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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 33 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 34 */ 35 36 #include "opt_ktrace.h" 37 #include "opt_sctp.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/sysproto.h> 43 #include <sys/malloc.h> 44 #include <sys/filedesc.h> 45 #include <sys/event.h> 46 #include <sys/proc.h> 47 #include <sys/fcntl.h> 48 #include <sys/file.h> 49 #include <sys/filio.h> 50 #include <sys/kern_syscall.h> 51 #include <sys/mbuf.h> 52 #include <sys/protosw.h> 53 #include <sys/sfbuf.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/socketops.h> 57 #include <sys/uio.h> 58 #include <sys/vnode.h> 59 #include <sys/lock.h> 60 #include <sys/mount.h> 61 #ifdef KTRACE 62 #include <sys/ktrace.h> 63 #endif 64 #include <vm/vm.h> 65 #include <vm/vm_object.h> 66 #include <vm/vm_page.h> 67 #include <vm/vm_pageout.h> 68 #include <vm/vm_kern.h> 69 #include <vm/vm_extern.h> 70 #include <sys/file2.h> 71 #include <sys/signalvar.h> 72 #include <sys/serialize.h> 73 74 #include <sys/thread2.h> 75 #include <sys/msgport2.h> 76 #include <sys/socketvar2.h> 77 #include <net/netmsg2.h> 78 79 #ifdef SCTP 80 #include <netinet/sctp_peeloff.h> 81 #endif /* SCTP */ 82 83 extern int use_soaccept_pred_fast; 84 extern int use_sendfile_async; 85 86 /* 87 * System call interface to the socket abstraction. 88 */ 89 90 extern struct fileops socketops; 91 92 /* 93 * socket_args(int domain, int type, int protocol) 94 */ 95 int 96 kern_socket(int domain, int type, int protocol, int *res) 97 { 98 struct thread *td = curthread; 99 struct filedesc *fdp = td->td_proc->p_fd; 100 struct socket *so; 101 struct file *fp; 102 int fd, error; 103 104 KKASSERT(td->td_lwp); 105 106 error = falloc(td->td_lwp, &fp, &fd); 107 if (error) 108 return (error); 109 error = socreate(domain, &so, type, protocol, td); 110 if (error) { 111 fsetfd(fdp, NULL, fd); 112 } else { 113 fp->f_type = DTYPE_SOCKET; 114 fp->f_flag = FREAD | FWRITE; 115 fp->f_ops = &socketops; 116 fp->f_data = so; 117 *res = fd; 118 fsetfd(fdp, fp, fd); 119 } 120 fdrop(fp); 121 return (error); 122 } 123 124 /* 125 * MPALMOSTSAFE 126 */ 127 int 128 sys_socket(struct socket_args *uap) 129 { 130 int error; 131 132 error = kern_socket(uap->domain, uap->type, uap->protocol, 133 &uap->sysmsg_iresult); 134 135 return (error); 136 } 137 138 int 139 kern_bind(int s, struct sockaddr *sa) 140 { 141 struct thread *td = curthread; 142 struct proc *p = td->td_proc; 143 struct file *fp; 144 int error; 145 146 KKASSERT(p); 147 error = holdsock(p->p_fd, s, &fp); 148 if (error) 149 return (error); 150 error = sobind((struct socket *)fp->f_data, sa, td); 151 fdrop(fp); 152 return (error); 153 } 154 155 /* 156 * bind_args(int s, caddr_t name, int namelen) 157 * 158 * MPALMOSTSAFE 159 */ 160 int 161 sys_bind(struct bind_args *uap) 162 { 163 struct sockaddr *sa; 164 int error; 165 166 error = getsockaddr(&sa, uap->name, uap->namelen); 167 if (error) 168 return (error); 169 error = kern_bind(uap->s, sa); 170 kfree(sa, M_SONAME); 171 172 return (error); 173 } 174 175 int 176 kern_listen(int s, int backlog) 177 { 178 struct thread *td = curthread; 179 struct proc *p = td->td_proc; 180 struct file *fp; 181 int error; 182 183 KKASSERT(p); 184 error = holdsock(p->p_fd, s, &fp); 185 if (error) 186 return (error); 187 error = solisten((struct socket *)fp->f_data, backlog, td); 188 fdrop(fp); 189 return(error); 190 } 191 192 /* 193 * listen_args(int s, int backlog) 194 * 195 * MPALMOSTSAFE 196 */ 197 int 198 sys_listen(struct listen_args *uap) 199 { 200 int error; 201 202 error = kern_listen(uap->s, uap->backlog); 203 return (error); 204 } 205 206 /* 207 * Returns the accepted socket as well. 208 * 209 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 210 * pool token is absolutely required to avoid a sofree() race, 211 * as well as to avoid tailq handling races. 212 */ 213 static boolean_t 214 soaccept_predicate(struct netmsg_so_notify *msg) 215 { 216 struct socket *head = msg->base.nm_so; 217 struct socket *so; 218 219 if (head->so_error != 0) { 220 msg->base.lmsg.ms_error = head->so_error; 221 return (TRUE); 222 } 223 lwkt_getpooltoken(head); 224 if (!TAILQ_EMPTY(&head->so_comp)) { 225 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 226 so = TAILQ_FIRST(&head->so_comp); 227 TAILQ_REMOVE(&head->so_comp, so, so_list); 228 head->so_qlen--; 229 soclrstate(so, SS_COMP); 230 so->so_head = NULL; 231 soreference(so); 232 233 lwkt_relpooltoken(head); 234 235 msg->base.lmsg.ms_error = 0; 236 msg->base.nm_so = so; 237 return (TRUE); 238 } 239 lwkt_relpooltoken(head); 240 if (head->so_state & SS_CANTRCVMORE) { 241 msg->base.lmsg.ms_error = ECONNABORTED; 242 return (TRUE); 243 } 244 if (msg->nm_fflags & FNONBLOCK) { 245 msg->base.lmsg.ms_error = EWOULDBLOCK; 246 return (TRUE); 247 } 248 249 return (FALSE); 250 } 251 252 /* 253 * The second argument to kern_accept() is a handle to a struct sockaddr. 254 * This allows kern_accept() to return a pointer to an allocated struct 255 * sockaddr which must be freed later with FREE(). The caller must 256 * initialize *name to NULL. 257 */ 258 int 259 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 260 { 261 struct thread *td = curthread; 262 struct filedesc *fdp = td->td_proc->p_fd; 263 struct file *lfp = NULL; 264 struct file *nfp = NULL; 265 struct sockaddr *sa; 266 struct socket *head, *so; 267 struct netmsg_so_notify msg; 268 int fd; 269 u_int fflag; /* type must match fp->f_flag */ 270 int error, tmp; 271 272 *res = -1; 273 if (name && namelen && *namelen < 0) 274 return (EINVAL); 275 276 error = holdsock(td->td_proc->p_fd, s, &lfp); 277 if (error) 278 return (error); 279 280 error = falloc(td->td_lwp, &nfp, &fd); 281 if (error) { /* Probably ran out of file descriptors. */ 282 fdrop(lfp); 283 return (error); 284 } 285 head = (struct socket *)lfp->f_data; 286 if ((head->so_options & SO_ACCEPTCONN) == 0) { 287 error = EINVAL; 288 goto done; 289 } 290 291 if (fflags & O_FBLOCKING) 292 fflags |= lfp->f_flag & ~FNONBLOCK; 293 else if (fflags & O_FNONBLOCKING) 294 fflags |= lfp->f_flag | FNONBLOCK; 295 else 296 fflags = lfp->f_flag; 297 298 if (use_soaccept_pred_fast) { 299 boolean_t pred; 300 301 /* Initialize necessary parts for soaccept_predicate() */ 302 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 303 msg.nm_fflags = fflags; 304 305 lwkt_getpooltoken(head); 306 pred = soaccept_predicate(&msg); 307 lwkt_relpooltoken(head); 308 309 if (pred) { 310 error = msg.base.lmsg.ms_error; 311 if (error) 312 goto done; 313 else 314 goto accepted; 315 } 316 } 317 318 /* optimize for uniprocessor case later XXX JH */ 319 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 320 0, netmsg_so_notify, netmsg_so_notify_doabort); 321 msg.nm_predicate = soaccept_predicate; 322 msg.nm_fflags = fflags; 323 msg.nm_etype = NM_REVENT; 324 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 325 if (error) 326 goto done; 327 328 accepted: 329 /* 330 * At this point we have the connection that's ready to be accepted. 331 * 332 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 333 * to eat the ref and turn it into a descriptor. 334 */ 335 so = msg.base.nm_so; 336 337 fflag = lfp->f_flag; 338 339 /* connection has been removed from the listen queue */ 340 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 341 342 if (head->so_sigio != NULL) 343 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 344 345 nfp->f_type = DTYPE_SOCKET; 346 nfp->f_flag = fflag; 347 nfp->f_ops = &socketops; 348 nfp->f_data = so; 349 /* Sync socket nonblocking/async state with file flags */ 350 tmp = fflag & FNONBLOCK; 351 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 352 tmp = fflag & FASYNC; 353 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 354 355 sa = NULL; 356 if (so->so_faddr != NULL) { 357 sa = so->so_faddr; 358 so->so_faddr = NULL; 359 360 soaccept_generic(so); 361 error = 0; 362 } else { 363 error = soaccept(so, &sa); 364 } 365 366 /* 367 * Set the returned name and namelen as applicable. Set the returned 368 * namelen to 0 for older code which might ignore the return value 369 * from accept. 370 */ 371 if (error == 0) { 372 if (sa && name && namelen) { 373 if (*namelen > sa->sa_len) 374 *namelen = sa->sa_len; 375 *name = sa; 376 } else { 377 if (sa) 378 kfree(sa, M_SONAME); 379 } 380 } 381 382 done: 383 /* 384 * If an error occured clear the reserved descriptor, else associate 385 * nfp with it. 386 * 387 * Note that *res is normally ignored if an error is returned but 388 * a syscall message will still have access to the result code. 389 */ 390 if (error) { 391 fsetfd(fdp, NULL, fd); 392 } else { 393 *res = fd; 394 fsetfd(fdp, nfp, fd); 395 } 396 fdrop(nfp); 397 fdrop(lfp); 398 return (error); 399 } 400 401 /* 402 * accept(int s, caddr_t name, int *anamelen) 403 * 404 * MPALMOSTSAFE 405 */ 406 int 407 sys_accept(struct accept_args *uap) 408 { 409 struct sockaddr *sa = NULL; 410 int sa_len; 411 int error; 412 413 if (uap->name) { 414 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 415 if (error) 416 return (error); 417 418 error = kern_accept(uap->s, 0, &sa, &sa_len, 419 &uap->sysmsg_iresult); 420 421 if (error == 0) 422 error = copyout(sa, uap->name, sa_len); 423 if (error == 0) { 424 error = copyout(&sa_len, uap->anamelen, 425 sizeof(*uap->anamelen)); 426 } 427 if (sa) 428 kfree(sa, M_SONAME); 429 } else { 430 error = kern_accept(uap->s, 0, NULL, 0, 431 &uap->sysmsg_iresult); 432 } 433 return (error); 434 } 435 436 /* 437 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 438 * 439 * MPALMOSTSAFE 440 */ 441 int 442 sys_extaccept(struct extaccept_args *uap) 443 { 444 struct sockaddr *sa = NULL; 445 int sa_len; 446 int error; 447 int fflags = uap->flags & O_FMASK; 448 449 if (uap->name) { 450 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 451 if (error) 452 return (error); 453 454 error = kern_accept(uap->s, fflags, &sa, &sa_len, 455 &uap->sysmsg_iresult); 456 457 if (error == 0) 458 error = copyout(sa, uap->name, sa_len); 459 if (error == 0) { 460 error = copyout(&sa_len, uap->anamelen, 461 sizeof(*uap->anamelen)); 462 } 463 if (sa) 464 kfree(sa, M_SONAME); 465 } else { 466 error = kern_accept(uap->s, fflags, NULL, 0, 467 &uap->sysmsg_iresult); 468 } 469 return (error); 470 } 471 472 473 /* 474 * Returns TRUE if predicate satisfied. 475 */ 476 static boolean_t 477 soconnected_predicate(struct netmsg_so_notify *msg) 478 { 479 struct socket *so = msg->base.nm_so; 480 481 /* check predicate */ 482 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 483 msg->base.lmsg.ms_error = so->so_error; 484 return (TRUE); 485 } 486 487 return (FALSE); 488 } 489 490 int 491 kern_connect(int s, int fflags, struct sockaddr *sa) 492 { 493 struct thread *td = curthread; 494 struct proc *p = td->td_proc; 495 struct file *fp; 496 struct socket *so; 497 int error, interrupted = 0; 498 499 error = holdsock(p->p_fd, s, &fp); 500 if (error) 501 return (error); 502 so = (struct socket *)fp->f_data; 503 504 if (fflags & O_FBLOCKING) 505 /* fflags &= ~FNONBLOCK; */; 506 else if (fflags & O_FNONBLOCKING) 507 fflags |= FNONBLOCK; 508 else 509 fflags = fp->f_flag; 510 511 if (so->so_state & SS_ISCONNECTING) { 512 error = EALREADY; 513 goto done; 514 } 515 error = soconnect(so, sa, td); 516 if (error) 517 goto bad; 518 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 519 error = EINPROGRESS; 520 goto done; 521 } 522 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 523 struct netmsg_so_notify msg; 524 525 netmsg_init_abortable(&msg.base, so, 526 &curthread->td_msgport, 527 0, 528 netmsg_so_notify, 529 netmsg_so_notify_doabort); 530 msg.nm_predicate = soconnected_predicate; 531 msg.nm_etype = NM_REVENT; 532 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 533 if (error == EINTR || error == ERESTART) 534 interrupted = 1; 535 } 536 if (error == 0) { 537 error = so->so_error; 538 so->so_error = 0; 539 } 540 bad: 541 if (!interrupted) 542 soclrstate(so, SS_ISCONNECTING); 543 if (error == ERESTART) 544 error = EINTR; 545 done: 546 fdrop(fp); 547 return (error); 548 } 549 550 /* 551 * connect_args(int s, caddr_t name, int namelen) 552 * 553 * MPALMOSTSAFE 554 */ 555 int 556 sys_connect(struct connect_args *uap) 557 { 558 struct sockaddr *sa; 559 int error; 560 561 error = getsockaddr(&sa, uap->name, uap->namelen); 562 if (error) 563 return (error); 564 error = kern_connect(uap->s, 0, sa); 565 kfree(sa, M_SONAME); 566 567 return (error); 568 } 569 570 /* 571 * connect_args(int s, int fflags, caddr_t name, int namelen) 572 * 573 * MPALMOSTSAFE 574 */ 575 int 576 sys_extconnect(struct extconnect_args *uap) 577 { 578 struct sockaddr *sa; 579 int error; 580 int fflags = uap->flags & O_FMASK; 581 582 error = getsockaddr(&sa, uap->name, uap->namelen); 583 if (error) 584 return (error); 585 error = kern_connect(uap->s, fflags, sa); 586 kfree(sa, M_SONAME); 587 588 return (error); 589 } 590 591 int 592 kern_socketpair(int domain, int type, int protocol, int *sv) 593 { 594 struct thread *td = curthread; 595 struct filedesc *fdp; 596 struct file *fp1, *fp2; 597 struct socket *so1, *so2; 598 int fd1, fd2, error; 599 600 fdp = td->td_proc->p_fd; 601 error = socreate(domain, &so1, type, protocol, td); 602 if (error) 603 return (error); 604 error = socreate(domain, &so2, type, protocol, td); 605 if (error) 606 goto free1; 607 error = falloc(td->td_lwp, &fp1, &fd1); 608 if (error) 609 goto free2; 610 sv[0] = fd1; 611 fp1->f_data = so1; 612 error = falloc(td->td_lwp, &fp2, &fd2); 613 if (error) 614 goto free3; 615 fp2->f_data = so2; 616 sv[1] = fd2; 617 error = soconnect2(so1, so2); 618 if (error) 619 goto free4; 620 if (type == SOCK_DGRAM) { 621 /* 622 * Datagram socket connection is asymmetric. 623 */ 624 error = soconnect2(so2, so1); 625 if (error) 626 goto free4; 627 } 628 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 629 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 630 fp1->f_ops = fp2->f_ops = &socketops; 631 fsetfd(fdp, fp1, fd1); 632 fsetfd(fdp, fp2, fd2); 633 fdrop(fp1); 634 fdrop(fp2); 635 return (error); 636 free4: 637 fsetfd(fdp, NULL, fd2); 638 fdrop(fp2); 639 free3: 640 fsetfd(fdp, NULL, fd1); 641 fdrop(fp1); 642 free2: 643 (void)soclose(so2, 0); 644 free1: 645 (void)soclose(so1, 0); 646 return (error); 647 } 648 649 /* 650 * socketpair(int domain, int type, int protocol, int *rsv) 651 */ 652 int 653 sys_socketpair(struct socketpair_args *uap) 654 { 655 int error, sockv[2]; 656 657 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 658 659 if (error == 0) { 660 error = copyout(sockv, uap->rsv, sizeof(sockv)); 661 662 if (error != 0) { 663 kern_close(sockv[0]); 664 kern_close(sockv[1]); 665 } 666 } 667 668 return (error); 669 } 670 671 int 672 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 673 struct mbuf *control, int flags, size_t *res) 674 { 675 struct thread *td = curthread; 676 struct lwp *lp = td->td_lwp; 677 struct proc *p = td->td_proc; 678 struct file *fp; 679 size_t len; 680 int error; 681 struct socket *so; 682 #ifdef KTRACE 683 struct iovec *ktriov = NULL; 684 struct uio ktruio; 685 #endif 686 687 error = holdsock(p->p_fd, s, &fp); 688 if (error) 689 return (error); 690 #ifdef KTRACE 691 if (KTRPOINT(td, KTR_GENIO)) { 692 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 693 694 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 695 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 696 ktruio = *auio; 697 } 698 #endif 699 len = auio->uio_resid; 700 so = (struct socket *)fp->f_data; 701 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 702 if (fp->f_flag & FNONBLOCK) 703 flags |= MSG_FNONBLOCKING; 704 } 705 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 706 if (error) { 707 if (auio->uio_resid != len && (error == ERESTART || 708 error == EINTR || error == EWOULDBLOCK)) 709 error = 0; 710 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 711 !(so->so_options & SO_NOSIGPIPE)) 712 lwpsignal(p, lp, SIGPIPE); 713 } 714 #ifdef KTRACE 715 if (ktriov != NULL) { 716 if (error == 0) { 717 ktruio.uio_iov = ktriov; 718 ktruio.uio_resid = len - auio->uio_resid; 719 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 720 } 721 kfree(ktriov, M_TEMP); 722 } 723 #endif 724 if (error == 0) 725 *res = len - auio->uio_resid; 726 fdrop(fp); 727 return (error); 728 } 729 730 /* 731 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 732 * 733 * MPALMOSTSAFE 734 */ 735 int 736 sys_sendto(struct sendto_args *uap) 737 { 738 struct thread *td = curthread; 739 struct uio auio; 740 struct iovec aiov; 741 struct sockaddr *sa = NULL; 742 int error; 743 744 if (uap->to) { 745 error = getsockaddr(&sa, uap->to, uap->tolen); 746 if (error) 747 return (error); 748 } 749 aiov.iov_base = uap->buf; 750 aiov.iov_len = uap->len; 751 auio.uio_iov = &aiov; 752 auio.uio_iovcnt = 1; 753 auio.uio_offset = 0; 754 auio.uio_resid = uap->len; 755 auio.uio_segflg = UIO_USERSPACE; 756 auio.uio_rw = UIO_WRITE; 757 auio.uio_td = td; 758 759 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 760 &uap->sysmsg_szresult); 761 762 if (sa) 763 kfree(sa, M_SONAME); 764 return (error); 765 } 766 767 /* 768 * sendmsg_args(int s, caddr_t msg, int flags) 769 * 770 * MPALMOSTSAFE 771 */ 772 int 773 sys_sendmsg(struct sendmsg_args *uap) 774 { 775 struct thread *td = curthread; 776 struct msghdr msg; 777 struct uio auio; 778 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 779 struct sockaddr *sa = NULL; 780 struct mbuf *control = NULL; 781 int error; 782 783 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 784 if (error) 785 return (error); 786 787 /* 788 * Conditionally copyin msg.msg_name. 789 */ 790 if (msg.msg_name) { 791 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 792 if (error) 793 return (error); 794 } 795 796 /* 797 * Populate auio. 798 */ 799 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 800 &auio.uio_resid); 801 if (error) 802 goto cleanup2; 803 auio.uio_iov = iov; 804 auio.uio_iovcnt = msg.msg_iovlen; 805 auio.uio_offset = 0; 806 auio.uio_segflg = UIO_USERSPACE; 807 auio.uio_rw = UIO_WRITE; 808 auio.uio_td = td; 809 810 /* 811 * Conditionally copyin msg.msg_control. 812 */ 813 if (msg.msg_control) { 814 if (msg.msg_controllen < sizeof(struct cmsghdr) || 815 msg.msg_controllen > MLEN) { 816 error = EINVAL; 817 goto cleanup; 818 } 819 control = m_get(MB_WAIT, MT_CONTROL); 820 if (control == NULL) { 821 error = ENOBUFS; 822 goto cleanup; 823 } 824 control->m_len = msg.msg_controllen; 825 error = copyin(msg.msg_control, mtod(control, caddr_t), 826 msg.msg_controllen); 827 if (error) { 828 m_free(control); 829 goto cleanup; 830 } 831 } 832 833 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 834 &uap->sysmsg_szresult); 835 836 cleanup: 837 iovec_free(&iov, aiov); 838 cleanup2: 839 if (sa) 840 kfree(sa, M_SONAME); 841 return (error); 842 } 843 844 /* 845 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 846 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 847 * Don't forget to FREE() and m_free() these if they are returned. 848 */ 849 int 850 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 851 struct mbuf **control, int *flags, size_t *res) 852 { 853 struct thread *td = curthread; 854 struct proc *p = td->td_proc; 855 struct file *fp; 856 size_t len; 857 int error; 858 int lflags; 859 struct socket *so; 860 #ifdef KTRACE 861 struct iovec *ktriov = NULL; 862 struct uio ktruio; 863 #endif 864 865 error = holdsock(p->p_fd, s, &fp); 866 if (error) 867 return (error); 868 #ifdef KTRACE 869 if (KTRPOINT(td, KTR_GENIO)) { 870 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 871 872 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 873 bcopy(auio->uio_iov, ktriov, iovlen); 874 ktruio = *auio; 875 } 876 #endif 877 len = auio->uio_resid; 878 so = (struct socket *)fp->f_data; 879 880 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 881 if (fp->f_flag & FNONBLOCK) { 882 if (flags) { 883 *flags |= MSG_FNONBLOCKING; 884 } else { 885 lflags = MSG_FNONBLOCKING; 886 flags = &lflags; 887 } 888 } 889 } 890 891 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 892 if (error) { 893 if (auio->uio_resid != len && (error == ERESTART || 894 error == EINTR || error == EWOULDBLOCK)) 895 error = 0; 896 } 897 #ifdef KTRACE 898 if (ktriov != NULL) { 899 if (error == 0) { 900 ktruio.uio_iov = ktriov; 901 ktruio.uio_resid = len - auio->uio_resid; 902 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 903 } 904 kfree(ktriov, M_TEMP); 905 } 906 #endif 907 if (error == 0) 908 *res = len - auio->uio_resid; 909 fdrop(fp); 910 return (error); 911 } 912 913 /* 914 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 915 * caddr_t from, int *fromlenaddr) 916 * 917 * MPALMOSTSAFE 918 */ 919 int 920 sys_recvfrom(struct recvfrom_args *uap) 921 { 922 struct thread *td = curthread; 923 struct uio auio; 924 struct iovec aiov; 925 struct sockaddr *sa = NULL; 926 int error, fromlen; 927 928 if (uap->from && uap->fromlenaddr) { 929 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 930 if (error) 931 return (error); 932 if (fromlen < 0) 933 return (EINVAL); 934 } else { 935 fromlen = 0; 936 } 937 aiov.iov_base = uap->buf; 938 aiov.iov_len = uap->len; 939 auio.uio_iov = &aiov; 940 auio.uio_iovcnt = 1; 941 auio.uio_offset = 0; 942 auio.uio_resid = uap->len; 943 auio.uio_segflg = UIO_USERSPACE; 944 auio.uio_rw = UIO_READ; 945 auio.uio_td = td; 946 947 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 948 &uap->flags, &uap->sysmsg_szresult); 949 950 if (error == 0 && uap->from) { 951 /* note: sa may still be NULL */ 952 if (sa) { 953 fromlen = MIN(fromlen, sa->sa_len); 954 error = copyout(sa, uap->from, fromlen); 955 } else { 956 fromlen = 0; 957 } 958 if (error == 0) { 959 error = copyout(&fromlen, uap->fromlenaddr, 960 sizeof(fromlen)); 961 } 962 } 963 if (sa) 964 kfree(sa, M_SONAME); 965 966 return (error); 967 } 968 969 /* 970 * recvmsg_args(int s, struct msghdr *msg, int flags) 971 * 972 * MPALMOSTSAFE 973 */ 974 int 975 sys_recvmsg(struct recvmsg_args *uap) 976 { 977 struct thread *td = curthread; 978 struct msghdr msg; 979 struct uio auio; 980 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 981 struct mbuf *m, *control = NULL; 982 struct sockaddr *sa = NULL; 983 caddr_t ctlbuf; 984 socklen_t *ufromlenp, *ucontrollenp; 985 int error, fromlen, controllen, len, flags, *uflagsp; 986 987 /* 988 * This copyin handles everything except the iovec. 989 */ 990 error = copyin(uap->msg, &msg, sizeof(msg)); 991 if (error) 992 return (error); 993 994 if (msg.msg_name && msg.msg_namelen < 0) 995 return (EINVAL); 996 if (msg.msg_control && msg.msg_controllen < 0) 997 return (EINVAL); 998 999 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1000 msg_namelen)); 1001 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1002 msg_controllen)); 1003 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1004 msg_flags)); 1005 1006 /* 1007 * Populate auio. 1008 */ 1009 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1010 &auio.uio_resid); 1011 if (error) 1012 return (error); 1013 auio.uio_iov = iov; 1014 auio.uio_iovcnt = msg.msg_iovlen; 1015 auio.uio_offset = 0; 1016 auio.uio_segflg = UIO_USERSPACE; 1017 auio.uio_rw = UIO_READ; 1018 auio.uio_td = td; 1019 1020 flags = uap->flags; 1021 1022 error = kern_recvmsg(uap->s, 1023 (msg.msg_name ? &sa : NULL), &auio, 1024 (msg.msg_control ? &control : NULL), &flags, 1025 &uap->sysmsg_szresult); 1026 1027 /* 1028 * Conditionally copyout the name and populate the namelen field. 1029 */ 1030 if (error == 0 && msg.msg_name) { 1031 /* note: sa may still be NULL */ 1032 if (sa != NULL) { 1033 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1034 error = copyout(sa, msg.msg_name, fromlen); 1035 } else { 1036 fromlen = 0; 1037 } 1038 if (error == 0) 1039 error = copyout(&fromlen, ufromlenp, 1040 sizeof(*ufromlenp)); 1041 } 1042 1043 /* 1044 * Copyout msg.msg_control and msg.msg_controllen. 1045 */ 1046 if (error == 0 && msg.msg_control) { 1047 len = msg.msg_controllen; 1048 m = control; 1049 ctlbuf = (caddr_t)msg.msg_control; 1050 1051 while(m && len > 0) { 1052 unsigned int tocopy; 1053 1054 if (len >= m->m_len) { 1055 tocopy = m->m_len; 1056 } else { 1057 msg.msg_flags |= MSG_CTRUNC; 1058 tocopy = len; 1059 } 1060 1061 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1062 if (error) 1063 goto cleanup; 1064 1065 ctlbuf += tocopy; 1066 len -= tocopy; 1067 m = m->m_next; 1068 } 1069 controllen = ctlbuf - (caddr_t)msg.msg_control; 1070 error = copyout(&controllen, ucontrollenp, 1071 sizeof(*ucontrollenp)); 1072 } 1073 1074 if (error == 0) 1075 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1076 1077 cleanup: 1078 if (sa) 1079 kfree(sa, M_SONAME); 1080 iovec_free(&iov, aiov); 1081 if (control) 1082 m_freem(control); 1083 return (error); 1084 } 1085 1086 /* 1087 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1088 * in kernel pointer instead of a userland pointer. This allows us 1089 * to manipulate socket options in the emulation code. 1090 */ 1091 int 1092 kern_setsockopt(int s, struct sockopt *sopt) 1093 { 1094 struct thread *td = curthread; 1095 struct proc *p = td->td_proc; 1096 struct file *fp; 1097 int error; 1098 1099 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1100 return (EFAULT); 1101 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1102 return (EINVAL); 1103 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1104 return (EINVAL); 1105 1106 error = holdsock(p->p_fd, s, &fp); 1107 if (error) 1108 return (error); 1109 1110 error = sosetopt((struct socket *)fp->f_data, sopt); 1111 fdrop(fp); 1112 return (error); 1113 } 1114 1115 /* 1116 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1117 * 1118 * MPALMOSTSAFE 1119 */ 1120 int 1121 sys_setsockopt(struct setsockopt_args *uap) 1122 { 1123 struct thread *td = curthread; 1124 struct sockopt sopt; 1125 int error; 1126 1127 sopt.sopt_level = uap->level; 1128 sopt.sopt_name = uap->name; 1129 sopt.sopt_valsize = uap->valsize; 1130 sopt.sopt_td = td; 1131 sopt.sopt_val = NULL; 1132 1133 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1134 return (EINVAL); 1135 if (uap->val) { 1136 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1137 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1138 if (error) 1139 goto out; 1140 } 1141 1142 error = kern_setsockopt(uap->s, &sopt); 1143 out: 1144 if (uap->val) 1145 kfree(sopt.sopt_val, M_TEMP); 1146 return(error); 1147 } 1148 1149 /* 1150 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1151 * in kernel pointer instead of a userland pointer. This allows us 1152 * to manipulate socket options in the emulation code. 1153 */ 1154 int 1155 kern_getsockopt(int s, struct sockopt *sopt) 1156 { 1157 struct thread *td = curthread; 1158 struct proc *p = td->td_proc; 1159 struct file *fp; 1160 int error; 1161 1162 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1163 return (EFAULT); 1164 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1165 return (EINVAL); 1166 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1167 return (EINVAL); 1168 1169 error = holdsock(p->p_fd, s, &fp); 1170 if (error) 1171 return (error); 1172 1173 error = sogetopt((struct socket *)fp->f_data, sopt); 1174 fdrop(fp); 1175 return (error); 1176 } 1177 1178 /* 1179 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1180 * 1181 * MPALMOSTSAFE 1182 */ 1183 int 1184 sys_getsockopt(struct getsockopt_args *uap) 1185 { 1186 struct thread *td = curthread; 1187 struct sockopt sopt; 1188 int error, valsize; 1189 1190 if (uap->val) { 1191 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1192 if (error) 1193 return (error); 1194 } else { 1195 valsize = 0; 1196 } 1197 1198 sopt.sopt_level = uap->level; 1199 sopt.sopt_name = uap->name; 1200 sopt.sopt_valsize = valsize; 1201 sopt.sopt_td = td; 1202 sopt.sopt_val = NULL; 1203 1204 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1205 return (EINVAL); 1206 if (uap->val) { 1207 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1208 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1209 if (error) 1210 goto out; 1211 } 1212 1213 error = kern_getsockopt(uap->s, &sopt); 1214 if (error) 1215 goto out; 1216 valsize = sopt.sopt_valsize; 1217 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1218 if (error) 1219 goto out; 1220 if (uap->val) 1221 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1222 out: 1223 if (uap->val) 1224 kfree(sopt.sopt_val, M_TEMP); 1225 return (error); 1226 } 1227 1228 /* 1229 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1230 * This allows kern_getsockname() to return a pointer to an allocated struct 1231 * sockaddr which must be freed later with FREE(). The caller must 1232 * initialize *name to NULL. 1233 */ 1234 int 1235 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1236 { 1237 struct thread *td = curthread; 1238 struct proc *p = td->td_proc; 1239 struct file *fp; 1240 struct socket *so; 1241 struct sockaddr *sa = NULL; 1242 int error; 1243 1244 error = holdsock(p->p_fd, s, &fp); 1245 if (error) 1246 return (error); 1247 if (*namelen < 0) { 1248 fdrop(fp); 1249 return (EINVAL); 1250 } 1251 so = (struct socket *)fp->f_data; 1252 error = so_pru_sockaddr(so, &sa); 1253 if (error == 0) { 1254 if (sa == NULL) { 1255 *namelen = 0; 1256 } else { 1257 *namelen = MIN(*namelen, sa->sa_len); 1258 *name = sa; 1259 } 1260 } 1261 1262 fdrop(fp); 1263 return (error); 1264 } 1265 1266 /* 1267 * getsockname_args(int fdes, caddr_t asa, int *alen) 1268 * 1269 * Get socket name. 1270 * 1271 * MPALMOSTSAFE 1272 */ 1273 int 1274 sys_getsockname(struct getsockname_args *uap) 1275 { 1276 struct sockaddr *sa = NULL; 1277 int error, sa_len; 1278 1279 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1280 if (error) 1281 return (error); 1282 1283 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1284 1285 if (error == 0) 1286 error = copyout(sa, uap->asa, sa_len); 1287 if (error == 0) 1288 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1289 if (sa) 1290 kfree(sa, M_SONAME); 1291 return (error); 1292 } 1293 1294 /* 1295 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1296 * This allows kern_getpeername() to return a pointer to an allocated struct 1297 * sockaddr which must be freed later with FREE(). The caller must 1298 * initialize *name to NULL. 1299 */ 1300 int 1301 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1302 { 1303 struct thread *td = curthread; 1304 struct proc *p = td->td_proc; 1305 struct file *fp; 1306 struct socket *so; 1307 struct sockaddr *sa = NULL; 1308 int error; 1309 1310 error = holdsock(p->p_fd, s, &fp); 1311 if (error) 1312 return (error); 1313 if (*namelen < 0) { 1314 fdrop(fp); 1315 return (EINVAL); 1316 } 1317 so = (struct socket *)fp->f_data; 1318 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1319 fdrop(fp); 1320 return (ENOTCONN); 1321 } 1322 error = so_pru_peeraddr(so, &sa); 1323 if (error == 0) { 1324 if (sa == NULL) { 1325 *namelen = 0; 1326 } else { 1327 *namelen = MIN(*namelen, sa->sa_len); 1328 *name = sa; 1329 } 1330 } 1331 1332 fdrop(fp); 1333 return (error); 1334 } 1335 1336 /* 1337 * getpeername_args(int fdes, caddr_t asa, int *alen) 1338 * 1339 * Get name of peer for connected socket. 1340 * 1341 * MPALMOSTSAFE 1342 */ 1343 int 1344 sys_getpeername(struct getpeername_args *uap) 1345 { 1346 struct sockaddr *sa = NULL; 1347 int error, sa_len; 1348 1349 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1350 if (error) 1351 return (error); 1352 1353 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1354 1355 if (error == 0) 1356 error = copyout(sa, uap->asa, sa_len); 1357 if (error == 0) 1358 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1359 if (sa) 1360 kfree(sa, M_SONAME); 1361 return (error); 1362 } 1363 1364 int 1365 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1366 { 1367 struct sockaddr *sa; 1368 int error; 1369 1370 *namp = NULL; 1371 if (len > SOCK_MAXADDRLEN) 1372 return ENAMETOOLONG; 1373 if (len < offsetof(struct sockaddr, sa_data[0])) 1374 return EDOM; 1375 sa = kmalloc(len, M_SONAME, M_WAITOK); 1376 error = copyin(uaddr, sa, len); 1377 if (error) { 1378 kfree(sa, M_SONAME); 1379 } else { 1380 #if BYTE_ORDER != BIG_ENDIAN 1381 /* 1382 * The bind(), connect(), and sendto() syscalls were not 1383 * versioned for COMPAT_43. Thus, this check must stay. 1384 */ 1385 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1386 sa->sa_family = sa->sa_len; 1387 #endif 1388 sa->sa_len = len; 1389 *namp = sa; 1390 } 1391 return error; 1392 } 1393 1394 /* 1395 * Detach a mapped page and release resources back to the system. 1396 * We must release our wiring and if the object is ripped out 1397 * from under the vm_page we become responsible for freeing the 1398 * page. 1399 * 1400 * MPSAFE 1401 */ 1402 static void 1403 sf_buf_mfree(void *arg) 1404 { 1405 struct sf_buf *sf = arg; 1406 vm_page_t m; 1407 1408 m = sf_buf_page(sf); 1409 if (sf_buf_free(sf)) { 1410 /* sf invalid now */ 1411 vm_page_busy_wait(m, FALSE, "sockpgf"); 1412 vm_page_unwire(m, 0); 1413 if (m->object == NULL && 1414 m->wire_count == 0 && 1415 (m->flags & PG_NEED_COMMIT) == 0) { 1416 vm_page_free(m); 1417 } else { 1418 vm_page_wakeup(m); 1419 } 1420 } 1421 } 1422 1423 /* 1424 * sendfile(2). 1425 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1426 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1427 * 1428 * Send a file specified by 'fd' and starting at 'offset' to a socket 1429 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1430 * nbytes == 0. Optionally add a header and/or trailer to the socket 1431 * output. If specified, write the total number of bytes sent into *sbytes. 1432 * 1433 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1434 * the headers to count against the remaining bytes to be sent from 1435 * the file descriptor. We may wish to implement a compatibility syscall 1436 * in the future. 1437 * 1438 * MPALMOSTSAFE 1439 */ 1440 int 1441 sys_sendfile(struct sendfile_args *uap) 1442 { 1443 struct thread *td = curthread; 1444 struct proc *p = td->td_proc; 1445 struct file *fp; 1446 struct vnode *vp = NULL; 1447 struct sf_hdtr hdtr; 1448 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1449 struct uio auio; 1450 struct mbuf *mheader = NULL; 1451 size_t hbytes = 0; 1452 size_t tbytes; 1453 off_t hdtr_size = 0; 1454 off_t sbytes; 1455 int error; 1456 1457 KKASSERT(p); 1458 1459 /* 1460 * Do argument checking. Must be a regular file in, stream 1461 * type and connected socket out, positive offset. 1462 */ 1463 fp = holdfp(p->p_fd, uap->fd, FREAD); 1464 if (fp == NULL) { 1465 return (EBADF); 1466 } 1467 if (fp->f_type != DTYPE_VNODE) { 1468 fdrop(fp); 1469 return (EINVAL); 1470 } 1471 vp = (struct vnode *)fp->f_data; 1472 vref(vp); 1473 fdrop(fp); 1474 1475 /* 1476 * If specified, get the pointer to the sf_hdtr struct for 1477 * any headers/trailers. 1478 */ 1479 if (uap->hdtr) { 1480 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1481 if (error) 1482 goto done; 1483 /* 1484 * Send any headers. 1485 */ 1486 if (hdtr.headers) { 1487 error = iovec_copyin(hdtr.headers, &iov, aiov, 1488 hdtr.hdr_cnt, &hbytes); 1489 if (error) 1490 goto done; 1491 auio.uio_iov = iov; 1492 auio.uio_iovcnt = hdtr.hdr_cnt; 1493 auio.uio_offset = 0; 1494 auio.uio_segflg = UIO_USERSPACE; 1495 auio.uio_rw = UIO_WRITE; 1496 auio.uio_td = td; 1497 auio.uio_resid = hbytes; 1498 1499 mheader = m_uiomove(&auio); 1500 1501 iovec_free(&iov, aiov); 1502 if (mheader == NULL) 1503 goto done; 1504 } 1505 } 1506 1507 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1508 &sbytes, uap->flags); 1509 if (error) 1510 goto done; 1511 1512 /* 1513 * Send trailers. Wimp out and use writev(2). 1514 */ 1515 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1516 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1517 hdtr.trl_cnt, &auio.uio_resid); 1518 if (error) 1519 goto done; 1520 auio.uio_iov = iov; 1521 auio.uio_iovcnt = hdtr.trl_cnt; 1522 auio.uio_offset = 0; 1523 auio.uio_segflg = UIO_USERSPACE; 1524 auio.uio_rw = UIO_WRITE; 1525 auio.uio_td = td; 1526 1527 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1528 1529 iovec_free(&iov, aiov); 1530 if (error) 1531 goto done; 1532 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1533 } 1534 1535 done: 1536 if (vp) 1537 vrele(vp); 1538 if (uap->sbytes != NULL) { 1539 sbytes += hdtr_size; 1540 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1541 } 1542 return (error); 1543 } 1544 1545 int 1546 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1547 struct mbuf *mheader, off_t *sbytes, int flags) 1548 { 1549 struct thread *td = curthread; 1550 struct proc *p = td->td_proc; 1551 struct vm_object *obj; 1552 struct socket *so; 1553 struct file *fp; 1554 struct mbuf *m, *mp; 1555 struct sf_buf *sf; 1556 struct vm_page *pg; 1557 off_t off, xfsize; 1558 off_t hbytes = 0; 1559 int error = 0; 1560 1561 if (vp->v_type != VREG) { 1562 error = EINVAL; 1563 goto done0; 1564 } 1565 if ((obj = vp->v_object) == NULL) { 1566 error = EINVAL; 1567 goto done0; 1568 } 1569 error = holdsock(p->p_fd, sfd, &fp); 1570 if (error) 1571 goto done0; 1572 so = (struct socket *)fp->f_data; 1573 if (so->so_type != SOCK_STREAM) { 1574 error = EINVAL; 1575 goto done; 1576 } 1577 if ((so->so_state & SS_ISCONNECTED) == 0) { 1578 error = ENOTCONN; 1579 goto done; 1580 } 1581 if (offset < 0) { 1582 error = EINVAL; 1583 goto done; 1584 } 1585 1586 *sbytes = 0; 1587 /* 1588 * Protect against multiple writers to the socket. 1589 */ 1590 ssb_lock(&so->so_snd, M_WAITOK); 1591 1592 /* 1593 * Loop through the pages in the file, starting with the requested 1594 * offset. Get a file page (do I/O if necessary), map the file page 1595 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1596 * it on the socket. 1597 */ 1598 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1599 vm_pindex_t pindex; 1600 vm_offset_t pgoff; 1601 int space; 1602 1603 pindex = OFF_TO_IDX(off); 1604 retry_lookup: 1605 /* 1606 * Calculate the amount to transfer. Not to exceed a page, 1607 * the EOF, or the passed in nbytes. 1608 */ 1609 xfsize = vp->v_filesize - off; 1610 if (xfsize > PAGE_SIZE) 1611 xfsize = PAGE_SIZE; 1612 pgoff = (vm_offset_t)(off & PAGE_MASK); 1613 if (PAGE_SIZE - pgoff < xfsize) 1614 xfsize = PAGE_SIZE - pgoff; 1615 if (nbytes && xfsize > (nbytes - *sbytes)) 1616 xfsize = nbytes - *sbytes; 1617 if (xfsize <= 0) 1618 break; 1619 /* 1620 * Optimize the non-blocking case by looking at the socket space 1621 * before going to the extra work of constituting the sf_buf. 1622 */ 1623 if ((fp->f_flag & FNONBLOCK) && 1624 ssb_space_prealloc(&so->so_snd) <= 0) { 1625 if (so->so_state & SS_CANTSENDMORE) 1626 error = EPIPE; 1627 else 1628 error = EAGAIN; 1629 ssb_unlock(&so->so_snd); 1630 goto done; 1631 } 1632 /* 1633 * Attempt to look up the page. 1634 * 1635 * Allocate if not found, wait and loop if busy, then 1636 * wire the page. critical section protection is 1637 * required to maintain the object association (an 1638 * interrupt can free the page) through to the 1639 * vm_page_wire() call. 1640 */ 1641 vm_object_hold(obj); 1642 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error); 1643 if (error) { 1644 vm_page_sleep_busy(pg, TRUE, "sfpbsy"); 1645 vm_object_drop(obj); 1646 goto retry_lookup; 1647 } 1648 if (pg == NULL) { 1649 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL | 1650 VM_ALLOC_NULL_OK); 1651 if (pg == NULL) { 1652 vm_wait(0); 1653 vm_object_drop(obj); 1654 goto retry_lookup; 1655 } 1656 } 1657 vm_page_wire(pg); 1658 vm_object_drop(obj); 1659 1660 /* 1661 * If page is not valid for what we need, initiate I/O 1662 */ 1663 1664 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1665 struct uio auio; 1666 struct iovec aiov; 1667 int bsize; 1668 1669 /* 1670 * Ensure that our page is still around when the I/O 1671 * completes. 1672 * 1673 * Ensure that our page is not modified while part of 1674 * a mbuf as this could mess up tcp checksums, DMA, 1675 * etc (XXX NEEDS WORK). The softbusy is supposed to 1676 * help here but it actually doesn't. 1677 * 1678 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying 1679 * VM pages are not protected by the soft-busy 1680 * unless we vm_page_protect... READ them, and 1681 * they STILL aren't protected against 1682 * modification via the buffer cache (VOP_WRITE). 1683 * 1684 * Fixing the second issue is particularly 1685 * difficult. 1686 */ 1687 vm_page_io_start(pg); 1688 vm_page_wakeup(pg); 1689 1690 /* 1691 * Get the page from backing store. 1692 */ 1693 bsize = vp->v_mount->mnt_stat.f_iosize; 1694 auio.uio_iov = &aiov; 1695 auio.uio_iovcnt = 1; 1696 aiov.iov_base = 0; 1697 aiov.iov_len = MAXBSIZE; 1698 auio.uio_resid = MAXBSIZE; 1699 auio.uio_offset = trunc_page(off); 1700 auio.uio_segflg = UIO_NOCOPY; 1701 auio.uio_rw = UIO_READ; 1702 auio.uio_td = td; 1703 vn_lock(vp, LK_SHARED | LK_RETRY); 1704 error = VOP_READ(vp, &auio, 1705 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1706 td->td_ucred); 1707 vn_unlock(vp); 1708 vm_page_flag_clear(pg, PG_ZERO); 1709 vm_page_busy_wait(pg, FALSE, "sockpg"); 1710 vm_page_io_finish(pg); 1711 if (error) { 1712 vm_page_unwire(pg, 0); 1713 vm_page_wakeup(pg); 1714 vm_page_try_to_free(pg); 1715 ssb_unlock(&so->so_snd); 1716 goto done; 1717 } 1718 } 1719 1720 1721 /* 1722 * Get a sendfile buf. We usually wait as long as necessary, 1723 * but this wait can be interrupted. 1724 */ 1725 if ((sf = sf_buf_alloc(pg)) == NULL) { 1726 vm_page_unwire(pg, 0); 1727 vm_page_wakeup(pg); 1728 vm_page_try_to_free(pg); 1729 ssb_unlock(&so->so_snd); 1730 error = EINTR; 1731 goto done; 1732 } 1733 vm_page_wakeup(pg); 1734 1735 /* 1736 * Get an mbuf header and set it up as having external storage. 1737 */ 1738 MGETHDR(m, MB_WAIT, MT_DATA); 1739 if (m == NULL) { 1740 error = ENOBUFS; 1741 sf_buf_free(sf); 1742 ssb_unlock(&so->so_snd); 1743 goto done; 1744 } 1745 1746 m->m_ext.ext_free = sf_buf_mfree; 1747 m->m_ext.ext_ref = sf_buf_ref; 1748 m->m_ext.ext_arg = sf; 1749 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1750 m->m_ext.ext_size = PAGE_SIZE; 1751 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1752 m->m_flags |= M_EXT; 1753 m->m_pkthdr.len = m->m_len = xfsize; 1754 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1755 1756 if (mheader != NULL) { 1757 hbytes = mheader->m_pkthdr.len; 1758 mheader->m_pkthdr.len += m->m_pkthdr.len; 1759 m_cat(mheader, m); 1760 m = mheader; 1761 mheader = NULL; 1762 } else 1763 hbytes = 0; 1764 1765 /* 1766 * Add the buffer to the socket buffer chain. 1767 */ 1768 crit_enter(); 1769 retry_space: 1770 /* 1771 * Make sure that the socket is still able to take more data. 1772 * CANTSENDMORE being true usually means that the connection 1773 * was closed. so_error is true when an error was sensed after 1774 * a previous send. 1775 * The state is checked after the page mapping and buffer 1776 * allocation above since those operations may block and make 1777 * any socket checks stale. From this point forward, nothing 1778 * blocks before the pru_send (or more accurately, any blocking 1779 * results in a loop back to here to re-check). 1780 */ 1781 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1782 if (so->so_state & SS_CANTSENDMORE) { 1783 error = EPIPE; 1784 } else { 1785 error = so->so_error; 1786 so->so_error = 0; 1787 } 1788 m_freem(m); 1789 ssb_unlock(&so->so_snd); 1790 crit_exit(); 1791 goto done; 1792 } 1793 /* 1794 * Wait for socket space to become available. We do this just 1795 * after checking the connection state above in order to avoid 1796 * a race condition with ssb_wait(). 1797 */ 1798 space = ssb_space_prealloc(&so->so_snd); 1799 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1800 if (fp->f_flag & FNONBLOCK) { 1801 m_freem(m); 1802 ssb_unlock(&so->so_snd); 1803 crit_exit(); 1804 error = EAGAIN; 1805 goto done; 1806 } 1807 error = ssb_wait(&so->so_snd); 1808 /* 1809 * An error from ssb_wait usually indicates that we've 1810 * been interrupted by a signal. If we've sent anything 1811 * then return bytes sent, otherwise return the error. 1812 */ 1813 if (error) { 1814 m_freem(m); 1815 ssb_unlock(&so->so_snd); 1816 crit_exit(); 1817 goto done; 1818 } 1819 goto retry_space; 1820 } 1821 1822 for (mp = m; mp != NULL; mp = mp->m_next) 1823 ssb_preallocstream(&so->so_snd, mp); 1824 if (use_sendfile_async) 1825 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1826 else 1827 error = so_pru_send(so, 0, m, NULL, NULL, td); 1828 1829 crit_exit(); 1830 if (error) { 1831 ssb_unlock(&so->so_snd); 1832 goto done; 1833 } 1834 } 1835 if (mheader != NULL) { 1836 *sbytes += mheader->m_pkthdr.len; 1837 1838 for (mp = mheader; mp != NULL; mp = mp->m_next) 1839 ssb_preallocstream(&so->so_snd, mp); 1840 if (use_sendfile_async) 1841 error = so_pru_senda(so, 0, mheader, NULL, NULL, td); 1842 else 1843 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1844 1845 mheader = NULL; 1846 } 1847 ssb_unlock(&so->so_snd); 1848 1849 done: 1850 fdrop(fp); 1851 done0: 1852 if (mheader != NULL) 1853 m_freem(mheader); 1854 return (error); 1855 } 1856 1857 /* 1858 * MPALMOSTSAFE 1859 */ 1860 int 1861 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1862 { 1863 #ifdef SCTP 1864 struct thread *td = curthread; 1865 struct filedesc *fdp = td->td_proc->p_fd; 1866 struct file *lfp = NULL; 1867 struct file *nfp = NULL; 1868 int error; 1869 struct socket *head, *so; 1870 caddr_t assoc_id; 1871 int fd; 1872 short fflag; /* type must match fp->f_flag */ 1873 1874 assoc_id = uap->name; 1875 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp); 1876 if (error) 1877 return (error); 1878 1879 crit_enter(); 1880 head = (struct socket *)lfp->f_data; 1881 error = sctp_can_peel_off(head, assoc_id); 1882 if (error) { 1883 crit_exit(); 1884 goto done; 1885 } 1886 /* 1887 * At this point we know we do have a assoc to pull 1888 * we proceed to get the fd setup. This may block 1889 * but that is ok. 1890 */ 1891 1892 fflag = lfp->f_flag; 1893 error = falloc(td->td_lwp, &nfp, &fd); 1894 if (error) { 1895 /* 1896 * Probably ran out of file descriptors. Put the 1897 * unaccepted connection back onto the queue and 1898 * do another wakeup so some other process might 1899 * have a chance at it. 1900 */ 1901 crit_exit(); 1902 goto done; 1903 } 1904 uap->sysmsg_iresult = fd; 1905 1906 so = sctp_get_peeloff(head, assoc_id, &error); 1907 if (so == NULL) { 1908 /* 1909 * Either someone else peeled it off OR 1910 * we can't get a socket. 1911 */ 1912 goto noconnection; 1913 } 1914 soreference(so); /* reference needed */ 1915 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */ 1916 so->so_head = NULL; 1917 if (head->so_sigio != NULL) 1918 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 1919 1920 nfp->f_type = DTYPE_SOCKET; 1921 nfp->f_flag = fflag; 1922 nfp->f_ops = &socketops; 1923 nfp->f_data = so; 1924 1925 noconnection: 1926 /* 1927 * Assign the file pointer to the reserved descriptor, or clear 1928 * the reserved descriptor if an error occured. 1929 */ 1930 if (error) 1931 fsetfd(fdp, NULL, fd); 1932 else 1933 fsetfd(fdp, nfp, fd); 1934 crit_exit(); 1935 /* 1936 * Release explicitly held references before returning. 1937 */ 1938 done: 1939 if (nfp != NULL) 1940 fdrop(nfp); 1941 fdrop(lfp); 1942 return (error); 1943 #else /* SCTP */ 1944 return(EOPNOTSUPP); 1945 #endif /* SCTP */ 1946 } 1947