1 /* 2 * Copyright (c) 1982, 1986, 1989, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * sendfile(2) and related extensions: 6 * Copyright (c) 1998, David Greenman. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 37 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 38 * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.89 2008/07/26 15:36:28 sephe Exp $ 39 */ 40 41 #include "opt_ktrace.h" 42 #include "opt_sctp.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/kernel.h> 47 #include <sys/sysproto.h> 48 #include <sys/malloc.h> 49 #include <sys/filedesc.h> 50 #include <sys/event.h> 51 #include <sys/proc.h> 52 #include <sys/fcntl.h> 53 #include <sys/file.h> 54 #include <sys/filio.h> 55 #include <sys/kern_syscall.h> 56 #include <sys/mbuf.h> 57 #include <sys/protosw.h> 58 #include <sys/sfbuf.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/socketops.h> 62 #include <sys/uio.h> 63 #include <sys/vnode.h> 64 #include <sys/lock.h> 65 #include <sys/mount.h> 66 #ifdef KTRACE 67 #include <sys/ktrace.h> 68 #endif 69 #include <vm/vm.h> 70 #include <vm/vm_object.h> 71 #include <vm/vm_page.h> 72 #include <vm/vm_pageout.h> 73 #include <vm/vm_kern.h> 74 #include <vm/vm_extern.h> 75 #include <sys/file2.h> 76 #include <sys/signalvar.h> 77 #include <sys/serialize.h> 78 79 #include <sys/thread2.h> 80 #include <sys/msgport2.h> 81 #include <sys/socketvar2.h> 82 #include <net/netmsg2.h> 83 84 #ifdef SCTP 85 #include <netinet/sctp_peeloff.h> 86 #endif /* SCTP */ 87 88 struct sfbuf_mref { 89 struct sf_buf *sf; 90 int mref_count; 91 struct lwkt_serialize serializer; 92 }; 93 94 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures"); 95 96 /* 97 * System call interface to the socket abstraction. 98 */ 99 100 extern struct fileops socketops; 101 102 /* 103 * socket_args(int domain, int type, int protocol) 104 */ 105 int 106 kern_socket(int domain, int type, int protocol, int *res) 107 { 108 struct thread *td = curthread; 109 struct proc *p = td->td_proc; 110 struct socket *so; 111 struct file *fp; 112 int fd, error; 113 114 KKASSERT(p); 115 116 error = falloc(p, &fp, &fd); 117 if (error) 118 return (error); 119 error = socreate(domain, &so, type, protocol, td); 120 if (error) { 121 fsetfd(p, NULL, fd); 122 } else { 123 fp->f_type = DTYPE_SOCKET; 124 fp->f_flag = FREAD | FWRITE; 125 fp->f_ops = &socketops; 126 fp->f_data = so; 127 *res = fd; 128 fsetfd(p, fp, fd); 129 } 130 fdrop(fp); 131 return (error); 132 } 133 134 int 135 sys_socket(struct socket_args *uap) 136 { 137 int error; 138 139 error = kern_socket(uap->domain, uap->type, uap->protocol, 140 &uap->sysmsg_result); 141 142 return (error); 143 } 144 145 int 146 kern_bind(int s, struct sockaddr *sa) 147 { 148 struct thread *td = curthread; 149 struct proc *p = td->td_proc; 150 struct file *fp; 151 int error; 152 153 KKASSERT(p); 154 error = holdsock(p->p_fd, s, &fp); 155 if (error) 156 return (error); 157 error = sobind((struct socket *)fp->f_data, sa, td); 158 fdrop(fp); 159 return (error); 160 } 161 162 /* 163 * bind_args(int s, caddr_t name, int namelen) 164 */ 165 int 166 sys_bind(struct bind_args *uap) 167 { 168 struct sockaddr *sa; 169 int error; 170 171 error = getsockaddr(&sa, uap->name, uap->namelen); 172 if (error) 173 return (error); 174 error = kern_bind(uap->s, sa); 175 FREE(sa, M_SONAME); 176 177 return (error); 178 } 179 180 int 181 kern_listen(int s, int backlog) 182 { 183 struct thread *td = curthread; 184 struct proc *p = td->td_proc; 185 struct file *fp; 186 int error; 187 188 KKASSERT(p); 189 error = holdsock(p->p_fd, s, &fp); 190 if (error) 191 return (error); 192 error = solisten((struct socket *)fp->f_data, backlog, td); 193 fdrop(fp); 194 return(error); 195 } 196 197 /* 198 * listen_args(int s, int backlog) 199 */ 200 int 201 sys_listen(struct listen_args *uap) 202 { 203 int error; 204 205 error = kern_listen(uap->s, uap->backlog); 206 return (error); 207 } 208 209 /* 210 * Returns the accepted socket as well. 211 */ 212 static boolean_t 213 soaccept_predicate(struct netmsg *msg0) 214 { 215 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 216 struct socket *head = msg->nm_so; 217 218 if (head->so_error != 0) { 219 msg->nm_netmsg.nm_lmsg.ms_error = head->so_error; 220 return (TRUE); 221 } 222 if (!TAILQ_EMPTY(&head->so_comp)) { 223 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 224 msg->nm_so = TAILQ_FIRST(&head->so_comp); 225 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list); 226 head->so_qlen--; 227 228 msg->nm_netmsg.nm_lmsg.ms_error = 0; 229 return (TRUE); 230 } 231 if (head->so_state & SS_CANTRCVMORE) { 232 msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED; 233 return (TRUE); 234 } 235 if (msg->nm_fflags & FNONBLOCK) { 236 msg->nm_netmsg.nm_lmsg.ms_error = EWOULDBLOCK; 237 return (TRUE); 238 } 239 240 return (FALSE); 241 } 242 243 /* 244 * The second argument to kern_accept() is a handle to a struct sockaddr. 245 * This allows kern_accept() to return a pointer to an allocated struct 246 * sockaddr which must be freed later with FREE(). The caller must 247 * initialize *name to NULL. 248 */ 249 int 250 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 251 { 252 struct thread *td = curthread; 253 struct proc *p = td->td_proc; 254 struct file *lfp = NULL; 255 struct file *nfp = NULL; 256 struct sockaddr *sa; 257 struct socket *head, *so; 258 struct netmsg_so_notify msg; 259 lwkt_port_t port; 260 int fd; 261 u_int fflag; /* type must match fp->f_flag */ 262 int error, tmp; 263 264 *res = -1; 265 if (name && namelen && *namelen < 0) 266 return (EINVAL); 267 268 error = holdsock(p->p_fd, s, &lfp); 269 if (error) 270 return (error); 271 272 error = falloc(p, &nfp, &fd); 273 if (error) { /* Probably ran out of file descriptors. */ 274 fdrop(lfp); 275 return (error); 276 } 277 head = (struct socket *)lfp->f_data; 278 if ((head->so_options & SO_ACCEPTCONN) == 0) { 279 error = EINVAL; 280 goto done; 281 } 282 283 if (fflags & O_FBLOCKING) 284 fflags |= lfp->f_flag & ~FNONBLOCK; 285 else if (fflags & O_FNONBLOCKING) 286 fflags |= lfp->f_flag | FNONBLOCK; 287 else 288 fflags = lfp->f_flag; 289 290 /* optimize for uniprocessor case later XXX JH */ 291 port = head->so_proto->pr_mport(head, NULL, NULL, PRU_PRED); 292 netmsg_init_abortable(&msg.nm_netmsg, &curthread->td_msgport, 293 0, 294 netmsg_so_notify, 295 netmsg_so_notify_doabort); 296 msg.nm_predicate = soaccept_predicate; 297 msg.nm_fflags = fflags; 298 msg.nm_so = head; 299 msg.nm_etype = NM_REVENT; 300 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, PCATCH); 301 if (error) 302 goto done; 303 304 /* 305 * At this point we have the connection that's ready to be accepted. 306 */ 307 so = msg.nm_so; 308 309 fflag = lfp->f_flag; 310 311 /* connection has been removed from the listen queue */ 312 KNOTE(&head->so_rcv.ssb_sel.si_note, 0); 313 314 so->so_state &= ~SS_COMP; 315 so->so_head = NULL; 316 if (head->so_sigio != NULL) 317 fsetown(fgetown(head->so_sigio), &so->so_sigio); 318 319 nfp->f_type = DTYPE_SOCKET; 320 nfp->f_flag = fflag; 321 nfp->f_ops = &socketops; 322 nfp->f_data = so; 323 /* Sync socket nonblocking/async state with file flags */ 324 tmp = fflag & FNONBLOCK; 325 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred); 326 tmp = fflag & FASYNC; 327 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred); 328 329 sa = NULL; 330 error = soaccept(so, &sa); 331 332 /* 333 * Set the returned name and namelen as applicable. Set the returned 334 * namelen to 0 for older code which might ignore the return value 335 * from accept. 336 */ 337 if (error == 0) { 338 if (sa && name && namelen) { 339 if (*namelen > sa->sa_len) 340 *namelen = sa->sa_len; 341 *name = sa; 342 } else { 343 if (sa) 344 FREE(sa, M_SONAME); 345 } 346 } 347 348 done: 349 /* 350 * If an error occured clear the reserved descriptor, else associate 351 * nfp with it. 352 * 353 * Note that *res is normally ignored if an error is returned but 354 * a syscall message will still have access to the result code. 355 */ 356 if (error) { 357 fsetfd(p, NULL, fd); 358 } else { 359 *res = fd; 360 fsetfd(p, nfp, fd); 361 } 362 fdrop(nfp); 363 fdrop(lfp); 364 return (error); 365 } 366 367 /* 368 * accept(int s, caddr_t name, int *anamelen) 369 */ 370 int 371 sys_accept(struct accept_args *uap) 372 { 373 struct sockaddr *sa = NULL; 374 int sa_len; 375 int error; 376 377 if (uap->name) { 378 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 379 if (error) 380 return (error); 381 382 error = kern_accept(uap->s, 0, &sa, &sa_len, &uap->sysmsg_result); 383 384 if (error == 0) 385 error = copyout(sa, uap->name, sa_len); 386 if (error == 0) { 387 error = copyout(&sa_len, uap->anamelen, 388 sizeof(*uap->anamelen)); 389 } 390 if (sa) 391 FREE(sa, M_SONAME); 392 } else { 393 error = kern_accept(uap->s, 0, NULL, 0, &uap->sysmsg_result); 394 } 395 return (error); 396 } 397 398 /* 399 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 400 */ 401 int 402 sys_extaccept(struct extaccept_args *uap) 403 { 404 struct sockaddr *sa = NULL; 405 int sa_len; 406 int error; 407 int fflags = uap->flags & O_FMASK; 408 409 if (uap->name) { 410 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 411 if (error) 412 return (error); 413 414 error = kern_accept(uap->s, fflags, &sa, &sa_len, &uap->sysmsg_result); 415 416 if (error == 0) 417 error = copyout(sa, uap->name, sa_len); 418 if (error == 0) { 419 error = copyout(&sa_len, uap->anamelen, 420 sizeof(*uap->anamelen)); 421 } 422 if (sa) 423 FREE(sa, M_SONAME); 424 } else { 425 error = kern_accept(uap->s, fflags, NULL, 0, &uap->sysmsg_result); 426 } 427 return (error); 428 } 429 430 431 /* 432 * Returns TRUE if predicate satisfied. 433 */ 434 static boolean_t 435 soconnected_predicate(struct netmsg *msg0) 436 { 437 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 438 struct socket *so = msg->nm_so; 439 440 /* check predicate */ 441 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 442 msg->nm_netmsg.nm_lmsg.ms_error = so->so_error; 443 return (TRUE); 444 } 445 446 return (FALSE); 447 } 448 449 int 450 kern_connect(int s, int fflags, struct sockaddr *sa) 451 { 452 struct thread *td = curthread; 453 struct proc *p = td->td_proc; 454 struct file *fp; 455 struct socket *so; 456 int error, interrupted = 0; 457 458 error = holdsock(p->p_fd, s, &fp); 459 if (error) 460 return (error); 461 so = (struct socket *)fp->f_data; 462 463 if (fflags & O_FBLOCKING) 464 /* fflags &= ~FNONBLOCK; */; 465 else if (fflags & O_FNONBLOCKING) 466 fflags |= FNONBLOCK; 467 else 468 fflags = fp->f_flag; 469 470 if (so->so_state & SS_ISCONNECTING) { 471 error = EALREADY; 472 goto done; 473 } 474 error = soconnect(so, sa, td); 475 if (error) 476 goto bad; 477 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 478 error = EINPROGRESS; 479 goto done; 480 } 481 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 482 struct netmsg_so_notify msg; 483 lwkt_port_t port; 484 485 port = so->so_proto->pr_mport(so, sa, NULL, PRU_PRED); 486 netmsg_init_abortable(&msg.nm_netmsg, 487 &curthread->td_msgport, 488 0, 489 netmsg_so_notify, 490 netmsg_so_notify_doabort); 491 msg.nm_predicate = soconnected_predicate; 492 msg.nm_so = so; 493 msg.nm_etype = NM_REVENT; 494 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, PCATCH); 495 if (error == EINTR || error == ERESTART) 496 interrupted = 1; 497 } 498 if (error == 0) { 499 error = so->so_error; 500 so->so_error = 0; 501 } 502 bad: 503 if (!interrupted) 504 so->so_state &= ~SS_ISCONNECTING; 505 if (error == ERESTART) 506 error = EINTR; 507 done: 508 fdrop(fp); 509 return (error); 510 } 511 512 /* 513 * connect_args(int s, caddr_t name, int namelen) 514 */ 515 int 516 sys_connect(struct connect_args *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 error = kern_connect(uap->s, 0, sa); 525 FREE(sa, M_SONAME); 526 527 return (error); 528 } 529 530 /* 531 * connect_args(int s, int fflags, caddr_t name, int namelen) 532 */ 533 int 534 sys_extconnect(struct extconnect_args *uap) 535 { 536 struct sockaddr *sa; 537 int error; 538 int fflags = uap->flags & O_FMASK; 539 540 error = getsockaddr(&sa, uap->name, uap->namelen); 541 if (error) 542 return (error); 543 error = kern_connect(uap->s, fflags, sa); 544 FREE(sa, M_SONAME); 545 546 return (error); 547 } 548 549 int 550 kern_socketpair(int domain, int type, int protocol, int *sv) 551 { 552 struct thread *td = curthread; 553 struct proc *p = td->td_proc; 554 struct file *fp1, *fp2; 555 struct socket *so1, *so2; 556 int fd1, fd2, error; 557 558 KKASSERT(p); 559 error = socreate(domain, &so1, type, protocol, td); 560 if (error) 561 return (error); 562 error = socreate(domain, &so2, type, protocol, td); 563 if (error) 564 goto free1; 565 error = falloc(p, &fp1, &fd1); 566 if (error) 567 goto free2; 568 sv[0] = fd1; 569 fp1->f_data = so1; 570 error = falloc(p, &fp2, &fd2); 571 if (error) 572 goto free3; 573 fp2->f_data = so2; 574 sv[1] = fd2; 575 error = soconnect2(so1, so2); 576 if (error) 577 goto free4; 578 if (type == SOCK_DGRAM) { 579 /* 580 * Datagram socket connection is asymmetric. 581 */ 582 error = soconnect2(so2, so1); 583 if (error) 584 goto free4; 585 } 586 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 587 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 588 fp1->f_ops = fp2->f_ops = &socketops; 589 fsetfd(p, fp1, fd1); 590 fsetfd(p, fp2, fd2); 591 fdrop(fp1); 592 fdrop(fp2); 593 return (error); 594 free4: 595 fsetfd(p, NULL, fd2); 596 fdrop(fp2); 597 free3: 598 fsetfd(p, NULL, fd1); 599 fdrop(fp1); 600 free2: 601 (void)soclose(so2, 0); 602 free1: 603 (void)soclose(so1, 0); 604 return (error); 605 } 606 607 /* 608 * socketpair(int domain, int type, int protocol, int *rsv) 609 */ 610 int 611 sys_socketpair(struct socketpair_args *uap) 612 { 613 int error, sockv[2]; 614 615 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 616 617 if (error == 0) 618 error = copyout(sockv, uap->rsv, sizeof(sockv)); 619 return (error); 620 } 621 622 int 623 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 624 struct mbuf *control, int flags, int *res) 625 { 626 struct thread *td = curthread; 627 struct lwp *lp = td->td_lwp; 628 struct proc *p = td->td_proc; 629 struct file *fp; 630 int len, error; 631 struct socket *so; 632 #ifdef KTRACE 633 struct iovec *ktriov = NULL; 634 struct uio ktruio; 635 #endif 636 637 error = holdsock(p->p_fd, s, &fp); 638 if (error) 639 return (error); 640 if (auio->uio_resid < 0) { 641 error = EINVAL; 642 goto done; 643 } 644 #ifdef KTRACE 645 if (KTRPOINT(td, KTR_GENIO)) { 646 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 647 648 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 649 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 650 ktruio = *auio; 651 } 652 #endif 653 len = auio->uio_resid; 654 so = (struct socket *)fp->f_data; 655 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 656 if (fp->f_flag & FNONBLOCK) 657 flags |= MSG_FNONBLOCKING; 658 } 659 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 660 if (error) { 661 if (auio->uio_resid != len && (error == ERESTART || 662 error == EINTR || error == EWOULDBLOCK)) 663 error = 0; 664 if (error == EPIPE) 665 lwpsignal(p, lp, SIGPIPE); 666 } 667 #ifdef KTRACE 668 if (ktriov != NULL) { 669 if (error == 0) { 670 ktruio.uio_iov = ktriov; 671 ktruio.uio_resid = len - auio->uio_resid; 672 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 673 } 674 FREE(ktriov, M_TEMP); 675 } 676 #endif 677 if (error == 0) 678 *res = len - auio->uio_resid; 679 done: 680 fdrop(fp); 681 return (error); 682 } 683 684 /* 685 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 686 */ 687 int 688 sys_sendto(struct sendto_args *uap) 689 { 690 struct thread *td = curthread; 691 struct uio auio; 692 struct iovec aiov; 693 struct sockaddr *sa = NULL; 694 int error; 695 696 if (uap->to) { 697 error = getsockaddr(&sa, uap->to, uap->tolen); 698 if (error) 699 return (error); 700 } 701 aiov.iov_base = uap->buf; 702 aiov.iov_len = uap->len; 703 auio.uio_iov = &aiov; 704 auio.uio_iovcnt = 1; 705 auio.uio_offset = 0; 706 auio.uio_resid = uap->len; 707 auio.uio_segflg = UIO_USERSPACE; 708 auio.uio_rw = UIO_WRITE; 709 auio.uio_td = td; 710 711 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 712 &uap->sysmsg_result); 713 714 if (sa) 715 FREE(sa, M_SONAME); 716 return (error); 717 } 718 719 /* 720 * sendmsg_args(int s, caddr_t msg, int flags) 721 */ 722 int 723 sys_sendmsg(struct sendmsg_args *uap) 724 { 725 struct thread *td = curthread; 726 struct msghdr msg; 727 struct uio auio; 728 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 729 struct sockaddr *sa = NULL; 730 struct mbuf *control = NULL; 731 int error; 732 733 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 734 if (error) 735 return (error); 736 737 /* 738 * Conditionally copyin msg.msg_name. 739 */ 740 if (msg.msg_name) { 741 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 742 if (error) 743 return (error); 744 } 745 746 /* 747 * Populate auio. 748 */ 749 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 750 &auio.uio_resid); 751 if (error) 752 goto cleanup2; 753 auio.uio_iov = iov; 754 auio.uio_iovcnt = msg.msg_iovlen; 755 auio.uio_offset = 0; 756 auio.uio_segflg = UIO_USERSPACE; 757 auio.uio_rw = UIO_WRITE; 758 auio.uio_td = td; 759 760 /* 761 * Conditionally copyin msg.msg_control. 762 */ 763 if (msg.msg_control) { 764 if (msg.msg_controllen < sizeof(struct cmsghdr) || 765 msg.msg_controllen > MLEN) { 766 error = EINVAL; 767 goto cleanup; 768 } 769 control = m_get(MB_WAIT, MT_CONTROL); 770 if (control == NULL) { 771 error = ENOBUFS; 772 goto cleanup; 773 } 774 control->m_len = msg.msg_controllen; 775 error = copyin(msg.msg_control, mtod(control, caddr_t), 776 msg.msg_controllen); 777 if (error) { 778 m_free(control); 779 goto cleanup; 780 } 781 } 782 783 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 784 &uap->sysmsg_result); 785 786 cleanup: 787 iovec_free(&iov, aiov); 788 cleanup2: 789 if (sa) 790 FREE(sa, M_SONAME); 791 return (error); 792 } 793 794 /* 795 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 796 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 797 * Don't forget to FREE() and m_free() these if they are returned. 798 */ 799 int 800 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 801 struct mbuf **control, int *flags, int *res) 802 { 803 struct thread *td = curthread; 804 struct proc *p = td->td_proc; 805 struct file *fp; 806 int len, error; 807 int lflags; 808 struct socket *so; 809 #ifdef KTRACE 810 struct iovec *ktriov = NULL; 811 struct uio ktruio; 812 #endif 813 814 error = holdsock(p->p_fd, s, &fp); 815 if (error) 816 return (error); 817 if (auio->uio_resid < 0) { 818 error = EINVAL; 819 goto done; 820 } 821 #ifdef KTRACE 822 if (KTRPOINT(td, KTR_GENIO)) { 823 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 824 825 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 826 bcopy(auio->uio_iov, ktriov, iovlen); 827 ktruio = *auio; 828 } 829 #endif 830 len = auio->uio_resid; 831 so = (struct socket *)fp->f_data; 832 833 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 834 if (fp->f_flag & FNONBLOCK) { 835 if (flags) { 836 *flags |= MSG_FNONBLOCKING; 837 } else { 838 lflags = MSG_FNONBLOCKING; 839 flags = &lflags; 840 } 841 } 842 } 843 844 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 845 if (error) { 846 if (auio->uio_resid != len && (error == ERESTART || 847 error == EINTR || error == EWOULDBLOCK)) 848 error = 0; 849 } 850 #ifdef KTRACE 851 if (ktriov != NULL) { 852 if (error == 0) { 853 ktruio.uio_iov = ktriov; 854 ktruio.uio_resid = len - auio->uio_resid; 855 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 856 } 857 FREE(ktriov, M_TEMP); 858 } 859 #endif 860 if (error == 0) 861 *res = len - auio->uio_resid; 862 done: 863 fdrop(fp); 864 return (error); 865 } 866 867 /* 868 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 869 * caddr_t from, int *fromlenaddr) 870 */ 871 int 872 sys_recvfrom(struct recvfrom_args *uap) 873 { 874 struct thread *td = curthread; 875 struct uio auio; 876 struct iovec aiov; 877 struct sockaddr *sa = NULL; 878 int error, fromlen; 879 880 if (uap->from && uap->fromlenaddr) { 881 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 882 if (error) 883 return (error); 884 if (fromlen < 0) 885 return (EINVAL); 886 } else { 887 fromlen = 0; 888 } 889 aiov.iov_base = uap->buf; 890 aiov.iov_len = uap->len; 891 auio.uio_iov = &aiov; 892 auio.uio_iovcnt = 1; 893 auio.uio_offset = 0; 894 auio.uio_resid = uap->len; 895 auio.uio_segflg = UIO_USERSPACE; 896 auio.uio_rw = UIO_READ; 897 auio.uio_td = td; 898 899 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 900 &uap->flags, &uap->sysmsg_result); 901 902 if (error == 0 && uap->from) { 903 /* note: sa may still be NULL */ 904 if (sa) { 905 fromlen = MIN(fromlen, sa->sa_len); 906 error = copyout(sa, uap->from, fromlen); 907 } else { 908 fromlen = 0; 909 } 910 if (error == 0) { 911 error = copyout(&fromlen, uap->fromlenaddr, 912 sizeof(fromlen)); 913 } 914 } 915 if (sa) 916 FREE(sa, M_SONAME); 917 918 return (error); 919 } 920 921 /* 922 * recvmsg_args(int s, struct msghdr *msg, int flags) 923 */ 924 int 925 sys_recvmsg(struct recvmsg_args *uap) 926 { 927 struct thread *td = curthread; 928 struct msghdr msg; 929 struct uio auio; 930 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 931 struct mbuf *m, *control = NULL; 932 struct sockaddr *sa = NULL; 933 caddr_t ctlbuf; 934 socklen_t *ufromlenp, *ucontrollenp; 935 int error, fromlen, controllen, len, flags, *uflagsp; 936 937 /* 938 * This copyin handles everything except the iovec. 939 */ 940 error = copyin(uap->msg, &msg, sizeof(msg)); 941 if (error) 942 return (error); 943 944 if (msg.msg_name && msg.msg_namelen < 0) 945 return (EINVAL); 946 if (msg.msg_control && msg.msg_controllen < 0) 947 return (EINVAL); 948 949 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 950 msg_namelen)); 951 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 952 msg_controllen)); 953 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 954 msg_flags)); 955 956 /* 957 * Populate auio. 958 */ 959 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 960 &auio.uio_resid); 961 if (error) 962 return (error); 963 auio.uio_iov = iov; 964 auio.uio_iovcnt = msg.msg_iovlen; 965 auio.uio_offset = 0; 966 auio.uio_segflg = UIO_USERSPACE; 967 auio.uio_rw = UIO_READ; 968 auio.uio_td = td; 969 970 flags = uap->flags; 971 972 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio, 973 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result); 974 975 /* 976 * Conditionally copyout the name and populate the namelen field. 977 */ 978 if (error == 0 && msg.msg_name) { 979 /* note: sa may still be NULL */ 980 if (sa != NULL) { 981 fromlen = MIN(msg.msg_namelen, sa->sa_len); 982 error = copyout(sa, msg.msg_name, fromlen); 983 } else 984 fromlen = 0; 985 if (error == 0) 986 error = copyout(&fromlen, ufromlenp, 987 sizeof(*ufromlenp)); 988 } 989 990 /* 991 * Copyout msg.msg_control and msg.msg_controllen. 992 */ 993 if (error == 0 && msg.msg_control) { 994 len = msg.msg_controllen; 995 m = control; 996 ctlbuf = (caddr_t)msg.msg_control; 997 998 while(m && len > 0) { 999 unsigned int tocopy; 1000 1001 if (len >= m->m_len) { 1002 tocopy = m->m_len; 1003 } else { 1004 msg.msg_flags |= MSG_CTRUNC; 1005 tocopy = len; 1006 } 1007 1008 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1009 if (error) 1010 goto cleanup; 1011 1012 ctlbuf += tocopy; 1013 len -= tocopy; 1014 m = m->m_next; 1015 } 1016 controllen = ctlbuf - (caddr_t)msg.msg_control; 1017 error = copyout(&controllen, ucontrollenp, 1018 sizeof(*ucontrollenp)); 1019 } 1020 1021 if (error == 0) 1022 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1023 1024 cleanup: 1025 if (sa) 1026 FREE(sa, M_SONAME); 1027 iovec_free(&iov, aiov); 1028 if (control) 1029 m_freem(control); 1030 return (error); 1031 } 1032 1033 /* 1034 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1035 * in kernel pointer instead of a userland pointer. This allows us 1036 * to manipulate socket options in the emulation code. 1037 */ 1038 int 1039 kern_setsockopt(int s, struct sockopt *sopt) 1040 { 1041 struct thread *td = curthread; 1042 struct proc *p = td->td_proc; 1043 struct file *fp; 1044 int error; 1045 1046 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1047 return (EFAULT); 1048 if (sopt->sopt_valsize < 0) 1049 return (EINVAL); 1050 1051 error = holdsock(p->p_fd, s, &fp); 1052 if (error) 1053 return (error); 1054 1055 error = sosetopt((struct socket *)fp->f_data, sopt); 1056 fdrop(fp); 1057 return (error); 1058 } 1059 1060 /* 1061 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1062 */ 1063 int 1064 sys_setsockopt(struct setsockopt_args *uap) 1065 { 1066 struct thread *td = curthread; 1067 struct sockopt sopt; 1068 int error; 1069 1070 sopt.sopt_level = uap->level; 1071 sopt.sopt_name = uap->name; 1072 sopt.sopt_valsize = uap->valsize; 1073 sopt.sopt_td = td; 1074 1075 if (uap->val) { 1076 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1077 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1078 if (error) 1079 goto out; 1080 } else { 1081 sopt.sopt_val = NULL; 1082 } 1083 error = kern_setsockopt(uap->s, &sopt); 1084 if (error) 1085 goto out; 1086 if (uap->val) 1087 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1088 out: 1089 if (uap->val) 1090 kfree(sopt.sopt_val, M_TEMP); 1091 return(error); 1092 } 1093 1094 /* 1095 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1096 * in kernel pointer instead of a userland pointer. This allows us 1097 * to manipulate socket options in the emulation code. 1098 */ 1099 int 1100 kern_getsockopt(int s, struct sockopt *sopt) 1101 { 1102 struct thread *td = curthread; 1103 struct proc *p = td->td_proc; 1104 struct file *fp; 1105 int error; 1106 1107 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1108 return (EFAULT); 1109 if (sopt->sopt_valsize < 0) 1110 return (EINVAL); 1111 1112 error = holdsock(p->p_fd, s, &fp); 1113 if (error) 1114 return (error); 1115 1116 error = sogetopt((struct socket *)fp->f_data, sopt); 1117 fdrop(fp); 1118 return (error); 1119 } 1120 1121 /* 1122 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1123 */ 1124 int 1125 sys_getsockopt(struct getsockopt_args *uap) 1126 { 1127 struct thread *td = curthread; 1128 struct sockopt sopt; 1129 int error, valsize; 1130 1131 if (uap->val) { 1132 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1133 if (error) 1134 return (error); 1135 if (valsize < 0) 1136 return (EINVAL); 1137 } else { 1138 valsize = 0; 1139 } 1140 1141 sopt.sopt_level = uap->level; 1142 sopt.sopt_name = uap->name; 1143 sopt.sopt_valsize = valsize; 1144 sopt.sopt_td = td; 1145 1146 if (uap->val) { 1147 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1148 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1149 if (error) 1150 goto out; 1151 } else { 1152 sopt.sopt_val = NULL; 1153 } 1154 error = kern_getsockopt(uap->s, &sopt); 1155 if (error) 1156 goto out; 1157 valsize = sopt.sopt_valsize; 1158 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1159 if (error) 1160 goto out; 1161 if (uap->val) 1162 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1163 out: 1164 if (uap->val) 1165 kfree(sopt.sopt_val, M_TEMP); 1166 return (error); 1167 } 1168 1169 /* 1170 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1171 * This allows kern_getsockname() to return a pointer to an allocated struct 1172 * sockaddr which must be freed later with FREE(). The caller must 1173 * initialize *name to NULL. 1174 */ 1175 int 1176 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1177 { 1178 struct thread *td = curthread; 1179 struct proc *p = td->td_proc; 1180 struct file *fp; 1181 struct socket *so; 1182 struct sockaddr *sa = NULL; 1183 int error; 1184 1185 error = holdsock(p->p_fd, s, &fp); 1186 if (error) 1187 return (error); 1188 if (*namelen < 0) { 1189 fdrop(fp); 1190 return (EINVAL); 1191 } 1192 so = (struct socket *)fp->f_data; 1193 error = so_pru_sockaddr(so, &sa); 1194 if (error == 0) { 1195 if (sa == 0) { 1196 *namelen = 0; 1197 } else { 1198 *namelen = MIN(*namelen, sa->sa_len); 1199 *name = sa; 1200 } 1201 } 1202 1203 fdrop(fp); 1204 return (error); 1205 } 1206 1207 /* 1208 * getsockname_args(int fdes, caddr_t asa, int *alen) 1209 * 1210 * Get socket name. 1211 */ 1212 int 1213 sys_getsockname(struct getsockname_args *uap) 1214 { 1215 struct sockaddr *sa = NULL; 1216 int error, sa_len; 1217 1218 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1219 if (error) 1220 return (error); 1221 1222 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1223 1224 if (error == 0) 1225 error = copyout(sa, uap->asa, sa_len); 1226 if (error == 0) 1227 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1228 if (sa) 1229 FREE(sa, M_SONAME); 1230 return (error); 1231 } 1232 1233 /* 1234 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1235 * This allows kern_getpeername() to return a pointer to an allocated struct 1236 * sockaddr which must be freed later with FREE(). The caller must 1237 * initialize *name to NULL. 1238 */ 1239 int 1240 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1241 { 1242 struct thread *td = curthread; 1243 struct proc *p = td->td_proc; 1244 struct file *fp; 1245 struct socket *so; 1246 struct sockaddr *sa = NULL; 1247 int error; 1248 1249 error = holdsock(p->p_fd, s, &fp); 1250 if (error) 1251 return (error); 1252 if (*namelen < 0) { 1253 fdrop(fp); 1254 return (EINVAL); 1255 } 1256 so = (struct socket *)fp->f_data; 1257 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1258 fdrop(fp); 1259 return (ENOTCONN); 1260 } 1261 error = so_pru_peeraddr(so, &sa); 1262 if (error == 0) { 1263 if (sa == 0) { 1264 *namelen = 0; 1265 } else { 1266 *namelen = MIN(*namelen, sa->sa_len); 1267 *name = sa; 1268 } 1269 } 1270 1271 fdrop(fp); 1272 return (error); 1273 } 1274 1275 /* 1276 * getpeername_args(int fdes, caddr_t asa, int *alen) 1277 * 1278 * Get name of peer for connected socket. 1279 */ 1280 int 1281 sys_getpeername(struct getpeername_args *uap) 1282 { 1283 struct sockaddr *sa = NULL; 1284 int error, sa_len; 1285 1286 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1287 if (error) 1288 return (error); 1289 1290 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1291 1292 if (error == 0) 1293 error = copyout(sa, uap->asa, sa_len); 1294 if (error == 0) 1295 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1296 if (sa) 1297 FREE(sa, M_SONAME); 1298 return (error); 1299 } 1300 1301 int 1302 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1303 { 1304 struct sockaddr *sa; 1305 int error; 1306 1307 *namp = NULL; 1308 if (len > SOCK_MAXADDRLEN) 1309 return ENAMETOOLONG; 1310 if (len < offsetof(struct sockaddr, sa_data[0])) 1311 return EDOM; 1312 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1313 error = copyin(uaddr, sa, len); 1314 if (error) { 1315 FREE(sa, M_SONAME); 1316 } else { 1317 #if BYTE_ORDER != BIG_ENDIAN 1318 /* 1319 * The bind(), connect(), and sendto() syscalls were not 1320 * versioned for COMPAT_43. Thus, this check must stay. 1321 */ 1322 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1323 sa->sa_family = sa->sa_len; 1324 #endif 1325 sa->sa_len = len; 1326 *namp = sa; 1327 } 1328 return error; 1329 } 1330 1331 /* 1332 * Detach a mapped page and release resources back to the system. 1333 * We must release our wiring and if the object is ripped out 1334 * from under the vm_page we become responsible for freeing the 1335 * page. These routines must be MPSAFE. 1336 * 1337 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1338 * 1339 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required. 1340 */ 1341 static void 1342 sf_buf_mref(void *arg) 1343 { 1344 struct sfbuf_mref *sfm = arg; 1345 1346 /* 1347 * We must already hold a ref so there is no race to 0, just 1348 * atomically increment the count. 1349 */ 1350 atomic_add_int(&sfm->mref_count, 1); 1351 } 1352 1353 static void 1354 sf_buf_mfree(void *arg) 1355 { 1356 struct sfbuf_mref *sfm = arg; 1357 vm_page_t m; 1358 1359 KKASSERT(sfm->mref_count > 0); 1360 if (sfm->mref_count == 1) { 1361 /* 1362 * We are the only holder so no further locking is required, 1363 * the sfbuf can simply be freed. 1364 */ 1365 sfm->mref_count = 0; 1366 goto freeit; 1367 } else { 1368 /* 1369 * There may be other holders, we must obtain the serializer 1370 * to protect against a sf_buf_mfree() race to 0. An atomic 1371 * operation is still required for races against 1372 * sf_buf_mref(). 1373 * 1374 * XXX vm_page_*() and SFBUF routines not MPSAFE yet. 1375 */ 1376 lwkt_serialize_enter(&sfm->serializer); 1377 atomic_subtract_int(&sfm->mref_count, 1); 1378 if (sfm->mref_count == 0) { 1379 lwkt_serialize_exit(&sfm->serializer); 1380 freeit: 1381 get_mplock(); 1382 crit_enter(); 1383 m = sf_buf_page(sfm->sf); 1384 sf_buf_free(sfm->sf); 1385 vm_page_unwire(m, 0); 1386 if (m->wire_count == 0 && m->object == NULL) 1387 vm_page_try_to_free(m); 1388 crit_exit(); 1389 rel_mplock(); 1390 kfree(sfm, M_SENDFILE); 1391 } else { 1392 lwkt_serialize_exit(&sfm->serializer); 1393 } 1394 } 1395 } 1396 1397 /* 1398 * sendfile(2). 1399 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1400 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1401 * 1402 * Send a file specified by 'fd' and starting at 'offset' to a socket 1403 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1404 * nbytes == 0. Optionally add a header and/or trailer to the socket 1405 * output. If specified, write the total number of bytes sent into *sbytes. 1406 * 1407 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1408 * the headers to count against the remaining bytes to be sent from 1409 * the file descriptor. We may wish to implement a compatibility syscall 1410 * in the future. 1411 */ 1412 int 1413 sys_sendfile(struct sendfile_args *uap) 1414 { 1415 struct thread *td = curthread; 1416 struct proc *p = td->td_proc; 1417 struct file *fp; 1418 struct vnode *vp = NULL; 1419 struct sf_hdtr hdtr; 1420 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1421 struct uio auio; 1422 struct mbuf *mheader = NULL; 1423 off_t hdtr_size = 0, sbytes; 1424 int error, hbytes = 0, tbytes; 1425 1426 KKASSERT(p); 1427 1428 /* 1429 * Do argument checking. Must be a regular file in, stream 1430 * type and connected socket out, positive offset. 1431 */ 1432 fp = holdfp(p->p_fd, uap->fd, FREAD); 1433 if (fp == NULL) { 1434 return (EBADF); 1435 } 1436 if (fp->f_type != DTYPE_VNODE) { 1437 fdrop(fp); 1438 return (EINVAL); 1439 } 1440 vp = (struct vnode *)fp->f_data; 1441 vref(vp); 1442 fdrop(fp); 1443 1444 /* 1445 * If specified, get the pointer to the sf_hdtr struct for 1446 * any headers/trailers. 1447 */ 1448 if (uap->hdtr) { 1449 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1450 if (error) 1451 goto done; 1452 /* 1453 * Send any headers. 1454 */ 1455 if (hdtr.headers) { 1456 error = iovec_copyin(hdtr.headers, &iov, aiov, 1457 hdtr.hdr_cnt, &hbytes); 1458 if (error) 1459 goto done; 1460 auio.uio_iov = iov; 1461 auio.uio_iovcnt = hdtr.hdr_cnt; 1462 auio.uio_offset = 0; 1463 auio.uio_segflg = UIO_USERSPACE; 1464 auio.uio_rw = UIO_WRITE; 1465 auio.uio_td = td; 1466 auio.uio_resid = hbytes; 1467 1468 mheader = m_uiomove(&auio); 1469 1470 iovec_free(&iov, aiov); 1471 if (mheader == NULL) 1472 goto done; 1473 } 1474 } 1475 1476 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1477 &sbytes, uap->flags); 1478 if (error) 1479 goto done; 1480 1481 /* 1482 * Send trailers. Wimp out and use writev(2). 1483 */ 1484 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1485 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1486 hdtr.trl_cnt, &auio.uio_resid); 1487 if (error) 1488 goto done; 1489 auio.uio_iov = iov; 1490 auio.uio_iovcnt = hdtr.trl_cnt; 1491 auio.uio_offset = 0; 1492 auio.uio_segflg = UIO_USERSPACE; 1493 auio.uio_rw = UIO_WRITE; 1494 auio.uio_td = td; 1495 1496 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1497 1498 iovec_free(&iov, aiov); 1499 if (error) 1500 goto done; 1501 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1502 } 1503 1504 done: 1505 if (uap->sbytes != NULL) { 1506 sbytes += hdtr_size; 1507 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1508 } 1509 if (vp) 1510 vrele(vp); 1511 return (error); 1512 } 1513 1514 int 1515 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1516 struct mbuf *mheader, off_t *sbytes, int flags) 1517 { 1518 struct thread *td = curthread; 1519 struct proc *p = td->td_proc; 1520 struct vm_object *obj; 1521 struct socket *so; 1522 struct file *fp; 1523 struct mbuf *m; 1524 struct sf_buf *sf; 1525 struct sfbuf_mref *sfm; 1526 struct vm_page *pg; 1527 off_t off, xfsize; 1528 off_t hbytes = 0; 1529 int error = 0; 1530 1531 if (vp->v_type != VREG) { 1532 error = EINVAL; 1533 goto done0; 1534 } 1535 if ((obj = vp->v_object) == NULL) { 1536 error = EINVAL; 1537 goto done0; 1538 } 1539 error = holdsock(p->p_fd, sfd, &fp); 1540 if (error) 1541 goto done0; 1542 so = (struct socket *)fp->f_data; 1543 if (so->so_type != SOCK_STREAM) { 1544 error = EINVAL; 1545 goto done; 1546 } 1547 if ((so->so_state & SS_ISCONNECTED) == 0) { 1548 error = ENOTCONN; 1549 goto done; 1550 } 1551 if (offset < 0) { 1552 error = EINVAL; 1553 goto done; 1554 } 1555 1556 *sbytes = 0; 1557 /* 1558 * Protect against multiple writers to the socket. 1559 */ 1560 ssb_lock(&so->so_snd, M_WAITOK); 1561 1562 /* 1563 * Loop through the pages in the file, starting with the requested 1564 * offset. Get a file page (do I/O if necessary), map the file page 1565 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1566 * it on the socket. 1567 */ 1568 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1569 vm_pindex_t pindex; 1570 vm_offset_t pgoff; 1571 1572 pindex = OFF_TO_IDX(off); 1573 retry_lookup: 1574 /* 1575 * Calculate the amount to transfer. Not to exceed a page, 1576 * the EOF, or the passed in nbytes. 1577 */ 1578 xfsize = vp->v_filesize - off; 1579 if (xfsize > PAGE_SIZE) 1580 xfsize = PAGE_SIZE; 1581 pgoff = (vm_offset_t)(off & PAGE_MASK); 1582 if (PAGE_SIZE - pgoff < xfsize) 1583 xfsize = PAGE_SIZE - pgoff; 1584 if (nbytes && xfsize > (nbytes - *sbytes)) 1585 xfsize = nbytes - *sbytes; 1586 if (xfsize <= 0) 1587 break; 1588 /* 1589 * Optimize the non-blocking case by looking at the socket space 1590 * before going to the extra work of constituting the sf_buf. 1591 */ 1592 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) { 1593 if (so->so_state & SS_CANTSENDMORE) 1594 error = EPIPE; 1595 else 1596 error = EAGAIN; 1597 ssb_unlock(&so->so_snd); 1598 goto done; 1599 } 1600 /* 1601 * Attempt to look up the page. 1602 * 1603 * Allocate if not found, wait and loop if busy, then 1604 * wire the page. critical section protection is 1605 * required to maintain the object association (an 1606 * interrupt can free the page) through to the 1607 * vm_page_wire() call. 1608 */ 1609 crit_enter(); 1610 pg = vm_page_lookup(obj, pindex); 1611 if (pg == NULL) { 1612 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1613 if (pg == NULL) { 1614 vm_wait(0); 1615 crit_exit(); 1616 goto retry_lookup; 1617 } 1618 vm_page_wakeup(pg); 1619 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1620 crit_exit(); 1621 goto retry_lookup; 1622 } 1623 vm_page_wire(pg); 1624 crit_exit(); 1625 1626 /* 1627 * If page is not valid for what we need, initiate I/O 1628 */ 1629 1630 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1631 struct uio auio; 1632 struct iovec aiov; 1633 int bsize; 1634 1635 /* 1636 * Ensure that our page is still around when the I/O 1637 * completes. 1638 */ 1639 vm_page_io_start(pg); 1640 1641 /* 1642 * Get the page from backing store. 1643 */ 1644 bsize = vp->v_mount->mnt_stat.f_iosize; 1645 auio.uio_iov = &aiov; 1646 auio.uio_iovcnt = 1; 1647 aiov.iov_base = 0; 1648 aiov.iov_len = MAXBSIZE; 1649 auio.uio_resid = MAXBSIZE; 1650 auio.uio_offset = trunc_page(off); 1651 auio.uio_segflg = UIO_NOCOPY; 1652 auio.uio_rw = UIO_READ; 1653 auio.uio_td = td; 1654 vn_lock(vp, LK_SHARED | LK_RETRY); 1655 error = VOP_READ(vp, &auio, 1656 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1657 p->p_ucred); 1658 vn_unlock(vp); 1659 vm_page_flag_clear(pg, PG_ZERO); 1660 vm_page_io_finish(pg); 1661 if (error) { 1662 crit_enter(); 1663 vm_page_unwire(pg, 0); 1664 vm_page_try_to_free(pg); 1665 crit_exit(); 1666 ssb_unlock(&so->so_snd); 1667 goto done; 1668 } 1669 } 1670 1671 1672 /* 1673 * Get a sendfile buf. We usually wait as long as necessary, 1674 * but this wait can be interrupted. 1675 */ 1676 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) { 1677 crit_enter(); 1678 vm_page_unwire(pg, 0); 1679 vm_page_try_to_free(pg); 1680 crit_exit(); 1681 ssb_unlock(&so->so_snd); 1682 error = EINTR; 1683 goto done; 1684 } 1685 1686 /* 1687 * Get an mbuf header and set it up as having external storage. 1688 */ 1689 MGETHDR(m, MB_WAIT, MT_DATA); 1690 if (m == NULL) { 1691 error = ENOBUFS; 1692 sf_buf_free(sf); 1693 ssb_unlock(&so->so_snd); 1694 goto done; 1695 } 1696 1697 /* 1698 * sfm is a temporary hack, use a per-cpu cache for this. 1699 */ 1700 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK); 1701 sfm->sf = sf; 1702 sfm->mref_count = 1; 1703 lwkt_serialize_init(&sfm->serializer); 1704 1705 m->m_ext.ext_free = sf_buf_mfree; 1706 m->m_ext.ext_ref = sf_buf_mref; 1707 m->m_ext.ext_arg = sfm; 1708 m->m_ext.ext_buf = (void *)sf->kva; 1709 m->m_ext.ext_size = PAGE_SIZE; 1710 m->m_data = (char *) sf->kva + pgoff; 1711 m->m_flags |= M_EXT; 1712 m->m_pkthdr.len = m->m_len = xfsize; 1713 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1714 1715 if (mheader != NULL) { 1716 hbytes = mheader->m_pkthdr.len; 1717 mheader->m_pkthdr.len += m->m_pkthdr.len; 1718 m_cat(mheader, m); 1719 m = mheader; 1720 mheader = NULL; 1721 } else 1722 hbytes = 0; 1723 1724 /* 1725 * Add the buffer to the socket buffer chain. 1726 */ 1727 crit_enter(); 1728 retry_space: 1729 /* 1730 * Make sure that the socket is still able to take more data. 1731 * CANTSENDMORE being true usually means that the connection 1732 * was closed. so_error is true when an error was sensed after 1733 * a previous send. 1734 * The state is checked after the page mapping and buffer 1735 * allocation above since those operations may block and make 1736 * any socket checks stale. From this point forward, nothing 1737 * blocks before the pru_send (or more accurately, any blocking 1738 * results in a loop back to here to re-check). 1739 */ 1740 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1741 if (so->so_state & SS_CANTSENDMORE) { 1742 error = EPIPE; 1743 } else { 1744 error = so->so_error; 1745 so->so_error = 0; 1746 } 1747 m_freem(m); 1748 ssb_unlock(&so->so_snd); 1749 crit_exit(); 1750 goto done; 1751 } 1752 /* 1753 * Wait for socket space to become available. We do this just 1754 * after checking the connection state above in order to avoid 1755 * a race condition with ssb_wait(). 1756 */ 1757 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) { 1758 if (fp->f_flag & FNONBLOCK) { 1759 m_freem(m); 1760 ssb_unlock(&so->so_snd); 1761 crit_exit(); 1762 error = EAGAIN; 1763 goto done; 1764 } 1765 error = ssb_wait(&so->so_snd); 1766 /* 1767 * An error from ssb_wait usually indicates that we've 1768 * been interrupted by a signal. If we've sent anything 1769 * then return bytes sent, otherwise return the error. 1770 */ 1771 if (error) { 1772 m_freem(m); 1773 ssb_unlock(&so->so_snd); 1774 crit_exit(); 1775 goto done; 1776 } 1777 goto retry_space; 1778 } 1779 error = so_pru_send(so, 0, m, NULL, NULL, td); 1780 crit_exit(); 1781 if (error) { 1782 ssb_unlock(&so->so_snd); 1783 goto done; 1784 } 1785 } 1786 if (mheader != NULL) { 1787 *sbytes += mheader->m_pkthdr.len; 1788 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1789 mheader = NULL; 1790 } 1791 ssb_unlock(&so->so_snd); 1792 1793 done: 1794 fdrop(fp); 1795 done0: 1796 if (mheader != NULL) 1797 m_freem(mheader); 1798 return (error); 1799 } 1800 1801 int 1802 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1803 { 1804 #ifdef SCTP 1805 struct thread *td = curthread; 1806 struct proc *p = td->td_proc; 1807 struct file *lfp = NULL; 1808 struct file *nfp = NULL; 1809 int error; 1810 struct socket *head, *so; 1811 caddr_t assoc_id; 1812 int fd; 1813 short fflag; /* type must match fp->f_flag */ 1814 1815 assoc_id = uap->name; 1816 error = holdsock(p->p_fd, uap->sd, &lfp); 1817 if (error) { 1818 return (error); 1819 } 1820 crit_enter(); 1821 head = (struct socket *)lfp->f_data; 1822 error = sctp_can_peel_off(head, assoc_id); 1823 if (error) { 1824 crit_exit(); 1825 goto done; 1826 } 1827 /* 1828 * At this point we know we do have a assoc to pull 1829 * we proceed to get the fd setup. This may block 1830 * but that is ok. 1831 */ 1832 1833 fflag = lfp->f_flag; 1834 error = falloc(p, &nfp, &fd); 1835 if (error) { 1836 /* 1837 * Probably ran out of file descriptors. Put the 1838 * unaccepted connection back onto the queue and 1839 * do another wakeup so some other process might 1840 * have a chance at it. 1841 */ 1842 crit_exit(); 1843 goto done; 1844 } 1845 uap->sysmsg_result = fd; 1846 1847 so = sctp_get_peeloff(head, assoc_id, &error); 1848 if (so == NULL) { 1849 /* 1850 * Either someone else peeled it off OR 1851 * we can't get a socket. 1852 */ 1853 goto noconnection; 1854 } 1855 so->so_state &= ~SS_COMP; 1856 so->so_state &= ~SS_NOFDREF; 1857 so->so_head = NULL; 1858 if (head->so_sigio != NULL) 1859 fsetown(fgetown(head->so_sigio), &so->so_sigio); 1860 1861 nfp->f_type = DTYPE_SOCKET; 1862 nfp->f_flag = fflag; 1863 nfp->f_ops = &socketops; 1864 nfp->f_data = so; 1865 1866 noconnection: 1867 /* 1868 * Assign the file pointer to the reserved descriptor, or clear 1869 * the reserved descriptor if an error occured. 1870 */ 1871 if (error) 1872 fsetfd(p, NULL, fd); 1873 else 1874 fsetfd(p, nfp, fd); 1875 crit_exit(); 1876 /* 1877 * Release explicitly held references before returning. 1878 */ 1879 done: 1880 if (nfp != NULL) 1881 fdrop(nfp); 1882 fdrop(lfp); 1883 return (error); 1884 #else /* SCTP */ 1885 return(EOPNOTSUPP); 1886 #endif /* SCTP */ 1887 } 1888