef KTRACE
980 if (KTRPOINT(td, KTR_GENIO))
981 ktruio = cloneuio(&auio);
982#endif
983 len = auio.uio_resid;
984 CURVNET_SET(so->so_vnet);
985 error = soreceive(so, &fromsa, &auio, (struct mbuf **)0,
986 (mp->msg_control || controlp) ? &control : (struct mbuf **)0,
987 &mp->msg_flags);
988 CURVNET_RESTORE();
989 if (error) {
990 if (auio.uio_resid != (int)len && (error == ERESTART ||
991 error == EINTR || error == EWOULDBLOCK))
992 error = 0;
993 }
994#ifdef KTRACE
995 if (ktruio != NULL) {
996 ktruio->uio_resid = (int)len - auio.uio_resid;
997 ktrgenio(s, UIO_READ, ktruio, error);
998 }
999#endif
1000 if (error)
1001 goto out;
1002 td->td_retval[0] = (int)len - auio.uio_resid;
1003 if (mp->msg_name) {
1004 len = mp->msg_namelen;
1005 if (len <= 0 || fromsa == 0)
1006 len = 0;
1007 else {
1008 /* save sa_len before it is destroyed by MSG_COMPAT */
1009 len = MIN(len, fromsa->sa_len);
1010#ifdef COMPAT_OLDSOCK
1011 if (mp->msg_flags & MSG_COMPAT)
1012 ((struct osockaddr *)fromsa)->sa_family =
1013 fromsa->sa_family;
1014#endif
1015 if (fromseg == UIO_USERSPACE) {
1016 error = copyout(fromsa, mp->msg_name,
1017 (unsigned)len);
1018 if (error)
1019 goto out;
1020 } else
1021 bcopy(fromsa, mp->msg_name, len);
1022 }
1023 mp->msg_namelen = len;
1024 }
1025 if (mp->msg_control && controlp == NULL) {
1026#ifdef COMPAT_OLDSOCK
1027 /*
1028 * We assume that old recvmsg calls won't receive access
1029 * rights and other control info, esp. as control info
1030 * is always optional and those options didn't exist in 4.3.
1031 * If we receive rights, trim the cmsghdr; anything else
1032 * is tossed.
1033 */
1034 if (control && mp->msg_flags & MSG_COMPAT) {
1035 if (mtod(control, struct cmsghdr *)->cmsg_level !=
1036 SOL_SOCKET ||
1037 mtod(control, struct cmsghdr *)->cmsg_type !=
1038 SCM_RIGHTS) {
1039 mp->msg_controllen = 0;
1040 goto out;
1041 }
1042 control->m_len -= sizeof (struct cmsghdr);
1043 control->m_data += sizeof (struct cmsghdr);
1044 }
1045#endif
1046 len = mp->msg_controllen;
1047 m = control;
1048 mp->msg_controllen = 0;
1049 ctlbuf = mp->msg_control;
1050
1051 while (m && len > 0) {
1052 unsigned int tocopy;
1053
1054 if (len >= m->m_len)
1055 tocopy = m->m_len;
1056 else {
1057 mp->msg_flags |= MSG_CTRUNC;
1058 tocopy = len;
1059 }
1060
1061 if ((error = copyout(mtod(m, caddr_t),
1062 ctlbuf, tocopy)) != 0)
1063 goto out;
1064
1065 ctlbuf += tocopy;
1066 len -= tocopy;
1067 m = m->m_next;
1068 }
1069 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
1070 }
1071out:
1072 fdrop(fp, td);
1073#ifdef KTRACE
1074 if (fromsa && KTRPOINT(td, KTR_STRUCT))
1075 ktrsockaddr(fromsa);
1076#endif
1077 if (fromsa)
1078 free(fromsa, M_SONAME);
1079
1080 if (error == 0 && controlp != NULL)
1081 *controlp = control;
1082 else if (control)
1083 m_freem(control);
1084
1085 return (error);
1086}
1087
1088static int
1089recvit(td, s, mp, namelenp)
1090 struct thread *td;
1091 int s;
1092 struct msghdr *mp;
1093 void *namelenp;
1094{
1095 int error;
1096
1097 error = kern_recvit(td, s, mp, UIO_USERSPACE, NULL);
1098 if (error)
1099 return (error);
1100 if (namelenp) {
1101 error = copyout(&mp->msg_namelen, namelenp, sizeof (socklen_t));
1102#ifdef COMPAT_OLDSOCK
1103 if (mp->msg_flags & MSG_COMPAT)
1104 error = 0; /* old recvfrom didn't check */
1105#endif
1106 }
1107 return (error);
1108}
1109
1110int
1111recvfrom(td, uap)
1112 struct thread *td;
1113 struct recvfrom_args /* {
1114 int s;
1115 caddr_t buf;
1116 size_t len;
1117 int flags;
1118 struct sockaddr * __restrict from;
1119 socklen_t * __restrict fromlenaddr;
1120 } */ *uap;
1121{
1122 struct msghdr msg;
1123 struct iovec aiov;
1124 int error;
1125
1126 if (uap->fromlenaddr) {
1127 error = copyin(uap->fromlenaddr,
1128 &msg.msg_namelen, sizeof (msg.msg_namelen));
1129 if (error)
1130 goto done2;
1131 } else {
1132 msg.msg_namelen = 0;
1133 }
1134 msg.msg_name = uap->from;
1135 msg.msg_iov = &aiov;
1136 msg.msg_iovlen = 1;
1137 aiov.iov_base = uap->buf;
1138 aiov.iov_len = uap->len;
1139 msg.msg_control = 0;
1140 msg.msg_flags = uap->flags;
1141 error = recvit(td, uap->s, &msg, uap->fromlenaddr);
1142done2:
1143 return(error);
1144}
1145
1146#ifdef COMPAT_OLDSOCK
1147int
1148orecvfrom(td, uap)
1149 struct thread *td;
1150 struct recvfrom_args *uap;
1151{
1152
1153 uap->flags |= MSG_COMPAT;
1154 return (recvfrom(td, uap));
1155}
1156#endif
1157
1158#ifdef COMPAT_OLDSOCK
1159int
1160orecv(td, uap)
1161 struct thread *td;
1162 struct orecv_args /* {
1163 int s;
1164 caddr_t buf;
1165 int len;
1166 int flags;
1167 } */ *uap;
1168{
1169 struct msghdr msg;
1170 struct iovec aiov;
1171 int error;
1172
1173 msg.msg_name = 0;
1174 msg.msg_namelen = 0;
1175 msg.msg_iov = &aiov;
1176 msg.msg_iovlen = 1;
1177 aiov.iov_base = uap->buf;
1178 aiov.iov_len = uap->len;
1179 msg.msg_control = 0;
1180 msg.msg_flags = uap->flags;
1181 error = recvit(td, uap->s, &msg, NULL);
1182 return (error);
1183}
1184
1185/*
1186 * Old recvmsg. This code takes advantage of the fact that the old msghdr
1187 * overlays the new one, missing only the flags, and with the (old) access
1188 * rights where the control fields are now.
1189 */
1190int
1191orecvmsg(td, uap)
1192 struct thread *td;
1193 struct orecvmsg_args /* {
1194 int s;
1195 struct omsghdr *msg;
1196 int flags;
1197 } */ *uap;
1198{
1199 struct msghdr msg;
1200 struct iovec *iov;
1201 int error;
1202
1203 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
1204 if (error)
1205 return (error);
1206 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1207 if (error)
1208 return (error);
1209 msg.msg_flags = uap->flags | MSG_COMPAT;
1210 msg.msg_iov = iov;
1211 error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen);
1212 if (msg.msg_controllen && error == 0)
1213 error = copyout(&msg.msg_controllen,
1214 &uap->msg->msg_accrightslen, sizeof (int));
1215 free(iov, M_IOV);
1216 return (error);
1217}
1218#endif
1219
1220int
1221recvmsg(td, uap)
1222 struct thread *td;
1223 struct recvmsg_args /* {
1224 int s;
1225 struct msghdr *msg;
1226 int flags;
1227 } */ *uap;
1228{
1229 struct msghdr msg;
1230 struct iovec *uiov, *iov;
1231 int error;
1232
1233 error = copyin(uap->msg, &msg, sizeof (msg));
1234 if (error)
1235 return (error);
1236 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1237 if (error)
1238 return (error);
1239 msg.msg_flags = uap->flags;
1240#ifdef COMPAT_OLDSOCK
1241 msg.msg_flags &= ~MSG_COMPAT;
1242#endif
1243 uiov = msg.msg_iov;
1244 msg.msg_iov = iov;
1245 error = recvit(td, uap->s, &msg, NULL);
1246 if (error == 0) {
1247 msg.msg_iov = uiov;
1248 error = copyout(&msg, uap->msg, sizeof(msg));
1249 }
1250 free(iov, M_IOV);
1251 return (error);
1252}
1253
1254/* ARGSUSED */
1255int
1256shutdown(td, uap)
1257 struct thread *td;
1258 struct shutdown_args /* {
1259 int s;
1260 int how;
1261 } */ *uap;
1262{
1263 struct socket *so;
1264 struct file *fp;
1265 int error;
1266
1267 AUDIT_ARG_FD(uap->s);
1268 error = getsock(td->td_proc->p_fd, uap->s, &fp, NULL);
1269 if (error == 0) {
1270 so = fp->f_data;
1271 error = soshutdown(so, uap->how);
1272 fdrop(fp, td);
1273 }
1274 return (error);
1275}
1276
1277/* ARGSUSED */
1278int
1279setsockopt(td, uap)
1280 struct thread *td;
1281 struct setsockopt_args /* {
1282 int s;
1283 int level;
1284 int name;
1285 caddr_t val;
1286 int valsize;
1287 } */ *uap;
1288{
1289
1290 return (kern_setsockopt(td, uap->s, uap->level, uap->name,
1291 uap->val, UIO_USERSPACE, uap->valsize));
1292}
1293
1294int
1295kern_setsockopt(td, s, level, name, val, valseg, valsize)
1296 struct thread *td;
1297 int s;
1298 int level;
1299 int name;
1300 void *val;
1301 enum uio_seg valseg;
1302 socklen_t valsize;
1303{
1304 int error;
1305 struct socket *so;
1306 struct file *fp;
1307 struct sockopt sopt;
1308
1309 if (val == NULL && valsize != 0)
1310 return (EFAULT);
1311 if ((int)valsize < 0)
1312 return (EINVAL);
1313
1314 sopt.sopt_dir = SOPT_SET;
1315 sopt.sopt_level = level;
1316 sopt.sopt_name = name;
1317 sopt.sopt_val = val;
1318 sopt.sopt_valsize = valsize;
1319 switch (valseg) {
1320 case UIO_USERSPACE:
1321 sopt.sopt_td = td;
1322 break;
1323 case UIO_SYSSPACE:
1324 sopt.sopt_td = NULL;
1325 break;
1326 default:
1327 panic("kern_setsockopt called with bad valseg");
1328 }
1329
1330 AUDIT_ARG_FD(s);
1331 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
1332 if (error == 0) {
1333 so = fp->f_data;
1334 CURVNET_SET(so->so_vnet);
1335 error = sosetopt(so, &sopt);
1336 CURVNET_RESTORE();
1337 fdrop(fp, td);
1338 }
1339 return(error);
1340}
1341
1342/* ARGSUSED */
1343int
1344getsockopt(td, uap)
1345 struct thread *td;
1346 struct getsockopt_args /* {
1347 int s;
1348 int level;
1349 int name;
1350 void * __restrict val;
1351 socklen_t * __restrict avalsize;
1352 } */ *uap;
1353{
1354 socklen_t valsize;
1355 int error;
1356
1357 if (uap->val) {
1358 error = copyin(uap->avalsize, &valsize, sizeof (valsize));
1359 if (error)
1360 return (error);
1361 }
1362
1363 error = kern_getsockopt(td, uap->s, uap->level, uap->name,
1364 uap->val, UIO_USERSPACE, &valsize);
1365
1366 if (error == 0)
1367 error = copyout(&valsize, uap->avalsize, sizeof (valsize));
1368 return (error);
1369}
1370
1371/*
1372 * Kernel version of getsockopt.
1373 * optval can be a userland or userspace. optlen is always a kernel pointer.
1374 */
1375int
1376kern_getsockopt(td, s, level, name, val, valseg, valsize)
1377 struct thread *td;
1378 int s;
1379 int level;
1380 int name;
1381 void *val;
1382 enum uio_seg valseg;
1383 socklen_t *valsize;
1384{
1385 int error;
1386 struct socket *so;
1387 struct file *fp;
1388 struct sockopt sopt;
1389
1390 if (val == NULL)
1391 *valsize = 0;
1392 if ((int)*valsize < 0)
1393 return (EINVAL);
1394
1395 sopt.sopt_dir = SOPT_GET;
1396 sopt.sopt_level = level;
1397 sopt.sopt_name = name;
1398 sopt.sopt_val = val;
1399 sopt.sopt_valsize = (size_t)*valsize; /* checked non-negative above */
1400 switch (valseg) {
1401 case UIO_USERSPACE:
1402 sopt.sopt_td = td;
1403 break;
1404 case UIO_SYSSPACE:
1405 sopt.sopt_td = NULL;
1406 break;
1407 default:
1408 panic("kern_getsockopt called with bad valseg");
1409 }
1410
1411 AUDIT_ARG_FD(s);
1412 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
1413 if (error == 0) {
1414 so = fp->f_data;
1415 CURVNET_SET(so->so_vnet);
1416 error = sogetopt(so, &sopt);
1417 CURVNET_RESTORE();
1418 *valsize = sopt.sopt_valsize;
1419 fdrop(fp, td);
1420 }
1421 return (error);
1422}
1423
1424/*
1425 * getsockname1() - Get socket name.
1426 */
1427/* ARGSUSED */
1428static int
1429getsockname1(td, uap, compat)
1430 struct thread *td;
1431 struct getsockname_args /* {
1432 int fdes;
1433 struct sockaddr * __restrict asa;
1434 socklen_t * __restrict alen;
1435 } */ *uap;
1436 int compat;
1437{
1438 struct sockaddr *sa;
1439 socklen_t len;
1440 int error;
1441
1442 error = copyin(uap->alen, &len, sizeof(len));
1443 if (error)
1444 return (error);
1445
1446 error = kern_getsockname(td, uap->fdes, &sa, &len);
1447 if (error)
1448 return (error);
1449
1450 if (len != 0) {
1451#ifdef COMPAT_OLDSOCK
1452 if (compat)
1453 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1454#endif
1455 error = copyout(sa, uap->asa, (u_int)len);
1456 }
1457 free(sa, M_SONAME);
1458 if (error == 0)
1459 error = copyout(&len, uap->alen, sizeof(len));
1460 return (error);
1461}
1462
1463int
1464kern_getsockname(struct thread *td, int fd, struct sockaddr **sa,
1465 socklen_t *alen)
1466{
1467 struct socket *so;
1468 struct file *fp;
1469 socklen_t len;
1470 int error;
1471
1472 if (*alen < 0)
1473 return (EINVAL);
1474
1475 AUDIT_ARG_FD(fd);
1476 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
1477 if (error)
1478 return (error);
1479 so = fp->f_data;
1480 *sa = NULL;
1481 CURVNET_SET(so->so_vnet);
1482 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, sa);
1483 CURVNET_RESTORE();
1484 if (error)
1485 goto bad;
1486 if (*sa == NULL)
1487 len = 0;
1488 else
1489 len = MIN(*alen, (*sa)->sa_len);
1490 *alen = len;
1491#ifdef KTRACE
1492 if (KTRPOINT(td, KTR_STRUCT))
1493 ktrsockaddr(*sa);
1494#endif
1495bad:
1496 fdrop(fp, td);
1497 if (error && *sa) {
1498 free(*sa, M_SONAME);
1499 *sa = NULL;
1500 }
1501 return (error);
1502}
1503
1504int
1505getsockname(td, uap)
1506 struct thread *td;
1507 struct getsockname_args *uap;
1508{
1509
1510 return (getsockname1(td, uap, 0));
1511}
1512
1513#ifdef COMPAT_OLDSOCK
1514int
1515ogetsockname(td, uap)
1516 struct thread *td;
1517 struct getsockname_args *uap;
1518{
1519
1520 return (getsockname1(td, uap, 1));
1521}
1522#endif /* COMPAT_OLDSOCK */
1523
1524/*
1525 * getpeername1() - Get name of peer for connected socket.
1526 */
1527/* ARGSUSED */
1528static int
1529getpeername1(td, uap, compat)
1530 struct thread *td;
1531 struct getpeername_args /* {
1532 int fdes;
1533 struct sockaddr * __restrict asa;
1534 socklen_t * __restrict alen;
1535 } */ *uap;
1536 int compat;
1537{
1538 struct sockaddr *sa;
1539 socklen_t len;
1540 int error;
1541
1542 error = copyin(uap->alen, &len, sizeof (len));
1543 if (error)
1544 return (error);
1545
1546 error = kern_getpeername(td, uap->fdes, &sa, &len);
1547 if (error)
1548 return (error);
1549
1550 if (len != 0) {
1551#ifdef COMPAT_OLDSOCK
1552 if (compat)
1553 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1554#endif
1555 error = copyout(sa, uap->asa, (u_int)len);
1556 }
1557 free(sa, M_SONAME);
1558 if (error == 0)
1559 error = copyout(&len, uap->alen, sizeof(len));
1560 return (error);
1561}
1562
1563int
1564kern_getpeername(struct thread *td, int fd, struct sockaddr **sa,
1565 socklen_t *alen)
1566{
1567 struct socket *so;
1568 struct file *fp;
1569 socklen_t len;
1570 int error;
1571
1572 if (*alen < 0)
1573 return (EINVAL);
1574
1575 AUDIT_ARG_FD(fd);
1576 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
1577 if (error)
1578 return (error);
1579 so = fp->f_data;
1580 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1581 error = ENOTCONN;
1582 goto done;
1583 }
1584 *sa = NULL;
1585 CURVNET_SET(so->so_vnet);
1586 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, sa);
1587 CURVNET_RESTORE();
1588 if (error)
1589 goto bad;
1590 if (*sa == NULL)
1591 len = 0;
1592 else
1593 len = MIN(*alen, (*sa)->sa_len);
1594 *alen = len;
1595#ifdef KTRACE
1596 if (KTRPOINT(td, KTR_STRUCT))
1597 ktrsockaddr(*sa);
1598#endif
1599bad:
1600 if (error && *sa) {
1601 free(*sa, M_SONAME);
1602 *sa = NULL;
1603 }
1604done:
1605 fdrop(fp, td);
1606 return (error);
1607}
1608
1609int
1610getpeername(td, uap)
1611 struct thread *td;
1612 struct getpeername_args *uap;
1613{
1614
1615 return (getpeername1(td, uap, 0));
1616}
1617
1618#ifdef COMPAT_OLDSOCK
1619int
1620ogetpeername(td, uap)
1621 struct thread *td;
1622 struct ogetpeername_args *uap;
1623{
1624
1625 /* XXX uap should have type `getpeername_args *' to begin with. */
1626 return (getpeername1(td, (struct getpeername_args *)uap, 1));
1627}
1628#endif /* COMPAT_OLDSOCK */
1629
1630int
1631sockargs(mp, buf, buflen, type)
1632 struct mbuf **mp;
1633 caddr_t buf;
1634 int buflen, type;
1635{
1636 struct sockaddr *sa;
1637 struct mbuf *m;
1638 int error;
1639
1640 if ((u_int)buflen > MLEN) {
1641#ifdef COMPAT_OLDSOCK
1642 if (type == MT_SONAME && (u_int)buflen <= 112)
1643 buflen = MLEN; /* unix domain compat. hack */
1644 else
1645#endif
1646 if ((u_int)buflen > MCLBYTES)
1647 return (EINVAL);
1648 }
1649 m = m_get(M_WAIT, type);
1650 if ((u_int)buflen > MLEN)
1651 MCLGET(m, M_WAIT);
1652 m->m_len = buflen;
1653 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1654 if (error)
1655 (void) m_free(m);
1656 else {
1657 *mp = m;
1658 if (type == MT_SONAME) {
1659 sa = mtod(m, struct sockaddr *);
1660
1661#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1662 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1663 sa->sa_family = sa->sa_len;
1664#endif
1665 sa->sa_len = buflen;
1666 }
1667 }
1668 return (error);
1669}
1670
1671int
1672getsockaddr(namp, uaddr, len)
1673 struct sockaddr **namp;
1674 caddr_t uaddr;
1675 size_t len;
1676{
1677 struct sockaddr *sa;
1678 int error;
1679
1680 if (len > SOCK_MAXADDRLEN)
1681 return (ENAMETOOLONG);
1682 if (len < offsetof(struct sockaddr, sa_data[0]))
1683 return (EINVAL);
1684 sa = malloc(len, M_SONAME, M_WAITOK);
1685 error = copyin(uaddr, sa, len);
1686 if (error) {
1687 free(sa, M_SONAME);
1688 } else {
1689#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1690 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1691 sa->sa_family = sa->sa_len;
1692#endif
1693 sa->sa_len = len;
1694 *namp = sa;
1695 }
1696 return (error);
1697}
1698
1699#include <sys/condvar.h>
1700
1701struct sendfile_sync {
1702 struct mtx mtx;
1703 struct cv cv;
1704 unsigned count;
1705};
1706
1707/*
1708 * Detach mapped page and release resources back to the system.
1709 */
1710void
1711sf_buf_mext(void *addr, void *args)
1712{
1713 vm_page_t m;
1714 struct sendfile_sync *sfs;
1715
1716 m = sf_buf_page(args);
1717 sf_buf_free(args);
1718 vm_page_lock(m);
1719 vm_page_unwire(m, 0);
1720 /*
1721 * Check for the object going away on us. This can
1722 * happen since we don't hold a reference to it.
1723 * If so, we're responsible for freeing the page.
1724 */
1725 if (m->wire_count == 0 && m->object == NULL)
1726 vm_page_free(m);
1727 vm_page_unlock(m);
1728 if (addr == NULL)
1729 return;
1730 sfs = addr;
1731 mtx_lock(&sfs->mtx);
1732 KASSERT(sfs->count> 0, ("Sendfile sync botchup count == 0"));
1733 if (--sfs->count == 0)
1734 cv_signal(&sfs->cv);
1735 mtx_unlock(&sfs->mtx);
1736}
1737
1738/*
1739 * sendfile(2)
1740 *
1741 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1742 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1743 *
1744 * Send a file specified by 'fd' and starting at 'offset' to a socket
1745 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1746 * 0. Optionally add a header and/or trailer to the socket output. If
1747 * specified, write the total number of bytes sent into *sbytes.
1748 */
1749int
1750sendfile(struct thread *td, struct sendfile_args *uap)
1751{
1752
1753 return (do_sendfile(td, uap, 0));
1754}
1755
1756static int
1757do_sendfile(struct thread *td, struct sendfile_args *uap, int compat)
1758{
1759 struct sf_hdtr hdtr;
1760 struct uio *hdr_uio, *trl_uio;
1761 int error;
1762
1763 hdr_uio = trl_uio = NULL;
1764
1765 if (uap->hdtr != NULL) {
1766 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1767 if (error)
1768 goto out;
1769 if (hdtr.headers != NULL) {
1770 error = copyinuio(hdtr.headers, hdtr.hdr_cnt, &hdr_uio);
1771 if (error)
1772 goto out;
1773 }
1774 if (hdtr.trailers != NULL) {
1775 error = copyinuio(hdtr.trailers, hdtr.trl_cnt, &trl_uio);
1776 if (error)
1777 goto out;
1778
1779 }
1780 }
1781
1782 error = kern_sendfile(td, uap, hdr_uio, trl_uio, compat);
1783out:
1784 if (hdr_uio)
1785 free(hdr_uio, M_IOV);
1786 if (trl_uio)
1787 free(trl_uio, M_IOV);
1788 return (error);
1789}
1790
1791#ifdef COMPAT_FREEBSD4
1792int
1793freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1794{
1795 struct sendfile_args args;
1796
1797 args.fd = uap->fd;
1798 args.s = uap->s;
1799 args.offset = uap->offset;
1800 args.nbytes = uap->nbytes;
1801 args.hdtr = uap->hdtr;
1802 args.sbytes = uap->sbytes;
1803 args.flags = uap->flags;
1804
1805 return (do_sendfile(td, &args, 1));
1806}
1807#endif /* COMPAT_FREEBSD4 */
1808
1809int
1810kern_sendfile(struct thread *td, struct sendfile_args *uap,
1811 struct uio *hdr_uio, struct uio *trl_uio, int compat)
1812{
1813 struct file *sock_fp;
1814 struct vnode *vp;
1815 struct vm_object *obj = NULL;
1816 struct socket *so = NULL;
1817 struct mbuf *m = NULL;
1818 struct sf_buf *sf;
1819 struct vm_page *pg;
1820 off_t off, xfsize, fsbytes = 0, sbytes = 0, rem = 0;
1821 int error, hdrlen = 0, mnw = 0;
1822 int vfslocked;
1823 struct sendfile_sync *sfs = NULL;
1824
1825 /*
1826 * The file descriptor must be a regular file and have a
1827 * backing VM object.
1828 * File offset must be positive. If it goes beyond EOF
1829 * we send only the header/trailer and no payload data.
1830 */
1831 AUDIT_ARG_FD(uap->fd);
1832 if ((error = fgetvp_read(td, uap->fd, &vp)) != 0)
1833 goto out;
1834 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1835 vn_lock(vp, LK_SHARED | LK_RETRY);
1836 if (vp->v_type == VREG) {
1837 obj = vp->v_object;
1838 if (obj != NULL) {
1839 /*
1840 * Temporarily increase the backing VM
1841 * object's reference count so that a forced
1842 * reclamation of its vnode does not
1843 * immediately destroy it.
1844 */
1845 VM_OBJECT_LOCK(obj);
1846 if ((obj->flags & OBJ_DEAD) == 0) {
1847 vm_object_reference_locked(obj);
1848 VM_OBJECT_UNLOCK(obj);
1849 } else {
1850 VM_OBJECT_UNLOCK(obj);
1851 obj = NULL;
1852 }
1853 }
1854 }
1855 VOP_UNLOCK(vp, 0);
1856 VFS_UNLOCK_GIANT(vfslocked);
1857 if (obj == NULL) {
1858 error = EINVAL;
1859 goto out;
1860 }
1861 if (uap->offset < 0) {
1862 error = EINVAL;
1863 goto out;
1864 }
1865
1866 /*
1867 * The socket must be a stream socket and connected.
1868 * Remember if it a blocking or non-blocking socket.
1869 */
1870 if ((error = getsock(td->td_proc->p_fd, uap->s, &sock_fp,
1871 NULL)) != 0)
1872 goto out;
1873 so = sock_fp->f_data;
1874 if (so->so_type != SOCK_STREAM) {
1875 error = EINVAL;
1876 goto out;
1877 }
1878 if ((so->so_state & SS_ISCONNECTED) == 0) {
1879 error = ENOTCONN;
1880 goto out;
1881 }
1882 /*
1883 * Do not wait on memory allocations but return ENOMEM for
1884 * caller to retry later.
1885 * XXX: Experimental.
1886 */
1887 if (uap->flags & SF_MNOWAIT)
1888 mnw = 1;
1889
1890 if (uap->flags & SF_SYNC) {
1891 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK);
1892 memset(sfs, 0, sizeof *sfs);
1893 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
1894 cv_init(&sfs->cv, "sendfile");
1895 }
1896
1897#ifdef MAC
1898 error = mac_socket_check_send(td->td_ucred, so);
1899 if (error)
1900 goto out;
1901#endif
1902
1903 /* If headers are specified copy them into mbufs. */
1904 if (hdr_uio != NULL) {
1905 hdr_uio->uio_td = td;
1906 hdr_uio->uio_rw = UIO_WRITE;
1907 if (hdr_uio->uio_resid > 0) {
1908 /*
1909 * In FBSD < 5.0 the nbytes to send also included
1910 * the header. If compat is specified subtract the
1911 * header size from nbytes.
1912 */
1913 if (compat) {
1914 if (uap->nbytes > hdr_uio->uio_resid)
1915 uap->nbytes -= hdr_uio->uio_resid;
1916 else
1917 uap->nbytes = 0;
1918 }
1919 m = m_uiotombuf(hdr_uio, (mnw ? M_NOWAIT : M_WAITOK),
1920 0, 0, 0);
1921 if (m == NULL) {
1922 error = mnw ? EAGAIN : ENOBUFS;
1923 goto out;
1924 }
1925 hdrlen = m_length(m, NULL);
1926 }
1927 }
1928
1929 /*
1930 * Protect against multiple writers to the socket.
1931 *
1932 * XXXRW: Historically this has assumed non-interruptibility, so now
1933 * we implement that, but possibly shouldn't.
1934 */
1935 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
1936
1937 /*
1938 * Loop through the pages of the file, starting with the requested
1939 * offset. Get a file page (do I/O if necessary), map the file page
1940 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1941 * it on the socket.
1942 * This is done in two loops. The inner loop turns as many pages
1943 * as it can, up to available socket buffer space, without blocking
1944 * into mbufs to have it bulk delivered into the socket send buffer.
1945 * The outer loop checks the state and available space of the socket
1946 * and takes care of the overall progress.
1947 */
1948 for (off = uap->offset, rem = uap->nbytes; ; ) {
1949 int loopbytes = 0;
1950 int space = 0;
1951 int done = 0;
1952
1953 /*
1954 * Check the socket state for ongoing connection,
1955 * no errors and space in socket buffer.
1956 * If space is low allow for the remainder of the
1957 * file to be processed if it fits the socket buffer.
1958 * Otherwise block in waiting for sufficient space
1959 * to proceed, or if the socket is nonblocking, return
1960 * to userland with EAGAIN while reporting how far
1961 * we've come.
1962 * We wait until the socket buffer has significant free
1963 * space to do bulk sends. This makes good use of file
1964 * system read ahead and allows packet segmentation
1965 * offloading hardware to take over lots of work. If
1966 * we were not careful here we would send off only one
1967 * sfbuf at a time.
1968 */
1969 SOCKBUF_LOCK(&so->so_snd);
1970 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
1971 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
1972retry_space:
1973 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1974 error = EPIPE;
1975 SOCKBUF_UNLOCK(&so->so_snd);
1976 goto done;
1977 } else if (so->so_error) {
1978 error = so->so_error;
1979 so->so_error = 0;
1980 SOCKBUF_UNLOCK(&so->so_snd);
1981 goto done;
1982 }
1983 space = sbspace(&so->so_snd);
1984 if (space < rem &&
1985 (space <= 0 ||
1986 space < so->so_snd.sb_lowat)) {
1987 if (so->so_state & SS_NBIO) {
1988 SOCKBUF_UNLOCK(&so->so_snd);
1989 error = EAGAIN;
1990 goto done;
1991 }
1992 /*
1993 * sbwait drops the lock while sleeping.
1994 * When we loop back to retry_space the
1995 * state may have changed and we retest
1996 * for it.
1997 */
1998 error = sbwait(&so->so_snd);
1999 /*
2000 * An error from sbwait usually indicates that we've
2001 * been interrupted by a signal. If we've sent anything
2002 * then return bytes sent, otherwise return the error.
2003 */
2004 if (error) {
2005 SOCKBUF_UNLOCK(&so->so_snd);
2006 goto done;
2007 }
2008 goto retry_space;
2009 }
2010 SOCKBUF_UNLOCK(&so->so_snd);
2011
2012 /*
2013 * Reduce space in the socket buffer by the size of
2014 * the header mbuf chain.
2015 * hdrlen is set to 0 after the first loop.
2016 */
2017 space -= hdrlen;
2018
2019 /*
2020 * Loop and construct maximum sized mbuf chain to be bulk
2021 * dumped into socket buffer.
2022 */
2023 while (space > loopbytes) {
2024 vm_pindex_t pindex;
2025 vm_offset_t pgoff;
2026 struct mbuf *m0;
2027
2028 VM_OBJECT_LOCK(obj);
2029 /*
2030 * Calculate the amount to transfer.
2031 * Not to exceed a page, the EOF,
2032 * or the passed in nbytes.
2033 */
2034 pgoff = (vm_offset_t)(off & PAGE_MASK);
2035 xfsize = omin(PAGE_SIZE - pgoff,
2036 obj->un_pager.vnp.vnp_size - uap->offset -
2037 fsbytes - loopbytes);
2038 if (uap->nbytes)
2039 rem = (uap->nbytes - fsbytes - loopbytes);
2040 else
2041 rem = obj->un_pager.vnp.vnp_size -
2042 uap->offset - fsbytes - loopbytes;
2043 xfsize = omin(rem, xfsize);
2044 xfsize = omin(space - loopbytes, xfsize);
2045 if (xfsize <= 0) {
2046 VM_OBJECT_UNLOCK(obj);
2047 done = 1; /* all data sent */
2048 break;
2049 }
2050
2051 /*
2052 * Attempt to look up the page. Allocate
2053 * if not found or wait and loop if busy.
2054 */
2055 pindex = OFF_TO_IDX(off);
2056 pg = vm_page_grab(obj, pindex, VM_ALLOC_NOBUSY |
2057 VM_ALLOC_NORMAL | VM_ALLOC_WIRED | VM_ALLOC_RETRY);
2058
2059 /*
2060 * Check if page is valid for what we need,
2061 * otherwise initiate I/O.
2062 * If we already turned some pages into mbufs,
2063 * send them off before we come here again and
2064 * block.
2065 */
2066 if (pg->valid && vm_page_is_valid(pg, pgoff, xfsize))
2067 VM_OBJECT_UNLOCK(obj);
2068 else if (m != NULL)
2069 error = EAGAIN; /* send what we already got */
2070 else if (uap->flags & SF_NODISKIO)
2071 error = EBUSY;
2072 else {
2073 int bsize, resid;
2074
2075 /*
2076 * Ensure that our page is still around
2077 * when the I/O completes.
2078 */
2079 vm_page_io_start(pg);
2080 VM_OBJECT_UNLOCK(obj);
2081
2082 /*
2083 * Get the page from backing store.
2084 */
2085 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2086 error = vn_lock(vp, LK_SHARED);
2087 if (error != 0)
2088 goto after_read;
2089 bsize = vp->v_mount->mnt_stat.f_iosize;
2090
2091 /*
2092 * XXXMAC: Because we don't have fp->f_cred
2093 * here, we pass in NOCRED. This is probably
2094 * wrong, but is consistent with our original
2095 * implementation.
2096 */
2097 error = vn_rdwr(UIO_READ, vp, NULL, MAXBSIZE,
2098 trunc_page(off), UIO_NOCOPY, IO_NODELOCKED |
2099 IO_VMIO | ((MAXBSIZE / bsize) << IO_SEQSHIFT),
2100 td->td_ucred, NOCRED, &resid, td);
2101 VOP_UNLOCK(vp, 0);
2102 after_read:
2103 VFS_UNLOCK_GIANT(vfslocked);
2104 VM_OBJECT_LOCK(obj);
2105 vm_page_io_finish(pg);
2106 if (!error)
2107 VM_OBJECT_UNLOCK(obj);
2108 mbstat.sf_iocnt++;
2109 }
2110 if (error) {
2111 vm_page_lock(pg);
2112 vm_page_unwire(pg, 0);
2113 /*
2114 * See if anyone else might know about
2115 * this page. If not and it is not valid,
2116 * then free it.
2117 */
2118 if (pg->wire_count == 0 && pg->valid == 0 &&
2119 pg->busy == 0 && !(pg->oflags & VPO_BUSY) &&
2120 pg->hold_count == 0)
2121 vm_page_free(pg);
2122 vm_page_unlock(pg);
2123 VM_OBJECT_UNLOCK(obj);
2124 if (error == EAGAIN)
2125 error = 0; /* not a real error */
2126 break;
2127 }
2128
2129 /*
2130 * Get a sendfile buf. We usually wait as long
2131 * as necessary, but this wait can be interrupted.
2132 */
2133 if ((sf = sf_buf_alloc(pg,
2134 (mnw ? SFB_NOWAIT : SFB_CATCH))) == NULL) {
2135 mbstat.sf_allocfail++;
2136 vm_page_lock(pg);
2137 vm_page_unwire(pg, 0);
2138 KASSERT(pg->object != NULL,
2139 ("kern_sendfile: object disappeared"));
2140 vm_page_unlock(pg);
2141 error = (mnw ? EAGAIN : EINTR);
2142 break;
2143 }
2144
2145 /*
2146 * Get an mbuf and set it up as having
2147 * external storage.
2148 */
2149 m0 = m_get((mnw ? M_NOWAIT : M_WAITOK), MT_DATA);
2150 if (m0 == NULL) {
2151 error = (mnw ? EAGAIN : ENOBUFS);
2152 sf_buf_mext((void *)sf_buf_kva(sf), sf);
2153 break;
2154 }
2155 MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, sf_buf_mext,
2156 sfs, sf, M_RDONLY, EXT_SFBUF);
2157 m0->m_data = (char *)sf_buf_kva(sf) + pgoff;
2158 m0->m_len = xfsize;
2159
2160 /* Append to mbuf chain. */
2161 if (m != NULL)
2162 m_cat(m, m0);
2163 else
2164 m = m0;
2165
2166 /* Keep track of bits processed. */
2167 loopbytes += xfsize;
2168 off += xfsize;
2169
2170 if (sfs != NULL) {
2171 mtx_lock(&sfs->mtx);
2172 sfs->count++;
2173 mtx_unlock(&sfs->mtx);
2174 }
2175 }
2176
2177 /* Add the buffer chain to the socket buffer. */
2178 if (m != NULL) {
2179 int mlen, err;
2180
2181 mlen = m_length(m, NULL);
2182 SOCKBUF_LOCK(&so->so_snd);
2183 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2184 error = EPIPE;
2185 SOCKBUF_UNLOCK(&so->so_snd);
2186 goto done;
2187 }
2188 SOCKBUF_UNLOCK(&so->so_snd);
2189 CURVNET_SET(so->so_vnet);
2190 /* Avoid error aliasing. */
2191 err = (*so->so_proto->pr_usrreqs->pru_send)
2192 (so, 0, m, NULL, NULL, td);
2193 CURVNET_RESTORE();
2194 if (err == 0) {
2195 /*
2196 * We need two counters to get the
2197 * file offset and nbytes to send
2198 * right:
2199 * - sbytes contains the total amount
2200 * of bytes sent, including headers.
2201 * - fsbytes contains the total amount
2202 * of bytes sent from the file.
2203 */
2204 sbytes += mlen;
2205 fsbytes += mlen;
2206 if (hdrlen) {
2207 fsbytes -= hdrlen;
2208 hdrlen = 0;
2209 }
2210 } else if (error == 0)
2211 error = err;
2212 m = NULL; /* pru_send always consumes */
2213 }
2214
2215 /* Quit outer loop on error or when we're done. */
2216 if (done)
2217 break;
2218 if (error)
2219 goto done;
2220 }
2221
2222 /*
2223 * Send trailers. Wimp out and use writev(2).
2224 */
2225 if (trl_uio != NULL) {
2226 sbunlock(&so->so_snd);
2227 error = kern_writev(td, uap->s, trl_uio);
2228 if (error == 0)
2229 sbytes += td->td_retval[0];
2230 goto out;
2231 }
2232
2233done:
2234 sbunlock(&so->so_snd);
2235out:
2236 /*
2237 * If there was no error we have to clear td->td_retval[0]
2238 * because it may have been set by writev.
2239 */
2240 if (error == 0) {
2241 td->td_retval[0] = 0;
2242 }
2243 if (uap->sbytes != NULL) {
2244 copyout(&sbytes, uap->sbytes, sizeof(off_t));
2245 }
2246 if (obj != NULL)
2247 vm_object_deallocate(obj);
2248 if (vp != NULL) {
2249 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2250 vrele(vp);
2251 VFS_UNLOCK_GIANT(vfslocked);
2252 }
2253 if (so)
2254 fdrop(sock_fp, td);
2255 if (m)
2256 m_freem(m);
2257
2258 if (sfs != NULL) {
2259 mtx_lock(&sfs->mtx);
2260 if (sfs->count != 0)
2261 cv_wait(&sfs->cv, &sfs->mtx);
2262 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
2263 cv_destroy(&sfs->cv);
2264 mtx_destroy(&sfs->mtx);
2265 free(sfs, M_TEMP);
2266 }
2267
2268 if (error == ERESTART)
2269 error = EINTR;
2270
2271 return (error);
2272}
2273
2274/*
2275 * SCTP syscalls.
2276 * Functionality only compiled in if SCTP is defined in the kernel Makefile,
2277 * otherwise all return EOPNOTSUPP.
2278 * XXX: We should make this loadable one day.
2279 */
2280int
2281sctp_peeloff(td, uap)
2282 struct thread *td;
2283 struct sctp_peeloff_args /* {
2284 int sd;
2285 caddr_t name;
2286 } */ *uap;
2287{
2288#if (defined(INET) || defined(INET6)) && defined(SCTP)
2289 struct filedesc *fdp;
2290 struct file *nfp = NULL;
2291 int error;
2292 struct socket *head, *so;
2293 int fd;
2294 u_int fflag;
2295
2296 fdp = td->td_proc->p_fd;
2297 AUDIT_ARG_FD(uap->sd);
2298 error = fgetsock(td, uap->sd, &head, &fflag);
2299 if (error)
2300 goto done2;
2301 error = sctp_can_peel_off(head, (sctp_assoc_t)uap->name);
2302 if (error)
2303 goto done2;
2304 /*
2305 * At this point we know we do have a assoc to pull
2306 * we proceed to get the fd setup. This may block
2307 * but that is ok.
2308 */
2309
2310 error = falloc(td, &nfp, &fd);
2311 if (error)
2312 goto done;
2313 td->td_retval[0] = fd;
2314
2315 CURVNET_SET(head->so_vnet);
2316 so = sonewconn(head, SS_ISCONNECTED);
2317 if (so == NULL)
2318 goto noconnection;
2319 /*
2320 * Before changing the flags on the socket, we have to bump the
2321 * reference count. Otherwise, if the protocol calls sofree(),
2322 * the socket will be released due to a zero refcount.
2323 */
2324 SOCK_LOCK(so);
2325 soref(so); /* file descriptor reference */
2326 SOCK_UNLOCK(so);
2327
2328 ACCEPT_LOCK();
2329
2330 TAILQ_REMOVE(&head->so_comp, so, so_list);
2331 head->so_qlen--;
2332 so->so_state |= (head->so_state & SS_NBIO);
2333 so->so_state &= ~SS_NOFDREF;
2334 so->so_qstate &= ~SQ_COMP;
2335 so->so_head = NULL;
2336 ACCEPT_UNLOCK();
2337 finit(nfp, fflag, DTYPE_SOCKET, so, &socketops);
2338 error = sctp_do_peeloff(head, so, (sctp_assoc_t)uap->name);
2339 if (error)
2340 goto noconnection;
2341 if (head->so_sigio != NULL)
2342 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
2343
2344noconnection:
2345 /*
2346 * close the new descriptor, assuming someone hasn't ripped it
2347 * out from under us.
2348 */
2349 if (error)
2350 fdclose(fdp, nfp, fd, td);
2351
2352 /*
2353 * Release explicitly held references before returning.
2354 */
2355 CURVNET_RESTORE();
2356done:
2357 if (nfp != NULL)
2358 fdrop(nfp, td);
2359 fputsock(head);
2360done2:
2361 return (error);
2362#else /* SCTP */
2363 return (EOPNOTSUPP);
2364#endif /* SCTP */
2365}
2366
2367int
2368sctp_generic_sendmsg (td, uap)
2369 struct thread *td;
2370 struct sctp_generic_sendmsg_args /* {
2371 int sd,
2372 caddr_t msg,
2373 int mlen,
2374 caddr_t to,
2375 __socklen_t tolen,
2376 struct sctp_sndrcvinfo *sinfo,
2377 int flags
2378 } */ *uap;
2379{
2380#if (defined(INET) || defined(INET6)) && defined(SCTP)
2381 struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL;
2382 struct socket *so;
2383 struct file *fp = NULL;
2384 int use_rcvinfo = 1;
2385 int error = 0, len;
2386 struct sockaddr *to = NULL;
2387#ifdef KTRACE
2388 struct uio *ktruio = NULL;
2389#endif
2390 struct uio auio;
2391 struct iovec iov[1];
2392
2393 if (uap->sinfo) {
2394 error = copyin(uap->sinfo, &sinfo, sizeof (sinfo));
2395 if (error)
2396 return (error);
2397 u_sinfo = &sinfo;
2398 }
2399 if (uap->tolen) {
2400 error = getsockaddr(&to, uap->to, uap->tolen);
2401 if (error) {
2402 to = NULL;
2403 goto sctp_bad2;
2404 }
2405 }
2406
2407 AUDIT_ARG_FD(uap->sd);
2408 error = getsock(td->td_proc->p_fd, uap->sd, &fp, NULL);
2409 if (error)
2410 goto sctp_bad;
2411#ifdef KTRACE
2412 if (KTRPOINT(td, KTR_STRUCT))
2413 ktrsockaddr(to);
2414#endif
2415
2416 iov[0].iov_base = uap->msg;
2417 iov[0].iov_len = uap->mlen;
2418
2419 so = (struct socket *)fp->f_data;
2420#ifdef MAC
2421 error = mac_socket_check_send(td->td_ucred, so);
2422 if (error)
2423 goto sctp_bad;
2424#endif /* MAC */
2425
2426 auio.uio_iov = iov;
2427 auio.uio_iovcnt = 1;
2428 auio.uio_segflg = UIO_USERSPACE;
2429 auio.uio_rw = UIO_WRITE;
2430 auio.uio_td = td;
2431 auio.uio_offset = 0; /* XXX */
2432 auio.uio_resid = 0;
2433 len = auio.uio_resid = uap->mlen;
2434 CURVNET_SET(so->so_vnet);
2435 error = sctp_lower_sosend(so, to, &auio,
2436 (struct mbuf *)NULL, (struct mbuf *)NULL,
2437 uap->flags, use_rcvinfo, u_sinfo, td);
2438 CURVNET_RESTORE();
2439 if (error) {
2440 if (auio.uio_resid != len && (error == ERESTART ||
2441 error == EINTR || error == EWOULDBLOCK))
2442 error = 0;
2443 /* Generation of SIGPIPE can be controlled per socket. */
2444 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
2445 !(uap->flags & MSG_NOSIGNAL)) {
2446 PROC_LOCK(td->td_proc);
2447 psignal(td->td_proc, SIGPIPE);
2448 PROC_UNLOCK(td->td_proc);
2449 }
2450 }
2451 if (error == 0)
2452 td->td_retval[0] = len - auio.uio_resid;
2453#ifdef KTRACE
2454 if (ktruio != NULL) {
2455 ktruio->uio_resid = td->td_retval[0];
2456 ktrgenio(uap->sd, UIO_WRITE, ktruio, error);
2457 }
2458#endif /* KTRACE */
2459sctp_bad:
2460 if (fp)
2461 fdrop(fp, td);
2462sctp_bad2:
2463 if (to)
2464 free(to, M_SONAME);
2465 return (error);
2466#else /* SCTP */
2467 return (EOPNOTSUPP);
2468#endif /* SCTP */
2469}
2470
2471int
2472sctp_generic_sendmsg_iov(td, uap)
2473 struct thread *td;
2474 struct sctp_generic_sendmsg_iov_args /* {
2475 int sd,
2476 struct iovec *iov,
2477 int iovlen,
2478 caddr_t to,
2479 __socklen_t tolen,
2480 struct sctp_sndrcvinfo *sinfo,
2481 int flags
2482 } */ *uap;
2483{
2484#if (defined(INET) || defined(INET6)) && defined(SCTP)
2485 struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL;
2486 struct socket *so;
2487 struct file *fp = NULL;
2488 int use_rcvinfo = 1;
2489 int error=0, len, i;
2490 struct sockaddr *to = NULL;
2491#ifdef KTRACE
2492 struct uio *ktruio = NULL;
2493#endif
2494 struct uio auio;
2495 struct iovec *iov, *tiov;
2496
2497 if (uap->sinfo) {
2498 error = copyin(uap->sinfo, &sinfo, sizeof (sinfo));
2499 if (error)
2500 return (error);
2501 u_sinfo = &sinfo;
2502 }
2503 if (uap->tolen) {
2504 error = getsockaddr(&to, uap->to, uap->tolen);
2505 if (error) {
2506 to = NULL;
2507 goto sctp_bad2;
2508 }
2509 }
2510
2511 AUDIT_ARG_FD(uap->sd);
2512 error = getsock(td->td_proc->p_fd, uap->sd, &fp, NULL);
2513 if (error)
2514 goto sctp_bad1;
2515
2516#ifdef COMPAT_FREEBSD32
2517 if (SV_CURPROC_FLAG(SV_ILP32))
2518 error = freebsd32_copyiniov((struct iovec32 *)uap->iov,
2519 uap->iovlen, &iov, EMSGSIZE);
2520 else
2521#endif
2522 error = copyiniov(uap->iov, uap->iovlen, &iov, EMSGSIZE);
2523 if (error)
2524 goto sctp_bad1;
2525#ifdef KTRACE
2526 if (KTRPOINT(td, KTR_STRUCT))
2527 ktrsockaddr(to);
2528#endif
2529
2530 so = (struct socket *)fp->f_data;
2531#ifdef MAC
2532 error = mac_socket_check_send(td->td_ucred, so);
2533 if (error)
2534 goto sctp_bad;
2535#endif /* MAC */
2536
2537 auio.uio_iov = iov;
2538 auio.uio_iovcnt = uap->iovlen;
2539 auio.uio_segflg = UIO_USERSPACE;
2540 auio.uio_rw = UIO_WRITE;
2541 auio.uio_td = td;
2542 auio.uio_offset = 0; /* XXX */
2543 auio.uio_resid = 0;
2544 tiov = iov;
2545 for (i = 0; i <uap->iovlen; i++, tiov++) {
2546 if ((auio.uio_resid += tiov->iov_len) < 0) {
2547 error = EINVAL;
2548 goto sctp_bad;
2549 }
2550 }
2551 len = auio.uio_resid;
2552 CURVNET_SET(so->so_vnet);
2553 error = sctp_lower_sosend(so, to, &auio,
2554 (struct mbuf *)NULL, (struct mbuf *)NULL,
2555 uap->flags, use_rcvinfo, u_sinfo, td);
2556 CURVNET_RESTORE();
2557 if (error) {
2558 if (auio.uio_resid != len && (error == ERESTART ||
2559 error == EINTR || error == EWOULDBLOCK))
2560 error = 0;
2561 /* Generation of SIGPIPE can be controlled per socket */
2562 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
2563 !(uap->flags & MSG_NOSIGNAL)) {
2564 PROC_LOCK(td->td_proc);
2565 psignal(td->td_proc, SIGPIPE);
2566 PROC_UNLOCK(td->td_proc);
2567 }
2568 }
2569 if (error == 0)
2570 td->td_retval[0] = len - auio.uio_resid;
2571#ifdef KTRACE
2572 if (ktruio != NULL) {
2573 ktruio->uio_resid = td->td_retval[0];
2574 ktrgenio(uap->sd, UIO_WRITE, ktruio, error);
2575 }
2576#endif /* KTRACE */
2577sctp_bad:
2578 free(iov, M_IOV);
2579sctp_bad1:
2580 if (fp)
2581 fdrop(fp, td);
2582sctp_bad2:
2583 if (to)
2584 free(to, M_SONAME);
2585 return (error);
2586#else /* SCTP */
2587 return (EOPNOTSUPP);
2588#endif /* SCTP */
2589}
2590
2591int
2592sctp_generic_recvmsg(td, uap)
2593 struct thread *td;
2594 struct sctp_generic_recvmsg_args /* {
2595 int sd,
2596 struct iovec *iov,
2597 int iovlen,
2598 struct sockaddr *from,
2599 __socklen_t *fromlenaddr,
2600 struct sctp_sndrcvinfo *sinfo,
2601 int *msg_flags
2602 } */ *uap;
2603{
2604#if (defined(INET) || defined(INET6)) && defined(SCTP)
| 37
38#include "opt_inet.h"
39#include "opt_inet6.h"
40#include "opt_sctp.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/kernel.h>
47#include <sys/lock.h>
48#include <sys/mutex.h>
49#include <sys/sysproto.h>
50#include <sys/malloc.h>
51#include <sys/filedesc.h>
52#include <sys/event.h>
53#include <sys/proc.h>
54#include <sys/fcntl.h>
55#include <sys/file.h>
56#include <sys/filio.h>
57#include <sys/jail.h>
58#include <sys/mount.h>
59#include <sys/mbuf.h>
60#include <sys/protosw.h>
61#include <sys/sf_buf.h>
62#include <sys/sysent.h>
63#include <sys/socket.h>
64#include <sys/socketvar.h>
65#include <sys/signalvar.h>
66#include <sys/syscallsubr.h>
67#include <sys/sysctl.h>
68#include <sys/uio.h>
69#include <sys/vnode.h>
70#ifdef KTRACE
71#include <sys/ktrace.h>
72#endif
73#ifdef COMPAT_FREEBSD32
74#include <compat/freebsd32/freebsd32_util.h>
75#endif
76
77#include <net/vnet.h>
78
79#include <security/audit/audit.h>
80#include <security/mac/mac_framework.h>
81
82#include <vm/vm.h>
83#include <vm/vm_object.h>
84#include <vm/vm_page.h>
85#include <vm/vm_pageout.h>
86#include <vm/vm_kern.h>
87#include <vm/vm_extern.h>
88
89#if defined(INET) || defined(INET6)
90#ifdef SCTP
91#include <netinet/sctp.h>
92#include <netinet/sctp_peeloff.h>
93#endif /* SCTP */
94#endif /* INET || INET6 */
95
96static int sendit(struct thread *td, int s, struct msghdr *mp, int flags);
97static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp);
98
99static int accept1(struct thread *td, struct accept_args *uap, int compat);
100static int do_sendfile(struct thread *td, struct sendfile_args *uap, int compat);
101static int getsockname1(struct thread *td, struct getsockname_args *uap,
102 int compat);
103static int getpeername1(struct thread *td, struct getpeername_args *uap,
104 int compat);
105
106/*
107 * NSFBUFS-related variables and associated sysctls
108 */
109int nsfbufs;
110int nsfbufspeak;
111int nsfbufsused;
112
113SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
114 "Maximum number of sendfile(2) sf_bufs available");
115SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
116 "Number of sendfile(2) sf_bufs at peak usage");
117SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
118 "Number of sendfile(2) sf_bufs in use");
119
120/*
121 * Convert a user file descriptor to a kernel file entry. A reference on the
122 * file entry is held upon returning. This is lighter weight than
123 * fgetsock(), which bumps the socket reference drops the file reference
124 * count instead, as this approach avoids several additional mutex operations
125 * associated with the additional reference count. If requested, return the
126 * open file flags.
127 */
128static int
129getsock(struct filedesc *fdp, int fd, struct file **fpp, u_int *fflagp)
130{
131 struct file *fp;
132 int error;
133
134 fp = NULL;
135 if (fdp == NULL || (fp = fget_unlocked(fdp, fd)) == NULL) {
136 error = EBADF;
137 } else if (fp->f_type != DTYPE_SOCKET) {
138 fdrop(fp, curthread);
139 fp = NULL;
140 error = ENOTSOCK;
141 } else {
142 if (fflagp != NULL)
143 *fflagp = fp->f_flag;
144 error = 0;
145 }
146 *fpp = fp;
147 return (error);
148}
149
150/*
151 * System call interface to the socket abstraction.
152 */
153#if defined(COMPAT_43)
154#define COMPAT_OLDSOCK
155#endif
156
157int
158socket(td, uap)
159 struct thread *td;
160 struct socket_args /* {
161 int domain;
162 int type;
163 int protocol;
164 } */ *uap;
165{
166 struct filedesc *fdp;
167 struct socket *so;
168 struct file *fp;
169 int fd, error;
170
171 AUDIT_ARG_SOCKET(uap->domain, uap->type, uap->protocol);
172#ifdef MAC
173 error = mac_socket_check_create(td->td_ucred, uap->domain, uap->type,
174 uap->protocol);
175 if (error)
176 return (error);
177#endif
178 fdp = td->td_proc->p_fd;
179 error = falloc(td, &fp, &fd);
180 if (error)
181 return (error);
182 /* An extra reference on `fp' has been held for us by falloc(). */
183 error = socreate(uap->domain, &so, uap->type, uap->protocol,
184 td->td_ucred, td);
185 if (error) {
186 fdclose(fdp, fp, fd, td);
187 } else {
188 finit(fp, FREAD | FWRITE, DTYPE_SOCKET, so, &socketops);
189 td->td_retval[0] = fd;
190 }
191 fdrop(fp, td);
192 return (error);
193}
194
195/* ARGSUSED */
196int
197bind(td, uap)
198 struct thread *td;
199 struct bind_args /* {
200 int s;
201 caddr_t name;
202 int namelen;
203 } */ *uap;
204{
205 struct sockaddr *sa;
206 int error;
207
208 if ((error = getsockaddr(&sa, uap->name, uap->namelen)) != 0)
209 return (error);
210
211 error = kern_bind(td, uap->s, sa);
212 free(sa, M_SONAME);
213 return (error);
214}
215
216int
217kern_bind(td, fd, sa)
218 struct thread *td;
219 int fd;
220 struct sockaddr *sa;
221{
222 struct socket *so;
223 struct file *fp;
224 int error;
225
226 AUDIT_ARG_FD(fd);
227 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
228 if (error)
229 return (error);
230 so = fp->f_data;
231#ifdef KTRACE
232 if (KTRPOINT(td, KTR_STRUCT))
233 ktrsockaddr(sa);
234#endif
235#ifdef MAC
236 error = mac_socket_check_bind(td->td_ucred, so, sa);
237 if (error == 0)
238#endif
239 error = sobind(so, sa, td);
240 fdrop(fp, td);
241 return (error);
242}
243
244/* ARGSUSED */
245int
246listen(td, uap)
247 struct thread *td;
248 struct listen_args /* {
249 int s;
250 int backlog;
251 } */ *uap;
252{
253 struct socket *so;
254 struct file *fp;
255 int error;
256
257 AUDIT_ARG_FD(uap->s);
258 error = getsock(td->td_proc->p_fd, uap->s, &fp, NULL);
259 if (error == 0) {
260 so = fp->f_data;
261#ifdef MAC
262 error = mac_socket_check_listen(td->td_ucred, so);
263 if (error == 0) {
264#endif
265 CURVNET_SET(so->so_vnet);
266 error = solisten(so, uap->backlog, td);
267 CURVNET_RESTORE();
268#ifdef MAC
269 }
270#endif
271 fdrop(fp, td);
272 }
273 return(error);
274}
275
276/*
277 * accept1()
278 */
279static int
280accept1(td, uap, compat)
281 struct thread *td;
282 struct accept_args /* {
283 int s;
284 struct sockaddr * __restrict name;
285 socklen_t * __restrict anamelen;
286 } */ *uap;
287 int compat;
288{
289 struct sockaddr *name;
290 socklen_t namelen;
291 struct file *fp;
292 int error;
293
294 if (uap->name == NULL)
295 return (kern_accept(td, uap->s, NULL, NULL, NULL));
296
297 error = copyin(uap->anamelen, &namelen, sizeof (namelen));
298 if (error)
299 return (error);
300
301 error = kern_accept(td, uap->s, &name, &namelen, &fp);
302
303 /*
304 * return a namelen of zero for older code which might
305 * ignore the return value from accept.
306 */
307 if (error) {
308 (void) copyout(&namelen,
309 uap->anamelen, sizeof(*uap->anamelen));
310 return (error);
311 }
312
313 if (error == 0 && name != NULL) {
314#ifdef COMPAT_OLDSOCK
315 if (compat)
316 ((struct osockaddr *)name)->sa_family =
317 name->sa_family;
318#endif
319 error = copyout(name, uap->name, namelen);
320 }
321 if (error == 0)
322 error = copyout(&namelen, uap->anamelen,
323 sizeof(namelen));
324 if (error)
325 fdclose(td->td_proc->p_fd, fp, td->td_retval[0], td);
326 fdrop(fp, td);
327 free(name, M_SONAME);
328 return (error);
329}
330
331int
332kern_accept(struct thread *td, int s, struct sockaddr **name,
333 socklen_t *namelen, struct file **fp)
334{
335 struct filedesc *fdp;
336 struct file *headfp, *nfp = NULL;
337 struct sockaddr *sa = NULL;
338 int error;
339 struct socket *head, *so;
340 int fd;
341 u_int fflag;
342 pid_t pgid;
343 int tmp;
344
345 if (name) {
346 *name = NULL;
347 if (*namelen < 0)
348 return (EINVAL);
349 }
350
351 AUDIT_ARG_FD(s);
352 fdp = td->td_proc->p_fd;
353 error = getsock(fdp, s, &headfp, &fflag);
354 if (error)
355 return (error);
356 head = headfp->f_data;
357 if ((head->so_options & SO_ACCEPTCONN) == 0) {
358 error = EINVAL;
359 goto done;
360 }
361#ifdef MAC
362 error = mac_socket_check_accept(td->td_ucred, head);
363 if (error != 0)
364 goto done;
365#endif
366 error = falloc(td, &nfp, &fd);
367 if (error)
368 goto done;
369 ACCEPT_LOCK();
370 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) {
371 ACCEPT_UNLOCK();
372 error = EWOULDBLOCK;
373 goto noconnection;
374 }
375 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
376 if (head->so_rcv.sb_state & SBS_CANTRCVMORE) {
377 head->so_error = ECONNABORTED;
378 break;
379 }
380 error = msleep(&head->so_timeo, &accept_mtx, PSOCK | PCATCH,
381 "accept", 0);
382 if (error) {
383 ACCEPT_UNLOCK();
384 goto noconnection;
385 }
386 }
387 if (head->so_error) {
388 error = head->so_error;
389 head->so_error = 0;
390 ACCEPT_UNLOCK();
391 goto noconnection;
392 }
393 so = TAILQ_FIRST(&head->so_comp);
394 KASSERT(!(so->so_qstate & SQ_INCOMP), ("accept1: so SQ_INCOMP"));
395 KASSERT(so->so_qstate & SQ_COMP, ("accept1: so not SQ_COMP"));
396
397 /*
398 * Before changing the flags on the socket, we have to bump the
399 * reference count. Otherwise, if the protocol calls sofree(),
400 * the socket will be released due to a zero refcount.
401 */
402 SOCK_LOCK(so); /* soref() and so_state update */
403 soref(so); /* file descriptor reference */
404
405 TAILQ_REMOVE(&head->so_comp, so, so_list);
406 head->so_qlen--;
407 so->so_state |= (head->so_state & SS_NBIO);
408 so->so_qstate &= ~SQ_COMP;
409 so->so_head = NULL;
410
411 SOCK_UNLOCK(so);
412 ACCEPT_UNLOCK();
413
414 /* An extra reference on `nfp' has been held for us by falloc(). */
415 td->td_retval[0] = fd;
416
417 /* connection has been removed from the listen queue */
418 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
419
420 pgid = fgetown(&head->so_sigio);
421 if (pgid != 0)
422 fsetown(pgid, &so->so_sigio);
423
424 finit(nfp, fflag, DTYPE_SOCKET, so, &socketops);
425 /* Sync socket nonblocking/async state with file flags */
426 tmp = fflag & FNONBLOCK;
427 (void) fo_ioctl(nfp, FIONBIO, &tmp, td->td_ucred, td);
428 tmp = fflag & FASYNC;
429 (void) fo_ioctl(nfp, FIOASYNC, &tmp, td->td_ucred, td);
430 sa = 0;
431 CURVNET_SET(so->so_vnet);
432 error = soaccept(so, &sa);
433 CURVNET_RESTORE();
434 if (error) {
435 /*
436 * return a namelen of zero for older code which might
437 * ignore the return value from accept.
438 */
439 if (name)
440 *namelen = 0;
441 goto noconnection;
442 }
443 if (sa == NULL) {
444 if (name)
445 *namelen = 0;
446 goto done;
447 }
448 if (name) {
449 /* check sa_len before it is destroyed */
450 if (*namelen > sa->sa_len)
451 *namelen = sa->sa_len;
452#ifdef KTRACE
453 if (KTRPOINT(td, KTR_STRUCT))
454 ktrsockaddr(sa);
455#endif
456 *name = sa;
457 sa = NULL;
458 }
459noconnection:
460 if (sa)
461 free(sa, M_SONAME);
462
463 /*
464 * close the new descriptor, assuming someone hasn't ripped it
465 * out from under us.
466 */
467 if (error)
468 fdclose(fdp, nfp, fd, td);
469
470 /*
471 * Release explicitly held references before returning. We return
472 * a reference on nfp to the caller on success if they request it.
473 */
474done:
475 if (fp != NULL) {
476 if (error == 0) {
477 *fp = nfp;
478 nfp = NULL;
479 } else
480 *fp = NULL;
481 }
482 if (nfp != NULL)
483 fdrop(nfp, td);
484 fdrop(headfp, td);
485 return (error);
486}
487
488int
489accept(td, uap)
490 struct thread *td;
491 struct accept_args *uap;
492{
493
494 return (accept1(td, uap, 0));
495}
496
497#ifdef COMPAT_OLDSOCK
498int
499oaccept(td, uap)
500 struct thread *td;
501 struct accept_args *uap;
502{
503
504 return (accept1(td, uap, 1));
505}
506#endif /* COMPAT_OLDSOCK */
507
508/* ARGSUSED */
509int
510connect(td, uap)
511 struct thread *td;
512 struct connect_args /* {
513 int s;
514 caddr_t name;
515 int namelen;
516 } */ *uap;
517{
518 struct sockaddr *sa;
519 int error;
520
521 error = getsockaddr(&sa, uap->name, uap->namelen);
522 if (error)
523 return (error);
524
525 error = kern_connect(td, uap->s, sa);
526 free(sa, M_SONAME);
527 return (error);
528}
529
530
531int
532kern_connect(td, fd, sa)
533 struct thread *td;
534 int fd;
535 struct sockaddr *sa;
536{
537 struct socket *so;
538 struct file *fp;
539 int error;
540 int interrupted = 0;
541
542 AUDIT_ARG_FD(fd);
543 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
544 if (error)
545 return (error);
546 so = fp->f_data;
547 if (so->so_state & SS_ISCONNECTING) {
548 error = EALREADY;
549 goto done1;
550 }
551#ifdef KTRACE
552 if (KTRPOINT(td, KTR_STRUCT))
553 ktrsockaddr(sa);
554#endif
555#ifdef MAC
556 error = mac_socket_check_connect(td->td_ucred, so, sa);
557 if (error)
558 goto bad;
559#endif
560 error = soconnect(so, sa, td);
561 if (error)
562 goto bad;
563 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
564 error = EINPROGRESS;
565 goto done1;
566 }
567 SOCK_LOCK(so);
568 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
569 error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH,
570 "connec", 0);
571 if (error) {
572 if (error == EINTR || error == ERESTART)
573 interrupted = 1;
574 break;
575 }
576 }
577 if (error == 0) {
578 error = so->so_error;
579 so->so_error = 0;
580 }
581 SOCK_UNLOCK(so);
582bad:
583 if (!interrupted)
584 so->so_state &= ~SS_ISCONNECTING;
585 if (error == ERESTART)
586 error = EINTR;
587done1:
588 fdrop(fp, td);
589 return (error);
590}
591
592int
593kern_socketpair(struct thread *td, int domain, int type, int protocol,
594 int *rsv)
595{
596 struct filedesc *fdp = td->td_proc->p_fd;
597 struct file *fp1, *fp2;
598 struct socket *so1, *so2;
599 int fd, error;
600
601 AUDIT_ARG_SOCKET(domain, type, protocol);
602#ifdef MAC
603 /* We might want to have a separate check for socket pairs. */
604 error = mac_socket_check_create(td->td_ucred, domain, type,
605 protocol);
606 if (error)
607 return (error);
608#endif
609 error = socreate(domain, &so1, type, protocol, td->td_ucred, td);
610 if (error)
611 return (error);
612 error = socreate(domain, &so2, type, protocol, td->td_ucred, td);
613 if (error)
614 goto free1;
615 /* On success extra reference to `fp1' and 'fp2' is set by falloc. */
616 error = falloc(td, &fp1, &fd);
617 if (error)
618 goto free2;
619 rsv[0] = fd;
620 fp1->f_data = so1; /* so1 already has ref count */
621 error = falloc(td, &fp2, &fd);
622 if (error)
623 goto free3;
624 fp2->f_data = so2; /* so2 already has ref count */
625 rsv[1] = fd;
626 error = soconnect2(so1, so2);
627 if (error)
628 goto free4;
629 if (type == SOCK_DGRAM) {
630 /*
631 * Datagram socket connection is asymmetric.
632 */
633 error = soconnect2(so2, so1);
634 if (error)
635 goto free4;
636 }
637 finit(fp1, FREAD | FWRITE, DTYPE_SOCKET, fp1->f_data, &socketops);
638 finit(fp2, FREAD | FWRITE, DTYPE_SOCKET, fp2->f_data, &socketops);
639 fdrop(fp1, td);
640 fdrop(fp2, td);
641 return (0);
642free4:
643 fdclose(fdp, fp2, rsv[1], td);
644 fdrop(fp2, td);
645free3:
646 fdclose(fdp, fp1, rsv[0], td);
647 fdrop(fp1, td);
648free2:
649 if (so2 != NULL)
650 (void)soclose(so2);
651free1:
652 if (so1 != NULL)
653 (void)soclose(so1);
654 return (error);
655}
656
657int
658socketpair(struct thread *td, struct socketpair_args *uap)
659{
660 int error, sv[2];
661
662 error = kern_socketpair(td, uap->domain, uap->type,
663 uap->protocol, sv);
664 if (error)
665 return (error);
666 error = copyout(sv, uap->rsv, 2 * sizeof(int));
667 if (error) {
668 (void)kern_close(td, sv[0]);
669 (void)kern_close(td, sv[1]);
670 }
671 return (error);
672}
673
674static int
675sendit(td, s, mp, flags)
676 struct thread *td;
677 int s;
678 struct msghdr *mp;
679 int flags;
680{
681 struct mbuf *control;
682 struct sockaddr *to;
683 int error;
684
685 if (mp->msg_name != NULL) {
686 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
687 if (error) {
688 to = NULL;
689 goto bad;
690 }
691 mp->msg_name = to;
692 } else {
693 to = NULL;
694 }
695
696 if (mp->msg_control) {
697 if (mp->msg_controllen < sizeof(struct cmsghdr)
698#ifdef COMPAT_OLDSOCK
699 && mp->msg_flags != MSG_COMPAT
700#endif
701 ) {
702 error = EINVAL;
703 goto bad;
704 }
705 error = sockargs(&control, mp->msg_control,
706 mp->msg_controllen, MT_CONTROL);
707 if (error)
708 goto bad;
709#ifdef COMPAT_OLDSOCK
710 if (mp->msg_flags == MSG_COMPAT) {
711 struct cmsghdr *cm;
712
713 M_PREPEND(control, sizeof(*cm), M_WAIT);
714 cm = mtod(control, struct cmsghdr *);
715 cm->cmsg_len = control->m_len;
716 cm->cmsg_level = SOL_SOCKET;
717 cm->cmsg_type = SCM_RIGHTS;
718 }
719#endif
720 } else {
721 control = NULL;
722 }
723
724 error = kern_sendit(td, s, mp, flags, control, UIO_USERSPACE);
725
726bad:
727 if (to)
728 free(to, M_SONAME);
729 return (error);
730}
731
732int
733kern_sendit(td, s, mp, flags, control, segflg)
734 struct thread *td;
735 int s;
736 struct msghdr *mp;
737 int flags;
738 struct mbuf *control;
739 enum uio_seg segflg;
740{
741 struct file *fp;
742 struct uio auio;
743 struct iovec *iov;
744 struct socket *so;
745 int i;
746 int len, error;
747#ifdef KTRACE
748 struct uio *ktruio = NULL;
749#endif
750
751 AUDIT_ARG_FD(s);
752 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
753 if (error)
754 return (error);
755 so = (struct socket *)fp->f_data;
756
757#ifdef MAC
758 if (mp->msg_name != NULL) {
759 error = mac_socket_check_connect(td->td_ucred, so,
760 mp->msg_name);
761 if (error)
762 goto bad;
763 }
764 error = mac_socket_check_send(td->td_ucred, so);
765 if (error)
766 goto bad;
767#endif
768
769 auio.uio_iov = mp->msg_iov;
770 auio.uio_iovcnt = mp->msg_iovlen;
771 auio.uio_segflg = segflg;
772 auio.uio_rw = UIO_WRITE;
773 auio.uio_td = td;
774 auio.uio_offset = 0; /* XXX */
775 auio.uio_resid = 0;
776 iov = mp->msg_iov;
777 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
778 if ((auio.uio_resid += iov->iov_len) < 0) {
779 error = EINVAL;
780 goto bad;
781 }
782 }
783#ifdef KTRACE
784 if (KTRPOINT(td, KTR_GENIO))
785 ktruio = cloneuio(&auio);
786#endif
787 len = auio.uio_resid;
788 error = sosend(so, mp->msg_name, &auio, 0, control, flags, td);
789 if (error) {
790 if (auio.uio_resid != len && (error == ERESTART ||
791 error == EINTR || error == EWOULDBLOCK))
792 error = 0;
793 /* Generation of SIGPIPE can be controlled per socket */
794 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
795 !(flags & MSG_NOSIGNAL)) {
796 PROC_LOCK(td->td_proc);
797 psignal(td->td_proc, SIGPIPE);
798 PROC_UNLOCK(td->td_proc);
799 }
800 }
801 if (error == 0)
802 td->td_retval[0] = len - auio.uio_resid;
803#ifdef KTRACE
804 if (ktruio != NULL) {
805 ktruio->uio_resid = td->td_retval[0];
806 ktrgenio(s, UIO_WRITE, ktruio, error);
807 }
808#endif
809bad:
810 fdrop(fp, td);
811 return (error);
812}
813
814int
815sendto(td, uap)
816 struct thread *td;
817 struct sendto_args /* {
818 int s;
819 caddr_t buf;
820 size_t len;
821 int flags;
822 caddr_t to;
823 int tolen;
824 } */ *uap;
825{
826 struct msghdr msg;
827 struct iovec aiov;
828 int error;
829
830 msg.msg_name = uap->to;
831 msg.msg_namelen = uap->tolen;
832 msg.msg_iov = &aiov;
833 msg.msg_iovlen = 1;
834 msg.msg_control = 0;
835#ifdef COMPAT_OLDSOCK
836 msg.msg_flags = 0;
837#endif
838 aiov.iov_base = uap->buf;
839 aiov.iov_len = uap->len;
840 error = sendit(td, uap->s, &msg, uap->flags);
841 return (error);
842}
843
844#ifdef COMPAT_OLDSOCK
845int
846osend(td, uap)
847 struct thread *td;
848 struct osend_args /* {
849 int s;
850 caddr_t buf;
851 int len;
852 int flags;
853 } */ *uap;
854{
855 struct msghdr msg;
856 struct iovec aiov;
857 int error;
858
859 msg.msg_name = 0;
860 msg.msg_namelen = 0;
861 msg.msg_iov = &aiov;
862 msg.msg_iovlen = 1;
863 aiov.iov_base = uap->buf;
864 aiov.iov_len = uap->len;
865 msg.msg_control = 0;
866 msg.msg_flags = 0;
867 error = sendit(td, uap->s, &msg, uap->flags);
868 return (error);
869}
870
871int
872osendmsg(td, uap)
873 struct thread *td;
874 struct osendmsg_args /* {
875 int s;
876 caddr_t msg;
877 int flags;
878 } */ *uap;
879{
880 struct msghdr msg;
881 struct iovec *iov;
882 int error;
883
884 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
885 if (error)
886 return (error);
887 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
888 if (error)
889 return (error);
890 msg.msg_iov = iov;
891 msg.msg_flags = MSG_COMPAT;
892 error = sendit(td, uap->s, &msg, uap->flags);
893 free(iov, M_IOV);
894 return (error);
895}
896#endif
897
898int
899sendmsg(td, uap)
900 struct thread *td;
901 struct sendmsg_args /* {
902 int s;
903 caddr_t msg;
904 int flags;
905 } */ *uap;
906{
907 struct msghdr msg;
908 struct iovec *iov;
909 int error;
910
911 error = copyin(uap->msg, &msg, sizeof (msg));
912 if (error)
913 return (error);
914 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
915 if (error)
916 return (error);
917 msg.msg_iov = iov;
918#ifdef COMPAT_OLDSOCK
919 msg.msg_flags = 0;
920#endif
921 error = sendit(td, uap->s, &msg, uap->flags);
922 free(iov, M_IOV);
923 return (error);
924}
925
926int
927kern_recvit(td, s, mp, fromseg, controlp)
928 struct thread *td;
929 int s;
930 struct msghdr *mp;
931 enum uio_seg fromseg;
932 struct mbuf **controlp;
933{
934 struct uio auio;
935 struct iovec *iov;
936 int i;
937 socklen_t len;
938 int error;
939 struct mbuf *m, *control = 0;
940 caddr_t ctlbuf;
941 struct file *fp;
942 struct socket *so;
943 struct sockaddr *fromsa = 0;
944#ifdef KTRACE
945 struct uio *ktruio = NULL;
946#endif
947
948 if (controlp != NULL)
949 *controlp = NULL;
950
951 AUDIT_ARG_FD(s);
952 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
953 if (error)
954 return (error);
955 so = fp->f_data;
956
957#ifdef MAC
958 error = mac_socket_check_receive(td->td_ucred, so);
959 if (error) {
960 fdrop(fp, td);
961 return (error);
962 }
963#endif
964
965 auio.uio_iov = mp->msg_iov;
966 auio.uio_iovcnt = mp->msg_iovlen;
967 auio.uio_segflg = UIO_USERSPACE;
968 auio.uio_rw = UIO_READ;
969 auio.uio_td = td;
970 auio.uio_offset = 0; /* XXX */
971 auio.uio_resid = 0;
972 iov = mp->msg_iov;
973 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
974 if ((auio.uio_resid += iov->iov_len) < 0) {
975 fdrop(fp, td);
976 return (EINVAL);
977 }
978 }
979#ifdef KTRACE
980 if (KTRPOINT(td, KTR_GENIO))
981 ktruio = cloneuio(&auio);
982#endif
983 len = auio.uio_resid;
984 CURVNET_SET(so->so_vnet);
985 error = soreceive(so, &fromsa, &auio, (struct mbuf **)0,
986 (mp->msg_control || controlp) ? &control : (struct mbuf **)0,
987 &mp->msg_flags);
988 CURVNET_RESTORE();
989 if (error) {
990 if (auio.uio_resid != (int)len && (error == ERESTART ||
991 error == EINTR || error == EWOULDBLOCK))
992 error = 0;
993 }
994#ifdef KTRACE
995 if (ktruio != NULL) {
996 ktruio->uio_resid = (int)len - auio.uio_resid;
997 ktrgenio(s, UIO_READ, ktruio, error);
998 }
999#endif
1000 if (error)
1001 goto out;
1002 td->td_retval[0] = (int)len - auio.uio_resid;
1003 if (mp->msg_name) {
1004 len = mp->msg_namelen;
1005 if (len <= 0 || fromsa == 0)
1006 len = 0;
1007 else {
1008 /* save sa_len before it is destroyed by MSG_COMPAT */
1009 len = MIN(len, fromsa->sa_len);
1010#ifdef COMPAT_OLDSOCK
1011 if (mp->msg_flags & MSG_COMPAT)
1012 ((struct osockaddr *)fromsa)->sa_family =
1013 fromsa->sa_family;
1014#endif
1015 if (fromseg == UIO_USERSPACE) {
1016 error = copyout(fromsa, mp->msg_name,
1017 (unsigned)len);
1018 if (error)
1019 goto out;
1020 } else
1021 bcopy(fromsa, mp->msg_name, len);
1022 }
1023 mp->msg_namelen = len;
1024 }
1025 if (mp->msg_control && controlp == NULL) {
1026#ifdef COMPAT_OLDSOCK
1027 /*
1028 * We assume that old recvmsg calls won't receive access
1029 * rights and other control info, esp. as control info
1030 * is always optional and those options didn't exist in 4.3.
1031 * If we receive rights, trim the cmsghdr; anything else
1032 * is tossed.
1033 */
1034 if (control && mp->msg_flags & MSG_COMPAT) {
1035 if (mtod(control, struct cmsghdr *)->cmsg_level !=
1036 SOL_SOCKET ||
1037 mtod(control, struct cmsghdr *)->cmsg_type !=
1038 SCM_RIGHTS) {
1039 mp->msg_controllen = 0;
1040 goto out;
1041 }
1042 control->m_len -= sizeof (struct cmsghdr);
1043 control->m_data += sizeof (struct cmsghdr);
1044 }
1045#endif
1046 len = mp->msg_controllen;
1047 m = control;
1048 mp->msg_controllen = 0;
1049 ctlbuf = mp->msg_control;
1050
1051 while (m && len > 0) {
1052 unsigned int tocopy;
1053
1054 if (len >= m->m_len)
1055 tocopy = m->m_len;
1056 else {
1057 mp->msg_flags |= MSG_CTRUNC;
1058 tocopy = len;
1059 }
1060
1061 if ((error = copyout(mtod(m, caddr_t),
1062 ctlbuf, tocopy)) != 0)
1063 goto out;
1064
1065 ctlbuf += tocopy;
1066 len -= tocopy;
1067 m = m->m_next;
1068 }
1069 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
1070 }
1071out:
1072 fdrop(fp, td);
1073#ifdef KTRACE
1074 if (fromsa && KTRPOINT(td, KTR_STRUCT))
1075 ktrsockaddr(fromsa);
1076#endif
1077 if (fromsa)
1078 free(fromsa, M_SONAME);
1079
1080 if (error == 0 && controlp != NULL)
1081 *controlp = control;
1082 else if (control)
1083 m_freem(control);
1084
1085 return (error);
1086}
1087
1088static int
1089recvit(td, s, mp, namelenp)
1090 struct thread *td;
1091 int s;
1092 struct msghdr *mp;
1093 void *namelenp;
1094{
1095 int error;
1096
1097 error = kern_recvit(td, s, mp, UIO_USERSPACE, NULL);
1098 if (error)
1099 return (error);
1100 if (namelenp) {
1101 error = copyout(&mp->msg_namelen, namelenp, sizeof (socklen_t));
1102#ifdef COMPAT_OLDSOCK
1103 if (mp->msg_flags & MSG_COMPAT)
1104 error = 0; /* old recvfrom didn't check */
1105#endif
1106 }
1107 return (error);
1108}
1109
1110int
1111recvfrom(td, uap)
1112 struct thread *td;
1113 struct recvfrom_args /* {
1114 int s;
1115 caddr_t buf;
1116 size_t len;
1117 int flags;
1118 struct sockaddr * __restrict from;
1119 socklen_t * __restrict fromlenaddr;
1120 } */ *uap;
1121{
1122 struct msghdr msg;
1123 struct iovec aiov;
1124 int error;
1125
1126 if (uap->fromlenaddr) {
1127 error = copyin(uap->fromlenaddr,
1128 &msg.msg_namelen, sizeof (msg.msg_namelen));
1129 if (error)
1130 goto done2;
1131 } else {
1132 msg.msg_namelen = 0;
1133 }
1134 msg.msg_name = uap->from;
1135 msg.msg_iov = &aiov;
1136 msg.msg_iovlen = 1;
1137 aiov.iov_base = uap->buf;
1138 aiov.iov_len = uap->len;
1139 msg.msg_control = 0;
1140 msg.msg_flags = uap->flags;
1141 error = recvit(td, uap->s, &msg, uap->fromlenaddr);
1142done2:
1143 return(error);
1144}
1145
1146#ifdef COMPAT_OLDSOCK
1147int
1148orecvfrom(td, uap)
1149 struct thread *td;
1150 struct recvfrom_args *uap;
1151{
1152
1153 uap->flags |= MSG_COMPAT;
1154 return (recvfrom(td, uap));
1155}
1156#endif
1157
1158#ifdef COMPAT_OLDSOCK
1159int
1160orecv(td, uap)
1161 struct thread *td;
1162 struct orecv_args /* {
1163 int s;
1164 caddr_t buf;
1165 int len;
1166 int flags;
1167 } */ *uap;
1168{
1169 struct msghdr msg;
1170 struct iovec aiov;
1171 int error;
1172
1173 msg.msg_name = 0;
1174 msg.msg_namelen = 0;
1175 msg.msg_iov = &aiov;
1176 msg.msg_iovlen = 1;
1177 aiov.iov_base = uap->buf;
1178 aiov.iov_len = uap->len;
1179 msg.msg_control = 0;
1180 msg.msg_flags = uap->flags;
1181 error = recvit(td, uap->s, &msg, NULL);
1182 return (error);
1183}
1184
1185/*
1186 * Old recvmsg. This code takes advantage of the fact that the old msghdr
1187 * overlays the new one, missing only the flags, and with the (old) access
1188 * rights where the control fields are now.
1189 */
1190int
1191orecvmsg(td, uap)
1192 struct thread *td;
1193 struct orecvmsg_args /* {
1194 int s;
1195 struct omsghdr *msg;
1196 int flags;
1197 } */ *uap;
1198{
1199 struct msghdr msg;
1200 struct iovec *iov;
1201 int error;
1202
1203 error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
1204 if (error)
1205 return (error);
1206 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1207 if (error)
1208 return (error);
1209 msg.msg_flags = uap->flags | MSG_COMPAT;
1210 msg.msg_iov = iov;
1211 error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen);
1212 if (msg.msg_controllen && error == 0)
1213 error = copyout(&msg.msg_controllen,
1214 &uap->msg->msg_accrightslen, sizeof (int));
1215 free(iov, M_IOV);
1216 return (error);
1217}
1218#endif
1219
1220int
1221recvmsg(td, uap)
1222 struct thread *td;
1223 struct recvmsg_args /* {
1224 int s;
1225 struct msghdr *msg;
1226 int flags;
1227 } */ *uap;
1228{
1229 struct msghdr msg;
1230 struct iovec *uiov, *iov;
1231 int error;
1232
1233 error = copyin(uap->msg, &msg, sizeof (msg));
1234 if (error)
1235 return (error);
1236 error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1237 if (error)
1238 return (error);
1239 msg.msg_flags = uap->flags;
1240#ifdef COMPAT_OLDSOCK
1241 msg.msg_flags &= ~MSG_COMPAT;
1242#endif
1243 uiov = msg.msg_iov;
1244 msg.msg_iov = iov;
1245 error = recvit(td, uap->s, &msg, NULL);
1246 if (error == 0) {
1247 msg.msg_iov = uiov;
1248 error = copyout(&msg, uap->msg, sizeof(msg));
1249 }
1250 free(iov, M_IOV);
1251 return (error);
1252}
1253
1254/* ARGSUSED */
1255int
1256shutdown(td, uap)
1257 struct thread *td;
1258 struct shutdown_args /* {
1259 int s;
1260 int how;
1261 } */ *uap;
1262{
1263 struct socket *so;
1264 struct file *fp;
1265 int error;
1266
1267 AUDIT_ARG_FD(uap->s);
1268 error = getsock(td->td_proc->p_fd, uap->s, &fp, NULL);
1269 if (error == 0) {
1270 so = fp->f_data;
1271 error = soshutdown(so, uap->how);
1272 fdrop(fp, td);
1273 }
1274 return (error);
1275}
1276
1277/* ARGSUSED */
1278int
1279setsockopt(td, uap)
1280 struct thread *td;
1281 struct setsockopt_args /* {
1282 int s;
1283 int level;
1284 int name;
1285 caddr_t val;
1286 int valsize;
1287 } */ *uap;
1288{
1289
1290 return (kern_setsockopt(td, uap->s, uap->level, uap->name,
1291 uap->val, UIO_USERSPACE, uap->valsize));
1292}
1293
1294int
1295kern_setsockopt(td, s, level, name, val, valseg, valsize)
1296 struct thread *td;
1297 int s;
1298 int level;
1299 int name;
1300 void *val;
1301 enum uio_seg valseg;
1302 socklen_t valsize;
1303{
1304 int error;
1305 struct socket *so;
1306 struct file *fp;
1307 struct sockopt sopt;
1308
1309 if (val == NULL && valsize != 0)
1310 return (EFAULT);
1311 if ((int)valsize < 0)
1312 return (EINVAL);
1313
1314 sopt.sopt_dir = SOPT_SET;
1315 sopt.sopt_level = level;
1316 sopt.sopt_name = name;
1317 sopt.sopt_val = val;
1318 sopt.sopt_valsize = valsize;
1319 switch (valseg) {
1320 case UIO_USERSPACE:
1321 sopt.sopt_td = td;
1322 break;
1323 case UIO_SYSSPACE:
1324 sopt.sopt_td = NULL;
1325 break;
1326 default:
1327 panic("kern_setsockopt called with bad valseg");
1328 }
1329
1330 AUDIT_ARG_FD(s);
1331 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
1332 if (error == 0) {
1333 so = fp->f_data;
1334 CURVNET_SET(so->so_vnet);
1335 error = sosetopt(so, &sopt);
1336 CURVNET_RESTORE();
1337 fdrop(fp, td);
1338 }
1339 return(error);
1340}
1341
1342/* ARGSUSED */
1343int
1344getsockopt(td, uap)
1345 struct thread *td;
1346 struct getsockopt_args /* {
1347 int s;
1348 int level;
1349 int name;
1350 void * __restrict val;
1351 socklen_t * __restrict avalsize;
1352 } */ *uap;
1353{
1354 socklen_t valsize;
1355 int error;
1356
1357 if (uap->val) {
1358 error = copyin(uap->avalsize, &valsize, sizeof (valsize));
1359 if (error)
1360 return (error);
1361 }
1362
1363 error = kern_getsockopt(td, uap->s, uap->level, uap->name,
1364 uap->val, UIO_USERSPACE, &valsize);
1365
1366 if (error == 0)
1367 error = copyout(&valsize, uap->avalsize, sizeof (valsize));
1368 return (error);
1369}
1370
1371/*
1372 * Kernel version of getsockopt.
1373 * optval can be a userland or userspace. optlen is always a kernel pointer.
1374 */
1375int
1376kern_getsockopt(td, s, level, name, val, valseg, valsize)
1377 struct thread *td;
1378 int s;
1379 int level;
1380 int name;
1381 void *val;
1382 enum uio_seg valseg;
1383 socklen_t *valsize;
1384{
1385 int error;
1386 struct socket *so;
1387 struct file *fp;
1388 struct sockopt sopt;
1389
1390 if (val == NULL)
1391 *valsize = 0;
1392 if ((int)*valsize < 0)
1393 return (EINVAL);
1394
1395 sopt.sopt_dir = SOPT_GET;
1396 sopt.sopt_level = level;
1397 sopt.sopt_name = name;
1398 sopt.sopt_val = val;
1399 sopt.sopt_valsize = (size_t)*valsize; /* checked non-negative above */
1400 switch (valseg) {
1401 case UIO_USERSPACE:
1402 sopt.sopt_td = td;
1403 break;
1404 case UIO_SYSSPACE:
1405 sopt.sopt_td = NULL;
1406 break;
1407 default:
1408 panic("kern_getsockopt called with bad valseg");
1409 }
1410
1411 AUDIT_ARG_FD(s);
1412 error = getsock(td->td_proc->p_fd, s, &fp, NULL);
1413 if (error == 0) {
1414 so = fp->f_data;
1415 CURVNET_SET(so->so_vnet);
1416 error = sogetopt(so, &sopt);
1417 CURVNET_RESTORE();
1418 *valsize = sopt.sopt_valsize;
1419 fdrop(fp, td);
1420 }
1421 return (error);
1422}
1423
1424/*
1425 * getsockname1() - Get socket name.
1426 */
1427/* ARGSUSED */
1428static int
1429getsockname1(td, uap, compat)
1430 struct thread *td;
1431 struct getsockname_args /* {
1432 int fdes;
1433 struct sockaddr * __restrict asa;
1434 socklen_t * __restrict alen;
1435 } */ *uap;
1436 int compat;
1437{
1438 struct sockaddr *sa;
1439 socklen_t len;
1440 int error;
1441
1442 error = copyin(uap->alen, &len, sizeof(len));
1443 if (error)
1444 return (error);
1445
1446 error = kern_getsockname(td, uap->fdes, &sa, &len);
1447 if (error)
1448 return (error);
1449
1450 if (len != 0) {
1451#ifdef COMPAT_OLDSOCK
1452 if (compat)
1453 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1454#endif
1455 error = copyout(sa, uap->asa, (u_int)len);
1456 }
1457 free(sa, M_SONAME);
1458 if (error == 0)
1459 error = copyout(&len, uap->alen, sizeof(len));
1460 return (error);
1461}
1462
1463int
1464kern_getsockname(struct thread *td, int fd, struct sockaddr **sa,
1465 socklen_t *alen)
1466{
1467 struct socket *so;
1468 struct file *fp;
1469 socklen_t len;
1470 int error;
1471
1472 if (*alen < 0)
1473 return (EINVAL);
1474
1475 AUDIT_ARG_FD(fd);
1476 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
1477 if (error)
1478 return (error);
1479 so = fp->f_data;
1480 *sa = NULL;
1481 CURVNET_SET(so->so_vnet);
1482 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, sa);
1483 CURVNET_RESTORE();
1484 if (error)
1485 goto bad;
1486 if (*sa == NULL)
1487 len = 0;
1488 else
1489 len = MIN(*alen, (*sa)->sa_len);
1490 *alen = len;
1491#ifdef KTRACE
1492 if (KTRPOINT(td, KTR_STRUCT))
1493 ktrsockaddr(*sa);
1494#endif
1495bad:
1496 fdrop(fp, td);
1497 if (error && *sa) {
1498 free(*sa, M_SONAME);
1499 *sa = NULL;
1500 }
1501 return (error);
1502}
1503
1504int
1505getsockname(td, uap)
1506 struct thread *td;
1507 struct getsockname_args *uap;
1508{
1509
1510 return (getsockname1(td, uap, 0));
1511}
1512
1513#ifdef COMPAT_OLDSOCK
1514int
1515ogetsockname(td, uap)
1516 struct thread *td;
1517 struct getsockname_args *uap;
1518{
1519
1520 return (getsockname1(td, uap, 1));
1521}
1522#endif /* COMPAT_OLDSOCK */
1523
1524/*
1525 * getpeername1() - Get name of peer for connected socket.
1526 */
1527/* ARGSUSED */
1528static int
1529getpeername1(td, uap, compat)
1530 struct thread *td;
1531 struct getpeername_args /* {
1532 int fdes;
1533 struct sockaddr * __restrict asa;
1534 socklen_t * __restrict alen;
1535 } */ *uap;
1536 int compat;
1537{
1538 struct sockaddr *sa;
1539 socklen_t len;
1540 int error;
1541
1542 error = copyin(uap->alen, &len, sizeof (len));
1543 if (error)
1544 return (error);
1545
1546 error = kern_getpeername(td, uap->fdes, &sa, &len);
1547 if (error)
1548 return (error);
1549
1550 if (len != 0) {
1551#ifdef COMPAT_OLDSOCK
1552 if (compat)
1553 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1554#endif
1555 error = copyout(sa, uap->asa, (u_int)len);
1556 }
1557 free(sa, M_SONAME);
1558 if (error == 0)
1559 error = copyout(&len, uap->alen, sizeof(len));
1560 return (error);
1561}
1562
1563int
1564kern_getpeername(struct thread *td, int fd, struct sockaddr **sa,
1565 socklen_t *alen)
1566{
1567 struct socket *so;
1568 struct file *fp;
1569 socklen_t len;
1570 int error;
1571
1572 if (*alen < 0)
1573 return (EINVAL);
1574
1575 AUDIT_ARG_FD(fd);
1576 error = getsock(td->td_proc->p_fd, fd, &fp, NULL);
1577 if (error)
1578 return (error);
1579 so = fp->f_data;
1580 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1581 error = ENOTCONN;
1582 goto done;
1583 }
1584 *sa = NULL;
1585 CURVNET_SET(so->so_vnet);
1586 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, sa);
1587 CURVNET_RESTORE();
1588 if (error)
1589 goto bad;
1590 if (*sa == NULL)
1591 len = 0;
1592 else
1593 len = MIN(*alen, (*sa)->sa_len);
1594 *alen = len;
1595#ifdef KTRACE
1596 if (KTRPOINT(td, KTR_STRUCT))
1597 ktrsockaddr(*sa);
1598#endif
1599bad:
1600 if (error && *sa) {
1601 free(*sa, M_SONAME);
1602 *sa = NULL;
1603 }
1604done:
1605 fdrop(fp, td);
1606 return (error);
1607}
1608
1609int
1610getpeername(td, uap)
1611 struct thread *td;
1612 struct getpeername_args *uap;
1613{
1614
1615 return (getpeername1(td, uap, 0));
1616}
1617
1618#ifdef COMPAT_OLDSOCK
1619int
1620ogetpeername(td, uap)
1621 struct thread *td;
1622 struct ogetpeername_args *uap;
1623{
1624
1625 /* XXX uap should have type `getpeername_args *' to begin with. */
1626 return (getpeername1(td, (struct getpeername_args *)uap, 1));
1627}
1628#endif /* COMPAT_OLDSOCK */
1629
1630int
1631sockargs(mp, buf, buflen, type)
1632 struct mbuf **mp;
1633 caddr_t buf;
1634 int buflen, type;
1635{
1636 struct sockaddr *sa;
1637 struct mbuf *m;
1638 int error;
1639
1640 if ((u_int)buflen > MLEN) {
1641#ifdef COMPAT_OLDSOCK
1642 if (type == MT_SONAME && (u_int)buflen <= 112)
1643 buflen = MLEN; /* unix domain compat. hack */
1644 else
1645#endif
1646 if ((u_int)buflen > MCLBYTES)
1647 return (EINVAL);
1648 }
1649 m = m_get(M_WAIT, type);
1650 if ((u_int)buflen > MLEN)
1651 MCLGET(m, M_WAIT);
1652 m->m_len = buflen;
1653 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1654 if (error)
1655 (void) m_free(m);
1656 else {
1657 *mp = m;
1658 if (type == MT_SONAME) {
1659 sa = mtod(m, struct sockaddr *);
1660
1661#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1662 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1663 sa->sa_family = sa->sa_len;
1664#endif
1665 sa->sa_len = buflen;
1666 }
1667 }
1668 return (error);
1669}
1670
1671int
1672getsockaddr(namp, uaddr, len)
1673 struct sockaddr **namp;
1674 caddr_t uaddr;
1675 size_t len;
1676{
1677 struct sockaddr *sa;
1678 int error;
1679
1680 if (len > SOCK_MAXADDRLEN)
1681 return (ENAMETOOLONG);
1682 if (len < offsetof(struct sockaddr, sa_data[0]))
1683 return (EINVAL);
1684 sa = malloc(len, M_SONAME, M_WAITOK);
1685 error = copyin(uaddr, sa, len);
1686 if (error) {
1687 free(sa, M_SONAME);
1688 } else {
1689#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1690 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1691 sa->sa_family = sa->sa_len;
1692#endif
1693 sa->sa_len = len;
1694 *namp = sa;
1695 }
1696 return (error);
1697}
1698
1699#include <sys/condvar.h>
1700
1701struct sendfile_sync {
1702 struct mtx mtx;
1703 struct cv cv;
1704 unsigned count;
1705};
1706
1707/*
1708 * Detach mapped page and release resources back to the system.
1709 */
1710void
1711sf_buf_mext(void *addr, void *args)
1712{
1713 vm_page_t m;
1714 struct sendfile_sync *sfs;
1715
1716 m = sf_buf_page(args);
1717 sf_buf_free(args);
1718 vm_page_lock(m);
1719 vm_page_unwire(m, 0);
1720 /*
1721 * Check for the object going away on us. This can
1722 * happen since we don't hold a reference to it.
1723 * If so, we're responsible for freeing the page.
1724 */
1725 if (m->wire_count == 0 && m->object == NULL)
1726 vm_page_free(m);
1727 vm_page_unlock(m);
1728 if (addr == NULL)
1729 return;
1730 sfs = addr;
1731 mtx_lock(&sfs->mtx);
1732 KASSERT(sfs->count> 0, ("Sendfile sync botchup count == 0"));
1733 if (--sfs->count == 0)
1734 cv_signal(&sfs->cv);
1735 mtx_unlock(&sfs->mtx);
1736}
1737
1738/*
1739 * sendfile(2)
1740 *
1741 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1742 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1743 *
1744 * Send a file specified by 'fd' and starting at 'offset' to a socket
1745 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1746 * 0. Optionally add a header and/or trailer to the socket output. If
1747 * specified, write the total number of bytes sent into *sbytes.
1748 */
1749int
1750sendfile(struct thread *td, struct sendfile_args *uap)
1751{
1752
1753 return (do_sendfile(td, uap, 0));
1754}
1755
1756static int
1757do_sendfile(struct thread *td, struct sendfile_args *uap, int compat)
1758{
1759 struct sf_hdtr hdtr;
1760 struct uio *hdr_uio, *trl_uio;
1761 int error;
1762
1763 hdr_uio = trl_uio = NULL;
1764
1765 if (uap->hdtr != NULL) {
1766 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1767 if (error)
1768 goto out;
1769 if (hdtr.headers != NULL) {
1770 error = copyinuio(hdtr.headers, hdtr.hdr_cnt, &hdr_uio);
1771 if (error)
1772 goto out;
1773 }
1774 if (hdtr.trailers != NULL) {
1775 error = copyinuio(hdtr.trailers, hdtr.trl_cnt, &trl_uio);
1776 if (error)
1777 goto out;
1778
1779 }
1780 }
1781
1782 error = kern_sendfile(td, uap, hdr_uio, trl_uio, compat);
1783out:
1784 if (hdr_uio)
1785 free(hdr_uio, M_IOV);
1786 if (trl_uio)
1787 free(trl_uio, M_IOV);
1788 return (error);
1789}
1790
1791#ifdef COMPAT_FREEBSD4
1792int
1793freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1794{
1795 struct sendfile_args args;
1796
1797 args.fd = uap->fd;
1798 args.s = uap->s;
1799 args.offset = uap->offset;
1800 args.nbytes = uap->nbytes;
1801 args.hdtr = uap->hdtr;
1802 args.sbytes = uap->sbytes;
1803 args.flags = uap->flags;
1804
1805 return (do_sendfile(td, &args, 1));
1806}
1807#endif /* COMPAT_FREEBSD4 */
1808
1809int
1810kern_sendfile(struct thread *td, struct sendfile_args *uap,
1811 struct uio *hdr_uio, struct uio *trl_uio, int compat)
1812{
1813 struct file *sock_fp;
1814 struct vnode *vp;
1815 struct vm_object *obj = NULL;
1816 struct socket *so = NULL;
1817 struct mbuf *m = NULL;
1818 struct sf_buf *sf;
1819 struct vm_page *pg;
1820 off_t off, xfsize, fsbytes = 0, sbytes = 0, rem = 0;
1821 int error, hdrlen = 0, mnw = 0;
1822 int vfslocked;
1823 struct sendfile_sync *sfs = NULL;
1824
1825 /*
1826 * The file descriptor must be a regular file and have a
1827 * backing VM object.
1828 * File offset must be positive. If it goes beyond EOF
1829 * we send only the header/trailer and no payload data.
1830 */
1831 AUDIT_ARG_FD(uap->fd);
1832 if ((error = fgetvp_read(td, uap->fd, &vp)) != 0)
1833 goto out;
1834 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1835 vn_lock(vp, LK_SHARED | LK_RETRY);
1836 if (vp->v_type == VREG) {
1837 obj = vp->v_object;
1838 if (obj != NULL) {
1839 /*
1840 * Temporarily increase the backing VM
1841 * object's reference count so that a forced
1842 * reclamation of its vnode does not
1843 * immediately destroy it.
1844 */
1845 VM_OBJECT_LOCK(obj);
1846 if ((obj->flags & OBJ_DEAD) == 0) {
1847 vm_object_reference_locked(obj);
1848 VM_OBJECT_UNLOCK(obj);
1849 } else {
1850 VM_OBJECT_UNLOCK(obj);
1851 obj = NULL;
1852 }
1853 }
1854 }
1855 VOP_UNLOCK(vp, 0);
1856 VFS_UNLOCK_GIANT(vfslocked);
1857 if (obj == NULL) {
1858 error = EINVAL;
1859 goto out;
1860 }
1861 if (uap->offset < 0) {
1862 error = EINVAL;
1863 goto out;
1864 }
1865
1866 /*
1867 * The socket must be a stream socket and connected.
1868 * Remember if it a blocking or non-blocking socket.
1869 */
1870 if ((error = getsock(td->td_proc->p_fd, uap->s, &sock_fp,
1871 NULL)) != 0)
1872 goto out;
1873 so = sock_fp->f_data;
1874 if (so->so_type != SOCK_STREAM) {
1875 error = EINVAL;
1876 goto out;
1877 }
1878 if ((so->so_state & SS_ISCONNECTED) == 0) {
1879 error = ENOTCONN;
1880 goto out;
1881 }
1882 /*
1883 * Do not wait on memory allocations but return ENOMEM for
1884 * caller to retry later.
1885 * XXX: Experimental.
1886 */
1887 if (uap->flags & SF_MNOWAIT)
1888 mnw = 1;
1889
1890 if (uap->flags & SF_SYNC) {
1891 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK);
1892 memset(sfs, 0, sizeof *sfs);
1893 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
1894 cv_init(&sfs->cv, "sendfile");
1895 }
1896
1897#ifdef MAC
1898 error = mac_socket_check_send(td->td_ucred, so);
1899 if (error)
1900 goto out;
1901#endif
1902
1903 /* If headers are specified copy them into mbufs. */
1904 if (hdr_uio != NULL) {
1905 hdr_uio->uio_td = td;
1906 hdr_uio->uio_rw = UIO_WRITE;
1907 if (hdr_uio->uio_resid > 0) {
1908 /*
1909 * In FBSD < 5.0 the nbytes to send also included
1910 * the header. If compat is specified subtract the
1911 * header size from nbytes.
1912 */
1913 if (compat) {
1914 if (uap->nbytes > hdr_uio->uio_resid)
1915 uap->nbytes -= hdr_uio->uio_resid;
1916 else
1917 uap->nbytes = 0;
1918 }
1919 m = m_uiotombuf(hdr_uio, (mnw ? M_NOWAIT : M_WAITOK),
1920 0, 0, 0);
1921 if (m == NULL) {
1922 error = mnw ? EAGAIN : ENOBUFS;
1923 goto out;
1924 }
1925 hdrlen = m_length(m, NULL);
1926 }
1927 }
1928
1929 /*
1930 * Protect against multiple writers to the socket.
1931 *
1932 * XXXRW: Historically this has assumed non-interruptibility, so now
1933 * we implement that, but possibly shouldn't.
1934 */
1935 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
1936
1937 /*
1938 * Loop through the pages of the file, starting with the requested
1939 * offset. Get a file page (do I/O if necessary), map the file page
1940 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1941 * it on the socket.
1942 * This is done in two loops. The inner loop turns as many pages
1943 * as it can, up to available socket buffer space, without blocking
1944 * into mbufs to have it bulk delivered into the socket send buffer.
1945 * The outer loop checks the state and available space of the socket
1946 * and takes care of the overall progress.
1947 */
1948 for (off = uap->offset, rem = uap->nbytes; ; ) {
1949 int loopbytes = 0;
1950 int space = 0;
1951 int done = 0;
1952
1953 /*
1954 * Check the socket state for ongoing connection,
1955 * no errors and space in socket buffer.
1956 * If space is low allow for the remainder of the
1957 * file to be processed if it fits the socket buffer.
1958 * Otherwise block in waiting for sufficient space
1959 * to proceed, or if the socket is nonblocking, return
1960 * to userland with EAGAIN while reporting how far
1961 * we've come.
1962 * We wait until the socket buffer has significant free
1963 * space to do bulk sends. This makes good use of file
1964 * system read ahead and allows packet segmentation
1965 * offloading hardware to take over lots of work. If
1966 * we were not careful here we would send off only one
1967 * sfbuf at a time.
1968 */
1969 SOCKBUF_LOCK(&so->so_snd);
1970 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
1971 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
1972retry_space:
1973 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1974 error = EPIPE;
1975 SOCKBUF_UNLOCK(&so->so_snd);
1976 goto done;
1977 } else if (so->so_error) {
1978 error = so->so_error;
1979 so->so_error = 0;
1980 SOCKBUF_UNLOCK(&so->so_snd);
1981 goto done;
1982 }
1983 space = sbspace(&so->so_snd);
1984 if (space < rem &&
1985 (space <= 0 ||
1986 space < so->so_snd.sb_lowat)) {
1987 if (so->so_state & SS_NBIO) {
1988 SOCKBUF_UNLOCK(&so->so_snd);
1989 error = EAGAIN;
1990 goto done;
1991 }
1992 /*
1993 * sbwait drops the lock while sleeping.
1994 * When we loop back to retry_space the
1995 * state may have changed and we retest
1996 * for it.
1997 */
1998 error = sbwait(&so->so_snd);
1999 /*
2000 * An error from sbwait usually indicates that we've
2001 * been interrupted by a signal. If we've sent anything
2002 * then return bytes sent, otherwise return the error.
2003 */
2004 if (error) {
2005 SOCKBUF_UNLOCK(&so->so_snd);
2006 goto done;
2007 }
2008 goto retry_space;
2009 }
2010 SOCKBUF_UNLOCK(&so->so_snd);
2011
2012 /*
2013 * Reduce space in the socket buffer by the size of
2014 * the header mbuf chain.
2015 * hdrlen is set to 0 after the first loop.
2016 */
2017 space -= hdrlen;
2018
2019 /*
2020 * Loop and construct maximum sized mbuf chain to be bulk
2021 * dumped into socket buffer.
2022 */
2023 while (space > loopbytes) {
2024 vm_pindex_t pindex;
2025 vm_offset_t pgoff;
2026 struct mbuf *m0;
2027
2028 VM_OBJECT_LOCK(obj);
2029 /*
2030 * Calculate the amount to transfer.
2031 * Not to exceed a page, the EOF,
2032 * or the passed in nbytes.
2033 */
2034 pgoff = (vm_offset_t)(off & PAGE_MASK);
2035 xfsize = omin(PAGE_SIZE - pgoff,
2036 obj->un_pager.vnp.vnp_size - uap->offset -
2037 fsbytes - loopbytes);
2038 if (uap->nbytes)
2039 rem = (uap->nbytes - fsbytes - loopbytes);
2040 else
2041 rem = obj->un_pager.vnp.vnp_size -
2042 uap->offset - fsbytes - loopbytes;
2043 xfsize = omin(rem, xfsize);
2044 xfsize = omin(space - loopbytes, xfsize);
2045 if (xfsize <= 0) {
2046 VM_OBJECT_UNLOCK(obj);
2047 done = 1; /* all data sent */
2048 break;
2049 }
2050
2051 /*
2052 * Attempt to look up the page. Allocate
2053 * if not found or wait and loop if busy.
2054 */
2055 pindex = OFF_TO_IDX(off);
2056 pg = vm_page_grab(obj, pindex, VM_ALLOC_NOBUSY |
2057 VM_ALLOC_NORMAL | VM_ALLOC_WIRED | VM_ALLOC_RETRY);
2058
2059 /*
2060 * Check if page is valid for what we need,
2061 * otherwise initiate I/O.
2062 * If we already turned some pages into mbufs,
2063 * send them off before we come here again and
2064 * block.
2065 */
2066 if (pg->valid && vm_page_is_valid(pg, pgoff, xfsize))
2067 VM_OBJECT_UNLOCK(obj);
2068 else if (m != NULL)
2069 error = EAGAIN; /* send what we already got */
2070 else if (uap->flags & SF_NODISKIO)
2071 error = EBUSY;
2072 else {
2073 int bsize, resid;
2074
2075 /*
2076 * Ensure that our page is still around
2077 * when the I/O completes.
2078 */
2079 vm_page_io_start(pg);
2080 VM_OBJECT_UNLOCK(obj);
2081
2082 /*
2083 * Get the page from backing store.
2084 */
2085 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2086 error = vn_lock(vp, LK_SHARED);
2087 if (error != 0)
2088 goto after_read;
2089 bsize = vp->v_mount->mnt_stat.f_iosize;
2090
2091 /*
2092 * XXXMAC: Because we don't have fp->f_cred
2093 * here, we pass in NOCRED. This is probably
2094 * wrong, but is consistent with our original
2095 * implementation.
2096 */
2097 error = vn_rdwr(UIO_READ, vp, NULL, MAXBSIZE,
2098 trunc_page(off), UIO_NOCOPY, IO_NODELOCKED |
2099 IO_VMIO | ((MAXBSIZE / bsize) << IO_SEQSHIFT),
2100 td->td_ucred, NOCRED, &resid, td);
2101 VOP_UNLOCK(vp, 0);
2102 after_read:
2103 VFS_UNLOCK_GIANT(vfslocked);
2104 VM_OBJECT_LOCK(obj);
2105 vm_page_io_finish(pg);
2106 if (!error)
2107 VM_OBJECT_UNLOCK(obj);
2108 mbstat.sf_iocnt++;
2109 }
2110 if (error) {
2111 vm_page_lock(pg);
2112 vm_page_unwire(pg, 0);
2113 /*
2114 * See if anyone else might know about
2115 * this page. If not and it is not valid,
2116 * then free it.
2117 */
2118 if (pg->wire_count == 0 && pg->valid == 0 &&
2119 pg->busy == 0 && !(pg->oflags & VPO_BUSY) &&
2120 pg->hold_count == 0)
2121 vm_page_free(pg);
2122 vm_page_unlock(pg);
2123 VM_OBJECT_UNLOCK(obj);
2124 if (error == EAGAIN)
2125 error = 0; /* not a real error */
2126 break;
2127 }
2128
2129 /*
2130 * Get a sendfile buf. We usually wait as long
2131 * as necessary, but this wait can be interrupted.
2132 */
2133 if ((sf = sf_buf_alloc(pg,
2134 (mnw ? SFB_NOWAIT : SFB_CATCH))) == NULL) {
2135 mbstat.sf_allocfail++;
2136 vm_page_lock(pg);
2137 vm_page_unwire(pg, 0);
2138 KASSERT(pg->object != NULL,
2139 ("kern_sendfile: object disappeared"));
2140 vm_page_unlock(pg);
2141 error = (mnw ? EAGAIN : EINTR);
2142 break;
2143 }
2144
2145 /*
2146 * Get an mbuf and set it up as having
2147 * external storage.
2148 */
2149 m0 = m_get((mnw ? M_NOWAIT : M_WAITOK), MT_DATA);
2150 if (m0 == NULL) {
2151 error = (mnw ? EAGAIN : ENOBUFS);
2152 sf_buf_mext((void *)sf_buf_kva(sf), sf);
2153 break;
2154 }
2155 MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, sf_buf_mext,
2156 sfs, sf, M_RDONLY, EXT_SFBUF);
2157 m0->m_data = (char *)sf_buf_kva(sf) + pgoff;
2158 m0->m_len = xfsize;
2159
2160 /* Append to mbuf chain. */
2161 if (m != NULL)
2162 m_cat(m, m0);
2163 else
2164 m = m0;
2165
2166 /* Keep track of bits processed. */
2167 loopbytes += xfsize;
2168 off += xfsize;
2169
2170 if (sfs != NULL) {
2171 mtx_lock(&sfs->mtx);
2172 sfs->count++;
2173 mtx_unlock(&sfs->mtx);
2174 }
2175 }
2176
2177 /* Add the buffer chain to the socket buffer. */
2178 if (m != NULL) {
2179 int mlen, err;
2180
2181 mlen = m_length(m, NULL);
2182 SOCKBUF_LOCK(&so->so_snd);
2183 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2184 error = EPIPE;
2185 SOCKBUF_UNLOCK(&so->so_snd);
2186 goto done;
2187 }
2188 SOCKBUF_UNLOCK(&so->so_snd);
2189 CURVNET_SET(so->so_vnet);
2190 /* Avoid error aliasing. */
2191 err = (*so->so_proto->pr_usrreqs->pru_send)
2192 (so, 0, m, NULL, NULL, td);
2193 CURVNET_RESTORE();
2194 if (err == 0) {
2195 /*
2196 * We need two counters to get the
2197 * file offset and nbytes to send
2198 * right:
2199 * - sbytes contains the total amount
2200 * of bytes sent, including headers.
2201 * - fsbytes contains the total amount
2202 * of bytes sent from the file.
2203 */
2204 sbytes += mlen;
2205 fsbytes += mlen;
2206 if (hdrlen) {
2207 fsbytes -= hdrlen;
2208 hdrlen = 0;
2209 }
2210 } else if (error == 0)
2211 error = err;
2212 m = NULL; /* pru_send always consumes */
2213 }
2214
2215 /* Quit outer loop on error or when we're done. */
2216 if (done)
2217 break;
2218 if (error)
2219 goto done;
2220 }
2221
2222 /*
2223 * Send trailers. Wimp out and use writev(2).
2224 */
2225 if (trl_uio != NULL) {
2226 sbunlock(&so->so_snd);
2227 error = kern_writev(td, uap->s, trl_uio);
2228 if (error == 0)
2229 sbytes += td->td_retval[0];
2230 goto out;
2231 }
2232
2233done:
2234 sbunlock(&so->so_snd);
2235out:
2236 /*
2237 * If there was no error we have to clear td->td_retval[0]
2238 * because it may have been set by writev.
2239 */
2240 if (error == 0) {
2241 td->td_retval[0] = 0;
2242 }
2243 if (uap->sbytes != NULL) {
2244 copyout(&sbytes, uap->sbytes, sizeof(off_t));
2245 }
2246 if (obj != NULL)
2247 vm_object_deallocate(obj);
2248 if (vp != NULL) {
2249 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2250 vrele(vp);
2251 VFS_UNLOCK_GIANT(vfslocked);
2252 }
2253 if (so)
2254 fdrop(sock_fp, td);
2255 if (m)
2256 m_freem(m);
2257
2258 if (sfs != NULL) {
2259 mtx_lock(&sfs->mtx);
2260 if (sfs->count != 0)
2261 cv_wait(&sfs->cv, &sfs->mtx);
2262 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
2263 cv_destroy(&sfs->cv);
2264 mtx_destroy(&sfs->mtx);
2265 free(sfs, M_TEMP);
2266 }
2267
2268 if (error == ERESTART)
2269 error = EINTR;
2270
2271 return (error);
2272}
2273
2274/*
2275 * SCTP syscalls.
2276 * Functionality only compiled in if SCTP is defined in the kernel Makefile,
2277 * otherwise all return EOPNOTSUPP.
2278 * XXX: We should make this loadable one day.
2279 */
2280int
2281sctp_peeloff(td, uap)
2282 struct thread *td;
2283 struct sctp_peeloff_args /* {
2284 int sd;
2285 caddr_t name;
2286 } */ *uap;
2287{
2288#if (defined(INET) || defined(INET6)) && defined(SCTP)
2289 struct filedesc *fdp;
2290 struct file *nfp = NULL;
2291 int error;
2292 struct socket *head, *so;
2293 int fd;
2294 u_int fflag;
2295
2296 fdp = td->td_proc->p_fd;
2297 AUDIT_ARG_FD(uap->sd);
2298 error = fgetsock(td, uap->sd, &head, &fflag);
2299 if (error)
2300 goto done2;
2301 error = sctp_can_peel_off(head, (sctp_assoc_t)uap->name);
2302 if (error)
2303 goto done2;
2304 /*
2305 * At this point we know we do have a assoc to pull
2306 * we proceed to get the fd setup. This may block
2307 * but that is ok.
2308 */
2309
2310 error = falloc(td, &nfp, &fd);
2311 if (error)
2312 goto done;
2313 td->td_retval[0] = fd;
2314
2315 CURVNET_SET(head->so_vnet);
2316 so = sonewconn(head, SS_ISCONNECTED);
2317 if (so == NULL)
2318 goto noconnection;
2319 /*
2320 * Before changing the flags on the socket, we have to bump the
2321 * reference count. Otherwise, if the protocol calls sofree(),
2322 * the socket will be released due to a zero refcount.
2323 */
2324 SOCK_LOCK(so);
2325 soref(so); /* file descriptor reference */
2326 SOCK_UNLOCK(so);
2327
2328 ACCEPT_LOCK();
2329
2330 TAILQ_REMOVE(&head->so_comp, so, so_list);
2331 head->so_qlen--;
2332 so->so_state |= (head->so_state & SS_NBIO);
2333 so->so_state &= ~SS_NOFDREF;
2334 so->so_qstate &= ~SQ_COMP;
2335 so->so_head = NULL;
2336 ACCEPT_UNLOCK();
2337 finit(nfp, fflag, DTYPE_SOCKET, so, &socketops);
2338 error = sctp_do_peeloff(head, so, (sctp_assoc_t)uap->name);
2339 if (error)
2340 goto noconnection;
2341 if (head->so_sigio != NULL)
2342 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
2343
2344noconnection:
2345 /*
2346 * close the new descriptor, assuming someone hasn't ripped it
2347 * out from under us.
2348 */
2349 if (error)
2350 fdclose(fdp, nfp, fd, td);
2351
2352 /*
2353 * Release explicitly held references before returning.
2354 */
2355 CURVNET_RESTORE();
2356done:
2357 if (nfp != NULL)
2358 fdrop(nfp, td);
2359 fputsock(head);
2360done2:
2361 return (error);
2362#else /* SCTP */
2363 return (EOPNOTSUPP);
2364#endif /* SCTP */
2365}
2366
2367int
2368sctp_generic_sendmsg (td, uap)
2369 struct thread *td;
2370 struct sctp_generic_sendmsg_args /* {
2371 int sd,
2372 caddr_t msg,
2373 int mlen,
2374 caddr_t to,
2375 __socklen_t tolen,
2376 struct sctp_sndrcvinfo *sinfo,
2377 int flags
2378 } */ *uap;
2379{
2380#if (defined(INET) || defined(INET6)) && defined(SCTP)
2381 struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL;
2382 struct socket *so;
2383 struct file *fp = NULL;
2384 int use_rcvinfo = 1;
2385 int error = 0, len;
2386 struct sockaddr *to = NULL;
2387#ifdef KTRACE
2388 struct uio *ktruio = NULL;
2389#endif
2390 struct uio auio;
2391 struct iovec iov[1];
2392
2393 if (uap->sinfo) {
2394 error = copyin(uap->sinfo, &sinfo, sizeof (sinfo));
2395 if (error)
2396 return (error);
2397 u_sinfo = &sinfo;
2398 }
2399 if (uap->tolen) {
2400 error = getsockaddr(&to, uap->to, uap->tolen);
2401 if (error) {
2402 to = NULL;
2403 goto sctp_bad2;
2404 }
2405 }
2406
2407 AUDIT_ARG_FD(uap->sd);
2408 error = getsock(td->td_proc->p_fd, uap->sd, &fp, NULL);
2409 if (error)
2410 goto sctp_bad;
2411#ifdef KTRACE
2412 if (KTRPOINT(td, KTR_STRUCT))
2413 ktrsockaddr(to);
2414#endif
2415
2416 iov[0].iov_base = uap->msg;
2417 iov[0].iov_len = uap->mlen;
2418
2419 so = (struct socket *)fp->f_data;
2420#ifdef MAC
2421 error = mac_socket_check_send(td->td_ucred, so);
2422 if (error)
2423 goto sctp_bad;
2424#endif /* MAC */
2425
2426 auio.uio_iov = iov;
2427 auio.uio_iovcnt = 1;
2428 auio.uio_segflg = UIO_USERSPACE;
2429 auio.uio_rw = UIO_WRITE;
2430 auio.uio_td = td;
2431 auio.uio_offset = 0; /* XXX */
2432 auio.uio_resid = 0;
2433 len = auio.uio_resid = uap->mlen;
2434 CURVNET_SET(so->so_vnet);
2435 error = sctp_lower_sosend(so, to, &auio,
2436 (struct mbuf *)NULL, (struct mbuf *)NULL,
2437 uap->flags, use_rcvinfo, u_sinfo, td);
2438 CURVNET_RESTORE();
2439 if (error) {
2440 if (auio.uio_resid != len && (error == ERESTART ||
2441 error == EINTR || error == EWOULDBLOCK))
2442 error = 0;
2443 /* Generation of SIGPIPE can be controlled per socket. */
2444 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
2445 !(uap->flags & MSG_NOSIGNAL)) {
2446 PROC_LOCK(td->td_proc);
2447 psignal(td->td_proc, SIGPIPE);
2448 PROC_UNLOCK(td->td_proc);
2449 }
2450 }
2451 if (error == 0)
2452 td->td_retval[0] = len - auio.uio_resid;
2453#ifdef KTRACE
2454 if (ktruio != NULL) {
2455 ktruio->uio_resid = td->td_retval[0];
2456 ktrgenio(uap->sd, UIO_WRITE, ktruio, error);
2457 }
2458#endif /* KTRACE */
2459sctp_bad:
2460 if (fp)
2461 fdrop(fp, td);
2462sctp_bad2:
2463 if (to)
2464 free(to, M_SONAME);
2465 return (error);
2466#else /* SCTP */
2467 return (EOPNOTSUPP);
2468#endif /* SCTP */
2469}
2470
2471int
2472sctp_generic_sendmsg_iov(td, uap)
2473 struct thread *td;
2474 struct sctp_generic_sendmsg_iov_args /* {
2475 int sd,
2476 struct iovec *iov,
2477 int iovlen,
2478 caddr_t to,
2479 __socklen_t tolen,
2480 struct sctp_sndrcvinfo *sinfo,
2481 int flags
2482 } */ *uap;
2483{
2484#if (defined(INET) || defined(INET6)) && defined(SCTP)
2485 struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL;
2486 struct socket *so;
2487 struct file *fp = NULL;
2488 int use_rcvinfo = 1;
2489 int error=0, len, i;
2490 struct sockaddr *to = NULL;
2491#ifdef KTRACE
2492 struct uio *ktruio = NULL;
2493#endif
2494 struct uio auio;
2495 struct iovec *iov, *tiov;
2496
2497 if (uap->sinfo) {
2498 error = copyin(uap->sinfo, &sinfo, sizeof (sinfo));
2499 if (error)
2500 return (error);
2501 u_sinfo = &sinfo;
2502 }
2503 if (uap->tolen) {
2504 error = getsockaddr(&to, uap->to, uap->tolen);
2505 if (error) {
2506 to = NULL;
2507 goto sctp_bad2;
2508 }
2509 }
2510
2511 AUDIT_ARG_FD(uap->sd);
2512 error = getsock(td->td_proc->p_fd, uap->sd, &fp, NULL);
2513 if (error)
2514 goto sctp_bad1;
2515
2516#ifdef COMPAT_FREEBSD32
2517 if (SV_CURPROC_FLAG(SV_ILP32))
2518 error = freebsd32_copyiniov((struct iovec32 *)uap->iov,
2519 uap->iovlen, &iov, EMSGSIZE);
2520 else
2521#endif
2522 error = copyiniov(uap->iov, uap->iovlen, &iov, EMSGSIZE);
2523 if (error)
2524 goto sctp_bad1;
2525#ifdef KTRACE
2526 if (KTRPOINT(td, KTR_STRUCT))
2527 ktrsockaddr(to);
2528#endif
2529
2530 so = (struct socket *)fp->f_data;
2531#ifdef MAC
2532 error = mac_socket_check_send(td->td_ucred, so);
2533 if (error)
2534 goto sctp_bad;
2535#endif /* MAC */
2536
2537 auio.uio_iov = iov;
2538 auio.uio_iovcnt = uap->iovlen;
2539 auio.uio_segflg = UIO_USERSPACE;
2540 auio.uio_rw = UIO_WRITE;
2541 auio.uio_td = td;
2542 auio.uio_offset = 0; /* XXX */
2543 auio.uio_resid = 0;
2544 tiov = iov;
2545 for (i = 0; i <uap->iovlen; i++, tiov++) {
2546 if ((auio.uio_resid += tiov->iov_len) < 0) {
2547 error = EINVAL;
2548 goto sctp_bad;
2549 }
2550 }
2551 len = auio.uio_resid;
2552 CURVNET_SET(so->so_vnet);
2553 error = sctp_lower_sosend(so, to, &auio,
2554 (struct mbuf *)NULL, (struct mbuf *)NULL,
2555 uap->flags, use_rcvinfo, u_sinfo, td);
2556 CURVNET_RESTORE();
2557 if (error) {
2558 if (auio.uio_resid != len && (error == ERESTART ||
2559 error == EINTR || error == EWOULDBLOCK))
2560 error = 0;
2561 /* Generation of SIGPIPE can be controlled per socket */
2562 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) &&
2563 !(uap->flags & MSG_NOSIGNAL)) {
2564 PROC_LOCK(td->td_proc);
2565 psignal(td->td_proc, SIGPIPE);
2566 PROC_UNLOCK(td->td_proc);
2567 }
2568 }
2569 if (error == 0)
2570 td->td_retval[0] = len - auio.uio_resid;
2571#ifdef KTRACE
2572 if (ktruio != NULL) {
2573 ktruio->uio_resid = td->td_retval[0];
2574 ktrgenio(uap->sd, UIO_WRITE, ktruio, error);
2575 }
2576#endif /* KTRACE */
2577sctp_bad:
2578 free(iov, M_IOV);
2579sctp_bad1:
2580 if (fp)
2581 fdrop(fp, td);
2582sctp_bad2:
2583 if (to)
2584 free(to, M_SONAME);
2585 return (error);
2586#else /* SCTP */
2587 return (EOPNOTSUPP);
2588#endif /* SCTP */
2589}
2590
2591int
2592sctp_generic_recvmsg(td, uap)
2593 struct thread *td;
2594 struct sctp_generic_recvmsg_args /* {
2595 int sd,
2596 struct iovec *iov,
2597 int iovlen,
2598 struct sockaddr *from,
2599 __socklen_t *fromlenaddr,
2600 struct sctp_sndrcvinfo *sinfo,
2601 int *msg_flags
2602 } */ *uap;
2603{
2604#if (defined(INET) || defined(INET6)) && defined(SCTP)
|