1 /* 2 * Copyright (c) 1982, 1986, 1989, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * sendfile(2) and related extensions: 6 * Copyright (c) 1998, David Greenman. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 37 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 38 * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.72 2006/08/08 03:52:40 dillon Exp $ 39 */ 40 41 #include "opt_ktrace.h" 42 #include "opt_sctp.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/kernel.h> 47 #include <sys/sysproto.h> 48 #include <sys/malloc.h> 49 #include <sys/filedesc.h> 50 #include <sys/event.h> 51 #include <sys/proc.h> 52 #include <sys/fcntl.h> 53 #include <sys/file.h> 54 #include <sys/filio.h> 55 #include <sys/kern_syscall.h> 56 #include <sys/mbuf.h> 57 #include <sys/protosw.h> 58 #include <sys/sfbuf.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/socketops.h> 62 #include <sys/uio.h> 63 #include <sys/vnode.h> 64 #include <sys/lock.h> 65 #include <sys/mount.h> 66 #ifdef KTRACE 67 #include <sys/ktrace.h> 68 #endif 69 #include <vm/vm.h> 70 #include <vm/vm_object.h> 71 #include <vm/vm_page.h> 72 #include <vm/vm_pageout.h> 73 #include <vm/vm_kern.h> 74 #include <vm/vm_extern.h> 75 #include <sys/file2.h> 76 #include <sys/signalvar.h> 77 #include <sys/serialize.h> 78 79 #include <sys/thread2.h> 80 #include <sys/msgport2.h> 81 82 #ifdef SCTP 83 #include <netinet/sctp_peeloff.h> 84 #endif /* SCTP */ 85 86 struct sfbuf_mref { 87 struct sf_buf *sf; 88 int mref_count; 89 struct lwkt_serialize serializer; 90 }; 91 92 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures"); 93 94 /* 95 * System call interface to the socket abstraction. 96 */ 97 98 extern struct fileops socketops; 99 100 /* 101 * socket_args(int domain, int type, int protocol) 102 */ 103 int 104 kern_socket(int domain, int type, int protocol, int *res) 105 { 106 struct thread *td = curthread; 107 struct proc *p = td->td_proc; 108 struct socket *so; 109 struct file *fp; 110 int fd, error; 111 112 KKASSERT(p); 113 114 error = falloc(p, &fp, &fd); 115 if (error) 116 return (error); 117 error = socreate(domain, &so, type, protocol, td); 118 if (error) { 119 fsetfd(p, NULL, fd); 120 } else { 121 fp->f_type = DTYPE_SOCKET; 122 fp->f_flag = FREAD | FWRITE; 123 fp->f_ops = &socketops; 124 fp->f_data = so; 125 *res = fd; 126 fsetfd(p, fp, fd); 127 } 128 fdrop(fp); 129 return (error); 130 } 131 132 int 133 sys_socket(struct socket_args *uap) 134 { 135 int error; 136 137 error = kern_socket(uap->domain, uap->type, uap->protocol, 138 &uap->sysmsg_result); 139 140 return (error); 141 } 142 143 int 144 kern_bind(int s, struct sockaddr *sa) 145 { 146 struct thread *td = curthread; 147 struct proc *p = td->td_proc; 148 struct file *fp; 149 int error; 150 151 KKASSERT(p); 152 error = holdsock(p->p_fd, s, &fp); 153 if (error) 154 return (error); 155 error = sobind((struct socket *)fp->f_data, sa, td); 156 fdrop(fp); 157 return (error); 158 } 159 160 /* 161 * bind_args(int s, caddr_t name, int namelen) 162 */ 163 int 164 sys_bind(struct bind_args *uap) 165 { 166 struct sockaddr *sa; 167 int error; 168 169 error = getsockaddr(&sa, uap->name, uap->namelen); 170 if (error) 171 return (error); 172 error = kern_bind(uap->s, sa); 173 FREE(sa, M_SONAME); 174 175 return (error); 176 } 177 178 int 179 kern_listen(int s, int backlog) 180 { 181 struct thread *td = curthread; 182 struct proc *p = td->td_proc; 183 struct file *fp; 184 int error; 185 186 KKASSERT(p); 187 error = holdsock(p->p_fd, s, &fp); 188 if (error) 189 return (error); 190 error = solisten((struct socket *)fp->f_data, backlog, td); 191 fdrop(fp); 192 return(error); 193 } 194 195 /* 196 * listen_args(int s, int backlog) 197 */ 198 int 199 sys_listen(struct listen_args *uap) 200 { 201 int error; 202 203 error = kern_listen(uap->s, uap->backlog); 204 return (error); 205 } 206 207 /* 208 * Returns the accepted socket as well. 209 */ 210 static boolean_t 211 soaccept_predicate(struct netmsg *msg0) 212 { 213 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 214 struct socket *head = msg->nm_so; 215 216 if (head->so_error != 0) { 217 msg->nm_lmsg.ms_error = head->so_error; 218 return (TRUE); 219 } 220 if (!TAILQ_EMPTY(&head->so_comp)) { 221 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 222 msg->nm_so = TAILQ_FIRST(&head->so_comp); 223 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list); 224 head->so_qlen--; 225 226 msg->nm_lmsg.ms_error = 0; 227 return (TRUE); 228 } 229 if (head->so_state & SS_CANTRCVMORE) { 230 msg->nm_lmsg.ms_error = ECONNABORTED; 231 return (TRUE); 232 } 233 if (msg->nm_fflags & FNONBLOCK) { 234 msg->nm_lmsg.ms_error = EWOULDBLOCK; 235 return (TRUE); 236 } 237 238 return (FALSE); 239 } 240 241 /* 242 * The second argument to kern_accept() is a handle to a struct sockaddr. 243 * This allows kern_accept() to return a pointer to an allocated struct 244 * sockaddr which must be freed later with FREE(). The caller must 245 * initialize *name to NULL. 246 */ 247 int 248 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 249 { 250 struct thread *td = curthread; 251 struct proc *p = td->td_proc; 252 struct file *lfp = NULL; 253 struct file *nfp = NULL; 254 struct sockaddr *sa; 255 struct socket *head, *so; 256 struct netmsg_so_notify msg; 257 lwkt_port_t port; 258 int fd; 259 u_int fflag; /* type must match fp->f_flag */ 260 int error, tmp; 261 262 *res = -1; 263 if (name && namelen && *namelen < 0) 264 return (EINVAL); 265 266 error = holdsock(p->p_fd, s, &lfp); 267 if (error) 268 return (error); 269 270 error = falloc(p, &nfp, &fd); 271 if (error) { /* Probably ran out of file descriptors. */ 272 fdrop(lfp); 273 return (error); 274 } 275 head = (struct socket *)lfp->f_data; 276 if ((head->so_options & SO_ACCEPTCONN) == 0) { 277 error = EINVAL; 278 goto done; 279 } 280 281 if (fflags & O_FBLOCKING) 282 fflags |= lfp->f_flag & ~FNONBLOCK; 283 else if (fflags & O_FNONBLOCKING) 284 fflags |= lfp->f_flag | FNONBLOCK; 285 else 286 fflags = lfp->f_flag; 287 288 /* optimize for uniprocessor case later XXX JH */ 289 port = head->so_proto->pr_mport(head, NULL, PRU_PRED); 290 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport, 291 MSGF_PCATCH | MSGF_ABORTABLE, 292 lwkt_cmd_func(netmsg_so_notify), 293 lwkt_cmd_func(netmsg_so_notify_abort)); 294 msg.nm_predicate = soaccept_predicate; 295 msg.nm_fflags = fflags; 296 msg.nm_so = head; 297 msg.nm_etype = NM_REVENT; 298 error = lwkt_domsg(port, &msg.nm_lmsg); 299 if (error) 300 goto done; 301 302 /* 303 * At this point we have the connection that's ready to be accepted. 304 */ 305 so = msg.nm_so; 306 307 fflag = lfp->f_flag; 308 309 /* connection has been removed from the listen queue */ 310 KNOTE(&head->so_rcv.sb_sel.si_note, 0); 311 312 so->so_state &= ~SS_COMP; 313 so->so_head = NULL; 314 if (head->so_sigio != NULL) 315 fsetown(fgetown(head->so_sigio), &so->so_sigio); 316 317 nfp->f_type = DTYPE_SOCKET; 318 nfp->f_flag = fflag; 319 nfp->f_ops = &socketops; 320 nfp->f_data = so; 321 /* Sync socket nonblocking/async state with file flags */ 322 tmp = fflag & FNONBLOCK; 323 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred); 324 tmp = fflag & FASYNC; 325 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred); 326 327 sa = NULL; 328 error = soaccept(so, &sa); 329 330 /* 331 * Set the returned name and namelen as applicable. Set the returned 332 * namelen to 0 for older code which might ignore the return value 333 * from accept. 334 */ 335 if (error == 0) { 336 if (sa && name && namelen) { 337 if (*namelen > sa->sa_len) 338 *namelen = sa->sa_len; 339 *name = sa; 340 } else { 341 if (sa) 342 FREE(sa, M_SONAME); 343 } 344 } 345 346 done: 347 /* 348 * If an error occured clear the reserved descriptor, else associate 349 * nfp with it. 350 * 351 * Note that *res is normally ignored if an error is returned but 352 * a syscall message will still have access to the result code. 353 */ 354 if (error) { 355 fsetfd(p, NULL, fd); 356 } else { 357 *res = fd; 358 fsetfd(p, nfp, fd); 359 } 360 fdrop(nfp); 361 fdrop(lfp); 362 return (error); 363 } 364 365 /* 366 * accept(int s, caddr_t name, int *anamelen) 367 */ 368 int 369 sys_accept(struct accept_args *uap) 370 { 371 struct sockaddr *sa = NULL; 372 int sa_len; 373 int error; 374 375 if (uap->name) { 376 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 377 if (error) 378 return (error); 379 380 error = kern_accept(uap->s, 0, &sa, &sa_len, &uap->sysmsg_result); 381 382 if (error == 0) 383 error = copyout(sa, uap->name, sa_len); 384 if (error == 0) { 385 error = copyout(&sa_len, uap->anamelen, 386 sizeof(*uap->anamelen)); 387 } 388 if (sa) 389 FREE(sa, M_SONAME); 390 } else { 391 error = kern_accept(uap->s, 0, NULL, 0, &uap->sysmsg_result); 392 } 393 return (error); 394 } 395 396 /* 397 * __accept(int s, int fflags, caddr_t name, int *anamelen) 398 */ 399 int 400 sys___accept(struct __accept_args *uap) 401 { 402 struct sockaddr *sa = NULL; 403 int sa_len; 404 int error; 405 int fflags = uap->flags & O_FMASK; 406 407 if (uap->name) { 408 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 409 if (error) 410 return (error); 411 412 error = kern_accept(uap->s, fflags, &sa, &sa_len, &uap->sysmsg_result); 413 414 if (error == 0) 415 error = copyout(sa, uap->name, sa_len); 416 if (error == 0) { 417 error = copyout(&sa_len, uap->anamelen, 418 sizeof(*uap->anamelen)); 419 } 420 if (sa) 421 FREE(sa, M_SONAME); 422 } else { 423 error = kern_accept(uap->s, fflags, NULL, 0, &uap->sysmsg_result); 424 } 425 return (error); 426 } 427 428 429 /* 430 * Returns TRUE if predicate satisfied. 431 */ 432 static boolean_t 433 soconnected_predicate(struct netmsg *msg0) 434 { 435 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 436 struct socket *so = msg->nm_so; 437 438 /* check predicate */ 439 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 440 msg->nm_lmsg.ms_error = so->so_error; 441 return (TRUE); 442 } 443 444 return (FALSE); 445 } 446 447 int 448 kern_connect(int s, int fflags, struct sockaddr *sa) 449 { 450 struct thread *td = curthread; 451 struct proc *p = td->td_proc; 452 struct file *fp; 453 struct socket *so; 454 int error; 455 456 error = holdsock(p->p_fd, s, &fp); 457 if (error) 458 return (error); 459 so = (struct socket *)fp->f_data; 460 461 if (fflags & O_FBLOCKING) 462 /* fflags &= ~FNONBLOCK; */; 463 else if (fflags & O_FNONBLOCKING) 464 fflags |= FNONBLOCK; 465 else 466 fflags = fp->f_flag; 467 468 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 469 error = EALREADY; 470 goto done; 471 } 472 error = soconnect(so, sa, td); 473 if (error) 474 goto bad; 475 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 476 error = EINPROGRESS; 477 goto done; 478 } 479 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 480 struct netmsg_so_notify msg; 481 lwkt_port_t port; 482 483 port = so->so_proto->pr_mport(so, sa, PRU_PRED); 484 lwkt_initmsg(&msg.nm_lmsg, 485 &curthread->td_msgport, 486 MSGF_PCATCH | MSGF_ABORTABLE, 487 lwkt_cmd_func(netmsg_so_notify), 488 lwkt_cmd_func(netmsg_so_notify_abort)); 489 msg.nm_predicate = soconnected_predicate; 490 msg.nm_so = so; 491 msg.nm_etype = NM_REVENT; 492 error = lwkt_domsg(port, &msg.nm_lmsg); 493 } 494 if (error == 0) { 495 error = so->so_error; 496 so->so_error = 0; 497 } 498 bad: 499 so->so_state &= ~SS_ISCONNECTING; 500 if (error == ERESTART) 501 error = EINTR; 502 done: 503 fdrop(fp); 504 return (error); 505 } 506 507 /* 508 * connect_args(int s, caddr_t name, int namelen) 509 */ 510 int 511 sys_connect(struct connect_args *uap) 512 { 513 struct sockaddr *sa; 514 int error; 515 516 error = getsockaddr(&sa, uap->name, uap->namelen); 517 if (error) 518 return (error); 519 error = kern_connect(uap->s, 0, sa); 520 FREE(sa, M_SONAME); 521 522 return (error); 523 } 524 525 /* 526 * connect_args(int s, int fflags, caddr_t name, int namelen) 527 */ 528 int 529 sys___connect(struct __connect_args *uap) 530 { 531 struct sockaddr *sa; 532 int error; 533 int fflags = uap->flags & O_FMASK; 534 535 error = getsockaddr(&sa, uap->name, uap->namelen); 536 if (error) 537 return (error); 538 error = kern_connect(uap->s, fflags, sa); 539 FREE(sa, M_SONAME); 540 541 return (error); 542 } 543 544 int 545 kern_socketpair(int domain, int type, int protocol, int *sv) 546 { 547 struct thread *td = curthread; 548 struct proc *p = td->td_proc; 549 struct file *fp1, *fp2; 550 struct socket *so1, *so2; 551 int fd1, fd2, error; 552 553 KKASSERT(p); 554 error = socreate(domain, &so1, type, protocol, td); 555 if (error) 556 return (error); 557 error = socreate(domain, &so2, type, protocol, td); 558 if (error) 559 goto free1; 560 error = falloc(p, &fp1, &fd1); 561 if (error) 562 goto free2; 563 sv[0] = fd1; 564 fp1->f_data = so1; 565 error = falloc(p, &fp2, &fd2); 566 if (error) 567 goto free3; 568 fp2->f_data = so2; 569 sv[1] = fd2; 570 error = soconnect2(so1, so2); 571 if (error) 572 goto free4; 573 if (type == SOCK_DGRAM) { 574 /* 575 * Datagram socket connection is asymmetric. 576 */ 577 error = soconnect2(so2, so1); 578 if (error) 579 goto free4; 580 } 581 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 582 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 583 fp1->f_ops = fp2->f_ops = &socketops; 584 fsetfd(p, fp1, fd1); 585 fsetfd(p, fp2, fd2); 586 fdrop(fp1); 587 fdrop(fp2); 588 return (error); 589 free4: 590 fsetfd(p, NULL, fd2); 591 fdrop(fp2); 592 free3: 593 fsetfd(p, NULL, fd1); 594 fdrop(fp1); 595 free2: 596 (void)soclose(so2, 0); 597 free1: 598 (void)soclose(so1, 0); 599 return (error); 600 } 601 602 /* 603 * socketpair(int domain, int type, int protocol, int *rsv) 604 */ 605 int 606 sys_socketpair(struct socketpair_args *uap) 607 { 608 int error, sockv[2]; 609 610 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 611 612 if (error == 0) 613 error = copyout(sockv, uap->rsv, sizeof(sockv)); 614 return (error); 615 } 616 617 int 618 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 619 struct mbuf *control, int flags, int *res) 620 { 621 struct thread *td = curthread; 622 struct proc *p = td->td_proc; 623 struct file *fp; 624 int len, error; 625 struct socket *so; 626 #ifdef KTRACE 627 struct iovec *ktriov = NULL; 628 struct uio ktruio; 629 #endif 630 631 error = holdsock(p->p_fd, s, &fp); 632 if (error) 633 return (error); 634 if (auio->uio_resid < 0) { 635 error = EINVAL; 636 goto done; 637 } 638 #ifdef KTRACE 639 if (KTRPOINT(td, KTR_GENIO)) { 640 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 641 642 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 643 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 644 ktruio = *auio; 645 } 646 #endif 647 len = auio->uio_resid; 648 so = (struct socket *)fp->f_data; 649 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 650 if (fp->f_flag & FNONBLOCK) 651 flags |= MSG_FNONBLOCKING; 652 } 653 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 654 if (error) { 655 if (auio->uio_resid != len && (error == ERESTART || 656 error == EINTR || error == EWOULDBLOCK)) 657 error = 0; 658 if (error == EPIPE) 659 psignal(p, SIGPIPE); 660 } 661 #ifdef KTRACE 662 if (ktriov != NULL) { 663 if (error == 0) { 664 ktruio.uio_iov = ktriov; 665 ktruio.uio_resid = len - auio->uio_resid; 666 ktrgenio(p, s, UIO_WRITE, &ktruio, error); 667 } 668 FREE(ktriov, M_TEMP); 669 } 670 #endif 671 if (error == 0) 672 *res = len - auio->uio_resid; 673 done: 674 fdrop(fp); 675 return (error); 676 } 677 678 /* 679 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 680 */ 681 int 682 sys_sendto(struct sendto_args *uap) 683 { 684 struct thread *td = curthread; 685 struct uio auio; 686 struct iovec aiov; 687 struct sockaddr *sa = NULL; 688 int error; 689 690 if (uap->to) { 691 error = getsockaddr(&sa, uap->to, uap->tolen); 692 if (error) 693 return (error); 694 } 695 aiov.iov_base = uap->buf; 696 aiov.iov_len = uap->len; 697 auio.uio_iov = &aiov; 698 auio.uio_iovcnt = 1; 699 auio.uio_offset = 0; 700 auio.uio_resid = uap->len; 701 auio.uio_segflg = UIO_USERSPACE; 702 auio.uio_rw = UIO_WRITE; 703 auio.uio_td = td; 704 705 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 706 &uap->sysmsg_result); 707 708 if (sa) 709 FREE(sa, M_SONAME); 710 return (error); 711 } 712 713 /* 714 * sendmsg_args(int s, caddr_t msg, int flags) 715 */ 716 int 717 sys_sendmsg(struct sendmsg_args *uap) 718 { 719 struct thread *td = curthread; 720 struct msghdr msg; 721 struct uio auio; 722 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 723 struct sockaddr *sa = NULL; 724 struct mbuf *control = NULL; 725 int error; 726 727 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 728 if (error) 729 return (error); 730 731 /* 732 * Conditionally copyin msg.msg_name. 733 */ 734 if (msg.msg_name) { 735 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 736 if (error) 737 return (error); 738 } 739 740 /* 741 * Populate auio. 742 */ 743 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 744 &auio.uio_resid); 745 if (error) 746 goto cleanup; 747 auio.uio_iov = iov; 748 auio.uio_iovcnt = msg.msg_iovlen; 749 auio.uio_offset = 0; 750 auio.uio_segflg = UIO_USERSPACE; 751 auio.uio_rw = UIO_WRITE; 752 auio.uio_td = td; 753 754 /* 755 * Conditionally copyin msg.msg_control. 756 */ 757 if (msg.msg_control) { 758 if (msg.msg_controllen < sizeof(struct cmsghdr) || 759 msg.msg_controllen > MLEN) { 760 error = EINVAL; 761 goto cleanup; 762 } 763 control = m_get(MB_WAIT, MT_CONTROL); 764 if (control == NULL) { 765 error = ENOBUFS; 766 goto cleanup; 767 } 768 control->m_len = msg.msg_controllen; 769 error = copyin(msg.msg_control, mtod(control, caddr_t), 770 msg.msg_controllen); 771 if (error) { 772 m_free(control); 773 goto cleanup; 774 } 775 } 776 777 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 778 &uap->sysmsg_result); 779 780 cleanup: 781 if (sa) 782 FREE(sa, M_SONAME); 783 iovec_free(&iov, aiov); 784 return (error); 785 } 786 787 /* 788 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 789 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 790 * Don't forget to FREE() and m_free() these if they are returned. 791 */ 792 int 793 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 794 struct mbuf **control, int *flags, int *res) 795 { 796 struct thread *td = curthread; 797 struct proc *p = td->td_proc; 798 struct file *fp; 799 int len, error; 800 int lflags; 801 struct socket *so; 802 #ifdef KTRACE 803 struct iovec *ktriov = NULL; 804 struct uio ktruio; 805 #endif 806 807 error = holdsock(p->p_fd, s, &fp); 808 if (error) 809 return (error); 810 if (auio->uio_resid < 0) { 811 error = EINVAL; 812 goto done; 813 } 814 #ifdef KTRACE 815 if (KTRPOINT(td, KTR_GENIO)) { 816 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 817 818 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 819 bcopy(auio->uio_iov, ktriov, iovlen); 820 ktruio = *auio; 821 } 822 #endif 823 len = auio->uio_resid; 824 so = (struct socket *)fp->f_data; 825 826 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 827 if (fp->f_flag & FNONBLOCK) { 828 if (flags) { 829 *flags |= MSG_FNONBLOCKING; 830 } else { 831 lflags = MSG_FNONBLOCKING; 832 flags = &lflags; 833 } 834 } 835 } 836 837 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 838 if (error) { 839 if (auio->uio_resid != len && (error == ERESTART || 840 error == EINTR || error == EWOULDBLOCK)) 841 error = 0; 842 } 843 #ifdef KTRACE 844 if (ktriov != NULL) { 845 if (error == 0) { 846 ktruio.uio_iov = ktriov; 847 ktruio.uio_resid = len - auio->uio_resid; 848 ktrgenio(p, s, UIO_READ, &ktruio, error); 849 } 850 FREE(ktriov, M_TEMP); 851 } 852 #endif 853 if (error == 0) 854 *res = len - auio->uio_resid; 855 done: 856 fdrop(fp); 857 return (error); 858 } 859 860 /* 861 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 862 * caddr_t from, int *fromlenaddr) 863 */ 864 int 865 sys_recvfrom(struct recvfrom_args *uap) 866 { 867 struct thread *td = curthread; 868 struct uio auio; 869 struct iovec aiov; 870 struct sockaddr *sa = NULL; 871 int error, fromlen; 872 873 if (uap->from && uap->fromlenaddr) { 874 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 875 if (error) 876 return (error); 877 if (fromlen < 0) 878 return (EINVAL); 879 } else { 880 fromlen = 0; 881 } 882 aiov.iov_base = uap->buf; 883 aiov.iov_len = uap->len; 884 auio.uio_iov = &aiov; 885 auio.uio_iovcnt = 1; 886 auio.uio_offset = 0; 887 auio.uio_resid = uap->len; 888 auio.uio_segflg = UIO_USERSPACE; 889 auio.uio_rw = UIO_READ; 890 auio.uio_td = td; 891 892 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 893 &uap->flags, &uap->sysmsg_result); 894 895 if (error == 0 && uap->from) { 896 /* note: sa may still be NULL */ 897 if (sa) { 898 fromlen = MIN(fromlen, sa->sa_len); 899 error = copyout(sa, uap->from, fromlen); 900 } else { 901 fromlen = 0; 902 } 903 if (error == 0) { 904 error = copyout(&fromlen, uap->fromlenaddr, 905 sizeof(fromlen)); 906 } 907 } 908 if (sa) 909 FREE(sa, M_SONAME); 910 911 return (error); 912 } 913 914 /* 915 * recvmsg_args(int s, struct msghdr *msg, int flags) 916 */ 917 int 918 sys_recvmsg(struct recvmsg_args *uap) 919 { 920 struct thread *td = curthread; 921 struct msghdr msg; 922 struct uio auio; 923 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 924 struct mbuf *m, *control = NULL; 925 struct sockaddr *sa = NULL; 926 caddr_t ctlbuf; 927 socklen_t *ufromlenp, *ucontrollenp; 928 int error, fromlen, controllen, len, flags, *uflagsp; 929 930 /* 931 * This copyin handles everything except the iovec. 932 */ 933 error = copyin(uap->msg, &msg, sizeof(msg)); 934 if (error) 935 return (error); 936 937 if (msg.msg_name && msg.msg_namelen < 0) 938 return (EINVAL); 939 if (msg.msg_control && msg.msg_controllen < 0) 940 return (EINVAL); 941 942 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 943 msg_namelen)); 944 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 945 msg_controllen)); 946 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 947 msg_flags)); 948 949 /* 950 * Populate auio. 951 */ 952 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 953 &auio.uio_resid); 954 if (error) 955 return (error); 956 auio.uio_iov = iov; 957 auio.uio_iovcnt = msg.msg_iovlen; 958 auio.uio_offset = 0; 959 auio.uio_segflg = UIO_USERSPACE; 960 auio.uio_rw = UIO_READ; 961 auio.uio_td = td; 962 963 flags = uap->flags; 964 965 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio, 966 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result); 967 968 /* 969 * Conditionally copyout the name and populate the namelen field. 970 */ 971 if (error == 0 && msg.msg_name) { 972 fromlen = MIN(msg.msg_namelen, sa->sa_len); 973 error = copyout(sa, msg.msg_name, fromlen); 974 if (error == 0) 975 error = copyout(&fromlen, ufromlenp, 976 sizeof(*ufromlenp)); 977 } 978 979 /* 980 * Copyout msg.msg_control and msg.msg_controllen. 981 */ 982 if (error == 0 && msg.msg_control) { 983 len = msg.msg_controllen; 984 m = control; 985 ctlbuf = (caddr_t)msg.msg_control; 986 987 while(m && len > 0) { 988 unsigned int tocopy; 989 990 if (len >= m->m_len) { 991 tocopy = m->m_len; 992 } else { 993 msg.msg_flags |= MSG_CTRUNC; 994 tocopy = len; 995 } 996 997 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 998 if (error) 999 goto cleanup; 1000 1001 ctlbuf += tocopy; 1002 len -= tocopy; 1003 m = m->m_next; 1004 } 1005 controllen = ctlbuf - (caddr_t)msg.msg_control; 1006 error = copyout(&controllen, ucontrollenp, 1007 sizeof(*ucontrollenp)); 1008 } 1009 1010 if (error == 0) 1011 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1012 1013 cleanup: 1014 if (sa) 1015 FREE(sa, M_SONAME); 1016 iovec_free(&iov, aiov); 1017 if (control) 1018 m_freem(control); 1019 return (error); 1020 } 1021 1022 /* 1023 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1024 * in kernel pointer instead of a userland pointer. This allows us 1025 * to manipulate socket options in the emulation code. 1026 */ 1027 int 1028 kern_setsockopt(int s, struct sockopt *sopt) 1029 { 1030 struct thread *td = curthread; 1031 struct proc *p = td->td_proc; 1032 struct file *fp; 1033 int error; 1034 1035 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1036 return (EFAULT); 1037 if (sopt->sopt_valsize < 0) 1038 return (EINVAL); 1039 1040 error = holdsock(p->p_fd, s, &fp); 1041 if (error) 1042 return (error); 1043 1044 error = sosetopt((struct socket *)fp->f_data, sopt); 1045 fdrop(fp); 1046 return (error); 1047 } 1048 1049 /* 1050 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1051 */ 1052 int 1053 sys_setsockopt(struct setsockopt_args *uap) 1054 { 1055 struct thread *td = curthread; 1056 struct sockopt sopt; 1057 int error; 1058 1059 sopt.sopt_level = uap->level; 1060 sopt.sopt_name = uap->name; 1061 sopt.sopt_val = uap->val; 1062 sopt.sopt_valsize = uap->valsize; 1063 sopt.sopt_td = td; 1064 1065 error = kern_setsockopt(uap->s, &sopt); 1066 return(error); 1067 } 1068 1069 /* 1070 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1071 * in kernel pointer instead of a userland pointer. This allows us 1072 * to manipulate socket options in the emulation code. 1073 */ 1074 int 1075 kern_getsockopt(int s, struct sockopt *sopt) 1076 { 1077 struct thread *td = curthread; 1078 struct proc *p = td->td_proc; 1079 struct file *fp; 1080 int error; 1081 1082 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1083 return (EFAULT); 1084 if (sopt->sopt_valsize < 0) 1085 return (EINVAL); 1086 1087 error = holdsock(p->p_fd, s, &fp); 1088 if (error) 1089 return (error); 1090 1091 error = sogetopt((struct socket *)fp->f_data, sopt); 1092 fdrop(fp); 1093 return (error); 1094 } 1095 1096 /* 1097 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1098 */ 1099 int 1100 sys_getsockopt(struct getsockopt_args *uap) 1101 { 1102 struct thread *td = curthread; 1103 struct sockopt sopt; 1104 int error, valsize; 1105 1106 if (uap->val) { 1107 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1108 if (error) 1109 return (error); 1110 if (valsize < 0) 1111 return (EINVAL); 1112 } else { 1113 valsize = 0; 1114 } 1115 1116 sopt.sopt_level = uap->level; 1117 sopt.sopt_name = uap->name; 1118 sopt.sopt_val = uap->val; 1119 sopt.sopt_valsize = valsize; 1120 sopt.sopt_td = td; 1121 1122 error = kern_getsockopt(uap->s, &sopt); 1123 if (error == 0) { 1124 valsize = sopt.sopt_valsize; 1125 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1126 } 1127 return (error); 1128 } 1129 1130 /* 1131 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1132 * This allows kern_getsockname() to return a pointer to an allocated struct 1133 * sockaddr which must be freed later with FREE(). The caller must 1134 * initialize *name to NULL. 1135 */ 1136 int 1137 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1138 { 1139 struct thread *td = curthread; 1140 struct proc *p = td->td_proc; 1141 struct file *fp; 1142 struct socket *so; 1143 struct sockaddr *sa = NULL; 1144 int error; 1145 1146 error = holdsock(p->p_fd, s, &fp); 1147 if (error) 1148 return (error); 1149 if (*namelen < 0) { 1150 fdrop(fp); 1151 return (EINVAL); 1152 } 1153 so = (struct socket *)fp->f_data; 1154 error = so_pru_sockaddr(so, &sa); 1155 if (error == 0) { 1156 if (sa == 0) { 1157 *namelen = 0; 1158 } else { 1159 *namelen = MIN(*namelen, sa->sa_len); 1160 *name = sa; 1161 } 1162 } 1163 1164 fdrop(fp); 1165 return (error); 1166 } 1167 1168 /* 1169 * getsockname_args(int fdes, caddr_t asa, int *alen) 1170 * 1171 * Get socket name. 1172 */ 1173 int 1174 sys_getsockname(struct getsockname_args *uap) 1175 { 1176 struct sockaddr *sa = NULL; 1177 int error, sa_len; 1178 1179 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1180 if (error) 1181 return (error); 1182 1183 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1184 1185 if (error == 0) 1186 error = copyout(sa, uap->asa, sa_len); 1187 if (error == 0) 1188 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1189 if (sa) 1190 FREE(sa, M_SONAME); 1191 return (error); 1192 } 1193 1194 /* 1195 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1196 * This allows kern_getpeername() to return a pointer to an allocated struct 1197 * sockaddr which must be freed later with FREE(). The caller must 1198 * initialize *name to NULL. 1199 */ 1200 int 1201 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1202 { 1203 struct thread *td = curthread; 1204 struct proc *p = td->td_proc; 1205 struct file *fp; 1206 struct socket *so; 1207 struct sockaddr *sa = NULL; 1208 int error; 1209 1210 error = holdsock(p->p_fd, s, &fp); 1211 if (error) 1212 return (error); 1213 if (*namelen < 0) { 1214 fdrop(fp); 1215 return (EINVAL); 1216 } 1217 so = (struct socket *)fp->f_data; 1218 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1219 fdrop(fp); 1220 return (ENOTCONN); 1221 } 1222 error = so_pru_peeraddr(so, &sa); 1223 if (error == 0) { 1224 if (sa == 0) { 1225 *namelen = 0; 1226 } else { 1227 *namelen = MIN(*namelen, sa->sa_len); 1228 *name = sa; 1229 } 1230 } 1231 1232 fdrop(fp); 1233 return (error); 1234 } 1235 1236 /* 1237 * getpeername_args(int fdes, caddr_t asa, int *alen) 1238 * 1239 * Get name of peer for connected socket. 1240 */ 1241 int 1242 sys_getpeername(struct getpeername_args *uap) 1243 { 1244 struct sockaddr *sa = NULL; 1245 int error, sa_len; 1246 1247 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1248 if (error) 1249 return (error); 1250 1251 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1252 1253 if (error == 0) 1254 error = copyout(sa, uap->asa, sa_len); 1255 if (error == 0) 1256 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1257 if (sa) 1258 FREE(sa, M_SONAME); 1259 return (error); 1260 } 1261 1262 int 1263 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1264 { 1265 struct sockaddr *sa; 1266 int error; 1267 1268 *namp = NULL; 1269 if (len > SOCK_MAXADDRLEN) 1270 return ENAMETOOLONG; 1271 if (len < offsetof(struct sockaddr, sa_data[0])) 1272 return EDOM; 1273 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1274 error = copyin(uaddr, sa, len); 1275 if (error) { 1276 FREE(sa, M_SONAME); 1277 } else { 1278 #if BYTE_ORDER != BIG_ENDIAN 1279 /* 1280 * The bind(), connect(), and sendto() syscalls were not 1281 * versioned for COMPAT_43. Thus, this check must stay. 1282 */ 1283 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1284 sa->sa_family = sa->sa_len; 1285 #endif 1286 sa->sa_len = len; 1287 *namp = sa; 1288 } 1289 return error; 1290 } 1291 1292 /* 1293 * Detach a mapped page and release resources back to the system. 1294 * We must release our wiring and if the object is ripped out 1295 * from under the vm_page we become responsible for freeing the 1296 * page. These routines must be MPSAFE. 1297 * 1298 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1299 * 1300 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required. 1301 */ 1302 static void 1303 sf_buf_mref(void *arg) 1304 { 1305 struct sfbuf_mref *sfm = arg; 1306 1307 /* 1308 * We must already hold a ref so there is no race to 0, just 1309 * atomically increment the count. 1310 */ 1311 atomic_add_int(&sfm->mref_count, 1); 1312 } 1313 1314 static void 1315 sf_buf_mfree(void *arg) 1316 { 1317 struct sfbuf_mref *sfm = arg; 1318 vm_page_t m; 1319 1320 KKASSERT(sfm->mref_count > 0); 1321 if (sfm->mref_count == 1) { 1322 /* 1323 * We are the only holder so no further locking is required, 1324 * the sfbuf can simply be freed. 1325 */ 1326 sfm->mref_count = 0; 1327 goto freeit; 1328 } else { 1329 /* 1330 * There may be other holders, we must obtain the serializer 1331 * to protect against a sf_buf_mfree() race to 0. An atomic 1332 * operation is still required for races against 1333 * sf_buf_mref(). 1334 * 1335 * XXX vm_page_*() and SFBUF routines not MPSAFE yet. 1336 */ 1337 lwkt_serialize_enter(&sfm->serializer); 1338 atomic_subtract_int(&sfm->mref_count, 1); 1339 if (sfm->mref_count == 0) { 1340 lwkt_serialize_exit(&sfm->serializer); 1341 freeit: 1342 get_mplock(); 1343 crit_enter(); 1344 m = sf_buf_page(sfm->sf); 1345 sf_buf_free(sfm->sf); 1346 vm_page_unwire(m, 0); 1347 if (m->wire_count == 0 && m->object == NULL) 1348 vm_page_try_to_free(m); 1349 crit_exit(); 1350 rel_mplock(); 1351 free(sfm, M_SENDFILE); 1352 } else { 1353 lwkt_serialize_exit(&sfm->serializer); 1354 } 1355 } 1356 } 1357 1358 /* 1359 * sendfile(2). 1360 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1361 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1362 * 1363 * Send a file specified by 'fd' and starting at 'offset' to a socket 1364 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1365 * nbytes == 0. Optionally add a header and/or trailer to the socket 1366 * output. If specified, write the total number of bytes sent into *sbytes. 1367 * 1368 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1369 * the headers to count against the remaining bytes to be sent from 1370 * the file descriptor. We may wish to implement a compatibility syscall 1371 * in the future. 1372 */ 1373 int 1374 sys_sendfile(struct sendfile_args *uap) 1375 { 1376 struct thread *td = curthread; 1377 struct proc *p = td->td_proc; 1378 struct file *fp; 1379 struct vnode *vp = NULL; 1380 struct sf_hdtr hdtr; 1381 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1382 struct uio auio; 1383 struct mbuf *mheader = NULL; 1384 off_t hdtr_size = 0, sbytes; 1385 int error, hbytes = 0, tbytes; 1386 1387 KKASSERT(p); 1388 1389 /* 1390 * Do argument checking. Must be a regular file in, stream 1391 * type and connected socket out, positive offset. 1392 */ 1393 fp = holdfp(p->p_fd, uap->fd, FREAD); 1394 if (fp == NULL) { 1395 return (EBADF); 1396 } 1397 if (fp->f_type != DTYPE_VNODE) { 1398 fdrop(fp); 1399 return (EINVAL); 1400 } 1401 vp = (struct vnode *)fp->f_data; 1402 vref(vp); 1403 fdrop(fp); 1404 1405 /* 1406 * If specified, get the pointer to the sf_hdtr struct for 1407 * any headers/trailers. 1408 */ 1409 if (uap->hdtr) { 1410 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1411 if (error) 1412 goto done; 1413 /* 1414 * Send any headers. 1415 */ 1416 if (hdtr.headers) { 1417 error = iovec_copyin(hdtr.headers, &iov, aiov, 1418 hdtr.hdr_cnt, &hbytes); 1419 if (error) 1420 goto done; 1421 auio.uio_iov = iov; 1422 auio.uio_iovcnt = hdtr.hdr_cnt; 1423 auio.uio_offset = 0; 1424 auio.uio_segflg = UIO_USERSPACE; 1425 auio.uio_rw = UIO_WRITE; 1426 auio.uio_td = td; 1427 auio.uio_resid = hbytes; 1428 1429 mheader = m_uiomove(&auio); 1430 1431 iovec_free(&iov, aiov); 1432 if (mheader == NULL) 1433 goto done; 1434 } 1435 } 1436 1437 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1438 &sbytes, uap->flags); 1439 if (error) 1440 goto done; 1441 1442 /* 1443 * Send trailers. Wimp out and use writev(2). 1444 */ 1445 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1446 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1447 hdtr.trl_cnt, &auio.uio_resid); 1448 if (error) 1449 goto done; 1450 auio.uio_iov = iov; 1451 auio.uio_iovcnt = hdtr.trl_cnt; 1452 auio.uio_offset = 0; 1453 auio.uio_segflg = UIO_USERSPACE; 1454 auio.uio_rw = UIO_WRITE; 1455 auio.uio_td = td; 1456 1457 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1458 1459 iovec_free(&iov, aiov); 1460 if (error) 1461 goto done; 1462 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1463 } 1464 1465 done: 1466 if (uap->sbytes != NULL) { 1467 sbytes += hdtr_size; 1468 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1469 } 1470 if (vp) 1471 vrele(vp); 1472 return (error); 1473 } 1474 1475 int 1476 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1477 struct mbuf *mheader, off_t *sbytes, int flags) 1478 { 1479 struct thread *td = curthread; 1480 struct proc *p = td->td_proc; 1481 struct vm_object *obj; 1482 struct socket *so; 1483 struct file *fp; 1484 struct mbuf *m; 1485 struct sf_buf *sf; 1486 struct sfbuf_mref *sfm; 1487 struct vm_page *pg; 1488 off_t off, xfsize; 1489 off_t hbytes = 0; 1490 int error = 0; 1491 1492 if (vp->v_type != VREG) { 1493 error = EINVAL; 1494 goto done0; 1495 } 1496 if ((obj = vp->v_object) == NULL) { 1497 error = EINVAL; 1498 goto done0; 1499 } 1500 error = holdsock(p->p_fd, sfd, &fp); 1501 if (error) 1502 goto done0; 1503 so = (struct socket *)fp->f_data; 1504 if (so->so_type != SOCK_STREAM) { 1505 error = EINVAL; 1506 goto done; 1507 } 1508 if ((so->so_state & SS_ISCONNECTED) == 0) { 1509 error = ENOTCONN; 1510 goto done; 1511 } 1512 if (offset < 0) { 1513 error = EINVAL; 1514 goto done; 1515 } 1516 1517 *sbytes = 0; 1518 /* 1519 * Protect against multiple writers to the socket. 1520 */ 1521 (void) sblock(&so->so_snd, M_WAITOK); 1522 1523 /* 1524 * Loop through the pages in the file, starting with the requested 1525 * offset. Get a file page (do I/O if necessary), map the file page 1526 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1527 * it on the socket. 1528 */ 1529 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1530 vm_pindex_t pindex; 1531 vm_offset_t pgoff; 1532 1533 pindex = OFF_TO_IDX(off); 1534 retry_lookup: 1535 /* 1536 * Calculate the amount to transfer. Not to exceed a page, 1537 * the EOF, or the passed in nbytes. 1538 */ 1539 xfsize = vp->v_filesize - off; 1540 if (xfsize > PAGE_SIZE) 1541 xfsize = PAGE_SIZE; 1542 pgoff = (vm_offset_t)(off & PAGE_MASK); 1543 if (PAGE_SIZE - pgoff < xfsize) 1544 xfsize = PAGE_SIZE - pgoff; 1545 if (nbytes && xfsize > (nbytes - *sbytes)) 1546 xfsize = nbytes - *sbytes; 1547 if (xfsize <= 0) 1548 break; 1549 /* 1550 * Optimize the non-blocking case by looking at the socket space 1551 * before going to the extra work of constituting the sf_buf. 1552 */ 1553 if ((fp->f_flag & FNONBLOCK) && sbspace(&so->so_snd) <= 0) { 1554 if (so->so_state & SS_CANTSENDMORE) 1555 error = EPIPE; 1556 else 1557 error = EAGAIN; 1558 sbunlock(&so->so_snd); 1559 goto done; 1560 } 1561 /* 1562 * Attempt to look up the page. 1563 * 1564 * Allocate if not found, wait and loop if busy, then 1565 * wire the page. critical section protection is 1566 * required to maintain the object association (an 1567 * interrupt can free the page) through to the 1568 * vm_page_wire() call. 1569 */ 1570 crit_enter(); 1571 pg = vm_page_lookup(obj, pindex); 1572 if (pg == NULL) { 1573 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1574 if (pg == NULL) { 1575 vm_wait(); 1576 crit_exit(); 1577 goto retry_lookup; 1578 } 1579 vm_page_wakeup(pg); 1580 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1581 crit_exit(); 1582 goto retry_lookup; 1583 } 1584 vm_page_wire(pg); 1585 crit_exit(); 1586 1587 /* 1588 * If page is not valid for what we need, initiate I/O 1589 */ 1590 1591 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1592 struct uio auio; 1593 struct iovec aiov; 1594 int bsize; 1595 1596 /* 1597 * Ensure that our page is still around when the I/O 1598 * completes. 1599 */ 1600 vm_page_io_start(pg); 1601 1602 /* 1603 * Get the page from backing store. 1604 */ 1605 bsize = vp->v_mount->mnt_stat.f_iosize; 1606 auio.uio_iov = &aiov; 1607 auio.uio_iovcnt = 1; 1608 aiov.iov_base = 0; 1609 aiov.iov_len = MAXBSIZE; 1610 auio.uio_resid = MAXBSIZE; 1611 auio.uio_offset = trunc_page(off); 1612 auio.uio_segflg = UIO_NOCOPY; 1613 auio.uio_rw = UIO_READ; 1614 auio.uio_td = td; 1615 vn_lock(vp, LK_SHARED | LK_RETRY); 1616 error = VOP_READ(vp, &auio, 1617 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1618 p->p_ucred); 1619 VOP_UNLOCK(vp, 0); 1620 vm_page_flag_clear(pg, PG_ZERO); 1621 vm_page_io_finish(pg); 1622 if (error) { 1623 crit_enter(); 1624 vm_page_unwire(pg, 0); 1625 vm_page_try_to_free(pg); 1626 crit_exit(); 1627 sbunlock(&so->so_snd); 1628 goto done; 1629 } 1630 } 1631 1632 1633 /* 1634 * Get a sendfile buf. We usually wait as long as necessary, 1635 * but this wait can be interrupted. 1636 */ 1637 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) { 1638 crit_enter(); 1639 vm_page_unwire(pg, 0); 1640 vm_page_try_to_free(pg); 1641 crit_exit(); 1642 sbunlock(&so->so_snd); 1643 error = EINTR; 1644 goto done; 1645 } 1646 1647 /* 1648 * Get an mbuf header and set it up as having external storage. 1649 */ 1650 MGETHDR(m, MB_WAIT, MT_DATA); 1651 if (m == NULL) { 1652 error = ENOBUFS; 1653 sf_buf_free(sf); 1654 sbunlock(&so->so_snd); 1655 goto done; 1656 } 1657 1658 /* 1659 * sfm is a temporary hack, use a per-cpu cache for this. 1660 */ 1661 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK); 1662 sfm->sf = sf; 1663 sfm->mref_count = 1; 1664 lwkt_serialize_init(&sfm->serializer); 1665 1666 m->m_ext.ext_free = sf_buf_mfree; 1667 m->m_ext.ext_ref = sf_buf_mref; 1668 m->m_ext.ext_arg = sfm; 1669 m->m_ext.ext_buf = (void *)sf->kva; 1670 m->m_ext.ext_size = PAGE_SIZE; 1671 m->m_data = (char *) sf->kva + pgoff; 1672 m->m_flags |= M_EXT; 1673 m->m_pkthdr.len = m->m_len = xfsize; 1674 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1675 1676 if (mheader != NULL) { 1677 hbytes = mheader->m_pkthdr.len; 1678 mheader->m_pkthdr.len += m->m_pkthdr.len; 1679 m_cat(mheader, m); 1680 m = mheader; 1681 mheader = NULL; 1682 } else 1683 hbytes = 0; 1684 1685 /* 1686 * Add the buffer to the socket buffer chain. 1687 */ 1688 crit_enter(); 1689 retry_space: 1690 /* 1691 * Make sure that the socket is still able to take more data. 1692 * CANTSENDMORE being true usually means that the connection 1693 * was closed. so_error is true when an error was sensed after 1694 * a previous send. 1695 * The state is checked after the page mapping and buffer 1696 * allocation above since those operations may block and make 1697 * any socket checks stale. From this point forward, nothing 1698 * blocks before the pru_send (or more accurately, any blocking 1699 * results in a loop back to here to re-check). 1700 */ 1701 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1702 if (so->so_state & SS_CANTSENDMORE) { 1703 error = EPIPE; 1704 } else { 1705 error = so->so_error; 1706 so->so_error = 0; 1707 } 1708 m_freem(m); 1709 sbunlock(&so->so_snd); 1710 crit_exit(); 1711 goto done; 1712 } 1713 /* 1714 * Wait for socket space to become available. We do this just 1715 * after checking the connection state above in order to avoid 1716 * a race condition with sbwait(). 1717 */ 1718 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { 1719 if (fp->f_flag & FNONBLOCK) { 1720 m_freem(m); 1721 sbunlock(&so->so_snd); 1722 crit_exit(); 1723 error = EAGAIN; 1724 goto done; 1725 } 1726 error = sbwait(&so->so_snd); 1727 /* 1728 * An error from sbwait usually indicates that we've 1729 * been interrupted by a signal. If we've sent anything 1730 * then return bytes sent, otherwise return the error. 1731 */ 1732 if (error) { 1733 m_freem(m); 1734 sbunlock(&so->so_snd); 1735 crit_exit(); 1736 goto done; 1737 } 1738 goto retry_space; 1739 } 1740 error = so_pru_send(so, 0, m, NULL, NULL, td); 1741 crit_exit(); 1742 if (error) { 1743 sbunlock(&so->so_snd); 1744 goto done; 1745 } 1746 } 1747 if (mheader != NULL) { 1748 *sbytes += mheader->m_pkthdr.len; 1749 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1750 mheader = NULL; 1751 } 1752 sbunlock(&so->so_snd); 1753 1754 done: 1755 fdrop(fp); 1756 done0: 1757 if (mheader != NULL) 1758 m_freem(mheader); 1759 return (error); 1760 } 1761 1762 int 1763 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1764 { 1765 #ifdef SCTP 1766 struct thread *td = curthread; 1767 struct proc *p = td->td_proc; 1768 struct file *lfp = NULL; 1769 struct file *nfp = NULL; 1770 int error; 1771 struct socket *head, *so; 1772 caddr_t assoc_id; 1773 int fd; 1774 short fflag; /* type must match fp->f_flag */ 1775 1776 assoc_id = uap->name; 1777 error = holdsock(p->p_fd, uap->sd, &lfp); 1778 if (error) { 1779 return (error); 1780 } 1781 crit_enter(); 1782 head = (struct socket *)lfp->f_data; 1783 error = sctp_can_peel_off(head, assoc_id); 1784 if (error) { 1785 crit_exit(); 1786 goto done; 1787 } 1788 /* 1789 * At this point we know we do have a assoc to pull 1790 * we proceed to get the fd setup. This may block 1791 * but that is ok. 1792 */ 1793 1794 fflag = lfp->f_flag; 1795 error = falloc(p, &nfp, &fd); 1796 if (error) { 1797 /* 1798 * Probably ran out of file descriptors. Put the 1799 * unaccepted connection back onto the queue and 1800 * do another wakeup so some other process might 1801 * have a chance at it. 1802 */ 1803 crit_exit(); 1804 goto done; 1805 } 1806 uap->sysmsg_result = fd; 1807 1808 so = sctp_get_peeloff(head, assoc_id, &error); 1809 if (so == NULL) { 1810 /* 1811 * Either someone else peeled it off OR 1812 * we can't get a socket. 1813 */ 1814 goto noconnection; 1815 } 1816 so->so_state &= ~SS_COMP; 1817 so->so_state &= ~SS_NOFDREF; 1818 so->so_head = NULL; 1819 if (head->so_sigio != NULL) 1820 fsetown(fgetown(head->so_sigio), &so->so_sigio); 1821 1822 nfp->f_type = DTYPE_SOCKET; 1823 nfp->f_flag = fflag; 1824 nfp->f_ops = &socketops; 1825 nfp->f_data = so; 1826 1827 noconnection: 1828 /* 1829 * Assign the file pointer to the reserved descriptor, or clear 1830 * the reserved descriptor if an error occured. 1831 */ 1832 if (error) 1833 fsetfd(p, NULL, fd); 1834 else 1835 fsetfd(p, nfp, fd); 1836 crit_exit(); 1837 /* 1838 * Release explicitly held references before returning. 1839 */ 1840 done: 1841 if (nfp != NULL) 1842 fdrop(nfp); 1843 fdrop(lfp); 1844 return (error); 1845 #else /* SCTP */ 1846 return(EOPNOTSUPP); 1847 #endif /* SCTP */ 1848 } 1849