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 extern int use_soaccept_pred_fast; 88 extern int use_sendfile_async; 89 90 /* 91 * System call interface to the socket abstraction. 92 */ 93 94 extern struct fileops socketops; 95 96 /* 97 * socket_args(int domain, int type, int protocol) 98 */ 99 int 100 kern_socket(int domain, int type, int protocol, int *res) 101 { 102 struct thread *td = curthread; 103 struct filedesc *fdp = td->td_proc->p_fd; 104 struct socket *so; 105 struct file *fp; 106 int fd, error; 107 108 KKASSERT(td->td_lwp); 109 110 error = falloc(td->td_lwp, &fp, &fd); 111 if (error) 112 return (error); 113 error = socreate(domain, &so, type, protocol, td); 114 if (error) { 115 fsetfd(fdp, NULL, fd); 116 } else { 117 fp->f_type = DTYPE_SOCKET; 118 fp->f_flag = FREAD | FWRITE; 119 fp->f_ops = &socketops; 120 fp->f_data = so; 121 *res = fd; 122 fsetfd(fdp, fp, fd); 123 } 124 fdrop(fp); 125 return (error); 126 } 127 128 /* 129 * MPALMOSTSAFE 130 */ 131 int 132 sys_socket(struct socket_args *uap) 133 { 134 int error; 135 136 error = kern_socket(uap->domain, uap->type, uap->protocol, 137 &uap->sysmsg_iresult); 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 error = kern_bind(uap->s, sa); 174 kfree(sa, M_SONAME); 175 176 return (error); 177 } 178 179 int 180 kern_listen(int s, int backlog) 181 { 182 struct thread *td = curthread; 183 struct proc *p = td->td_proc; 184 struct file *fp; 185 int error; 186 187 KKASSERT(p); 188 error = holdsock(p->p_fd, s, &fp); 189 if (error) 190 return (error); 191 error = solisten((struct socket *)fp->f_data, backlog, td); 192 fdrop(fp); 193 return(error); 194 } 195 196 /* 197 * listen_args(int s, int backlog) 198 * 199 * MPALMOSTSAFE 200 */ 201 int 202 sys_listen(struct listen_args *uap) 203 { 204 int error; 205 206 error = kern_listen(uap->s, uap->backlog); 207 return (error); 208 } 209 210 /* 211 * Returns the accepted socket as well. 212 * 213 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 214 * pool token is absolutely required to avoid a sofree() race, 215 * as well as to avoid tailq handling races. 216 */ 217 static boolean_t 218 soaccept_predicate(struct netmsg_so_notify *msg) 219 { 220 struct socket *head = msg->base.nm_so; 221 struct socket *so; 222 223 if (head->so_error != 0) { 224 msg->base.lmsg.ms_error = head->so_error; 225 return (TRUE); 226 } 227 lwkt_getpooltoken(head); 228 if (!TAILQ_EMPTY(&head->so_comp)) { 229 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 230 so = TAILQ_FIRST(&head->so_comp); 231 TAILQ_REMOVE(&head->so_comp, so, so_list); 232 head->so_qlen--; 233 soclrstate(so, SS_COMP); 234 so->so_head = NULL; 235 soreference(so); 236 237 lwkt_relpooltoken(head); 238 239 msg->base.lmsg.ms_error = 0; 240 msg->base.nm_so = so; 241 return (TRUE); 242 } 243 lwkt_relpooltoken(head); 244 if (head->so_state & SS_CANTRCVMORE) { 245 msg->base.lmsg.ms_error = ECONNABORTED; 246 return (TRUE); 247 } 248 if (msg->nm_fflags & FNONBLOCK) { 249 msg->base.lmsg.ms_error = EWOULDBLOCK; 250 return (TRUE); 251 } 252 253 return (FALSE); 254 } 255 256 /* 257 * The second argument to kern_accept() is a handle to a struct sockaddr. 258 * This allows kern_accept() to return a pointer to an allocated struct 259 * sockaddr which must be freed later with FREE(). The caller must 260 * initialize *name to NULL. 261 */ 262 int 263 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 264 { 265 struct thread *td = curthread; 266 struct filedesc *fdp = td->td_proc->p_fd; 267 struct file *lfp = NULL; 268 struct file *nfp = NULL; 269 struct sockaddr *sa; 270 struct socket *head, *so; 271 struct netmsg_so_notify msg; 272 int fd; 273 u_int fflag; /* type must match fp->f_flag */ 274 int error, tmp; 275 276 *res = -1; 277 if (name && namelen && *namelen < 0) 278 return (EINVAL); 279 280 error = holdsock(td->td_proc->p_fd, s, &lfp); 281 if (error) 282 return (error); 283 284 error = falloc(td->td_lwp, &nfp, &fd); 285 if (error) { /* Probably ran out of file descriptors. */ 286 fdrop(lfp); 287 return (error); 288 } 289 head = (struct socket *)lfp->f_data; 290 if ((head->so_options & SO_ACCEPTCONN) == 0) { 291 error = EINVAL; 292 goto done; 293 } 294 295 if (fflags & O_FBLOCKING) 296 fflags |= lfp->f_flag & ~FNONBLOCK; 297 else if (fflags & O_FNONBLOCKING) 298 fflags |= lfp->f_flag | FNONBLOCK; 299 else 300 fflags = lfp->f_flag; 301 302 if (use_soaccept_pred_fast) { 303 boolean_t pred; 304 305 /* Initialize necessary parts for soaccept_predicate() */ 306 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 307 msg.nm_fflags = fflags; 308 309 lwkt_getpooltoken(head); 310 pred = soaccept_predicate(&msg); 311 lwkt_relpooltoken(head); 312 313 if (pred) { 314 error = msg.base.lmsg.ms_error; 315 if (error) 316 goto done; 317 else 318 goto accepted; 319 } 320 } 321 322 /* optimize for uniprocessor case later XXX JH */ 323 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 324 0, netmsg_so_notify, netmsg_so_notify_doabort); 325 msg.nm_predicate = soaccept_predicate; 326 msg.nm_fflags = fflags; 327 msg.nm_etype = NM_REVENT; 328 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 329 if (error) 330 goto done; 331 332 accepted: 333 /* 334 * At this point we have the connection that's ready to be accepted. 335 * 336 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 337 * to eat the ref and turn it into a descriptor. 338 */ 339 so = msg.base.nm_so; 340 341 fflag = lfp->f_flag; 342 343 /* connection has been removed from the listen queue */ 344 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 345 346 if (head->so_sigio != NULL) 347 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 348 349 nfp->f_type = DTYPE_SOCKET; 350 nfp->f_flag = fflag; 351 nfp->f_ops = &socketops; 352 nfp->f_data = so; 353 /* Sync socket nonblocking/async state with file flags */ 354 tmp = fflag & FNONBLOCK; 355 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 356 tmp = fflag & FASYNC; 357 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 358 359 sa = NULL; 360 if (so->so_faddr != NULL) { 361 sa = so->so_faddr; 362 so->so_faddr = NULL; 363 364 soaccept_generic(so); 365 error = 0; 366 } else { 367 error = soaccept(so, &sa); 368 } 369 370 /* 371 * Set the returned name and namelen as applicable. Set the returned 372 * namelen to 0 for older code which might ignore the return value 373 * from accept. 374 */ 375 if (error == 0) { 376 if (sa && name && namelen) { 377 if (*namelen > sa->sa_len) 378 *namelen = sa->sa_len; 379 *name = sa; 380 } else { 381 if (sa) 382 kfree(sa, M_SONAME); 383 } 384 } 385 386 done: 387 /* 388 * If an error occured clear the reserved descriptor, else associate 389 * nfp with it. 390 * 391 * Note that *res is normally ignored if an error is returned but 392 * a syscall message will still have access to the result code. 393 */ 394 if (error) { 395 fsetfd(fdp, NULL, fd); 396 } else { 397 *res = fd; 398 fsetfd(fdp, nfp, fd); 399 } 400 fdrop(nfp); 401 fdrop(lfp); 402 return (error); 403 } 404 405 /* 406 * accept(int s, caddr_t name, int *anamelen) 407 * 408 * MPALMOSTSAFE 409 */ 410 int 411 sys_accept(struct accept_args *uap) 412 { 413 struct sockaddr *sa = NULL; 414 int sa_len; 415 int error; 416 417 if (uap->name) { 418 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 419 if (error) 420 return (error); 421 422 error = kern_accept(uap->s, 0, &sa, &sa_len, 423 &uap->sysmsg_iresult); 424 425 if (error == 0) 426 error = copyout(sa, uap->name, sa_len); 427 if (error == 0) { 428 error = copyout(&sa_len, uap->anamelen, 429 sizeof(*uap->anamelen)); 430 } 431 if (sa) 432 kfree(sa, M_SONAME); 433 } else { 434 error = kern_accept(uap->s, 0, NULL, 0, 435 &uap->sysmsg_iresult); 436 } 437 return (error); 438 } 439 440 /* 441 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 442 * 443 * MPALMOSTSAFE 444 */ 445 int 446 sys_extaccept(struct extaccept_args *uap) 447 { 448 struct sockaddr *sa = NULL; 449 int sa_len; 450 int error; 451 int fflags = uap->flags & O_FMASK; 452 453 if (uap->name) { 454 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 455 if (error) 456 return (error); 457 458 error = kern_accept(uap->s, fflags, &sa, &sa_len, 459 &uap->sysmsg_iresult); 460 461 if (error == 0) 462 error = copyout(sa, uap->name, sa_len); 463 if (error == 0) { 464 error = copyout(&sa_len, uap->anamelen, 465 sizeof(*uap->anamelen)); 466 } 467 if (sa) 468 kfree(sa, M_SONAME); 469 } else { 470 error = kern_accept(uap->s, fflags, NULL, 0, 471 &uap->sysmsg_iresult); 472 } 473 return (error); 474 } 475 476 477 /* 478 * Returns TRUE if predicate satisfied. 479 */ 480 static boolean_t 481 soconnected_predicate(struct netmsg_so_notify *msg) 482 { 483 struct socket *so = msg->base.nm_so; 484 485 /* check predicate */ 486 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 487 msg->base.lmsg.ms_error = so->so_error; 488 return (TRUE); 489 } 490 491 return (FALSE); 492 } 493 494 int 495 kern_connect(int s, int fflags, struct sockaddr *sa) 496 { 497 struct thread *td = curthread; 498 struct proc *p = td->td_proc; 499 struct file *fp; 500 struct socket *so; 501 int error, interrupted = 0; 502 503 error = holdsock(p->p_fd, s, &fp); 504 if (error) 505 return (error); 506 so = (struct socket *)fp->f_data; 507 508 if (fflags & O_FBLOCKING) 509 /* fflags &= ~FNONBLOCK; */; 510 else if (fflags & O_FNONBLOCKING) 511 fflags |= FNONBLOCK; 512 else 513 fflags = fp->f_flag; 514 515 if (so->so_state & SS_ISCONNECTING) { 516 error = EALREADY; 517 goto done; 518 } 519 error = soconnect(so, sa, td); 520 if (error) 521 goto bad; 522 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 523 error = EINPROGRESS; 524 goto done; 525 } 526 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 527 struct netmsg_so_notify msg; 528 529 netmsg_init_abortable(&msg.base, so, 530 &curthread->td_msgport, 531 0, 532 netmsg_so_notify, 533 netmsg_so_notify_doabort); 534 msg.nm_predicate = soconnected_predicate; 535 msg.nm_etype = NM_REVENT; 536 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 537 if (error == EINTR || error == ERESTART) 538 interrupted = 1; 539 } 540 if (error == 0) { 541 error = so->so_error; 542 so->so_error = 0; 543 } 544 bad: 545 if (!interrupted) 546 soclrstate(so, SS_ISCONNECTING); 547 if (error == ERESTART) 548 error = EINTR; 549 done: 550 fdrop(fp); 551 return (error); 552 } 553 554 /* 555 * connect_args(int s, caddr_t name, int namelen) 556 * 557 * MPALMOSTSAFE 558 */ 559 int 560 sys_connect(struct connect_args *uap) 561 { 562 struct sockaddr *sa; 563 int error; 564 565 error = getsockaddr(&sa, uap->name, uap->namelen); 566 if (error) 567 return (error); 568 error = kern_connect(uap->s, 0, sa); 569 kfree(sa, M_SONAME); 570 571 return (error); 572 } 573 574 /* 575 * connect_args(int s, int fflags, caddr_t name, int namelen) 576 * 577 * MPALMOSTSAFE 578 */ 579 int 580 sys_extconnect(struct extconnect_args *uap) 581 { 582 struct sockaddr *sa; 583 int error; 584 int fflags = uap->flags & O_FMASK; 585 586 error = getsockaddr(&sa, uap->name, uap->namelen); 587 if (error) 588 return (error); 589 error = kern_connect(uap->s, fflags, sa); 590 kfree(sa, M_SONAME); 591 592 return (error); 593 } 594 595 int 596 kern_socketpair(int domain, int type, int protocol, int *sv) 597 { 598 struct thread *td = curthread; 599 struct filedesc *fdp; 600 struct file *fp1, *fp2; 601 struct socket *so1, *so2; 602 int fd1, fd2, error; 603 604 fdp = td->td_proc->p_fd; 605 error = socreate(domain, &so1, type, protocol, td); 606 if (error) 607 return (error); 608 error = socreate(domain, &so2, type, protocol, td); 609 if (error) 610 goto free1; 611 error = falloc(td->td_lwp, &fp1, &fd1); 612 if (error) 613 goto free2; 614 sv[0] = fd1; 615 fp1->f_data = so1; 616 error = falloc(td->td_lwp, &fp2, &fd2); 617 if (error) 618 goto free3; 619 fp2->f_data = so2; 620 sv[1] = fd2; 621 error = soconnect2(so1, so2); 622 if (error) 623 goto free4; 624 if (type == SOCK_DGRAM) { 625 /* 626 * Datagram socket connection is asymmetric. 627 */ 628 error = soconnect2(so2, so1); 629 if (error) 630 goto free4; 631 } 632 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 633 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 634 fp1->f_ops = fp2->f_ops = &socketops; 635 fsetfd(fdp, fp1, fd1); 636 fsetfd(fdp, fp2, fd2); 637 fdrop(fp1); 638 fdrop(fp2); 639 return (error); 640 free4: 641 fsetfd(fdp, NULL, fd2); 642 fdrop(fp2); 643 free3: 644 fsetfd(fdp, NULL, fd1); 645 fdrop(fp1); 646 free2: 647 (void)soclose(so2, 0); 648 free1: 649 (void)soclose(so1, 0); 650 return (error); 651 } 652 653 /* 654 * socketpair(int domain, int type, int protocol, int *rsv) 655 */ 656 int 657 sys_socketpair(struct socketpair_args *uap) 658 { 659 int error, sockv[2]; 660 661 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 662 663 if (error == 0) { 664 error = copyout(sockv, uap->rsv, sizeof(sockv)); 665 666 if (error != 0) { 667 kern_close(sockv[0]); 668 kern_close(sockv[1]); 669 } 670 } 671 672 return (error); 673 } 674 675 int 676 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 677 struct mbuf *control, int flags, size_t *res) 678 { 679 struct thread *td = curthread; 680 struct lwp *lp = td->td_lwp; 681 struct proc *p = td->td_proc; 682 struct file *fp; 683 size_t len; 684 int error; 685 struct socket *so; 686 #ifdef KTRACE 687 struct iovec *ktriov = NULL; 688 struct uio ktruio; 689 #endif 690 691 error = holdsock(p->p_fd, s, &fp); 692 if (error) 693 return (error); 694 #ifdef KTRACE 695 if (KTRPOINT(td, KTR_GENIO)) { 696 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 697 698 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 699 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 700 ktruio = *auio; 701 } 702 #endif 703 len = auio->uio_resid; 704 so = (struct socket *)fp->f_data; 705 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 706 if (fp->f_flag & FNONBLOCK) 707 flags |= MSG_FNONBLOCKING; 708 } 709 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 710 if (error) { 711 if (auio->uio_resid != len && (error == ERESTART || 712 error == EINTR || error == EWOULDBLOCK)) 713 error = 0; 714 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 715 !(so->so_options & SO_NOSIGPIPE)) 716 lwpsignal(p, lp, SIGPIPE); 717 } 718 #ifdef KTRACE 719 if (ktriov != NULL) { 720 if (error == 0) { 721 ktruio.uio_iov = ktriov; 722 ktruio.uio_resid = len - auio->uio_resid; 723 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 724 } 725 kfree(ktriov, M_TEMP); 726 } 727 #endif 728 if (error == 0) 729 *res = len - auio->uio_resid; 730 fdrop(fp); 731 return (error); 732 } 733 734 /* 735 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 736 * 737 * MPALMOSTSAFE 738 */ 739 int 740 sys_sendto(struct sendto_args *uap) 741 { 742 struct thread *td = curthread; 743 struct uio auio; 744 struct iovec aiov; 745 struct sockaddr *sa = NULL; 746 int error; 747 748 if (uap->to) { 749 error = getsockaddr(&sa, uap->to, uap->tolen); 750 if (error) 751 return (error); 752 } 753 aiov.iov_base = uap->buf; 754 aiov.iov_len = uap->len; 755 auio.uio_iov = &aiov; 756 auio.uio_iovcnt = 1; 757 auio.uio_offset = 0; 758 auio.uio_resid = uap->len; 759 auio.uio_segflg = UIO_USERSPACE; 760 auio.uio_rw = UIO_WRITE; 761 auio.uio_td = td; 762 763 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 764 &uap->sysmsg_szresult); 765 766 if (sa) 767 kfree(sa, M_SONAME); 768 return (error); 769 } 770 771 /* 772 * sendmsg_args(int s, caddr_t msg, int flags) 773 * 774 * MPALMOSTSAFE 775 */ 776 int 777 sys_sendmsg(struct sendmsg_args *uap) 778 { 779 struct thread *td = curthread; 780 struct msghdr msg; 781 struct uio auio; 782 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 783 struct sockaddr *sa = NULL; 784 struct mbuf *control = NULL; 785 int error; 786 787 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 788 if (error) 789 return (error); 790 791 /* 792 * Conditionally copyin msg.msg_name. 793 */ 794 if (msg.msg_name) { 795 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 796 if (error) 797 return (error); 798 } 799 800 /* 801 * Populate auio. 802 */ 803 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 804 &auio.uio_resid); 805 if (error) 806 goto cleanup2; 807 auio.uio_iov = iov; 808 auio.uio_iovcnt = msg.msg_iovlen; 809 auio.uio_offset = 0; 810 auio.uio_segflg = UIO_USERSPACE; 811 auio.uio_rw = UIO_WRITE; 812 auio.uio_td = td; 813 814 /* 815 * Conditionally copyin msg.msg_control. 816 */ 817 if (msg.msg_control) { 818 if (msg.msg_controllen < sizeof(struct cmsghdr) || 819 msg.msg_controllen > MLEN) { 820 error = EINVAL; 821 goto cleanup; 822 } 823 control = m_get(MB_WAIT, MT_CONTROL); 824 if (control == NULL) { 825 error = ENOBUFS; 826 goto cleanup; 827 } 828 control->m_len = msg.msg_controllen; 829 error = copyin(msg.msg_control, mtod(control, caddr_t), 830 msg.msg_controllen); 831 if (error) { 832 m_free(control); 833 goto cleanup; 834 } 835 } 836 837 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 838 &uap->sysmsg_szresult); 839 840 cleanup: 841 iovec_free(&iov, aiov); 842 cleanup2: 843 if (sa) 844 kfree(sa, M_SONAME); 845 return (error); 846 } 847 848 /* 849 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 850 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 851 * Don't forget to FREE() and m_free() these if they are returned. 852 */ 853 int 854 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 855 struct mbuf **control, int *flags, size_t *res) 856 { 857 struct thread *td = curthread; 858 struct proc *p = td->td_proc; 859 struct file *fp; 860 size_t len; 861 int error; 862 int lflags; 863 struct socket *so; 864 #ifdef KTRACE 865 struct iovec *ktriov = NULL; 866 struct uio ktruio; 867 #endif 868 869 error = holdsock(p->p_fd, s, &fp); 870 if (error) 871 return (error); 872 #ifdef KTRACE 873 if (KTRPOINT(td, KTR_GENIO)) { 874 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 875 876 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 877 bcopy(auio->uio_iov, ktriov, iovlen); 878 ktruio = *auio; 879 } 880 #endif 881 len = auio->uio_resid; 882 so = (struct socket *)fp->f_data; 883 884 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 885 if (fp->f_flag & FNONBLOCK) { 886 if (flags) { 887 *flags |= MSG_FNONBLOCKING; 888 } else { 889 lflags = MSG_FNONBLOCKING; 890 flags = &lflags; 891 } 892 } 893 } 894 895 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 896 if (error) { 897 if (auio->uio_resid != len && (error == ERESTART || 898 error == EINTR || error == EWOULDBLOCK)) 899 error = 0; 900 } 901 #ifdef KTRACE 902 if (ktriov != NULL) { 903 if (error == 0) { 904 ktruio.uio_iov = ktriov; 905 ktruio.uio_resid = len - auio->uio_resid; 906 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 907 } 908 kfree(ktriov, M_TEMP); 909 } 910 #endif 911 if (error == 0) 912 *res = len - auio->uio_resid; 913 fdrop(fp); 914 return (error); 915 } 916 917 /* 918 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 919 * caddr_t from, int *fromlenaddr) 920 * 921 * MPALMOSTSAFE 922 */ 923 int 924 sys_recvfrom(struct recvfrom_args *uap) 925 { 926 struct thread *td = curthread; 927 struct uio auio; 928 struct iovec aiov; 929 struct sockaddr *sa = NULL; 930 int error, fromlen; 931 932 if (uap->from && uap->fromlenaddr) { 933 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 934 if (error) 935 return (error); 936 if (fromlen < 0) 937 return (EINVAL); 938 } else { 939 fromlen = 0; 940 } 941 aiov.iov_base = uap->buf; 942 aiov.iov_len = uap->len; 943 auio.uio_iov = &aiov; 944 auio.uio_iovcnt = 1; 945 auio.uio_offset = 0; 946 auio.uio_resid = uap->len; 947 auio.uio_segflg = UIO_USERSPACE; 948 auio.uio_rw = UIO_READ; 949 auio.uio_td = td; 950 951 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 952 &uap->flags, &uap->sysmsg_szresult); 953 954 if (error == 0 && uap->from) { 955 /* note: sa may still be NULL */ 956 if (sa) { 957 fromlen = MIN(fromlen, sa->sa_len); 958 error = copyout(sa, uap->from, fromlen); 959 } else { 960 fromlen = 0; 961 } 962 if (error == 0) { 963 error = copyout(&fromlen, uap->fromlenaddr, 964 sizeof(fromlen)); 965 } 966 } 967 if (sa) 968 kfree(sa, M_SONAME); 969 970 return (error); 971 } 972 973 /* 974 * recvmsg_args(int s, struct msghdr *msg, int flags) 975 * 976 * MPALMOSTSAFE 977 */ 978 int 979 sys_recvmsg(struct recvmsg_args *uap) 980 { 981 struct thread *td = curthread; 982 struct msghdr msg; 983 struct uio auio; 984 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 985 struct mbuf *m, *control = NULL; 986 struct sockaddr *sa = NULL; 987 caddr_t ctlbuf; 988 socklen_t *ufromlenp, *ucontrollenp; 989 int error, fromlen, controllen, len, flags, *uflagsp; 990 991 /* 992 * This copyin handles everything except the iovec. 993 */ 994 error = copyin(uap->msg, &msg, sizeof(msg)); 995 if (error) 996 return (error); 997 998 if (msg.msg_name && msg.msg_namelen < 0) 999 return (EINVAL); 1000 if (msg.msg_control && msg.msg_controllen < 0) 1001 return (EINVAL); 1002 1003 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1004 msg_namelen)); 1005 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1006 msg_controllen)); 1007 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1008 msg_flags)); 1009 1010 /* 1011 * Populate auio. 1012 */ 1013 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1014 &auio.uio_resid); 1015 if (error) 1016 return (error); 1017 auio.uio_iov = iov; 1018 auio.uio_iovcnt = msg.msg_iovlen; 1019 auio.uio_offset = 0; 1020 auio.uio_segflg = UIO_USERSPACE; 1021 auio.uio_rw = UIO_READ; 1022 auio.uio_td = td; 1023 1024 flags = uap->flags; 1025 1026 error = kern_recvmsg(uap->s, 1027 (msg.msg_name ? &sa : NULL), &auio, 1028 (msg.msg_control ? &control : NULL), &flags, 1029 &uap->sysmsg_szresult); 1030 1031 /* 1032 * Conditionally copyout the name and populate the namelen field. 1033 */ 1034 if (error == 0 && msg.msg_name) { 1035 /* note: sa may still be NULL */ 1036 if (sa != NULL) { 1037 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1038 error = copyout(sa, msg.msg_name, fromlen); 1039 } else { 1040 fromlen = 0; 1041 } 1042 if (error == 0) 1043 error = copyout(&fromlen, ufromlenp, 1044 sizeof(*ufromlenp)); 1045 } 1046 1047 /* 1048 * Copyout msg.msg_control and msg.msg_controllen. 1049 */ 1050 if (error == 0 && msg.msg_control) { 1051 len = msg.msg_controllen; 1052 m = control; 1053 ctlbuf = (caddr_t)msg.msg_control; 1054 1055 while(m && len > 0) { 1056 unsigned int tocopy; 1057 1058 if (len >= m->m_len) { 1059 tocopy = m->m_len; 1060 } else { 1061 msg.msg_flags |= MSG_CTRUNC; 1062 tocopy = len; 1063 } 1064 1065 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1066 if (error) 1067 goto cleanup; 1068 1069 ctlbuf += tocopy; 1070 len -= tocopy; 1071 m = m->m_next; 1072 } 1073 controllen = ctlbuf - (caddr_t)msg.msg_control; 1074 error = copyout(&controllen, ucontrollenp, 1075 sizeof(*ucontrollenp)); 1076 } 1077 1078 if (error == 0) 1079 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1080 1081 cleanup: 1082 if (sa) 1083 kfree(sa, M_SONAME); 1084 iovec_free(&iov, aiov); 1085 if (control) 1086 m_freem(control); 1087 return (error); 1088 } 1089 1090 /* 1091 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1092 * in kernel pointer instead of a userland pointer. This allows us 1093 * to manipulate socket options in the emulation code. 1094 */ 1095 int 1096 kern_setsockopt(int s, struct sockopt *sopt) 1097 { 1098 struct thread *td = curthread; 1099 struct proc *p = td->td_proc; 1100 struct file *fp; 1101 int error; 1102 1103 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1104 return (EFAULT); 1105 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1106 return (EINVAL); 1107 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1108 return (EINVAL); 1109 1110 error = holdsock(p->p_fd, s, &fp); 1111 if (error) 1112 return (error); 1113 1114 error = sosetopt((struct socket *)fp->f_data, sopt); 1115 fdrop(fp); 1116 return (error); 1117 } 1118 1119 /* 1120 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1121 * 1122 * MPALMOSTSAFE 1123 */ 1124 int 1125 sys_setsockopt(struct setsockopt_args *uap) 1126 { 1127 struct thread *td = curthread; 1128 struct sockopt sopt; 1129 int error; 1130 1131 sopt.sopt_level = uap->level; 1132 sopt.sopt_name = uap->name; 1133 sopt.sopt_valsize = uap->valsize; 1134 sopt.sopt_td = td; 1135 sopt.sopt_val = NULL; 1136 1137 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1138 return (EINVAL); 1139 if (uap->val) { 1140 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1141 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1142 if (error) 1143 goto out; 1144 } 1145 1146 error = kern_setsockopt(uap->s, &sopt); 1147 out: 1148 if (uap->val) 1149 kfree(sopt.sopt_val, M_TEMP); 1150 return(error); 1151 } 1152 1153 /* 1154 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1155 * in kernel pointer instead of a userland pointer. This allows us 1156 * to manipulate socket options in the emulation code. 1157 */ 1158 int 1159 kern_getsockopt(int s, struct sockopt *sopt) 1160 { 1161 struct thread *td = curthread; 1162 struct proc *p = td->td_proc; 1163 struct file *fp; 1164 int error; 1165 1166 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1167 return (EFAULT); 1168 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1169 return (EINVAL); 1170 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1171 return (EINVAL); 1172 1173 error = holdsock(p->p_fd, s, &fp); 1174 if (error) 1175 return (error); 1176 1177 error = sogetopt((struct socket *)fp->f_data, sopt); 1178 fdrop(fp); 1179 return (error); 1180 } 1181 1182 /* 1183 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1184 * 1185 * MPALMOSTSAFE 1186 */ 1187 int 1188 sys_getsockopt(struct getsockopt_args *uap) 1189 { 1190 struct thread *td = curthread; 1191 struct sockopt sopt; 1192 int error, valsize; 1193 1194 if (uap->val) { 1195 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1196 if (error) 1197 return (error); 1198 } else { 1199 valsize = 0; 1200 } 1201 1202 sopt.sopt_level = uap->level; 1203 sopt.sopt_name = uap->name; 1204 sopt.sopt_valsize = valsize; 1205 sopt.sopt_td = td; 1206 sopt.sopt_val = NULL; 1207 1208 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1209 return (EINVAL); 1210 if (uap->val) { 1211 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1212 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1213 if (error) 1214 goto out; 1215 } 1216 1217 error = kern_getsockopt(uap->s, &sopt); 1218 if (error) 1219 goto out; 1220 valsize = sopt.sopt_valsize; 1221 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1222 if (error) 1223 goto out; 1224 if (uap->val) 1225 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1226 out: 1227 if (uap->val) 1228 kfree(sopt.sopt_val, M_TEMP); 1229 return (error); 1230 } 1231 1232 /* 1233 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1234 * This allows kern_getsockname() to return a pointer to an allocated struct 1235 * sockaddr which must be freed later with FREE(). The caller must 1236 * initialize *name to NULL. 1237 */ 1238 int 1239 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1240 { 1241 struct thread *td = curthread; 1242 struct proc *p = td->td_proc; 1243 struct file *fp; 1244 struct socket *so; 1245 struct sockaddr *sa = NULL; 1246 int error; 1247 1248 error = holdsock(p->p_fd, s, &fp); 1249 if (error) 1250 return (error); 1251 if (*namelen < 0) { 1252 fdrop(fp); 1253 return (EINVAL); 1254 } 1255 so = (struct socket *)fp->f_data; 1256 error = so_pru_sockaddr(so, &sa); 1257 if (error == 0) { 1258 if (sa == NULL) { 1259 *namelen = 0; 1260 } else { 1261 *namelen = MIN(*namelen, sa->sa_len); 1262 *name = sa; 1263 } 1264 } 1265 1266 fdrop(fp); 1267 return (error); 1268 } 1269 1270 /* 1271 * getsockname_args(int fdes, caddr_t asa, int *alen) 1272 * 1273 * Get socket name. 1274 * 1275 * MPALMOSTSAFE 1276 */ 1277 int 1278 sys_getsockname(struct getsockname_args *uap) 1279 { 1280 struct sockaddr *sa = NULL; 1281 int error, sa_len; 1282 1283 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1284 if (error) 1285 return (error); 1286 1287 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1288 1289 if (error == 0) 1290 error = copyout(sa, uap->asa, sa_len); 1291 if (error == 0) 1292 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1293 if (sa) 1294 kfree(sa, M_SONAME); 1295 return (error); 1296 } 1297 1298 /* 1299 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1300 * This allows kern_getpeername() to return a pointer to an allocated struct 1301 * sockaddr which must be freed later with FREE(). The caller must 1302 * initialize *name to NULL. 1303 */ 1304 int 1305 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1306 { 1307 struct thread *td = curthread; 1308 struct proc *p = td->td_proc; 1309 struct file *fp; 1310 struct socket *so; 1311 struct sockaddr *sa = NULL; 1312 int error; 1313 1314 error = holdsock(p->p_fd, s, &fp); 1315 if (error) 1316 return (error); 1317 if (*namelen < 0) { 1318 fdrop(fp); 1319 return (EINVAL); 1320 } 1321 so = (struct socket *)fp->f_data; 1322 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1323 fdrop(fp); 1324 return (ENOTCONN); 1325 } 1326 error = so_pru_peeraddr(so, &sa); 1327 if (error == 0) { 1328 if (sa == NULL) { 1329 *namelen = 0; 1330 } else { 1331 *namelen = MIN(*namelen, sa->sa_len); 1332 *name = sa; 1333 } 1334 } 1335 1336 fdrop(fp); 1337 return (error); 1338 } 1339 1340 /* 1341 * getpeername_args(int fdes, caddr_t asa, int *alen) 1342 * 1343 * Get name of peer for connected socket. 1344 * 1345 * MPALMOSTSAFE 1346 */ 1347 int 1348 sys_getpeername(struct getpeername_args *uap) 1349 { 1350 struct sockaddr *sa = NULL; 1351 int error, sa_len; 1352 1353 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1354 if (error) 1355 return (error); 1356 1357 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1358 1359 if (error == 0) 1360 error = copyout(sa, uap->asa, sa_len); 1361 if (error == 0) 1362 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1363 if (sa) 1364 kfree(sa, M_SONAME); 1365 return (error); 1366 } 1367 1368 int 1369 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1370 { 1371 struct sockaddr *sa; 1372 int error; 1373 1374 *namp = NULL; 1375 if (len > SOCK_MAXADDRLEN) 1376 return ENAMETOOLONG; 1377 if (len < offsetof(struct sockaddr, sa_data[0])) 1378 return EDOM; 1379 sa = kmalloc(len, M_SONAME, M_WAITOK); 1380 error = copyin(uaddr, sa, len); 1381 if (error) { 1382 kfree(sa, M_SONAME); 1383 } else { 1384 #if BYTE_ORDER != BIG_ENDIAN 1385 /* 1386 * The bind(), connect(), and sendto() syscalls were not 1387 * versioned for COMPAT_43. Thus, this check must stay. 1388 */ 1389 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1390 sa->sa_family = sa->sa_len; 1391 #endif 1392 sa->sa_len = len; 1393 *namp = sa; 1394 } 1395 return error; 1396 } 1397 1398 /* 1399 * Detach a mapped page and release resources back to the system. 1400 * We must release our wiring and if the object is ripped out 1401 * from under the vm_page we become responsible for freeing the 1402 * page. 1403 * 1404 * MPSAFE 1405 */ 1406 static void 1407 sf_buf_mfree(void *arg) 1408 { 1409 struct sf_buf *sf = arg; 1410 vm_page_t m; 1411 1412 m = sf_buf_page(sf); 1413 if (sf_buf_free(sf)) { 1414 /* sf invalid now */ 1415 vm_page_busy_wait(m, FALSE, "sockpgf"); 1416 vm_page_unwire(m, 0); 1417 if (m->object == NULL && 1418 m->wire_count == 0 && 1419 (m->flags & PG_NEED_COMMIT) == 0) { 1420 vm_page_free(m); 1421 } else { 1422 vm_page_wakeup(m); 1423 } 1424 } 1425 } 1426 1427 /* 1428 * sendfile(2). 1429 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1430 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1431 * 1432 * Send a file specified by 'fd' and starting at 'offset' to a socket 1433 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1434 * nbytes == 0. Optionally add a header and/or trailer to the socket 1435 * output. If specified, write the total number of bytes sent into *sbytes. 1436 * 1437 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1438 * the headers to count against the remaining bytes to be sent from 1439 * the file descriptor. We may wish to implement a compatibility syscall 1440 * in the future. 1441 * 1442 * MPALMOSTSAFE 1443 */ 1444 int 1445 sys_sendfile(struct sendfile_args *uap) 1446 { 1447 struct thread *td = curthread; 1448 struct proc *p = td->td_proc; 1449 struct file *fp; 1450 struct vnode *vp = NULL; 1451 struct sf_hdtr hdtr; 1452 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1453 struct uio auio; 1454 struct mbuf *mheader = NULL; 1455 size_t hbytes = 0; 1456 size_t tbytes; 1457 off_t hdtr_size = 0; 1458 off_t sbytes; 1459 int error; 1460 1461 KKASSERT(p); 1462 1463 /* 1464 * Do argument checking. Must be a regular file in, stream 1465 * type and connected socket out, positive offset. 1466 */ 1467 fp = holdfp(p->p_fd, uap->fd, FREAD); 1468 if (fp == NULL) { 1469 return (EBADF); 1470 } 1471 if (fp->f_type != DTYPE_VNODE) { 1472 fdrop(fp); 1473 return (EINVAL); 1474 } 1475 vp = (struct vnode *)fp->f_data; 1476 vref(vp); 1477 fdrop(fp); 1478 1479 /* 1480 * If specified, get the pointer to the sf_hdtr struct for 1481 * any headers/trailers. 1482 */ 1483 if (uap->hdtr) { 1484 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1485 if (error) 1486 goto done; 1487 /* 1488 * Send any headers. 1489 */ 1490 if (hdtr.headers) { 1491 error = iovec_copyin(hdtr.headers, &iov, aiov, 1492 hdtr.hdr_cnt, &hbytes); 1493 if (error) 1494 goto done; 1495 auio.uio_iov = iov; 1496 auio.uio_iovcnt = hdtr.hdr_cnt; 1497 auio.uio_offset = 0; 1498 auio.uio_segflg = UIO_USERSPACE; 1499 auio.uio_rw = UIO_WRITE; 1500 auio.uio_td = td; 1501 auio.uio_resid = hbytes; 1502 1503 mheader = m_uiomove(&auio); 1504 1505 iovec_free(&iov, aiov); 1506 if (mheader == NULL) 1507 goto done; 1508 } 1509 } 1510 1511 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1512 &sbytes, uap->flags); 1513 if (error) 1514 goto done; 1515 1516 /* 1517 * Send trailers. Wimp out and use writev(2). 1518 */ 1519 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1520 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1521 hdtr.trl_cnt, &auio.uio_resid); 1522 if (error) 1523 goto done; 1524 auio.uio_iov = iov; 1525 auio.uio_iovcnt = hdtr.trl_cnt; 1526 auio.uio_offset = 0; 1527 auio.uio_segflg = UIO_USERSPACE; 1528 auio.uio_rw = UIO_WRITE; 1529 auio.uio_td = td; 1530 1531 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1532 1533 iovec_free(&iov, aiov); 1534 if (error) 1535 goto done; 1536 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1537 } 1538 1539 done: 1540 if (vp) 1541 vrele(vp); 1542 if (uap->sbytes != NULL) { 1543 sbytes += hdtr_size; 1544 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1545 } 1546 return (error); 1547 } 1548 1549 int 1550 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1551 struct mbuf *mheader, off_t *sbytes, int flags) 1552 { 1553 struct thread *td = curthread; 1554 struct proc *p = td->td_proc; 1555 struct vm_object *obj; 1556 struct socket *so; 1557 struct file *fp; 1558 struct mbuf *m, *mp; 1559 struct sf_buf *sf; 1560 struct vm_page *pg; 1561 off_t off, xfsize; 1562 off_t hbytes = 0; 1563 int error = 0; 1564 1565 if (vp->v_type != VREG) { 1566 error = EINVAL; 1567 goto done0; 1568 } 1569 if ((obj = vp->v_object) == NULL) { 1570 error = EINVAL; 1571 goto done0; 1572 } 1573 error = holdsock(p->p_fd, sfd, &fp); 1574 if (error) 1575 goto done0; 1576 so = (struct socket *)fp->f_data; 1577 if (so->so_type != SOCK_STREAM) { 1578 error = EINVAL; 1579 goto done; 1580 } 1581 if ((so->so_state & SS_ISCONNECTED) == 0) { 1582 error = ENOTCONN; 1583 goto done; 1584 } 1585 if (offset < 0) { 1586 error = EINVAL; 1587 goto done; 1588 } 1589 1590 *sbytes = 0; 1591 /* 1592 * Protect against multiple writers to the socket. 1593 */ 1594 ssb_lock(&so->so_snd, M_WAITOK); 1595 1596 /* 1597 * Loop through the pages in the file, starting with the requested 1598 * offset. Get a file page (do I/O if necessary), map the file page 1599 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1600 * it on the socket. 1601 */ 1602 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1603 vm_pindex_t pindex; 1604 vm_offset_t pgoff; 1605 int space; 1606 1607 pindex = OFF_TO_IDX(off); 1608 retry_lookup: 1609 /* 1610 * Calculate the amount to transfer. Not to exceed a page, 1611 * the EOF, or the passed in nbytes. 1612 */ 1613 xfsize = vp->v_filesize - off; 1614 if (xfsize > PAGE_SIZE) 1615 xfsize = PAGE_SIZE; 1616 pgoff = (vm_offset_t)(off & PAGE_MASK); 1617 if (PAGE_SIZE - pgoff < xfsize) 1618 xfsize = PAGE_SIZE - pgoff; 1619 if (nbytes && xfsize > (nbytes - *sbytes)) 1620 xfsize = nbytes - *sbytes; 1621 if (xfsize <= 0) 1622 break; 1623 /* 1624 * Optimize the non-blocking case by looking at the socket space 1625 * before going to the extra work of constituting the sf_buf. 1626 */ 1627 if ((fp->f_flag & FNONBLOCK) && 1628 ssb_space_prealloc(&so->so_snd) <= 0) { 1629 if (so->so_state & SS_CANTSENDMORE) 1630 error = EPIPE; 1631 else 1632 error = EAGAIN; 1633 ssb_unlock(&so->so_snd); 1634 goto done; 1635 } 1636 /* 1637 * Attempt to look up the page. 1638 * 1639 * Allocate if not found, wait and loop if busy, then 1640 * wire the page. critical section protection is 1641 * required to maintain the object association (an 1642 * interrupt can free the page) through to the 1643 * vm_page_wire() call. 1644 */ 1645 vm_object_hold(obj); 1646 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error); 1647 if (error) { 1648 vm_page_sleep_busy(pg, TRUE, "sfpbsy"); 1649 vm_object_drop(obj); 1650 goto retry_lookup; 1651 } 1652 if (pg == NULL) { 1653 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL | 1654 VM_ALLOC_NULL_OK); 1655 if (pg == NULL) { 1656 vm_wait(0); 1657 vm_object_drop(obj); 1658 goto retry_lookup; 1659 } 1660 } 1661 vm_page_wire(pg); 1662 vm_object_drop(obj); 1663 1664 /* 1665 * If page is not valid for what we need, initiate I/O 1666 */ 1667 1668 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1669 struct uio auio; 1670 struct iovec aiov; 1671 int bsize; 1672 1673 /* 1674 * Ensure that our page is still around when the I/O 1675 * completes. 1676 */ 1677 vm_page_io_start(pg); 1678 vm_page_wakeup(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_busy_wait(pg, FALSE, "sockpg"); 1700 vm_page_io_finish(pg); 1701 if (error) { 1702 vm_page_unwire(pg, 0); 1703 vm_page_wakeup(pg); 1704 vm_page_try_to_free(pg); 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 vm_page_unwire(pg, 0); 1717 vm_page_wakeup(pg); 1718 vm_page_try_to_free(pg); 1719 ssb_unlock(&so->so_snd); 1720 error = EINTR; 1721 goto done; 1722 } 1723 vm_page_wakeup(pg); 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 space = ssb_space_prealloc(&so->so_snd); 1789 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1790 if (fp->f_flag & FNONBLOCK) { 1791 m_freem(m); 1792 ssb_unlock(&so->so_snd); 1793 crit_exit(); 1794 error = EAGAIN; 1795 goto done; 1796 } 1797 error = ssb_wait(&so->so_snd); 1798 /* 1799 * An error from ssb_wait usually indicates that we've 1800 * been interrupted by a signal. If we've sent anything 1801 * then return bytes sent, otherwise return the error. 1802 */ 1803 if (error) { 1804 m_freem(m); 1805 ssb_unlock(&so->so_snd); 1806 crit_exit(); 1807 goto done; 1808 } 1809 goto retry_space; 1810 } 1811 1812 for (mp = m; mp != NULL; mp = mp->m_next) 1813 ssb_preallocstream(&so->so_snd, mp); 1814 if (use_sendfile_async) 1815 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1816 else 1817 error = so_pru_send(so, 0, m, NULL, NULL, td); 1818 1819 crit_exit(); 1820 if (error) { 1821 ssb_unlock(&so->so_snd); 1822 goto done; 1823 } 1824 } 1825 if (mheader != NULL) { 1826 *sbytes += mheader->m_pkthdr.len; 1827 1828 for (mp = mheader; mp != NULL; mp = mp->m_next) 1829 ssb_preallocstream(&so->so_snd, mp); 1830 if (use_sendfile_async) 1831 error = so_pru_senda(so, 0, mheader, NULL, NULL, td); 1832 else 1833 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1834 1835 mheader = NULL; 1836 } 1837 ssb_unlock(&so->so_snd); 1838 1839 done: 1840 fdrop(fp); 1841 done0: 1842 if (mheader != NULL) 1843 m_freem(mheader); 1844 return (error); 1845 } 1846 1847 /* 1848 * MPALMOSTSAFE 1849 */ 1850 int 1851 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1852 { 1853 #ifdef SCTP 1854 struct thread *td = curthread; 1855 struct filedesc *fdp = td->td_proc->p_fd; 1856 struct file *lfp = NULL; 1857 struct file *nfp = NULL; 1858 int error; 1859 struct socket *head, *so; 1860 caddr_t assoc_id; 1861 int fd; 1862 short fflag; /* type must match fp->f_flag */ 1863 1864 assoc_id = uap->name; 1865 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp); 1866 if (error) 1867 return (error); 1868 1869 crit_enter(); 1870 head = (struct socket *)lfp->f_data; 1871 error = sctp_can_peel_off(head, assoc_id); 1872 if (error) { 1873 crit_exit(); 1874 goto done; 1875 } 1876 /* 1877 * At this point we know we do have a assoc to pull 1878 * we proceed to get the fd setup. This may block 1879 * but that is ok. 1880 */ 1881 1882 fflag = lfp->f_flag; 1883 error = falloc(td->td_lwp, &nfp, &fd); 1884 if (error) { 1885 /* 1886 * Probably ran out of file descriptors. Put the 1887 * unaccepted connection back onto the queue and 1888 * do another wakeup so some other process might 1889 * have a chance at it. 1890 */ 1891 crit_exit(); 1892 goto done; 1893 } 1894 uap->sysmsg_iresult = fd; 1895 1896 so = sctp_get_peeloff(head, assoc_id, &error); 1897 if (so == NULL) { 1898 /* 1899 * Either someone else peeled it off OR 1900 * we can't get a socket. 1901 */ 1902 goto noconnection; 1903 } 1904 soreference(so); /* reference needed */ 1905 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */ 1906 so->so_head = NULL; 1907 if (head->so_sigio != NULL) 1908 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 1909 1910 nfp->f_type = DTYPE_SOCKET; 1911 nfp->f_flag = fflag; 1912 nfp->f_ops = &socketops; 1913 nfp->f_data = so; 1914 1915 noconnection: 1916 /* 1917 * Assign the file pointer to the reserved descriptor, or clear 1918 * the reserved descriptor if an error occured. 1919 */ 1920 if (error) 1921 fsetfd(fdp, NULL, fd); 1922 else 1923 fsetfd(fdp, nfp, fd); 1924 crit_exit(); 1925 /* 1926 * Release explicitly held references before returning. 1927 */ 1928 done: 1929 if (nfp != NULL) 1930 fdrop(nfp); 1931 fdrop(lfp); 1932 return (error); 1933 #else /* SCTP */ 1934 return(EOPNOTSUPP); 1935 #endif /* SCTP */ 1936 } 1937