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 <sys/mplock2.h> 82 #include <net/netmsg2.h> 83 84 #ifdef SCTP 85 #include <netinet/sctp_peeloff.h> 86 #endif /* SCTP */ 87 88 /* 89 * System call interface to the socket abstraction. 90 */ 91 92 extern struct fileops socketops; 93 94 /* 95 * socket_args(int domain, int type, int protocol) 96 */ 97 int 98 kern_socket(int domain, int type, int protocol, int *res) 99 { 100 struct thread *td = curthread; 101 struct filedesc *fdp = td->td_proc->p_fd; 102 struct socket *so; 103 struct file *fp; 104 int fd, error; 105 106 KKASSERT(td->td_lwp); 107 108 error = falloc(td->td_lwp, &fp, &fd); 109 if (error) 110 return (error); 111 error = socreate(domain, &so, type, protocol, td); 112 if (error) { 113 fsetfd(fdp, NULL, fd); 114 } else { 115 fp->f_type = DTYPE_SOCKET; 116 fp->f_flag = FREAD | FWRITE; 117 fp->f_ops = &socketops; 118 fp->f_data = so; 119 *res = fd; 120 fsetfd(fdp, fp, fd); 121 } 122 fdrop(fp); 123 return (error); 124 } 125 126 /* 127 * MPALMOSTSAFE 128 */ 129 int 130 sys_socket(struct socket_args *uap) 131 { 132 int error; 133 134 get_mplock(); 135 error = kern_socket(uap->domain, uap->type, uap->protocol, 136 &uap->sysmsg_iresult); 137 rel_mplock(); 138 139 return (error); 140 } 141 142 int 143 kern_bind(int s, struct sockaddr *sa) 144 { 145 struct thread *td = curthread; 146 struct proc *p = td->td_proc; 147 struct file *fp; 148 int error; 149 150 KKASSERT(p); 151 error = holdsock(p->p_fd, s, &fp); 152 if (error) 153 return (error); 154 error = sobind((struct socket *)fp->f_data, sa, td); 155 fdrop(fp); 156 return (error); 157 } 158 159 /* 160 * bind_args(int s, caddr_t name, int namelen) 161 * 162 * MPALMOSTSAFE 163 */ 164 int 165 sys_bind(struct bind_args *uap) 166 { 167 struct sockaddr *sa; 168 int error; 169 170 error = getsockaddr(&sa, uap->name, uap->namelen); 171 if (error) 172 return (error); 173 get_mplock(); 174 error = kern_bind(uap->s, sa); 175 rel_mplock(); 176 FREE(sa, M_SONAME); 177 178 return (error); 179 } 180 181 int 182 kern_listen(int s, int backlog) 183 { 184 struct thread *td = curthread; 185 struct proc *p = td->td_proc; 186 struct file *fp; 187 int error; 188 189 KKASSERT(p); 190 error = holdsock(p->p_fd, s, &fp); 191 if (error) 192 return (error); 193 error = solisten((struct socket *)fp->f_data, backlog, td); 194 fdrop(fp); 195 return(error); 196 } 197 198 /* 199 * listen_args(int s, int backlog) 200 * 201 * MPALMOSTSAFE 202 */ 203 int 204 sys_listen(struct listen_args *uap) 205 { 206 int error; 207 208 get_mplock(); 209 error = kern_listen(uap->s, uap->backlog); 210 rel_mplock(); 211 return (error); 212 } 213 214 /* 215 * Returns the accepted socket as well. 216 * 217 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 218 * pool token is absolutely required to avoid a sofree() race, 219 * as well as to avoid tailq handling races. 220 */ 221 static boolean_t 222 soaccept_predicate(struct netmsg_so_notify *msg) 223 { 224 struct socket *head = msg->base.nm_so; 225 struct socket *so; 226 227 if (head->so_error != 0) { 228 msg->base.lmsg.ms_error = head->so_error; 229 return (TRUE); 230 } 231 lwkt_getpooltoken(head); 232 if (!TAILQ_EMPTY(&head->so_comp)) { 233 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 234 so = TAILQ_FIRST(&head->so_comp); 235 TAILQ_REMOVE(&head->so_comp, so, so_list); 236 head->so_qlen--; 237 soclrstate(so, SS_COMP); 238 so->so_head = NULL; 239 soreference(so); 240 241 lwkt_relpooltoken(head); 242 243 msg->base.lmsg.ms_error = 0; 244 msg->base.nm_so = so; 245 return (TRUE); 246 } 247 lwkt_relpooltoken(head); 248 if (head->so_state & SS_CANTRCVMORE) { 249 msg->base.lmsg.ms_error = ECONNABORTED; 250 return (TRUE); 251 } 252 if (msg->nm_fflags & FNONBLOCK) { 253 msg->base.lmsg.ms_error = EWOULDBLOCK; 254 return (TRUE); 255 } 256 257 return (FALSE); 258 } 259 260 /* 261 * The second argument to kern_accept() is a handle to a struct sockaddr. 262 * This allows kern_accept() to return a pointer to an allocated struct 263 * sockaddr which must be freed later with FREE(). The caller must 264 * initialize *name to NULL. 265 */ 266 int 267 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 268 { 269 struct thread *td = curthread; 270 struct filedesc *fdp = td->td_proc->p_fd; 271 struct file *lfp = NULL; 272 struct file *nfp = NULL; 273 struct sockaddr *sa; 274 struct socket *head, *so; 275 struct netmsg_so_notify msg; 276 int fd; 277 u_int fflag; /* type must match fp->f_flag */ 278 int error, tmp; 279 280 *res = -1; 281 if (name && namelen && *namelen < 0) 282 return (EINVAL); 283 284 error = holdsock(td->td_proc->p_fd, s, &lfp); 285 if (error) 286 return (error); 287 288 error = falloc(td->td_lwp, &nfp, &fd); 289 if (error) { /* Probably ran out of file descriptors. */ 290 fdrop(lfp); 291 return (error); 292 } 293 head = (struct socket *)lfp->f_data; 294 if ((head->so_options & SO_ACCEPTCONN) == 0) { 295 error = EINVAL; 296 goto done; 297 } 298 299 if (fflags & O_FBLOCKING) 300 fflags |= lfp->f_flag & ~FNONBLOCK; 301 else if (fflags & O_FNONBLOCKING) 302 fflags |= lfp->f_flag | FNONBLOCK; 303 else 304 fflags = lfp->f_flag; 305 306 /* optimize for uniprocessor case later XXX JH */ 307 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 308 0, netmsg_so_notify, netmsg_so_notify_doabort); 309 msg.nm_predicate = soaccept_predicate; 310 msg.nm_fflags = fflags; 311 msg.nm_etype = NM_REVENT; 312 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 313 if (error) 314 goto done; 315 316 /* 317 * At this point we have the connection that's ready to be accepted. 318 * 319 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 320 * to eat the ref and turn it into a descriptor. 321 */ 322 so = msg.base.nm_so; 323 324 fflag = lfp->f_flag; 325 326 /* connection has been removed from the listen queue */ 327 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 328 329 if (head->so_sigio != NULL) 330 fsetown(fgetown(head->so_sigio), &so->so_sigio); 331 332 nfp->f_type = DTYPE_SOCKET; 333 nfp->f_flag = fflag; 334 nfp->f_ops = &socketops; 335 nfp->f_data = so; 336 /* Sync socket nonblocking/async state with file flags */ 337 tmp = fflag & FNONBLOCK; 338 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 339 tmp = fflag & FASYNC; 340 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 341 342 sa = NULL; 343 error = soaccept(so, &sa); 344 345 /* 346 * Set the returned name and namelen as applicable. Set the returned 347 * namelen to 0 for older code which might ignore the return value 348 * from accept. 349 */ 350 if (error == 0) { 351 if (sa && name && namelen) { 352 if (*namelen > sa->sa_len) 353 *namelen = sa->sa_len; 354 *name = sa; 355 } else { 356 if (sa) 357 FREE(sa, M_SONAME); 358 } 359 } 360 361 done: 362 /* 363 * If an error occured clear the reserved descriptor, else associate 364 * nfp with it. 365 * 366 * Note that *res is normally ignored if an error is returned but 367 * a syscall message will still have access to the result code. 368 */ 369 if (error) { 370 fsetfd(fdp, NULL, fd); 371 } else { 372 *res = fd; 373 fsetfd(fdp, nfp, fd); 374 } 375 fdrop(nfp); 376 fdrop(lfp); 377 return (error); 378 } 379 380 /* 381 * accept(int s, caddr_t name, int *anamelen) 382 * 383 * MPALMOSTSAFE 384 */ 385 int 386 sys_accept(struct accept_args *uap) 387 { 388 struct sockaddr *sa = NULL; 389 int sa_len; 390 int error; 391 392 if (uap->name) { 393 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 394 if (error) 395 return (error); 396 397 get_mplock(); 398 error = kern_accept(uap->s, 0, &sa, &sa_len, 399 &uap->sysmsg_iresult); 400 rel_mplock(); 401 402 if (error == 0) 403 error = copyout(sa, uap->name, sa_len); 404 if (error == 0) { 405 error = copyout(&sa_len, uap->anamelen, 406 sizeof(*uap->anamelen)); 407 } 408 if (sa) 409 FREE(sa, M_SONAME); 410 } else { 411 get_mplock(); 412 error = kern_accept(uap->s, 0, NULL, 0, 413 &uap->sysmsg_iresult); 414 rel_mplock(); 415 } 416 return (error); 417 } 418 419 /* 420 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 421 * 422 * MPALMOSTSAFE 423 */ 424 int 425 sys_extaccept(struct extaccept_args *uap) 426 { 427 struct sockaddr *sa = NULL; 428 int sa_len; 429 int error; 430 int fflags = uap->flags & O_FMASK; 431 432 if (uap->name) { 433 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 434 if (error) 435 return (error); 436 437 get_mplock(); 438 error = kern_accept(uap->s, fflags, &sa, &sa_len, 439 &uap->sysmsg_iresult); 440 rel_mplock(); 441 442 if (error == 0) 443 error = copyout(sa, uap->name, sa_len); 444 if (error == 0) { 445 error = copyout(&sa_len, uap->anamelen, 446 sizeof(*uap->anamelen)); 447 } 448 if (sa) 449 FREE(sa, M_SONAME); 450 } else { 451 get_mplock(); 452 error = kern_accept(uap->s, fflags, NULL, 0, 453 &uap->sysmsg_iresult); 454 rel_mplock(); 455 } 456 return (error); 457 } 458 459 460 /* 461 * Returns TRUE if predicate satisfied. 462 */ 463 static boolean_t 464 soconnected_predicate(struct netmsg_so_notify *msg) 465 { 466 struct socket *so = msg->base.nm_so; 467 468 /* check predicate */ 469 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 470 msg->base.lmsg.ms_error = so->so_error; 471 return (TRUE); 472 } 473 474 return (FALSE); 475 } 476 477 int 478 kern_connect(int s, int fflags, struct sockaddr *sa) 479 { 480 struct thread *td = curthread; 481 struct proc *p = td->td_proc; 482 struct file *fp; 483 struct socket *so; 484 int error, interrupted = 0; 485 486 error = holdsock(p->p_fd, s, &fp); 487 if (error) 488 return (error); 489 so = (struct socket *)fp->f_data; 490 491 if (fflags & O_FBLOCKING) 492 /* fflags &= ~FNONBLOCK; */; 493 else if (fflags & O_FNONBLOCKING) 494 fflags |= FNONBLOCK; 495 else 496 fflags = fp->f_flag; 497 498 if (so->so_state & SS_ISCONNECTING) { 499 error = EALREADY; 500 goto done; 501 } 502 error = soconnect(so, sa, td); 503 if (error) 504 goto bad; 505 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 506 error = EINPROGRESS; 507 goto done; 508 } 509 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 510 struct netmsg_so_notify msg; 511 512 netmsg_init_abortable(&msg.base, so, 513 &curthread->td_msgport, 514 0, 515 netmsg_so_notify, 516 netmsg_so_notify_doabort); 517 msg.nm_predicate = soconnected_predicate; 518 msg.nm_etype = NM_REVENT; 519 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 520 if (error == EINTR || error == ERESTART) 521 interrupted = 1; 522 } 523 if (error == 0) { 524 error = so->so_error; 525 so->so_error = 0; 526 } 527 bad: 528 if (!interrupted) 529 soclrstate(so, SS_ISCONNECTING); 530 if (error == ERESTART) 531 error = EINTR; 532 done: 533 fdrop(fp); 534 return (error); 535 } 536 537 /* 538 * connect_args(int s, caddr_t name, int namelen) 539 * 540 * MPALMOSTSAFE 541 */ 542 int 543 sys_connect(struct connect_args *uap) 544 { 545 struct sockaddr *sa; 546 int error; 547 548 error = getsockaddr(&sa, uap->name, uap->namelen); 549 if (error) 550 return (error); 551 get_mplock(); 552 error = kern_connect(uap->s, 0, sa); 553 rel_mplock(); 554 FREE(sa, M_SONAME); 555 556 return (error); 557 } 558 559 /* 560 * connect_args(int s, int fflags, caddr_t name, int namelen) 561 * 562 * MPALMOSTSAFE 563 */ 564 int 565 sys_extconnect(struct extconnect_args *uap) 566 { 567 struct sockaddr *sa; 568 int error; 569 int fflags = uap->flags & O_FMASK; 570 571 error = getsockaddr(&sa, uap->name, uap->namelen); 572 if (error) 573 return (error); 574 get_mplock(); 575 error = kern_connect(uap->s, fflags, sa); 576 rel_mplock(); 577 FREE(sa, M_SONAME); 578 579 return (error); 580 } 581 582 int 583 kern_socketpair(int domain, int type, int protocol, int *sv) 584 { 585 struct thread *td = curthread; 586 struct filedesc *fdp; 587 struct file *fp1, *fp2; 588 struct socket *so1, *so2; 589 int fd1, fd2, error; 590 591 fdp = td->td_proc->p_fd; 592 error = socreate(domain, &so1, type, protocol, td); 593 if (error) 594 return (error); 595 error = socreate(domain, &so2, type, protocol, td); 596 if (error) 597 goto free1; 598 error = falloc(td->td_lwp, &fp1, &fd1); 599 if (error) 600 goto free2; 601 sv[0] = fd1; 602 fp1->f_data = so1; 603 error = falloc(td->td_lwp, &fp2, &fd2); 604 if (error) 605 goto free3; 606 fp2->f_data = so2; 607 sv[1] = fd2; 608 error = soconnect2(so1, so2); 609 if (error) 610 goto free4; 611 if (type == SOCK_DGRAM) { 612 /* 613 * Datagram socket connection is asymmetric. 614 */ 615 error = soconnect2(so2, so1); 616 if (error) 617 goto free4; 618 } 619 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 620 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 621 fp1->f_ops = fp2->f_ops = &socketops; 622 fsetfd(fdp, fp1, fd1); 623 fsetfd(fdp, fp2, fd2); 624 fdrop(fp1); 625 fdrop(fp2); 626 return (error); 627 free4: 628 fsetfd(fdp, NULL, fd2); 629 fdrop(fp2); 630 free3: 631 fsetfd(fdp, NULL, fd1); 632 fdrop(fp1); 633 free2: 634 (void)soclose(so2, 0); 635 free1: 636 (void)soclose(so1, 0); 637 return (error); 638 } 639 640 /* 641 * socketpair(int domain, int type, int protocol, int *rsv) 642 * 643 * MPALMOSTSAFE 644 */ 645 int 646 sys_socketpair(struct socketpair_args *uap) 647 { 648 int error, sockv[2]; 649 650 get_mplock(); 651 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 652 rel_mplock(); 653 654 if (error == 0) 655 error = copyout(sockv, uap->rsv, sizeof(sockv)); 656 return (error); 657 } 658 659 int 660 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 661 struct mbuf *control, int flags, size_t *res) 662 { 663 struct thread *td = curthread; 664 struct lwp *lp = td->td_lwp; 665 struct proc *p = td->td_proc; 666 struct file *fp; 667 size_t len; 668 int error; 669 struct socket *so; 670 #ifdef KTRACE 671 struct iovec *ktriov = NULL; 672 struct uio ktruio; 673 #endif 674 675 error = holdsock(p->p_fd, s, &fp); 676 if (error) 677 return (error); 678 #ifdef KTRACE 679 if (KTRPOINT(td, KTR_GENIO)) { 680 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 681 682 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 683 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 684 ktruio = *auio; 685 } 686 #endif 687 len = auio->uio_resid; 688 so = (struct socket *)fp->f_data; 689 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 690 if (fp->f_flag & FNONBLOCK) 691 flags |= MSG_FNONBLOCKING; 692 } 693 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 694 if (error) { 695 if (auio->uio_resid != len && (error == ERESTART || 696 error == EINTR || error == EWOULDBLOCK)) 697 error = 0; 698 if (error == EPIPE && !(flags & MSG_NOSIGNAL)) 699 lwpsignal(p, lp, SIGPIPE); 700 } 701 #ifdef KTRACE 702 if (ktriov != NULL) { 703 if (error == 0) { 704 ktruio.uio_iov = ktriov; 705 ktruio.uio_resid = len - auio->uio_resid; 706 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 707 } 708 FREE(ktriov, M_TEMP); 709 } 710 #endif 711 if (error == 0) 712 *res = len - auio->uio_resid; 713 fdrop(fp); 714 return (error); 715 } 716 717 /* 718 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 719 * 720 * MPALMOSTSAFE 721 */ 722 int 723 sys_sendto(struct sendto_args *uap) 724 { 725 struct thread *td = curthread; 726 struct uio auio; 727 struct iovec aiov; 728 struct sockaddr *sa = NULL; 729 int error; 730 731 if (uap->to) { 732 error = getsockaddr(&sa, uap->to, uap->tolen); 733 if (error) 734 return (error); 735 } 736 aiov.iov_base = uap->buf; 737 aiov.iov_len = uap->len; 738 auio.uio_iov = &aiov; 739 auio.uio_iovcnt = 1; 740 auio.uio_offset = 0; 741 auio.uio_resid = uap->len; 742 auio.uio_segflg = UIO_USERSPACE; 743 auio.uio_rw = UIO_WRITE; 744 auio.uio_td = td; 745 746 get_mplock(); 747 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 748 &uap->sysmsg_szresult); 749 rel_mplock(); 750 751 if (sa) 752 FREE(sa, M_SONAME); 753 return (error); 754 } 755 756 /* 757 * sendmsg_args(int s, caddr_t msg, int flags) 758 * 759 * MPALMOSTSAFE 760 */ 761 int 762 sys_sendmsg(struct sendmsg_args *uap) 763 { 764 struct thread *td = curthread; 765 struct msghdr msg; 766 struct uio auio; 767 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 768 struct sockaddr *sa = NULL; 769 struct mbuf *control = NULL; 770 int error; 771 772 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 773 if (error) 774 return (error); 775 776 /* 777 * Conditionally copyin msg.msg_name. 778 */ 779 if (msg.msg_name) { 780 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 781 if (error) 782 return (error); 783 } 784 785 /* 786 * Populate auio. 787 */ 788 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 789 &auio.uio_resid); 790 if (error) 791 goto cleanup2; 792 auio.uio_iov = iov; 793 auio.uio_iovcnt = msg.msg_iovlen; 794 auio.uio_offset = 0; 795 auio.uio_segflg = UIO_USERSPACE; 796 auio.uio_rw = UIO_WRITE; 797 auio.uio_td = td; 798 799 /* 800 * Conditionally copyin msg.msg_control. 801 */ 802 if (msg.msg_control) { 803 if (msg.msg_controllen < sizeof(struct cmsghdr) || 804 msg.msg_controllen > MLEN) { 805 error = EINVAL; 806 goto cleanup; 807 } 808 control = m_get(MB_WAIT, MT_CONTROL); 809 if (control == NULL) { 810 error = ENOBUFS; 811 goto cleanup; 812 } 813 control->m_len = msg.msg_controllen; 814 error = copyin(msg.msg_control, mtod(control, caddr_t), 815 msg.msg_controllen); 816 if (error) { 817 m_free(control); 818 goto cleanup; 819 } 820 } 821 822 get_mplock(); 823 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 824 &uap->sysmsg_szresult); 825 rel_mplock(); 826 827 cleanup: 828 iovec_free(&iov, aiov); 829 cleanup2: 830 if (sa) 831 FREE(sa, M_SONAME); 832 return (error); 833 } 834 835 /* 836 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 837 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 838 * Don't forget to FREE() and m_free() these if they are returned. 839 */ 840 int 841 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 842 struct mbuf **control, int *flags, size_t *res) 843 { 844 struct thread *td = curthread; 845 struct proc *p = td->td_proc; 846 struct file *fp; 847 size_t len; 848 int error; 849 int lflags; 850 struct socket *so; 851 #ifdef KTRACE 852 struct iovec *ktriov = NULL; 853 struct uio ktruio; 854 #endif 855 856 error = holdsock(p->p_fd, s, &fp); 857 if (error) 858 return (error); 859 #ifdef KTRACE 860 if (KTRPOINT(td, KTR_GENIO)) { 861 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 862 863 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 864 bcopy(auio->uio_iov, ktriov, iovlen); 865 ktruio = *auio; 866 } 867 #endif 868 len = auio->uio_resid; 869 so = (struct socket *)fp->f_data; 870 871 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 872 if (fp->f_flag & FNONBLOCK) { 873 if (flags) { 874 *flags |= MSG_FNONBLOCKING; 875 } else { 876 lflags = MSG_FNONBLOCKING; 877 flags = &lflags; 878 } 879 } 880 } 881 882 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 883 if (error) { 884 if (auio->uio_resid != len && (error == ERESTART || 885 error == EINTR || error == EWOULDBLOCK)) 886 error = 0; 887 } 888 #ifdef KTRACE 889 if (ktriov != NULL) { 890 if (error == 0) { 891 ktruio.uio_iov = ktriov; 892 ktruio.uio_resid = len - auio->uio_resid; 893 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 894 } 895 FREE(ktriov, M_TEMP); 896 } 897 #endif 898 if (error == 0) 899 *res = len - auio->uio_resid; 900 fdrop(fp); 901 return (error); 902 } 903 904 /* 905 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 906 * caddr_t from, int *fromlenaddr) 907 * 908 * MPALMOSTSAFE 909 */ 910 int 911 sys_recvfrom(struct recvfrom_args *uap) 912 { 913 struct thread *td = curthread; 914 struct uio auio; 915 struct iovec aiov; 916 struct sockaddr *sa = NULL; 917 int error, fromlen; 918 919 if (uap->from && uap->fromlenaddr) { 920 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 921 if (error) 922 return (error); 923 if (fromlen < 0) 924 return (EINVAL); 925 } else { 926 fromlen = 0; 927 } 928 aiov.iov_base = uap->buf; 929 aiov.iov_len = uap->len; 930 auio.uio_iov = &aiov; 931 auio.uio_iovcnt = 1; 932 auio.uio_offset = 0; 933 auio.uio_resid = uap->len; 934 auio.uio_segflg = UIO_USERSPACE; 935 auio.uio_rw = UIO_READ; 936 auio.uio_td = td; 937 938 get_mplock(); 939 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 940 &uap->flags, &uap->sysmsg_szresult); 941 rel_mplock(); 942 943 if (error == 0 && uap->from) { 944 /* note: sa may still be NULL */ 945 if (sa) { 946 fromlen = MIN(fromlen, sa->sa_len); 947 error = copyout(sa, uap->from, fromlen); 948 } else { 949 fromlen = 0; 950 } 951 if (error == 0) { 952 error = copyout(&fromlen, uap->fromlenaddr, 953 sizeof(fromlen)); 954 } 955 } 956 if (sa) 957 FREE(sa, M_SONAME); 958 959 return (error); 960 } 961 962 /* 963 * recvmsg_args(int s, struct msghdr *msg, int flags) 964 * 965 * MPALMOSTSAFE 966 */ 967 int 968 sys_recvmsg(struct recvmsg_args *uap) 969 { 970 struct thread *td = curthread; 971 struct msghdr msg; 972 struct uio auio; 973 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 974 struct mbuf *m, *control = NULL; 975 struct sockaddr *sa = NULL; 976 caddr_t ctlbuf; 977 socklen_t *ufromlenp, *ucontrollenp; 978 int error, fromlen, controllen, len, flags, *uflagsp; 979 980 /* 981 * This copyin handles everything except the iovec. 982 */ 983 error = copyin(uap->msg, &msg, sizeof(msg)); 984 if (error) 985 return (error); 986 987 if (msg.msg_name && msg.msg_namelen < 0) 988 return (EINVAL); 989 if (msg.msg_control && msg.msg_controllen < 0) 990 return (EINVAL); 991 992 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 993 msg_namelen)); 994 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 995 msg_controllen)); 996 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 997 msg_flags)); 998 999 /* 1000 * Populate auio. 1001 */ 1002 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1003 &auio.uio_resid); 1004 if (error) 1005 return (error); 1006 auio.uio_iov = iov; 1007 auio.uio_iovcnt = msg.msg_iovlen; 1008 auio.uio_offset = 0; 1009 auio.uio_segflg = UIO_USERSPACE; 1010 auio.uio_rw = UIO_READ; 1011 auio.uio_td = td; 1012 1013 flags = uap->flags; 1014 1015 get_mplock(); 1016 error = kern_recvmsg(uap->s, 1017 (msg.msg_name ? &sa : NULL), &auio, 1018 (msg.msg_control ? &control : NULL), &flags, 1019 &uap->sysmsg_szresult); 1020 rel_mplock(); 1021 1022 /* 1023 * Conditionally copyout the name and populate the namelen field. 1024 */ 1025 if (error == 0 && msg.msg_name) { 1026 /* note: sa may still be NULL */ 1027 if (sa != NULL) { 1028 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1029 error = copyout(sa, msg.msg_name, fromlen); 1030 } else { 1031 fromlen = 0; 1032 } 1033 if (error == 0) 1034 error = copyout(&fromlen, ufromlenp, 1035 sizeof(*ufromlenp)); 1036 } 1037 1038 /* 1039 * Copyout msg.msg_control and msg.msg_controllen. 1040 */ 1041 if (error == 0 && msg.msg_control) { 1042 len = msg.msg_controllen; 1043 m = control; 1044 ctlbuf = (caddr_t)msg.msg_control; 1045 1046 while(m && len > 0) { 1047 unsigned int tocopy; 1048 1049 if (len >= m->m_len) { 1050 tocopy = m->m_len; 1051 } else { 1052 msg.msg_flags |= MSG_CTRUNC; 1053 tocopy = len; 1054 } 1055 1056 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1057 if (error) 1058 goto cleanup; 1059 1060 ctlbuf += tocopy; 1061 len -= tocopy; 1062 m = m->m_next; 1063 } 1064 controllen = ctlbuf - (caddr_t)msg.msg_control; 1065 error = copyout(&controllen, ucontrollenp, 1066 sizeof(*ucontrollenp)); 1067 } 1068 1069 if (error == 0) 1070 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1071 1072 cleanup: 1073 if (sa) 1074 FREE(sa, M_SONAME); 1075 iovec_free(&iov, aiov); 1076 if (control) 1077 m_freem(control); 1078 return (error); 1079 } 1080 1081 /* 1082 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1083 * in kernel pointer instead of a userland pointer. This allows us 1084 * to manipulate socket options in the emulation code. 1085 */ 1086 int 1087 kern_setsockopt(int s, struct sockopt *sopt) 1088 { 1089 struct thread *td = curthread; 1090 struct proc *p = td->td_proc; 1091 struct file *fp; 1092 int error; 1093 1094 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1095 return (EFAULT); 1096 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1097 return (EINVAL); 1098 if (sopt->sopt_valsize < 0) 1099 return (EINVAL); 1100 1101 error = holdsock(p->p_fd, s, &fp); 1102 if (error) 1103 return (error); 1104 1105 error = sosetopt((struct socket *)fp->f_data, sopt); 1106 fdrop(fp); 1107 return (error); 1108 } 1109 1110 /* 1111 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1112 * 1113 * MPALMOSTSAFE 1114 */ 1115 int 1116 sys_setsockopt(struct setsockopt_args *uap) 1117 { 1118 struct thread *td = curthread; 1119 struct sockopt sopt; 1120 int error; 1121 1122 sopt.sopt_level = uap->level; 1123 sopt.sopt_name = uap->name; 1124 sopt.sopt_valsize = uap->valsize; 1125 sopt.sopt_td = td; 1126 sopt.sopt_val = NULL; 1127 1128 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE) 1129 return (EINVAL); 1130 if (uap->val) { 1131 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1132 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1133 if (error) 1134 goto out; 1135 } 1136 1137 get_mplock(); 1138 error = kern_setsockopt(uap->s, &sopt); 1139 rel_mplock(); 1140 out: 1141 if (uap->val) 1142 kfree(sopt.sopt_val, M_TEMP); 1143 return(error); 1144 } 1145 1146 /* 1147 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1148 * in kernel pointer instead of a userland pointer. This allows us 1149 * to manipulate socket options in the emulation code. 1150 */ 1151 int 1152 kern_getsockopt(int s, struct sockopt *sopt) 1153 { 1154 struct thread *td = curthread; 1155 struct proc *p = td->td_proc; 1156 struct file *fp; 1157 int error; 1158 1159 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1160 return (EFAULT); 1161 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1162 return (EINVAL); 1163 if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE) 1164 return (EINVAL); 1165 1166 error = holdsock(p->p_fd, s, &fp); 1167 if (error) 1168 return (error); 1169 1170 error = sogetopt((struct socket *)fp->f_data, sopt); 1171 fdrop(fp); 1172 return (error); 1173 } 1174 1175 /* 1176 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1177 * 1178 * MPALMOSTSAFE 1179 */ 1180 int 1181 sys_getsockopt(struct getsockopt_args *uap) 1182 { 1183 struct thread *td = curthread; 1184 struct sockopt sopt; 1185 int error, valsize; 1186 1187 if (uap->val) { 1188 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1189 if (error) 1190 return (error); 1191 } else { 1192 valsize = 0; 1193 } 1194 1195 sopt.sopt_level = uap->level; 1196 sopt.sopt_name = uap->name; 1197 sopt.sopt_valsize = valsize; 1198 sopt.sopt_td = td; 1199 sopt.sopt_val = NULL; 1200 1201 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE) 1202 return (EINVAL); 1203 if (uap->val) { 1204 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1205 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1206 if (error) 1207 goto out; 1208 } 1209 1210 get_mplock(); 1211 error = kern_getsockopt(uap->s, &sopt); 1212 rel_mplock(); 1213 if (error) 1214 goto out; 1215 valsize = sopt.sopt_valsize; 1216 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1217 if (error) 1218 goto out; 1219 if (uap->val) 1220 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1221 out: 1222 if (uap->val) 1223 kfree(sopt.sopt_val, M_TEMP); 1224 return (error); 1225 } 1226 1227 /* 1228 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1229 * This allows kern_getsockname() to return a pointer to an allocated struct 1230 * sockaddr which must be freed later with FREE(). The caller must 1231 * initialize *name to NULL. 1232 */ 1233 int 1234 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1235 { 1236 struct thread *td = curthread; 1237 struct proc *p = td->td_proc; 1238 struct file *fp; 1239 struct socket *so; 1240 struct sockaddr *sa = NULL; 1241 int error; 1242 1243 error = holdsock(p->p_fd, s, &fp); 1244 if (error) 1245 return (error); 1246 if (*namelen < 0) { 1247 fdrop(fp); 1248 return (EINVAL); 1249 } 1250 so = (struct socket *)fp->f_data; 1251 error = so_pru_sockaddr(so, &sa); 1252 if (error == 0) { 1253 if (sa == NULL) { 1254 *namelen = 0; 1255 } else { 1256 *namelen = MIN(*namelen, sa->sa_len); 1257 *name = sa; 1258 } 1259 } 1260 1261 fdrop(fp); 1262 return (error); 1263 } 1264 1265 /* 1266 * getsockname_args(int fdes, caddr_t asa, int *alen) 1267 * 1268 * Get socket name. 1269 * 1270 * MPALMOSTSAFE 1271 */ 1272 int 1273 sys_getsockname(struct getsockname_args *uap) 1274 { 1275 struct sockaddr *sa = NULL; 1276 int error, sa_len; 1277 1278 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1279 if (error) 1280 return (error); 1281 1282 get_mplock(); 1283 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1284 rel_mplock(); 1285 1286 if (error == 0) 1287 error = copyout(sa, uap->asa, sa_len); 1288 if (error == 0) 1289 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1290 if (sa) 1291 FREE(sa, M_SONAME); 1292 return (error); 1293 } 1294 1295 /* 1296 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1297 * This allows kern_getpeername() to return a pointer to an allocated struct 1298 * sockaddr which must be freed later with FREE(). The caller must 1299 * initialize *name to NULL. 1300 */ 1301 int 1302 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1303 { 1304 struct thread *td = curthread; 1305 struct proc *p = td->td_proc; 1306 struct file *fp; 1307 struct socket *so; 1308 struct sockaddr *sa = NULL; 1309 int error; 1310 1311 error = holdsock(p->p_fd, s, &fp); 1312 if (error) 1313 return (error); 1314 if (*namelen < 0) { 1315 fdrop(fp); 1316 return (EINVAL); 1317 } 1318 so = (struct socket *)fp->f_data; 1319 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1320 fdrop(fp); 1321 return (ENOTCONN); 1322 } 1323 error = so_pru_peeraddr(so, &sa); 1324 if (error == 0) { 1325 if (sa == NULL) { 1326 *namelen = 0; 1327 } else { 1328 *namelen = MIN(*namelen, sa->sa_len); 1329 *name = sa; 1330 } 1331 } 1332 1333 fdrop(fp); 1334 return (error); 1335 } 1336 1337 /* 1338 * getpeername_args(int fdes, caddr_t asa, int *alen) 1339 * 1340 * Get name of peer for connected socket. 1341 * 1342 * MPALMOSTSAFE 1343 */ 1344 int 1345 sys_getpeername(struct getpeername_args *uap) 1346 { 1347 struct sockaddr *sa = NULL; 1348 int error, sa_len; 1349 1350 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1351 if (error) 1352 return (error); 1353 1354 get_mplock(); 1355 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1356 rel_mplock(); 1357 1358 if (error == 0) 1359 error = copyout(sa, uap->asa, sa_len); 1360 if (error == 0) 1361 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1362 if (sa) 1363 FREE(sa, M_SONAME); 1364 return (error); 1365 } 1366 1367 int 1368 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1369 { 1370 struct sockaddr *sa; 1371 int error; 1372 1373 *namp = NULL; 1374 if (len > SOCK_MAXADDRLEN) 1375 return ENAMETOOLONG; 1376 if (len < offsetof(struct sockaddr, sa_data[0])) 1377 return EDOM; 1378 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1379 error = copyin(uaddr, sa, len); 1380 if (error) { 1381 FREE(sa, M_SONAME); 1382 } else { 1383 #if BYTE_ORDER != BIG_ENDIAN 1384 /* 1385 * The bind(), connect(), and sendto() syscalls were not 1386 * versioned for COMPAT_43. Thus, this check must stay. 1387 */ 1388 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1389 sa->sa_family = sa->sa_len; 1390 #endif 1391 sa->sa_len = len; 1392 *namp = sa; 1393 } 1394 return error; 1395 } 1396 1397 /* 1398 * Detach a mapped page and release resources back to the system. 1399 * We must release our wiring and if the object is ripped out 1400 * from under the vm_page we become responsible for freeing the 1401 * page. These routines must be MPSAFE. 1402 * 1403 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1404 * 1405 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required. 1406 */ 1407 static void 1408 sf_buf_mfree(void *arg) 1409 { 1410 struct sf_buf *sf = arg; 1411 vm_page_t m; 1412 1413 /* 1414 * XXX vm_page_*() and SFBUF routines not MPSAFE yet. 1415 */ 1416 get_mplock(); 1417 crit_enter(); 1418 m = sf_buf_page(sf); 1419 if (sf_buf_free(sf) == 0) { 1420 vm_page_unwire(m, 0); 1421 if (m->wire_count == 0 && m->object == NULL) 1422 vm_page_try_to_free(m); 1423 } 1424 crit_exit(); 1425 rel_mplock(); 1426 } 1427 1428 /* 1429 * sendfile(2). 1430 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1431 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1432 * 1433 * Send a file specified by 'fd' and starting at 'offset' to a socket 1434 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1435 * nbytes == 0. Optionally add a header and/or trailer to the socket 1436 * output. If specified, write the total number of bytes sent into *sbytes. 1437 * 1438 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1439 * the headers to count against the remaining bytes to be sent from 1440 * the file descriptor. We may wish to implement a compatibility syscall 1441 * in the future. 1442 * 1443 * MPALMOSTSAFE 1444 */ 1445 int 1446 sys_sendfile(struct sendfile_args *uap) 1447 { 1448 struct thread *td = curthread; 1449 struct proc *p = td->td_proc; 1450 struct file *fp; 1451 struct vnode *vp = NULL; 1452 struct sf_hdtr hdtr; 1453 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1454 struct uio auio; 1455 struct mbuf *mheader = NULL; 1456 size_t hbytes = 0; 1457 size_t tbytes; 1458 off_t hdtr_size = 0; 1459 off_t sbytes; 1460 int error; 1461 1462 KKASSERT(p); 1463 1464 /* 1465 * Do argument checking. Must be a regular file in, stream 1466 * type and connected socket out, positive offset. 1467 */ 1468 fp = holdfp(p->p_fd, uap->fd, FREAD); 1469 if (fp == NULL) { 1470 return (EBADF); 1471 } 1472 if (fp->f_type != DTYPE_VNODE) { 1473 fdrop(fp); 1474 return (EINVAL); 1475 } 1476 get_mplock(); 1477 vp = (struct vnode *)fp->f_data; 1478 vref(vp); 1479 fdrop(fp); 1480 1481 /* 1482 * If specified, get the pointer to the sf_hdtr struct for 1483 * any headers/trailers. 1484 */ 1485 if (uap->hdtr) { 1486 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1487 if (error) 1488 goto done; 1489 /* 1490 * Send any headers. 1491 */ 1492 if (hdtr.headers) { 1493 error = iovec_copyin(hdtr.headers, &iov, aiov, 1494 hdtr.hdr_cnt, &hbytes); 1495 if (error) 1496 goto done; 1497 auio.uio_iov = iov; 1498 auio.uio_iovcnt = hdtr.hdr_cnt; 1499 auio.uio_offset = 0; 1500 auio.uio_segflg = UIO_USERSPACE; 1501 auio.uio_rw = UIO_WRITE; 1502 auio.uio_td = td; 1503 auio.uio_resid = hbytes; 1504 1505 mheader = m_uiomove(&auio); 1506 1507 iovec_free(&iov, aiov); 1508 if (mheader == NULL) 1509 goto done; 1510 } 1511 } 1512 1513 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1514 &sbytes, uap->flags); 1515 if (error) 1516 goto done; 1517 1518 /* 1519 * Send trailers. Wimp out and use writev(2). 1520 */ 1521 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1522 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1523 hdtr.trl_cnt, &auio.uio_resid); 1524 if (error) 1525 goto done; 1526 auio.uio_iov = iov; 1527 auio.uio_iovcnt = hdtr.trl_cnt; 1528 auio.uio_offset = 0; 1529 auio.uio_segflg = UIO_USERSPACE; 1530 auio.uio_rw = UIO_WRITE; 1531 auio.uio_td = td; 1532 1533 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1534 1535 iovec_free(&iov, aiov); 1536 if (error) 1537 goto done; 1538 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1539 } 1540 1541 done: 1542 if (vp) 1543 vrele(vp); 1544 rel_mplock(); 1545 if (uap->sbytes != NULL) { 1546 sbytes += hdtr_size; 1547 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1548 } 1549 return (error); 1550 } 1551 1552 int 1553 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1554 struct mbuf *mheader, off_t *sbytes, int flags) 1555 { 1556 struct thread *td = curthread; 1557 struct proc *p = td->td_proc; 1558 struct vm_object *obj; 1559 struct socket *so; 1560 struct file *fp; 1561 struct mbuf *m; 1562 struct sf_buf *sf; 1563 struct vm_page *pg; 1564 off_t off, xfsize; 1565 off_t hbytes = 0; 1566 int error = 0; 1567 1568 if (vp->v_type != VREG) { 1569 error = EINVAL; 1570 goto done0; 1571 } 1572 if ((obj = vp->v_object) == NULL) { 1573 error = EINVAL; 1574 goto done0; 1575 } 1576 error = holdsock(p->p_fd, sfd, &fp); 1577 if (error) 1578 goto done0; 1579 so = (struct socket *)fp->f_data; 1580 if (so->so_type != SOCK_STREAM) { 1581 error = EINVAL; 1582 goto done; 1583 } 1584 if ((so->so_state & SS_ISCONNECTED) == 0) { 1585 error = ENOTCONN; 1586 goto done; 1587 } 1588 if (offset < 0) { 1589 error = EINVAL; 1590 goto done; 1591 } 1592 1593 *sbytes = 0; 1594 /* 1595 * Protect against multiple writers to the socket. 1596 */ 1597 ssb_lock(&so->so_snd, M_WAITOK); 1598 1599 /* 1600 * Loop through the pages in the file, starting with the requested 1601 * offset. Get a file page (do I/O if necessary), map the file page 1602 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1603 * it on the socket. 1604 */ 1605 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1606 vm_pindex_t pindex; 1607 vm_offset_t pgoff; 1608 1609 pindex = OFF_TO_IDX(off); 1610 retry_lookup: 1611 /* 1612 * Calculate the amount to transfer. Not to exceed a page, 1613 * the EOF, or the passed in nbytes. 1614 */ 1615 xfsize = vp->v_filesize - off; 1616 if (xfsize > PAGE_SIZE) 1617 xfsize = PAGE_SIZE; 1618 pgoff = (vm_offset_t)(off & PAGE_MASK); 1619 if (PAGE_SIZE - pgoff < xfsize) 1620 xfsize = PAGE_SIZE - pgoff; 1621 if (nbytes && xfsize > (nbytes - *sbytes)) 1622 xfsize = nbytes - *sbytes; 1623 if (xfsize <= 0) 1624 break; 1625 /* 1626 * Optimize the non-blocking case by looking at the socket space 1627 * before going to the extra work of constituting the sf_buf. 1628 */ 1629 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) { 1630 if (so->so_state & SS_CANTSENDMORE) 1631 error = EPIPE; 1632 else 1633 error = EAGAIN; 1634 ssb_unlock(&so->so_snd); 1635 goto done; 1636 } 1637 /* 1638 * Attempt to look up the page. 1639 * 1640 * Allocate if not found, wait and loop if busy, then 1641 * wire the page. critical section protection is 1642 * required to maintain the object association (an 1643 * interrupt can free the page) through to the 1644 * vm_page_wire() call. 1645 */ 1646 lwkt_gettoken(&vm_token); 1647 pg = vm_page_lookup(obj, pindex); 1648 if (pg == NULL) { 1649 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1650 if (pg == NULL) { 1651 vm_wait(0); 1652 lwkt_reltoken(&vm_token); 1653 goto retry_lookup; 1654 } 1655 vm_page_wire(pg); 1656 vm_page_wakeup(pg); 1657 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1658 lwkt_reltoken(&vm_token); 1659 goto retry_lookup; 1660 } else { 1661 vm_page_wire(pg); 1662 } 1663 lwkt_reltoken(&vm_token); 1664 1665 /* 1666 * If page is not valid for what we need, initiate I/O 1667 */ 1668 1669 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1670 struct uio auio; 1671 struct iovec aiov; 1672 int bsize; 1673 1674 /* 1675 * Ensure that our page is still around when the I/O 1676 * completes. 1677 */ 1678 vm_page_io_start(pg); 1679 1680 /* 1681 * Get the page from backing store. 1682 */ 1683 bsize = vp->v_mount->mnt_stat.f_iosize; 1684 auio.uio_iov = &aiov; 1685 auio.uio_iovcnt = 1; 1686 aiov.iov_base = 0; 1687 aiov.iov_len = MAXBSIZE; 1688 auio.uio_resid = MAXBSIZE; 1689 auio.uio_offset = trunc_page(off); 1690 auio.uio_segflg = UIO_NOCOPY; 1691 auio.uio_rw = UIO_READ; 1692 auio.uio_td = td; 1693 vn_lock(vp, LK_SHARED | LK_RETRY); 1694 error = VOP_READ(vp, &auio, 1695 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1696 td->td_ucred); 1697 vn_unlock(vp); 1698 vm_page_flag_clear(pg, PG_ZERO); 1699 vm_page_io_finish(pg); 1700 if (error) { 1701 crit_enter(); 1702 vm_page_unwire(pg, 0); 1703 vm_page_try_to_free(pg); 1704 crit_exit(); 1705 ssb_unlock(&so->so_snd); 1706 goto done; 1707 } 1708 } 1709 1710 1711 /* 1712 * Get a sendfile buf. We usually wait as long as necessary, 1713 * but this wait can be interrupted. 1714 */ 1715 if ((sf = sf_buf_alloc(pg)) == NULL) { 1716 crit_enter(); 1717 vm_page_unwire(pg, 0); 1718 vm_page_try_to_free(pg); 1719 crit_exit(); 1720 ssb_unlock(&so->so_snd); 1721 error = EINTR; 1722 goto done; 1723 } 1724 1725 /* 1726 * Get an mbuf header and set it up as having external storage. 1727 */ 1728 MGETHDR(m, MB_WAIT, MT_DATA); 1729 if (m == NULL) { 1730 error = ENOBUFS; 1731 sf_buf_free(sf); 1732 ssb_unlock(&so->so_snd); 1733 goto done; 1734 } 1735 1736 m->m_ext.ext_free = sf_buf_mfree; 1737 m->m_ext.ext_ref = sf_buf_ref; 1738 m->m_ext.ext_arg = sf; 1739 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1740 m->m_ext.ext_size = PAGE_SIZE; 1741 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1742 m->m_flags |= M_EXT; 1743 m->m_pkthdr.len = m->m_len = xfsize; 1744 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1745 1746 if (mheader != NULL) { 1747 hbytes = mheader->m_pkthdr.len; 1748 mheader->m_pkthdr.len += m->m_pkthdr.len; 1749 m_cat(mheader, m); 1750 m = mheader; 1751 mheader = NULL; 1752 } else 1753 hbytes = 0; 1754 1755 /* 1756 * Add the buffer to the socket buffer chain. 1757 */ 1758 crit_enter(); 1759 retry_space: 1760 /* 1761 * Make sure that the socket is still able to take more data. 1762 * CANTSENDMORE being true usually means that the connection 1763 * was closed. so_error is true when an error was sensed after 1764 * a previous send. 1765 * The state is checked after the page mapping and buffer 1766 * allocation above since those operations may block and make 1767 * any socket checks stale. From this point forward, nothing 1768 * blocks before the pru_send (or more accurately, any blocking 1769 * results in a loop back to here to re-check). 1770 */ 1771 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1772 if (so->so_state & SS_CANTSENDMORE) { 1773 error = EPIPE; 1774 } else { 1775 error = so->so_error; 1776 so->so_error = 0; 1777 } 1778 m_freem(m); 1779 ssb_unlock(&so->so_snd); 1780 crit_exit(); 1781 goto done; 1782 } 1783 /* 1784 * Wait for socket space to become available. We do this just 1785 * after checking the connection state above in order to avoid 1786 * a race condition with ssb_wait(). 1787 */ 1788 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) { 1789 if (fp->f_flag & FNONBLOCK) { 1790 m_freem(m); 1791 ssb_unlock(&so->so_snd); 1792 crit_exit(); 1793 error = EAGAIN; 1794 goto done; 1795 } 1796 error = ssb_wait(&so->so_snd); 1797 /* 1798 * An error from ssb_wait usually indicates that we've 1799 * been interrupted by a signal. If we've sent anything 1800 * then return bytes sent, otherwise return the error. 1801 */ 1802 if (error) { 1803 m_freem(m); 1804 ssb_unlock(&so->so_snd); 1805 crit_exit(); 1806 goto done; 1807 } 1808 goto retry_space; 1809 } 1810 error = so_pru_send(so, 0, m, NULL, NULL, td); 1811 crit_exit(); 1812 if (error) { 1813 ssb_unlock(&so->so_snd); 1814 goto done; 1815 } 1816 } 1817 if (mheader != NULL) { 1818 *sbytes += mheader->m_pkthdr.len; 1819 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1820 mheader = NULL; 1821 } 1822 ssb_unlock(&so->so_snd); 1823 1824 done: 1825 fdrop(fp); 1826 done0: 1827 if (mheader != NULL) 1828 m_freem(mheader); 1829 return (error); 1830 } 1831 1832 /* 1833 * MPALMOSTSAFE 1834 */ 1835 int 1836 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1837 { 1838 #ifdef SCTP 1839 struct thread *td = curthread; 1840 struct filedesc *fdp = td->td_proc->p_fd; 1841 struct file *lfp = NULL; 1842 struct file *nfp = NULL; 1843 int error; 1844 struct socket *head, *so; 1845 caddr_t assoc_id; 1846 int fd; 1847 short fflag; /* type must match fp->f_flag */ 1848 1849 assoc_id = uap->name; 1850 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp); 1851 if (error) 1852 return (error); 1853 1854 get_mplock(); 1855 crit_enter(); 1856 head = (struct socket *)lfp->f_data; 1857 error = sctp_can_peel_off(head, assoc_id); 1858 if (error) { 1859 crit_exit(); 1860 goto done; 1861 } 1862 /* 1863 * At this point we know we do have a assoc to pull 1864 * we proceed to get the fd setup. This may block 1865 * but that is ok. 1866 */ 1867 1868 fflag = lfp->f_flag; 1869 error = falloc(td->td_lwp, &nfp, &fd); 1870 if (error) { 1871 /* 1872 * Probably ran out of file descriptors. Put the 1873 * unaccepted connection back onto the queue and 1874 * do another wakeup so some other process might 1875 * have a chance at it. 1876 */ 1877 crit_exit(); 1878 goto done; 1879 } 1880 uap->sysmsg_iresult = fd; 1881 1882 so = sctp_get_peeloff(head, assoc_id, &error); 1883 if (so == NULL) { 1884 /* 1885 * Either someone else peeled it off OR 1886 * we can't get a socket. 1887 */ 1888 goto noconnection; 1889 } 1890 soreference(so); /* reference needed */ 1891 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */ 1892 so->so_head = NULL; 1893 if (head->so_sigio != NULL) 1894 fsetown(fgetown(head->so_sigio), &so->so_sigio); 1895 1896 nfp->f_type = DTYPE_SOCKET; 1897 nfp->f_flag = fflag; 1898 nfp->f_ops = &socketops; 1899 nfp->f_data = so; 1900 1901 noconnection: 1902 /* 1903 * Assign the file pointer to the reserved descriptor, or clear 1904 * the reserved descriptor if an error occured. 1905 */ 1906 if (error) 1907 fsetfd(fdp, NULL, fd); 1908 else 1909 fsetfd(fdp, nfp, fd); 1910 crit_exit(); 1911 /* 1912 * Release explicitly held references before returning. 1913 */ 1914 done: 1915 rel_mplock(); 1916 if (nfp != NULL) 1917 fdrop(nfp); 1918 fdrop(lfp); 1919 return (error); 1920 #else /* SCTP */ 1921 return(EOPNOTSUPP); 1922 #endif /* SCTP */ 1923 } 1924