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 */ 39 40 #include "opt_ktrace.h" 41 #include "opt_sctp.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/sysproto.h> 47 #include <sys/malloc.h> 48 #include <sys/filedesc.h> 49 #include <sys/event.h> 50 #include <sys/proc.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filio.h> 54 #include <sys/kern_syscall.h> 55 #include <sys/mbuf.h> 56 #include <sys/protosw.h> 57 #include <sys/sfbuf.h> 58 #include <sys/socket.h> 59 #include <sys/socketvar.h> 60 #include <sys/socketops.h> 61 #include <sys/uio.h> 62 #include <sys/vnode.h> 63 #include <sys/lock.h> 64 #include <sys/mount.h> 65 #ifdef KTRACE 66 #include <sys/ktrace.h> 67 #endif 68 #include <vm/vm.h> 69 #include <vm/vm_object.h> 70 #include <vm/vm_page.h> 71 #include <vm/vm_pageout.h> 72 #include <vm/vm_kern.h> 73 #include <vm/vm_extern.h> 74 #include <sys/file2.h> 75 #include <sys/signalvar.h> 76 #include <sys/serialize.h> 77 78 #include <sys/thread2.h> 79 #include <sys/msgport2.h> 80 #include <sys/socketvar2.h> 81 #include <net/netmsg2.h> 82 83 #ifdef SCTP 84 #include <netinet/sctp_peeloff.h> 85 #endif /* SCTP */ 86 87 /* 88 * System call interface to the socket abstraction. 89 */ 90 91 extern struct fileops socketops; 92 93 /* 94 * socket_args(int domain, int type, int protocol) 95 */ 96 int 97 kern_socket(int domain, int type, int protocol, int *res) 98 { 99 struct thread *td = curthread; 100 struct filedesc *fdp = td->td_proc->p_fd; 101 struct socket *so; 102 struct file *fp; 103 int fd, error; 104 105 KKASSERT(td->td_lwp); 106 107 error = falloc(td->td_lwp, &fp, &fd); 108 if (error) 109 return (error); 110 error = socreate(domain, &so, type, protocol, td); 111 if (error) { 112 fsetfd(fdp, NULL, fd); 113 } else { 114 fp->f_type = DTYPE_SOCKET; 115 fp->f_flag = FREAD | FWRITE; 116 fp->f_ops = &socketops; 117 fp->f_data = so; 118 *res = fd; 119 fsetfd(fdp, fp, fd); 120 } 121 fdrop(fp); 122 return (error); 123 } 124 125 /* 126 * MPALMOSTSAFE 127 */ 128 int 129 sys_socket(struct socket_args *uap) 130 { 131 int error; 132 133 error = kern_socket(uap->domain, uap->type, uap->protocol, 134 &uap->sysmsg_iresult); 135 136 return (error); 137 } 138 139 int 140 kern_bind(int s, struct sockaddr *sa) 141 { 142 struct thread *td = curthread; 143 struct proc *p = td->td_proc; 144 struct file *fp; 145 int error; 146 147 KKASSERT(p); 148 error = holdsock(p->p_fd, s, &fp); 149 if (error) 150 return (error); 151 error = sobind((struct socket *)fp->f_data, sa, td); 152 fdrop(fp); 153 return (error); 154 } 155 156 /* 157 * bind_args(int s, caddr_t name, int namelen) 158 * 159 * MPALMOSTSAFE 160 */ 161 int 162 sys_bind(struct bind_args *uap) 163 { 164 struct sockaddr *sa; 165 int error; 166 167 error = getsockaddr(&sa, uap->name, uap->namelen); 168 if (error) 169 return (error); 170 error = kern_bind(uap->s, sa); 171 FREE(sa, M_SONAME); 172 173 return (error); 174 } 175 176 int 177 kern_listen(int s, int backlog) 178 { 179 struct thread *td = curthread; 180 struct proc *p = td->td_proc; 181 struct file *fp; 182 int error; 183 184 KKASSERT(p); 185 error = holdsock(p->p_fd, s, &fp); 186 if (error) 187 return (error); 188 error = solisten((struct socket *)fp->f_data, backlog, td); 189 fdrop(fp); 190 return(error); 191 } 192 193 /* 194 * listen_args(int s, int backlog) 195 * 196 * MPALMOSTSAFE 197 */ 198 int 199 sys_listen(struct listen_args *uap) 200 { 201 int error; 202 203 error = kern_listen(uap->s, uap->backlog); 204 return (error); 205 } 206 207 /* 208 * Returns the accepted socket as well. 209 * 210 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 211 * pool token is absolutely required to avoid a sofree() race, 212 * as well as to avoid tailq handling races. 213 */ 214 static boolean_t 215 soaccept_predicate(struct netmsg_so_notify *msg) 216 { 217 struct socket *head = msg->base.nm_so; 218 struct socket *so; 219 220 if (head->so_error != 0) { 221 msg->base.lmsg.ms_error = head->so_error; 222 return (TRUE); 223 } 224 lwkt_getpooltoken(head); 225 if (!TAILQ_EMPTY(&head->so_comp)) { 226 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 227 so = TAILQ_FIRST(&head->so_comp); 228 TAILQ_REMOVE(&head->so_comp, so, so_list); 229 head->so_qlen--; 230 soclrstate(so, SS_COMP); 231 so->so_head = NULL; 232 soreference(so); 233 234 lwkt_relpooltoken(head); 235 236 msg->base.lmsg.ms_error = 0; 237 msg->base.nm_so = so; 238 return (TRUE); 239 } 240 lwkt_relpooltoken(head); 241 if (head->so_state & SS_CANTRCVMORE) { 242 msg->base.lmsg.ms_error = ECONNABORTED; 243 return (TRUE); 244 } 245 if (msg->nm_fflags & FNONBLOCK) { 246 msg->base.lmsg.ms_error = EWOULDBLOCK; 247 return (TRUE); 248 } 249 250 return (FALSE); 251 } 252 253 /* 254 * The second argument to kern_accept() is a handle to a struct sockaddr. 255 * This allows kern_accept() to return a pointer to an allocated struct 256 * sockaddr which must be freed later with FREE(). The caller must 257 * initialize *name to NULL. 258 */ 259 int 260 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 261 { 262 struct thread *td = curthread; 263 struct filedesc *fdp = td->td_proc->p_fd; 264 struct file *lfp = NULL; 265 struct file *nfp = NULL; 266 struct sockaddr *sa; 267 struct socket *head, *so; 268 struct netmsg_so_notify msg; 269 int fd; 270 u_int fflag; /* type must match fp->f_flag */ 271 int error, tmp; 272 273 *res = -1; 274 if (name && namelen && *namelen < 0) 275 return (EINVAL); 276 277 error = holdsock(td->td_proc->p_fd, s, &lfp); 278 if (error) 279 return (error); 280 281 error = falloc(td->td_lwp, &nfp, &fd); 282 if (error) { /* Probably ran out of file descriptors. */ 283 fdrop(lfp); 284 return (error); 285 } 286 head = (struct socket *)lfp->f_data; 287 if ((head->so_options & SO_ACCEPTCONN) == 0) { 288 error = EINVAL; 289 goto done; 290 } 291 292 if (fflags & O_FBLOCKING) 293 fflags |= lfp->f_flag & ~FNONBLOCK; 294 else if (fflags & O_FNONBLOCKING) 295 fflags |= lfp->f_flag | FNONBLOCK; 296 else 297 fflags = lfp->f_flag; 298 299 /* optimize for uniprocessor case later XXX JH */ 300 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 301 0, netmsg_so_notify, netmsg_so_notify_doabort); 302 msg.nm_predicate = soaccept_predicate; 303 msg.nm_fflags = fflags; 304 msg.nm_etype = NM_REVENT; 305 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 306 if (error) 307 goto done; 308 309 /* 310 * At this point we have the connection that's ready to be accepted. 311 * 312 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 313 * to eat the ref and turn it into a descriptor. 314 */ 315 so = msg.base.nm_so; 316 317 fflag = lfp->f_flag; 318 319 /* connection has been removed from the listen queue */ 320 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 321 322 if (head->so_sigio != NULL) 323 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 324 325 nfp->f_type = DTYPE_SOCKET; 326 nfp->f_flag = fflag; 327 nfp->f_ops = &socketops; 328 nfp->f_data = so; 329 /* Sync socket nonblocking/async state with file flags */ 330 tmp = fflag & FNONBLOCK; 331 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 332 tmp = fflag & FASYNC; 333 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 334 335 sa = NULL; 336 error = soaccept(so, &sa); 337 338 /* 339 * Set the returned name and namelen as applicable. Set the returned 340 * namelen to 0 for older code which might ignore the return value 341 * from accept. 342 */ 343 if (error == 0) { 344 if (sa && name && namelen) { 345 if (*namelen > sa->sa_len) 346 *namelen = sa->sa_len; 347 *name = sa; 348 } else { 349 if (sa) 350 FREE(sa, M_SONAME); 351 } 352 } 353 354 done: 355 /* 356 * If an error occured clear the reserved descriptor, else associate 357 * nfp with it. 358 * 359 * Note that *res is normally ignored if an error is returned but 360 * a syscall message will still have access to the result code. 361 */ 362 if (error) { 363 fsetfd(fdp, NULL, fd); 364 } else { 365 *res = fd; 366 fsetfd(fdp, nfp, fd); 367 } 368 fdrop(nfp); 369 fdrop(lfp); 370 return (error); 371 } 372 373 /* 374 * accept(int s, caddr_t name, int *anamelen) 375 * 376 * MPALMOSTSAFE 377 */ 378 int 379 sys_accept(struct accept_args *uap) 380 { 381 struct sockaddr *sa = NULL; 382 int sa_len; 383 int error; 384 385 if (uap->name) { 386 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 387 if (error) 388 return (error); 389 390 error = kern_accept(uap->s, 0, &sa, &sa_len, 391 &uap->sysmsg_iresult); 392 393 if (error == 0) 394 error = copyout(sa, uap->name, sa_len); 395 if (error == 0) { 396 error = copyout(&sa_len, uap->anamelen, 397 sizeof(*uap->anamelen)); 398 } 399 if (sa) 400 FREE(sa, M_SONAME); 401 } else { 402 error = kern_accept(uap->s, 0, NULL, 0, 403 &uap->sysmsg_iresult); 404 } 405 return (error); 406 } 407 408 /* 409 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 410 * 411 * MPALMOSTSAFE 412 */ 413 int 414 sys_extaccept(struct extaccept_args *uap) 415 { 416 struct sockaddr *sa = NULL; 417 int sa_len; 418 int error; 419 int fflags = uap->flags & O_FMASK; 420 421 if (uap->name) { 422 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 423 if (error) 424 return (error); 425 426 error = kern_accept(uap->s, fflags, &sa, &sa_len, 427 &uap->sysmsg_iresult); 428 429 if (error == 0) 430 error = copyout(sa, uap->name, sa_len); 431 if (error == 0) { 432 error = copyout(&sa_len, uap->anamelen, 433 sizeof(*uap->anamelen)); 434 } 435 if (sa) 436 FREE(sa, M_SONAME); 437 } else { 438 error = kern_accept(uap->s, fflags, NULL, 0, 439 &uap->sysmsg_iresult); 440 } 441 return (error); 442 } 443 444 445 /* 446 * Returns TRUE if predicate satisfied. 447 */ 448 static boolean_t 449 soconnected_predicate(struct netmsg_so_notify *msg) 450 { 451 struct socket *so = msg->base.nm_so; 452 453 /* check predicate */ 454 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 455 msg->base.lmsg.ms_error = so->so_error; 456 return (TRUE); 457 } 458 459 return (FALSE); 460 } 461 462 int 463 kern_connect(int s, int fflags, struct sockaddr *sa) 464 { 465 struct thread *td = curthread; 466 struct proc *p = td->td_proc; 467 struct file *fp; 468 struct socket *so; 469 int error, interrupted = 0; 470 471 error = holdsock(p->p_fd, s, &fp); 472 if (error) 473 return (error); 474 so = (struct socket *)fp->f_data; 475 476 if (fflags & O_FBLOCKING) 477 /* fflags &= ~FNONBLOCK; */; 478 else if (fflags & O_FNONBLOCKING) 479 fflags |= FNONBLOCK; 480 else 481 fflags = fp->f_flag; 482 483 if (so->so_state & SS_ISCONNECTING) { 484 error = EALREADY; 485 goto done; 486 } 487 error = soconnect(so, sa, td); 488 if (error) 489 goto bad; 490 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 491 error = EINPROGRESS; 492 goto done; 493 } 494 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 495 struct netmsg_so_notify msg; 496 497 netmsg_init_abortable(&msg.base, so, 498 &curthread->td_msgport, 499 0, 500 netmsg_so_notify, 501 netmsg_so_notify_doabort); 502 msg.nm_predicate = soconnected_predicate; 503 msg.nm_etype = NM_REVENT; 504 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 505 if (error == EINTR || error == ERESTART) 506 interrupted = 1; 507 } 508 if (error == 0) { 509 error = so->so_error; 510 so->so_error = 0; 511 } 512 bad: 513 if (!interrupted) 514 soclrstate(so, SS_ISCONNECTING); 515 if (error == ERESTART) 516 error = EINTR; 517 done: 518 fdrop(fp); 519 return (error); 520 } 521 522 /* 523 * connect_args(int s, caddr_t name, int namelen) 524 * 525 * MPALMOSTSAFE 526 */ 527 int 528 sys_connect(struct connect_args *uap) 529 { 530 struct sockaddr *sa; 531 int error; 532 533 error = getsockaddr(&sa, uap->name, uap->namelen); 534 if (error) 535 return (error); 536 error = kern_connect(uap->s, 0, sa); 537 FREE(sa, M_SONAME); 538 539 return (error); 540 } 541 542 /* 543 * connect_args(int s, int fflags, caddr_t name, int namelen) 544 * 545 * MPALMOSTSAFE 546 */ 547 int 548 sys_extconnect(struct extconnect_args *uap) 549 { 550 struct sockaddr *sa; 551 int error; 552 int fflags = uap->flags & O_FMASK; 553 554 error = getsockaddr(&sa, uap->name, uap->namelen); 555 if (error) 556 return (error); 557 error = kern_connect(uap->s, fflags, sa); 558 FREE(sa, M_SONAME); 559 560 return (error); 561 } 562 563 int 564 kern_socketpair(int domain, int type, int protocol, int *sv) 565 { 566 struct thread *td = curthread; 567 struct filedesc *fdp; 568 struct file *fp1, *fp2; 569 struct socket *so1, *so2; 570 int fd1, fd2, error; 571 572 fdp = td->td_proc->p_fd; 573 error = socreate(domain, &so1, type, protocol, td); 574 if (error) 575 return (error); 576 error = socreate(domain, &so2, type, protocol, td); 577 if (error) 578 goto free1; 579 error = falloc(td->td_lwp, &fp1, &fd1); 580 if (error) 581 goto free2; 582 sv[0] = fd1; 583 fp1->f_data = so1; 584 error = falloc(td->td_lwp, &fp2, &fd2); 585 if (error) 586 goto free3; 587 fp2->f_data = so2; 588 sv[1] = fd2; 589 error = soconnect2(so1, so2); 590 if (error) 591 goto free4; 592 if (type == SOCK_DGRAM) { 593 /* 594 * Datagram socket connection is asymmetric. 595 */ 596 error = soconnect2(so2, so1); 597 if (error) 598 goto free4; 599 } 600 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 601 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 602 fp1->f_ops = fp2->f_ops = &socketops; 603 fsetfd(fdp, fp1, fd1); 604 fsetfd(fdp, fp2, fd2); 605 fdrop(fp1); 606 fdrop(fp2); 607 return (error); 608 free4: 609 fsetfd(fdp, NULL, fd2); 610 fdrop(fp2); 611 free3: 612 fsetfd(fdp, NULL, fd1); 613 fdrop(fp1); 614 free2: 615 (void)soclose(so2, 0); 616 free1: 617 (void)soclose(so1, 0); 618 return (error); 619 } 620 621 /* 622 * socketpair(int domain, int type, int protocol, int *rsv) 623 * 624 * MPALMOSTSAFE 625 */ 626 int 627 sys_socketpair(struct socketpair_args *uap) 628 { 629 int error, sockv[2]; 630 631 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 632 633 if (error == 0) 634 error = copyout(sockv, uap->rsv, sizeof(sockv)); 635 return (error); 636 } 637 638 int 639 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 640 struct mbuf *control, int flags, size_t *res) 641 { 642 struct thread *td = curthread; 643 struct lwp *lp = td->td_lwp; 644 struct proc *p = td->td_proc; 645 struct file *fp; 646 size_t len; 647 int error; 648 struct socket *so; 649 #ifdef KTRACE 650 struct iovec *ktriov = NULL; 651 struct uio ktruio; 652 #endif 653 654 error = holdsock(p->p_fd, s, &fp); 655 if (error) 656 return (error); 657 #ifdef KTRACE 658 if (KTRPOINT(td, KTR_GENIO)) { 659 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 660 661 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 662 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 663 ktruio = *auio; 664 } 665 #endif 666 len = auio->uio_resid; 667 so = (struct socket *)fp->f_data; 668 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 669 if (fp->f_flag & FNONBLOCK) 670 flags |= MSG_FNONBLOCKING; 671 } 672 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 673 if (error) { 674 if (auio->uio_resid != len && (error == ERESTART || 675 error == EINTR || error == EWOULDBLOCK)) 676 error = 0; 677 if (error == EPIPE && !(flags & MSG_NOSIGNAL)) 678 lwpsignal(p, lp, SIGPIPE); 679 } 680 #ifdef KTRACE 681 if (ktriov != NULL) { 682 if (error == 0) { 683 ktruio.uio_iov = ktriov; 684 ktruio.uio_resid = len - auio->uio_resid; 685 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 686 } 687 FREE(ktriov, M_TEMP); 688 } 689 #endif 690 if (error == 0) 691 *res = len - auio->uio_resid; 692 fdrop(fp); 693 return (error); 694 } 695 696 /* 697 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 698 * 699 * MPALMOSTSAFE 700 */ 701 int 702 sys_sendto(struct sendto_args *uap) 703 { 704 struct thread *td = curthread; 705 struct uio auio; 706 struct iovec aiov; 707 struct sockaddr *sa = NULL; 708 int error; 709 710 if (uap->to) { 711 error = getsockaddr(&sa, uap->to, uap->tolen); 712 if (error) 713 return (error); 714 } 715 aiov.iov_base = uap->buf; 716 aiov.iov_len = uap->len; 717 auio.uio_iov = &aiov; 718 auio.uio_iovcnt = 1; 719 auio.uio_offset = 0; 720 auio.uio_resid = uap->len; 721 auio.uio_segflg = UIO_USERSPACE; 722 auio.uio_rw = UIO_WRITE; 723 auio.uio_td = td; 724 725 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 726 &uap->sysmsg_szresult); 727 728 if (sa) 729 FREE(sa, M_SONAME); 730 return (error); 731 } 732 733 /* 734 * sendmsg_args(int s, caddr_t msg, int flags) 735 * 736 * MPALMOSTSAFE 737 */ 738 int 739 sys_sendmsg(struct sendmsg_args *uap) 740 { 741 struct thread *td = curthread; 742 struct msghdr msg; 743 struct uio auio; 744 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 745 struct sockaddr *sa = NULL; 746 struct mbuf *control = NULL; 747 int error; 748 749 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 750 if (error) 751 return (error); 752 753 /* 754 * Conditionally copyin msg.msg_name. 755 */ 756 if (msg.msg_name) { 757 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 758 if (error) 759 return (error); 760 } 761 762 /* 763 * Populate auio. 764 */ 765 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 766 &auio.uio_resid); 767 if (error) 768 goto cleanup2; 769 auio.uio_iov = iov; 770 auio.uio_iovcnt = msg.msg_iovlen; 771 auio.uio_offset = 0; 772 auio.uio_segflg = UIO_USERSPACE; 773 auio.uio_rw = UIO_WRITE; 774 auio.uio_td = td; 775 776 /* 777 * Conditionally copyin msg.msg_control. 778 */ 779 if (msg.msg_control) { 780 if (msg.msg_controllen < sizeof(struct cmsghdr) || 781 msg.msg_controllen > MLEN) { 782 error = EINVAL; 783 goto cleanup; 784 } 785 control = m_get(MB_WAIT, MT_CONTROL); 786 if (control == NULL) { 787 error = ENOBUFS; 788 goto cleanup; 789 } 790 control->m_len = msg.msg_controllen; 791 error = copyin(msg.msg_control, mtod(control, caddr_t), 792 msg.msg_controllen); 793 if (error) { 794 m_free(control); 795 goto cleanup; 796 } 797 } 798 799 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 800 &uap->sysmsg_szresult); 801 802 cleanup: 803 iovec_free(&iov, aiov); 804 cleanup2: 805 if (sa) 806 FREE(sa, M_SONAME); 807 return (error); 808 } 809 810 /* 811 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 812 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 813 * Don't forget to FREE() and m_free() these if they are returned. 814 */ 815 int 816 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 817 struct mbuf **control, int *flags, size_t *res) 818 { 819 struct thread *td = curthread; 820 struct proc *p = td->td_proc; 821 struct file *fp; 822 size_t len; 823 int error; 824 int lflags; 825 struct socket *so; 826 #ifdef KTRACE 827 struct iovec *ktriov = NULL; 828 struct uio ktruio; 829 #endif 830 831 error = holdsock(p->p_fd, s, &fp); 832 if (error) 833 return (error); 834 #ifdef KTRACE 835 if (KTRPOINT(td, KTR_GENIO)) { 836 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 837 838 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 839 bcopy(auio->uio_iov, ktriov, iovlen); 840 ktruio = *auio; 841 } 842 #endif 843 len = auio->uio_resid; 844 so = (struct socket *)fp->f_data; 845 846 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 847 if (fp->f_flag & FNONBLOCK) { 848 if (flags) { 849 *flags |= MSG_FNONBLOCKING; 850 } else { 851 lflags = MSG_FNONBLOCKING; 852 flags = &lflags; 853 } 854 } 855 } 856 857 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 858 if (error) { 859 if (auio->uio_resid != len && (error == ERESTART || 860 error == EINTR || error == EWOULDBLOCK)) 861 error = 0; 862 } 863 #ifdef KTRACE 864 if (ktriov != NULL) { 865 if (error == 0) { 866 ktruio.uio_iov = ktriov; 867 ktruio.uio_resid = len - auio->uio_resid; 868 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 869 } 870 FREE(ktriov, M_TEMP); 871 } 872 #endif 873 if (error == 0) 874 *res = len - auio->uio_resid; 875 fdrop(fp); 876 return (error); 877 } 878 879 /* 880 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 881 * caddr_t from, int *fromlenaddr) 882 * 883 * MPALMOSTSAFE 884 */ 885 int 886 sys_recvfrom(struct recvfrom_args *uap) 887 { 888 struct thread *td = curthread; 889 struct uio auio; 890 struct iovec aiov; 891 struct sockaddr *sa = NULL; 892 int error, fromlen; 893 894 if (uap->from && uap->fromlenaddr) { 895 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 896 if (error) 897 return (error); 898 if (fromlen < 0) 899 return (EINVAL); 900 } else { 901 fromlen = 0; 902 } 903 aiov.iov_base = uap->buf; 904 aiov.iov_len = uap->len; 905 auio.uio_iov = &aiov; 906 auio.uio_iovcnt = 1; 907 auio.uio_offset = 0; 908 auio.uio_resid = uap->len; 909 auio.uio_segflg = UIO_USERSPACE; 910 auio.uio_rw = UIO_READ; 911 auio.uio_td = td; 912 913 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 914 &uap->flags, &uap->sysmsg_szresult); 915 916 if (error == 0 && uap->from) { 917 /* note: sa may still be NULL */ 918 if (sa) { 919 fromlen = MIN(fromlen, sa->sa_len); 920 error = copyout(sa, uap->from, fromlen); 921 } else { 922 fromlen = 0; 923 } 924 if (error == 0) { 925 error = copyout(&fromlen, uap->fromlenaddr, 926 sizeof(fromlen)); 927 } 928 } 929 if (sa) 930 FREE(sa, M_SONAME); 931 932 return (error); 933 } 934 935 /* 936 * recvmsg_args(int s, struct msghdr *msg, int flags) 937 * 938 * MPALMOSTSAFE 939 */ 940 int 941 sys_recvmsg(struct recvmsg_args *uap) 942 { 943 struct thread *td = curthread; 944 struct msghdr msg; 945 struct uio auio; 946 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 947 struct mbuf *m, *control = NULL; 948 struct sockaddr *sa = NULL; 949 caddr_t ctlbuf; 950 socklen_t *ufromlenp, *ucontrollenp; 951 int error, fromlen, controllen, len, flags, *uflagsp; 952 953 /* 954 * This copyin handles everything except the iovec. 955 */ 956 error = copyin(uap->msg, &msg, sizeof(msg)); 957 if (error) 958 return (error); 959 960 if (msg.msg_name && msg.msg_namelen < 0) 961 return (EINVAL); 962 if (msg.msg_control && msg.msg_controllen < 0) 963 return (EINVAL); 964 965 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 966 msg_namelen)); 967 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 968 msg_controllen)); 969 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 970 msg_flags)); 971 972 /* 973 * Populate auio. 974 */ 975 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 976 &auio.uio_resid); 977 if (error) 978 return (error); 979 auio.uio_iov = iov; 980 auio.uio_iovcnt = msg.msg_iovlen; 981 auio.uio_offset = 0; 982 auio.uio_segflg = UIO_USERSPACE; 983 auio.uio_rw = UIO_READ; 984 auio.uio_td = td; 985 986 flags = uap->flags; 987 988 error = kern_recvmsg(uap->s, 989 (msg.msg_name ? &sa : NULL), &auio, 990 (msg.msg_control ? &control : NULL), &flags, 991 &uap->sysmsg_szresult); 992 993 /* 994 * Conditionally copyout the name and populate the namelen field. 995 */ 996 if (error == 0 && msg.msg_name) { 997 /* note: sa may still be NULL */ 998 if (sa != NULL) { 999 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1000 error = copyout(sa, msg.msg_name, fromlen); 1001 } else { 1002 fromlen = 0; 1003 } 1004 if (error == 0) 1005 error = copyout(&fromlen, ufromlenp, 1006 sizeof(*ufromlenp)); 1007 } 1008 1009 /* 1010 * Copyout msg.msg_control and msg.msg_controllen. 1011 */ 1012 if (error == 0 && msg.msg_control) { 1013 len = msg.msg_controllen; 1014 m = control; 1015 ctlbuf = (caddr_t)msg.msg_control; 1016 1017 while(m && len > 0) { 1018 unsigned int tocopy; 1019 1020 if (len >= m->m_len) { 1021 tocopy = m->m_len; 1022 } else { 1023 msg.msg_flags |= MSG_CTRUNC; 1024 tocopy = len; 1025 } 1026 1027 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1028 if (error) 1029 goto cleanup; 1030 1031 ctlbuf += tocopy; 1032 len -= tocopy; 1033 m = m->m_next; 1034 } 1035 controllen = ctlbuf - (caddr_t)msg.msg_control; 1036 error = copyout(&controllen, ucontrollenp, 1037 sizeof(*ucontrollenp)); 1038 } 1039 1040 if (error == 0) 1041 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1042 1043 cleanup: 1044 if (sa) 1045 FREE(sa, M_SONAME); 1046 iovec_free(&iov, aiov); 1047 if (control) 1048 m_freem(control); 1049 return (error); 1050 } 1051 1052 /* 1053 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1054 * in kernel pointer instead of a userland pointer. This allows us 1055 * to manipulate socket options in the emulation code. 1056 */ 1057 int 1058 kern_setsockopt(int s, struct sockopt *sopt) 1059 { 1060 struct thread *td = curthread; 1061 struct proc *p = td->td_proc; 1062 struct file *fp; 1063 int error; 1064 1065 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1066 return (EFAULT); 1067 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1068 return (EINVAL); 1069 if (sopt->sopt_valsize < 0) 1070 return (EINVAL); 1071 1072 error = holdsock(p->p_fd, s, &fp); 1073 if (error) 1074 return (error); 1075 1076 error = sosetopt((struct socket *)fp->f_data, sopt); 1077 fdrop(fp); 1078 return (error); 1079 } 1080 1081 /* 1082 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1083 * 1084 * MPALMOSTSAFE 1085 */ 1086 int 1087 sys_setsockopt(struct setsockopt_args *uap) 1088 { 1089 struct thread *td = curthread; 1090 struct sockopt sopt; 1091 int error; 1092 1093 sopt.sopt_level = uap->level; 1094 sopt.sopt_name = uap->name; 1095 sopt.sopt_valsize = uap->valsize; 1096 sopt.sopt_td = td; 1097 sopt.sopt_val = NULL; 1098 1099 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE) 1100 return (EINVAL); 1101 if (uap->val) { 1102 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1103 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1104 if (error) 1105 goto out; 1106 } 1107 1108 error = kern_setsockopt(uap->s, &sopt); 1109 out: 1110 if (uap->val) 1111 kfree(sopt.sopt_val, M_TEMP); 1112 return(error); 1113 } 1114 1115 /* 1116 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1117 * in kernel pointer instead of a userland pointer. This allows us 1118 * to manipulate socket options in the emulation code. 1119 */ 1120 int 1121 kern_getsockopt(int s, struct sockopt *sopt) 1122 { 1123 struct thread *td = curthread; 1124 struct proc *p = td->td_proc; 1125 struct file *fp; 1126 int error; 1127 1128 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1129 return (EFAULT); 1130 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1131 return (EINVAL); 1132 if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE) 1133 return (EINVAL); 1134 1135 error = holdsock(p->p_fd, s, &fp); 1136 if (error) 1137 return (error); 1138 1139 error = sogetopt((struct socket *)fp->f_data, sopt); 1140 fdrop(fp); 1141 return (error); 1142 } 1143 1144 /* 1145 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1146 * 1147 * MPALMOSTSAFE 1148 */ 1149 int 1150 sys_getsockopt(struct getsockopt_args *uap) 1151 { 1152 struct thread *td = curthread; 1153 struct sockopt sopt; 1154 int error, valsize; 1155 1156 if (uap->val) { 1157 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1158 if (error) 1159 return (error); 1160 } else { 1161 valsize = 0; 1162 } 1163 1164 sopt.sopt_level = uap->level; 1165 sopt.sopt_name = uap->name; 1166 sopt.sopt_valsize = valsize; 1167 sopt.sopt_td = td; 1168 sopt.sopt_val = NULL; 1169 1170 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE) 1171 return (EINVAL); 1172 if (uap->val) { 1173 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1174 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1175 if (error) 1176 goto out; 1177 } 1178 1179 error = kern_getsockopt(uap->s, &sopt); 1180 if (error) 1181 goto out; 1182 valsize = sopt.sopt_valsize; 1183 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1184 if (error) 1185 goto out; 1186 if (uap->val) 1187 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1188 out: 1189 if (uap->val) 1190 kfree(sopt.sopt_val, M_TEMP); 1191 return (error); 1192 } 1193 1194 /* 1195 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1196 * This allows kern_getsockname() to return a pointer to an allocated struct 1197 * sockaddr which must be freed later with FREE(). The caller must 1198 * initialize *name to NULL. 1199 */ 1200 int 1201 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1202 { 1203 struct thread *td = curthread; 1204 struct proc *p = td->td_proc; 1205 struct file *fp; 1206 struct socket *so; 1207 struct sockaddr *sa = NULL; 1208 int error; 1209 1210 error = holdsock(p->p_fd, s, &fp); 1211 if (error) 1212 return (error); 1213 if (*namelen < 0) { 1214 fdrop(fp); 1215 return (EINVAL); 1216 } 1217 so = (struct socket *)fp->f_data; 1218 error = so_pru_sockaddr(so, &sa); 1219 if (error == 0) { 1220 if (sa == NULL) { 1221 *namelen = 0; 1222 } else { 1223 *namelen = MIN(*namelen, sa->sa_len); 1224 *name = sa; 1225 } 1226 } 1227 1228 fdrop(fp); 1229 return (error); 1230 } 1231 1232 /* 1233 * getsockname_args(int fdes, caddr_t asa, int *alen) 1234 * 1235 * Get socket name. 1236 * 1237 * MPALMOSTSAFE 1238 */ 1239 int 1240 sys_getsockname(struct getsockname_args *uap) 1241 { 1242 struct sockaddr *sa = NULL; 1243 int error, sa_len; 1244 1245 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1246 if (error) 1247 return (error); 1248 1249 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1250 1251 if (error == 0) 1252 error = copyout(sa, uap->asa, sa_len); 1253 if (error == 0) 1254 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1255 if (sa) 1256 FREE(sa, M_SONAME); 1257 return (error); 1258 } 1259 1260 /* 1261 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1262 * This allows kern_getpeername() to return a pointer to an allocated struct 1263 * sockaddr which must be freed later with FREE(). The caller must 1264 * initialize *name to NULL. 1265 */ 1266 int 1267 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1268 { 1269 struct thread *td = curthread; 1270 struct proc *p = td->td_proc; 1271 struct file *fp; 1272 struct socket *so; 1273 struct sockaddr *sa = NULL; 1274 int error; 1275 1276 error = holdsock(p->p_fd, s, &fp); 1277 if (error) 1278 return (error); 1279 if (*namelen < 0) { 1280 fdrop(fp); 1281 return (EINVAL); 1282 } 1283 so = (struct socket *)fp->f_data; 1284 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1285 fdrop(fp); 1286 return (ENOTCONN); 1287 } 1288 error = so_pru_peeraddr(so, &sa); 1289 if (error == 0) { 1290 if (sa == NULL) { 1291 *namelen = 0; 1292 } else { 1293 *namelen = MIN(*namelen, sa->sa_len); 1294 *name = sa; 1295 } 1296 } 1297 1298 fdrop(fp); 1299 return (error); 1300 } 1301 1302 /* 1303 * getpeername_args(int fdes, caddr_t asa, int *alen) 1304 * 1305 * Get name of peer for connected socket. 1306 * 1307 * MPALMOSTSAFE 1308 */ 1309 int 1310 sys_getpeername(struct getpeername_args *uap) 1311 { 1312 struct sockaddr *sa = NULL; 1313 int error, sa_len; 1314 1315 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1316 if (error) 1317 return (error); 1318 1319 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1320 1321 if (error == 0) 1322 error = copyout(sa, uap->asa, sa_len); 1323 if (error == 0) 1324 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1325 if (sa) 1326 FREE(sa, M_SONAME); 1327 return (error); 1328 } 1329 1330 int 1331 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1332 { 1333 struct sockaddr *sa; 1334 int error; 1335 1336 *namp = NULL; 1337 if (len > SOCK_MAXADDRLEN) 1338 return ENAMETOOLONG; 1339 if (len < offsetof(struct sockaddr, sa_data[0])) 1340 return EDOM; 1341 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1342 error = copyin(uaddr, sa, len); 1343 if (error) { 1344 FREE(sa, M_SONAME); 1345 } else { 1346 #if BYTE_ORDER != BIG_ENDIAN 1347 /* 1348 * The bind(), connect(), and sendto() syscalls were not 1349 * versioned for COMPAT_43. Thus, this check must stay. 1350 */ 1351 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1352 sa->sa_family = sa->sa_len; 1353 #endif 1354 sa->sa_len = len; 1355 *namp = sa; 1356 } 1357 return error; 1358 } 1359 1360 /* 1361 * Detach a mapped page and release resources back to the system. 1362 * We must release our wiring and if the object is ripped out 1363 * from under the vm_page we become responsible for freeing the 1364 * page. 1365 * 1366 * MPSAFE 1367 */ 1368 static void 1369 sf_buf_mfree(void *arg) 1370 { 1371 struct sf_buf *sf = arg; 1372 vm_page_t m; 1373 1374 m = sf_buf_page(sf); 1375 if (sf_buf_free(sf)) { 1376 /* sf invalid now */ 1377 vm_page_unwire(m, 0); 1378 if (m->wire_count == 0 && m->object == NULL) 1379 vm_page_try_to_free(m); 1380 } 1381 } 1382 1383 /* 1384 * sendfile(2). 1385 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1386 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1387 * 1388 * Send a file specified by 'fd' and starting at 'offset' to a socket 1389 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1390 * nbytes == 0. Optionally add a header and/or trailer to the socket 1391 * output. If specified, write the total number of bytes sent into *sbytes. 1392 * 1393 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1394 * the headers to count against the remaining bytes to be sent from 1395 * the file descriptor. We may wish to implement a compatibility syscall 1396 * in the future. 1397 * 1398 * MPALMOSTSAFE 1399 */ 1400 int 1401 sys_sendfile(struct sendfile_args *uap) 1402 { 1403 struct thread *td = curthread; 1404 struct proc *p = td->td_proc; 1405 struct file *fp; 1406 struct vnode *vp = NULL; 1407 struct sf_hdtr hdtr; 1408 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1409 struct uio auio; 1410 struct mbuf *mheader = NULL; 1411 size_t hbytes = 0; 1412 size_t tbytes; 1413 off_t hdtr_size = 0; 1414 off_t sbytes; 1415 int error; 1416 1417 KKASSERT(p); 1418 1419 /* 1420 * Do argument checking. Must be a regular file in, stream 1421 * type and connected socket out, positive offset. 1422 */ 1423 fp = holdfp(p->p_fd, uap->fd, FREAD); 1424 if (fp == NULL) { 1425 return (EBADF); 1426 } 1427 if (fp->f_type != DTYPE_VNODE) { 1428 fdrop(fp); 1429 return (EINVAL); 1430 } 1431 vp = (struct vnode *)fp->f_data; 1432 vref(vp); 1433 fdrop(fp); 1434 1435 /* 1436 * If specified, get the pointer to the sf_hdtr struct for 1437 * any headers/trailers. 1438 */ 1439 if (uap->hdtr) { 1440 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1441 if (error) 1442 goto done; 1443 /* 1444 * Send any headers. 1445 */ 1446 if (hdtr.headers) { 1447 error = iovec_copyin(hdtr.headers, &iov, aiov, 1448 hdtr.hdr_cnt, &hbytes); 1449 if (error) 1450 goto done; 1451 auio.uio_iov = iov; 1452 auio.uio_iovcnt = hdtr.hdr_cnt; 1453 auio.uio_offset = 0; 1454 auio.uio_segflg = UIO_USERSPACE; 1455 auio.uio_rw = UIO_WRITE; 1456 auio.uio_td = td; 1457 auio.uio_resid = hbytes; 1458 1459 mheader = m_uiomove(&auio); 1460 1461 iovec_free(&iov, aiov); 1462 if (mheader == NULL) 1463 goto done; 1464 } 1465 } 1466 1467 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1468 &sbytes, uap->flags); 1469 if (error) 1470 goto done; 1471 1472 /* 1473 * Send trailers. Wimp out and use writev(2). 1474 */ 1475 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1476 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1477 hdtr.trl_cnt, &auio.uio_resid); 1478 if (error) 1479 goto done; 1480 auio.uio_iov = iov; 1481 auio.uio_iovcnt = hdtr.trl_cnt; 1482 auio.uio_offset = 0; 1483 auio.uio_segflg = UIO_USERSPACE; 1484 auio.uio_rw = UIO_WRITE; 1485 auio.uio_td = td; 1486 1487 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1488 1489 iovec_free(&iov, aiov); 1490 if (error) 1491 goto done; 1492 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1493 } 1494 1495 done: 1496 if (vp) 1497 vrele(vp); 1498 if (uap->sbytes != NULL) { 1499 sbytes += hdtr_size; 1500 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1501 } 1502 return (error); 1503 } 1504 1505 int 1506 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1507 struct mbuf *mheader, off_t *sbytes, int flags) 1508 { 1509 struct thread *td = curthread; 1510 struct proc *p = td->td_proc; 1511 struct vm_object *obj; 1512 struct socket *so; 1513 struct file *fp; 1514 struct mbuf *m; 1515 struct sf_buf *sf; 1516 struct vm_page *pg; 1517 off_t off, xfsize; 1518 off_t hbytes = 0; 1519 int error = 0; 1520 1521 if (vp->v_type != VREG) { 1522 error = EINVAL; 1523 goto done0; 1524 } 1525 if ((obj = vp->v_object) == NULL) { 1526 error = EINVAL; 1527 goto done0; 1528 } 1529 error = holdsock(p->p_fd, sfd, &fp); 1530 if (error) 1531 goto done0; 1532 so = (struct socket *)fp->f_data; 1533 if (so->so_type != SOCK_STREAM) { 1534 error = EINVAL; 1535 goto done; 1536 } 1537 if ((so->so_state & SS_ISCONNECTED) == 0) { 1538 error = ENOTCONN; 1539 goto done; 1540 } 1541 if (offset < 0) { 1542 error = EINVAL; 1543 goto done; 1544 } 1545 1546 *sbytes = 0; 1547 /* 1548 * Protect against multiple writers to the socket. 1549 */ 1550 ssb_lock(&so->so_snd, M_WAITOK); 1551 1552 /* 1553 * Loop through the pages in the file, starting with the requested 1554 * offset. Get a file page (do I/O if necessary), map the file page 1555 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1556 * it on the socket. 1557 */ 1558 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1559 vm_pindex_t pindex; 1560 vm_offset_t pgoff; 1561 1562 pindex = OFF_TO_IDX(off); 1563 retry_lookup: 1564 /* 1565 * Calculate the amount to transfer. Not to exceed a page, 1566 * the EOF, or the passed in nbytes. 1567 */ 1568 xfsize = vp->v_filesize - off; 1569 if (xfsize > PAGE_SIZE) 1570 xfsize = PAGE_SIZE; 1571 pgoff = (vm_offset_t)(off & PAGE_MASK); 1572 if (PAGE_SIZE - pgoff < xfsize) 1573 xfsize = PAGE_SIZE - pgoff; 1574 if (nbytes && xfsize > (nbytes - *sbytes)) 1575 xfsize = nbytes - *sbytes; 1576 if (xfsize <= 0) 1577 break; 1578 /* 1579 * Optimize the non-blocking case by looking at the socket space 1580 * before going to the extra work of constituting the sf_buf. 1581 */ 1582 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) { 1583 if (so->so_state & SS_CANTSENDMORE) 1584 error = EPIPE; 1585 else 1586 error = EAGAIN; 1587 ssb_unlock(&so->so_snd); 1588 goto done; 1589 } 1590 /* 1591 * Attempt to look up the page. 1592 * 1593 * Allocate if not found, wait and loop if busy, then 1594 * wire the page. critical section protection is 1595 * required to maintain the object association (an 1596 * interrupt can free the page) through to the 1597 * vm_page_wire() call. 1598 */ 1599 lwkt_gettoken(&vm_token); 1600 pg = vm_page_lookup(obj, pindex); 1601 if (pg == NULL) { 1602 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1603 if (pg == NULL) { 1604 vm_wait(0); 1605 lwkt_reltoken(&vm_token); 1606 goto retry_lookup; 1607 } 1608 vm_page_wire(pg); 1609 vm_page_wakeup(pg); 1610 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1611 lwkt_reltoken(&vm_token); 1612 goto retry_lookup; 1613 } else { 1614 vm_page_wire(pg); 1615 } 1616 lwkt_reltoken(&vm_token); 1617 1618 /* 1619 * If page is not valid for what we need, initiate I/O 1620 */ 1621 1622 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1623 struct uio auio; 1624 struct iovec aiov; 1625 int bsize; 1626 1627 /* 1628 * Ensure that our page is still around when the I/O 1629 * completes. 1630 */ 1631 vm_page_io_start(pg); 1632 1633 /* 1634 * Get the page from backing store. 1635 */ 1636 bsize = vp->v_mount->mnt_stat.f_iosize; 1637 auio.uio_iov = &aiov; 1638 auio.uio_iovcnt = 1; 1639 aiov.iov_base = 0; 1640 aiov.iov_len = MAXBSIZE; 1641 auio.uio_resid = MAXBSIZE; 1642 auio.uio_offset = trunc_page(off); 1643 auio.uio_segflg = UIO_NOCOPY; 1644 auio.uio_rw = UIO_READ; 1645 auio.uio_td = td; 1646 vn_lock(vp, LK_SHARED | LK_RETRY); 1647 error = VOP_READ(vp, &auio, 1648 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1649 td->td_ucred); 1650 vn_unlock(vp); 1651 vm_page_flag_clear(pg, PG_ZERO); 1652 vm_page_io_finish(pg); 1653 if (error) { 1654 crit_enter(); 1655 vm_page_unwire(pg, 0); 1656 vm_page_try_to_free(pg); 1657 crit_exit(); 1658 ssb_unlock(&so->so_snd); 1659 goto done; 1660 } 1661 } 1662 1663 1664 /* 1665 * Get a sendfile buf. We usually wait as long as necessary, 1666 * but this wait can be interrupted. 1667 */ 1668 if ((sf = sf_buf_alloc(pg)) == NULL) { 1669 crit_enter(); 1670 vm_page_unwire(pg, 0); 1671 vm_page_try_to_free(pg); 1672 crit_exit(); 1673 ssb_unlock(&so->so_snd); 1674 error = EINTR; 1675 goto done; 1676 } 1677 1678 /* 1679 * Get an mbuf header and set it up as having external storage. 1680 */ 1681 MGETHDR(m, MB_WAIT, MT_DATA); 1682 if (m == NULL) { 1683 error = ENOBUFS; 1684 sf_buf_free(sf); 1685 ssb_unlock(&so->so_snd); 1686 goto done; 1687 } 1688 1689 m->m_ext.ext_free = sf_buf_mfree; 1690 m->m_ext.ext_ref = sf_buf_ref; 1691 m->m_ext.ext_arg = sf; 1692 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1693 m->m_ext.ext_size = PAGE_SIZE; 1694 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1695 m->m_flags |= M_EXT; 1696 m->m_pkthdr.len = m->m_len = xfsize; 1697 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1698 1699 if (mheader != NULL) { 1700 hbytes = mheader->m_pkthdr.len; 1701 mheader->m_pkthdr.len += m->m_pkthdr.len; 1702 m_cat(mheader, m); 1703 m = mheader; 1704 mheader = NULL; 1705 } else 1706 hbytes = 0; 1707 1708 /* 1709 * Add the buffer to the socket buffer chain. 1710 */ 1711 crit_enter(); 1712 retry_space: 1713 /* 1714 * Make sure that the socket is still able to take more data. 1715 * CANTSENDMORE being true usually means that the connection 1716 * was closed. so_error is true when an error was sensed after 1717 * a previous send. 1718 * The state is checked after the page mapping and buffer 1719 * allocation above since those operations may block and make 1720 * any socket checks stale. From this point forward, nothing 1721 * blocks before the pru_send (or more accurately, any blocking 1722 * results in a loop back to here to re-check). 1723 */ 1724 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1725 if (so->so_state & SS_CANTSENDMORE) { 1726 error = EPIPE; 1727 } else { 1728 error = so->so_error; 1729 so->so_error = 0; 1730 } 1731 m_freem(m); 1732 ssb_unlock(&so->so_snd); 1733 crit_exit(); 1734 goto done; 1735 } 1736 /* 1737 * Wait for socket space to become available. We do this just 1738 * after checking the connection state above in order to avoid 1739 * a race condition with ssb_wait(). 1740 */ 1741 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) { 1742 if (fp->f_flag & FNONBLOCK) { 1743 m_freem(m); 1744 ssb_unlock(&so->so_snd); 1745 crit_exit(); 1746 error = EAGAIN; 1747 goto done; 1748 } 1749 error = ssb_wait(&so->so_snd); 1750 /* 1751 * An error from ssb_wait usually indicates that we've 1752 * been interrupted by a signal. If we've sent anything 1753 * then return bytes sent, otherwise return the error. 1754 */ 1755 if (error) { 1756 m_freem(m); 1757 ssb_unlock(&so->so_snd); 1758 crit_exit(); 1759 goto done; 1760 } 1761 goto retry_space; 1762 } 1763 error = so_pru_send(so, 0, m, NULL, NULL, td); 1764 crit_exit(); 1765 if (error) { 1766 ssb_unlock(&so->so_snd); 1767 goto done; 1768 } 1769 } 1770 if (mheader != NULL) { 1771 *sbytes += mheader->m_pkthdr.len; 1772 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1773 mheader = NULL; 1774 } 1775 ssb_unlock(&so->so_snd); 1776 1777 done: 1778 fdrop(fp); 1779 done0: 1780 if (mheader != NULL) 1781 m_freem(mheader); 1782 return (error); 1783 } 1784 1785 /* 1786 * MPALMOSTSAFE 1787 */ 1788 int 1789 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1790 { 1791 #ifdef SCTP 1792 struct thread *td = curthread; 1793 struct filedesc *fdp = td->td_proc->p_fd; 1794 struct file *lfp = NULL; 1795 struct file *nfp = NULL; 1796 int error; 1797 struct socket *head, *so; 1798 caddr_t assoc_id; 1799 int fd; 1800 short fflag; /* type must match fp->f_flag */ 1801 1802 assoc_id = uap->name; 1803 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp); 1804 if (error) 1805 return (error); 1806 1807 crit_enter(); 1808 head = (struct socket *)lfp->f_data; 1809 error = sctp_can_peel_off(head, assoc_id); 1810 if (error) { 1811 crit_exit(); 1812 goto done; 1813 } 1814 /* 1815 * At this point we know we do have a assoc to pull 1816 * we proceed to get the fd setup. This may block 1817 * but that is ok. 1818 */ 1819 1820 fflag = lfp->f_flag; 1821 error = falloc(td->td_lwp, &nfp, &fd); 1822 if (error) { 1823 /* 1824 * Probably ran out of file descriptors. Put the 1825 * unaccepted connection back onto the queue and 1826 * do another wakeup so some other process might 1827 * have a chance at it. 1828 */ 1829 crit_exit(); 1830 goto done; 1831 } 1832 uap->sysmsg_iresult = fd; 1833 1834 so = sctp_get_peeloff(head, assoc_id, &error); 1835 if (so == NULL) { 1836 /* 1837 * Either someone else peeled it off OR 1838 * we can't get a socket. 1839 */ 1840 goto noconnection; 1841 } 1842 soreference(so); /* reference needed */ 1843 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */ 1844 so->so_head = NULL; 1845 if (head->so_sigio != NULL) 1846 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 1847 1848 nfp->f_type = DTYPE_SOCKET; 1849 nfp->f_flag = fflag; 1850 nfp->f_ops = &socketops; 1851 nfp->f_data = so; 1852 1853 noconnection: 1854 /* 1855 * Assign the file pointer to the reserved descriptor, or clear 1856 * the reserved descriptor if an error occured. 1857 */ 1858 if (error) 1859 fsetfd(fdp, NULL, fd); 1860 else 1861 fsetfd(fdp, nfp, fd); 1862 crit_exit(); 1863 /* 1864 * Release explicitly held references before returning. 1865 */ 1866 done: 1867 if (nfp != NULL) 1868 fdrop(nfp); 1869 fdrop(lfp); 1870 return (error); 1871 #else /* SCTP */ 1872 return(EOPNOTSUPP); 1873 #endif /* SCTP */ 1874 } 1875