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 1.80 2007/04/22 01:13:10 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.ssb_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 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 398 */ 399 int 400 sys_extaccept(struct extaccept_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_extconnect(struct extconnect_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 lwp *lp = td->td_lwp; 623 struct proc *p = td->td_proc; 624 struct file *fp; 625 int len, error; 626 struct socket *so; 627 #ifdef KTRACE 628 struct iovec *ktriov = NULL; 629 struct uio ktruio; 630 #endif 631 632 error = holdsock(p->p_fd, s, &fp); 633 if (error) 634 return (error); 635 if (auio->uio_resid < 0) { 636 error = EINVAL; 637 goto done; 638 } 639 #ifdef KTRACE 640 if (KTRPOINT(td, KTR_GENIO)) { 641 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 642 643 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 644 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 645 ktruio = *auio; 646 } 647 #endif 648 len = auio->uio_resid; 649 so = (struct socket *)fp->f_data; 650 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 651 if (fp->f_flag & FNONBLOCK) 652 flags |= MSG_FNONBLOCKING; 653 } 654 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 655 if (error) { 656 if (auio->uio_resid != len && (error == ERESTART || 657 error == EINTR || error == EWOULDBLOCK)) 658 error = 0; 659 if (error == EPIPE) 660 lwpsignal(p, lp, SIGPIPE); 661 } 662 #ifdef KTRACE 663 if (ktriov != NULL) { 664 if (error == 0) { 665 ktruio.uio_iov = ktriov; 666 ktruio.uio_resid = len - auio->uio_resid; 667 ktrgenio(p, s, UIO_WRITE, &ktruio, error); 668 } 669 FREE(ktriov, M_TEMP); 670 } 671 #endif 672 if (error == 0) 673 *res = len - auio->uio_resid; 674 done: 675 fdrop(fp); 676 return (error); 677 } 678 679 /* 680 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 681 */ 682 int 683 sys_sendto(struct sendto_args *uap) 684 { 685 struct thread *td = curthread; 686 struct uio auio; 687 struct iovec aiov; 688 struct sockaddr *sa = NULL; 689 int error; 690 691 if (uap->to) { 692 error = getsockaddr(&sa, uap->to, uap->tolen); 693 if (error) 694 return (error); 695 } 696 aiov.iov_base = uap->buf; 697 aiov.iov_len = uap->len; 698 auio.uio_iov = &aiov; 699 auio.uio_iovcnt = 1; 700 auio.uio_offset = 0; 701 auio.uio_resid = uap->len; 702 auio.uio_segflg = UIO_USERSPACE; 703 auio.uio_rw = UIO_WRITE; 704 auio.uio_td = td; 705 706 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 707 &uap->sysmsg_result); 708 709 if (sa) 710 FREE(sa, M_SONAME); 711 return (error); 712 } 713 714 /* 715 * sendmsg_args(int s, caddr_t msg, int flags) 716 */ 717 int 718 sys_sendmsg(struct sendmsg_args *uap) 719 { 720 struct thread *td = curthread; 721 struct msghdr msg; 722 struct uio auio; 723 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 724 struct sockaddr *sa = NULL; 725 struct mbuf *control = NULL; 726 int error; 727 728 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 729 if (error) 730 return (error); 731 732 /* 733 * Conditionally copyin msg.msg_name. 734 */ 735 if (msg.msg_name) { 736 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 737 if (error) 738 return (error); 739 } 740 741 /* 742 * Populate auio. 743 */ 744 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 745 &auio.uio_resid); 746 if (error) 747 goto cleanup2; 748 auio.uio_iov = iov; 749 auio.uio_iovcnt = msg.msg_iovlen; 750 auio.uio_offset = 0; 751 auio.uio_segflg = UIO_USERSPACE; 752 auio.uio_rw = UIO_WRITE; 753 auio.uio_td = td; 754 755 /* 756 * Conditionally copyin msg.msg_control. 757 */ 758 if (msg.msg_control) { 759 if (msg.msg_controllen < sizeof(struct cmsghdr) || 760 msg.msg_controllen > MLEN) { 761 error = EINVAL; 762 goto cleanup; 763 } 764 control = m_get(MB_WAIT, MT_CONTROL); 765 if (control == NULL) { 766 error = ENOBUFS; 767 goto cleanup; 768 } 769 control->m_len = msg.msg_controllen; 770 error = copyin(msg.msg_control, mtod(control, caddr_t), 771 msg.msg_controllen); 772 if (error) { 773 m_free(control); 774 goto cleanup; 775 } 776 } 777 778 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 779 &uap->sysmsg_result); 780 781 cleanup: 782 iovec_free(&iov, aiov); 783 cleanup2: 784 if (sa) 785 FREE(sa, M_SONAME); 786 return (error); 787 } 788 789 /* 790 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 791 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 792 * Don't forget to FREE() and m_free() these if they are returned. 793 */ 794 int 795 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 796 struct mbuf **control, int *flags, int *res) 797 { 798 struct thread *td = curthread; 799 struct proc *p = td->td_proc; 800 struct file *fp; 801 int len, error; 802 int lflags; 803 struct socket *so; 804 #ifdef KTRACE 805 struct iovec *ktriov = NULL; 806 struct uio ktruio; 807 #endif 808 809 error = holdsock(p->p_fd, s, &fp); 810 if (error) 811 return (error); 812 if (auio->uio_resid < 0) { 813 error = EINVAL; 814 goto done; 815 } 816 #ifdef KTRACE 817 if (KTRPOINT(td, KTR_GENIO)) { 818 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 819 820 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 821 bcopy(auio->uio_iov, ktriov, iovlen); 822 ktruio = *auio; 823 } 824 #endif 825 len = auio->uio_resid; 826 so = (struct socket *)fp->f_data; 827 828 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 829 if (fp->f_flag & FNONBLOCK) { 830 if (flags) { 831 *flags |= MSG_FNONBLOCKING; 832 } else { 833 lflags = MSG_FNONBLOCKING; 834 flags = &lflags; 835 } 836 } 837 } 838 839 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 840 if (error) { 841 if (auio->uio_resid != len && (error == ERESTART || 842 error == EINTR || error == EWOULDBLOCK)) 843 error = 0; 844 } 845 #ifdef KTRACE 846 if (ktriov != NULL) { 847 if (error == 0) { 848 ktruio.uio_iov = ktriov; 849 ktruio.uio_resid = len - auio->uio_resid; 850 ktrgenio(p, s, UIO_READ, &ktruio, error); 851 } 852 FREE(ktriov, M_TEMP); 853 } 854 #endif 855 if (error == 0) 856 *res = len - auio->uio_resid; 857 done: 858 fdrop(fp); 859 return (error); 860 } 861 862 /* 863 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 864 * caddr_t from, int *fromlenaddr) 865 */ 866 int 867 sys_recvfrom(struct recvfrom_args *uap) 868 { 869 struct thread *td = curthread; 870 struct uio auio; 871 struct iovec aiov; 872 struct sockaddr *sa = NULL; 873 int error, fromlen; 874 875 if (uap->from && uap->fromlenaddr) { 876 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 877 if (error) 878 return (error); 879 if (fromlen < 0) 880 return (EINVAL); 881 } else { 882 fromlen = 0; 883 } 884 aiov.iov_base = uap->buf; 885 aiov.iov_len = uap->len; 886 auio.uio_iov = &aiov; 887 auio.uio_iovcnt = 1; 888 auio.uio_offset = 0; 889 auio.uio_resid = uap->len; 890 auio.uio_segflg = UIO_USERSPACE; 891 auio.uio_rw = UIO_READ; 892 auio.uio_td = td; 893 894 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 895 &uap->flags, &uap->sysmsg_result); 896 897 if (error == 0 && uap->from) { 898 /* note: sa may still be NULL */ 899 if (sa) { 900 fromlen = MIN(fromlen, sa->sa_len); 901 error = copyout(sa, uap->from, fromlen); 902 } else { 903 fromlen = 0; 904 } 905 if (error == 0) { 906 error = copyout(&fromlen, uap->fromlenaddr, 907 sizeof(fromlen)); 908 } 909 } 910 if (sa) 911 FREE(sa, M_SONAME); 912 913 return (error); 914 } 915 916 /* 917 * recvmsg_args(int s, struct msghdr *msg, int flags) 918 */ 919 int 920 sys_recvmsg(struct recvmsg_args *uap) 921 { 922 struct thread *td = curthread; 923 struct msghdr msg; 924 struct uio auio; 925 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 926 struct mbuf *m, *control = NULL; 927 struct sockaddr *sa = NULL; 928 caddr_t ctlbuf; 929 socklen_t *ufromlenp, *ucontrollenp; 930 int error, fromlen, controllen, len, flags, *uflagsp; 931 932 /* 933 * This copyin handles everything except the iovec. 934 */ 935 error = copyin(uap->msg, &msg, sizeof(msg)); 936 if (error) 937 return (error); 938 939 if (msg.msg_name && msg.msg_namelen < 0) 940 return (EINVAL); 941 if (msg.msg_control && msg.msg_controllen < 0) 942 return (EINVAL); 943 944 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 945 msg_namelen)); 946 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 947 msg_controllen)); 948 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 949 msg_flags)); 950 951 /* 952 * Populate auio. 953 */ 954 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 955 &auio.uio_resid); 956 if (error) 957 return (error); 958 auio.uio_iov = iov; 959 auio.uio_iovcnt = msg.msg_iovlen; 960 auio.uio_offset = 0; 961 auio.uio_segflg = UIO_USERSPACE; 962 auio.uio_rw = UIO_READ; 963 auio.uio_td = td; 964 965 flags = uap->flags; 966 967 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio, 968 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result); 969 970 /* 971 * Conditionally copyout the name and populate the namelen field. 972 */ 973 if (error == 0 && msg.msg_name) { 974 /* note: sa may still be NULL */ 975 if (sa != NULL) { 976 fromlen = MIN(msg.msg_namelen, sa->sa_len); 977 error = copyout(sa, msg.msg_name, fromlen); 978 } else 979 fromlen = 0; 980 if (error == 0) 981 error = copyout(&fromlen, ufromlenp, 982 sizeof(*ufromlenp)); 983 } 984 985 /* 986 * Copyout msg.msg_control and msg.msg_controllen. 987 */ 988 if (error == 0 && msg.msg_control) { 989 len = msg.msg_controllen; 990 m = control; 991 ctlbuf = (caddr_t)msg.msg_control; 992 993 while(m && len > 0) { 994 unsigned int tocopy; 995 996 if (len >= m->m_len) { 997 tocopy = m->m_len; 998 } else { 999 msg.msg_flags |= MSG_CTRUNC; 1000 tocopy = len; 1001 } 1002 1003 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1004 if (error) 1005 goto cleanup; 1006 1007 ctlbuf += tocopy; 1008 len -= tocopy; 1009 m = m->m_next; 1010 } 1011 controllen = ctlbuf - (caddr_t)msg.msg_control; 1012 error = copyout(&controllen, ucontrollenp, 1013 sizeof(*ucontrollenp)); 1014 } 1015 1016 if (error == 0) 1017 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1018 1019 cleanup: 1020 if (sa) 1021 FREE(sa, M_SONAME); 1022 iovec_free(&iov, aiov); 1023 if (control) 1024 m_freem(control); 1025 return (error); 1026 } 1027 1028 /* 1029 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1030 * in kernel pointer instead of a userland pointer. This allows us 1031 * to manipulate socket options in the emulation code. 1032 */ 1033 int 1034 kern_setsockopt(int s, struct sockopt *sopt) 1035 { 1036 struct thread *td = curthread; 1037 struct proc *p = td->td_proc; 1038 struct file *fp; 1039 int error; 1040 1041 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1042 return (EFAULT); 1043 if (sopt->sopt_valsize < 0) 1044 return (EINVAL); 1045 1046 error = holdsock(p->p_fd, s, &fp); 1047 if (error) 1048 return (error); 1049 1050 error = sosetopt((struct socket *)fp->f_data, sopt); 1051 fdrop(fp); 1052 return (error); 1053 } 1054 1055 /* 1056 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1057 */ 1058 int 1059 sys_setsockopt(struct setsockopt_args *uap) 1060 { 1061 struct thread *td = curthread; 1062 struct sockopt sopt; 1063 int error; 1064 1065 sopt.sopt_level = uap->level; 1066 sopt.sopt_name = uap->name; 1067 sopt.sopt_val = uap->val; 1068 sopt.sopt_valsize = uap->valsize; 1069 sopt.sopt_td = td; 1070 1071 error = kern_setsockopt(uap->s, &sopt); 1072 return(error); 1073 } 1074 1075 /* 1076 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1077 * in kernel pointer instead of a userland pointer. This allows us 1078 * to manipulate socket options in the emulation code. 1079 */ 1080 int 1081 kern_getsockopt(int s, struct sockopt *sopt) 1082 { 1083 struct thread *td = curthread; 1084 struct proc *p = td->td_proc; 1085 struct file *fp; 1086 int error; 1087 1088 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1089 return (EFAULT); 1090 if (sopt->sopt_valsize < 0) 1091 return (EINVAL); 1092 1093 error = holdsock(p->p_fd, s, &fp); 1094 if (error) 1095 return (error); 1096 1097 error = sogetopt((struct socket *)fp->f_data, sopt); 1098 fdrop(fp); 1099 return (error); 1100 } 1101 1102 /* 1103 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1104 */ 1105 int 1106 sys_getsockopt(struct getsockopt_args *uap) 1107 { 1108 struct thread *td = curthread; 1109 struct sockopt sopt; 1110 int error, valsize; 1111 1112 if (uap->val) { 1113 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1114 if (error) 1115 return (error); 1116 if (valsize < 0) 1117 return (EINVAL); 1118 } else { 1119 valsize = 0; 1120 } 1121 1122 sopt.sopt_level = uap->level; 1123 sopt.sopt_name = uap->name; 1124 sopt.sopt_val = uap->val; 1125 sopt.sopt_valsize = valsize; 1126 sopt.sopt_td = td; 1127 1128 error = kern_getsockopt(uap->s, &sopt); 1129 if (error == 0) { 1130 valsize = sopt.sopt_valsize; 1131 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1132 } 1133 return (error); 1134 } 1135 1136 /* 1137 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1138 * This allows kern_getsockname() to return a pointer to an allocated struct 1139 * sockaddr which must be freed later with FREE(). The caller must 1140 * initialize *name to NULL. 1141 */ 1142 int 1143 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1144 { 1145 struct thread *td = curthread; 1146 struct proc *p = td->td_proc; 1147 struct file *fp; 1148 struct socket *so; 1149 struct sockaddr *sa = NULL; 1150 int error; 1151 1152 error = holdsock(p->p_fd, s, &fp); 1153 if (error) 1154 return (error); 1155 if (*namelen < 0) { 1156 fdrop(fp); 1157 return (EINVAL); 1158 } 1159 so = (struct socket *)fp->f_data; 1160 error = so_pru_sockaddr(so, &sa); 1161 if (error == 0) { 1162 if (sa == 0) { 1163 *namelen = 0; 1164 } else { 1165 *namelen = MIN(*namelen, sa->sa_len); 1166 *name = sa; 1167 } 1168 } 1169 1170 fdrop(fp); 1171 return (error); 1172 } 1173 1174 /* 1175 * getsockname_args(int fdes, caddr_t asa, int *alen) 1176 * 1177 * Get socket name. 1178 */ 1179 int 1180 sys_getsockname(struct getsockname_args *uap) 1181 { 1182 struct sockaddr *sa = NULL; 1183 int error, sa_len; 1184 1185 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1186 if (error) 1187 return (error); 1188 1189 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1190 1191 if (error == 0) 1192 error = copyout(sa, uap->asa, sa_len); 1193 if (error == 0) 1194 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1195 if (sa) 1196 FREE(sa, M_SONAME); 1197 return (error); 1198 } 1199 1200 /* 1201 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1202 * This allows kern_getpeername() to return a pointer to an allocated struct 1203 * sockaddr which must be freed later with FREE(). The caller must 1204 * initialize *name to NULL. 1205 */ 1206 int 1207 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1208 { 1209 struct thread *td = curthread; 1210 struct proc *p = td->td_proc; 1211 struct file *fp; 1212 struct socket *so; 1213 struct sockaddr *sa = NULL; 1214 int error; 1215 1216 error = holdsock(p->p_fd, s, &fp); 1217 if (error) 1218 return (error); 1219 if (*namelen < 0) { 1220 fdrop(fp); 1221 return (EINVAL); 1222 } 1223 so = (struct socket *)fp->f_data; 1224 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1225 fdrop(fp); 1226 return (ENOTCONN); 1227 } 1228 error = so_pru_peeraddr(so, &sa); 1229 if (error == 0) { 1230 if (sa == 0) { 1231 *namelen = 0; 1232 } else { 1233 *namelen = MIN(*namelen, sa->sa_len); 1234 *name = sa; 1235 } 1236 } 1237 1238 fdrop(fp); 1239 return (error); 1240 } 1241 1242 /* 1243 * getpeername_args(int fdes, caddr_t asa, int *alen) 1244 * 1245 * Get name of peer for connected socket. 1246 */ 1247 int 1248 sys_getpeername(struct getpeername_args *uap) 1249 { 1250 struct sockaddr *sa = NULL; 1251 int error, sa_len; 1252 1253 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1254 if (error) 1255 return (error); 1256 1257 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1258 1259 if (error == 0) 1260 error = copyout(sa, uap->asa, sa_len); 1261 if (error == 0) 1262 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1263 if (sa) 1264 FREE(sa, M_SONAME); 1265 return (error); 1266 } 1267 1268 int 1269 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1270 { 1271 struct sockaddr *sa; 1272 int error; 1273 1274 *namp = NULL; 1275 if (len > SOCK_MAXADDRLEN) 1276 return ENAMETOOLONG; 1277 if (len < offsetof(struct sockaddr, sa_data[0])) 1278 return EDOM; 1279 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1280 error = copyin(uaddr, sa, len); 1281 if (error) { 1282 FREE(sa, M_SONAME); 1283 } else { 1284 #if BYTE_ORDER != BIG_ENDIAN 1285 /* 1286 * The bind(), connect(), and sendto() syscalls were not 1287 * versioned for COMPAT_43. Thus, this check must stay. 1288 */ 1289 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1290 sa->sa_family = sa->sa_len; 1291 #endif 1292 sa->sa_len = len; 1293 *namp = sa; 1294 } 1295 return error; 1296 } 1297 1298 /* 1299 * Detach a mapped page and release resources back to the system. 1300 * We must release our wiring and if the object is ripped out 1301 * from under the vm_page we become responsible for freeing the 1302 * page. These routines must be MPSAFE. 1303 * 1304 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1305 * 1306 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required. 1307 */ 1308 static void 1309 sf_buf_mref(void *arg) 1310 { 1311 struct sfbuf_mref *sfm = arg; 1312 1313 /* 1314 * We must already hold a ref so there is no race to 0, just 1315 * atomically increment the count. 1316 */ 1317 atomic_add_int(&sfm->mref_count, 1); 1318 } 1319 1320 static void 1321 sf_buf_mfree(void *arg) 1322 { 1323 struct sfbuf_mref *sfm = arg; 1324 vm_page_t m; 1325 1326 KKASSERT(sfm->mref_count > 0); 1327 if (sfm->mref_count == 1) { 1328 /* 1329 * We are the only holder so no further locking is required, 1330 * the sfbuf can simply be freed. 1331 */ 1332 sfm->mref_count = 0; 1333 goto freeit; 1334 } else { 1335 /* 1336 * There may be other holders, we must obtain the serializer 1337 * to protect against a sf_buf_mfree() race to 0. An atomic 1338 * operation is still required for races against 1339 * sf_buf_mref(). 1340 * 1341 * XXX vm_page_*() and SFBUF routines not MPSAFE yet. 1342 */ 1343 lwkt_serialize_enter(&sfm->serializer); 1344 atomic_subtract_int(&sfm->mref_count, 1); 1345 if (sfm->mref_count == 0) { 1346 lwkt_serialize_exit(&sfm->serializer); 1347 freeit: 1348 get_mplock(); 1349 crit_enter(); 1350 m = sf_buf_page(sfm->sf); 1351 sf_buf_free(sfm->sf); 1352 vm_page_unwire(m, 0); 1353 if (m->wire_count == 0 && m->object == NULL) 1354 vm_page_try_to_free(m); 1355 crit_exit(); 1356 rel_mplock(); 1357 kfree(sfm, M_SENDFILE); 1358 } else { 1359 lwkt_serialize_exit(&sfm->serializer); 1360 } 1361 } 1362 } 1363 1364 /* 1365 * sendfile(2). 1366 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1367 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1368 * 1369 * Send a file specified by 'fd' and starting at 'offset' to a socket 1370 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1371 * nbytes == 0. Optionally add a header and/or trailer to the socket 1372 * output. If specified, write the total number of bytes sent into *sbytes. 1373 * 1374 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1375 * the headers to count against the remaining bytes to be sent from 1376 * the file descriptor. We may wish to implement a compatibility syscall 1377 * in the future. 1378 */ 1379 int 1380 sys_sendfile(struct sendfile_args *uap) 1381 { 1382 struct thread *td = curthread; 1383 struct proc *p = td->td_proc; 1384 struct file *fp; 1385 struct vnode *vp = NULL; 1386 struct sf_hdtr hdtr; 1387 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1388 struct uio auio; 1389 struct mbuf *mheader = NULL; 1390 off_t hdtr_size = 0, sbytes; 1391 int error, hbytes = 0, tbytes; 1392 1393 KKASSERT(p); 1394 1395 /* 1396 * Do argument checking. Must be a regular file in, stream 1397 * type and connected socket out, positive offset. 1398 */ 1399 fp = holdfp(p->p_fd, uap->fd, FREAD); 1400 if (fp == NULL) { 1401 return (EBADF); 1402 } 1403 if (fp->f_type != DTYPE_VNODE) { 1404 fdrop(fp); 1405 return (EINVAL); 1406 } 1407 vp = (struct vnode *)fp->f_data; 1408 vref(vp); 1409 fdrop(fp); 1410 1411 /* 1412 * If specified, get the pointer to the sf_hdtr struct for 1413 * any headers/trailers. 1414 */ 1415 if (uap->hdtr) { 1416 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1417 if (error) 1418 goto done; 1419 /* 1420 * Send any headers. 1421 */ 1422 if (hdtr.headers) { 1423 error = iovec_copyin(hdtr.headers, &iov, aiov, 1424 hdtr.hdr_cnt, &hbytes); 1425 if (error) 1426 goto done; 1427 auio.uio_iov = iov; 1428 auio.uio_iovcnt = hdtr.hdr_cnt; 1429 auio.uio_offset = 0; 1430 auio.uio_segflg = UIO_USERSPACE; 1431 auio.uio_rw = UIO_WRITE; 1432 auio.uio_td = td; 1433 auio.uio_resid = hbytes; 1434 1435 mheader = m_uiomove(&auio); 1436 1437 iovec_free(&iov, aiov); 1438 if (mheader == NULL) 1439 goto done; 1440 } 1441 } 1442 1443 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1444 &sbytes, uap->flags); 1445 if (error) 1446 goto done; 1447 1448 /* 1449 * Send trailers. Wimp out and use writev(2). 1450 */ 1451 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1452 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1453 hdtr.trl_cnt, &auio.uio_resid); 1454 if (error) 1455 goto done; 1456 auio.uio_iov = iov; 1457 auio.uio_iovcnt = hdtr.trl_cnt; 1458 auio.uio_offset = 0; 1459 auio.uio_segflg = UIO_USERSPACE; 1460 auio.uio_rw = UIO_WRITE; 1461 auio.uio_td = td; 1462 1463 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1464 1465 iovec_free(&iov, aiov); 1466 if (error) 1467 goto done; 1468 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1469 } 1470 1471 done: 1472 if (uap->sbytes != NULL) { 1473 sbytes += hdtr_size; 1474 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1475 } 1476 if (vp) 1477 vrele(vp); 1478 return (error); 1479 } 1480 1481 int 1482 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1483 struct mbuf *mheader, off_t *sbytes, int flags) 1484 { 1485 struct thread *td = curthread; 1486 struct proc *p = td->td_proc; 1487 struct vm_object *obj; 1488 struct socket *so; 1489 struct file *fp; 1490 struct mbuf *m; 1491 struct sf_buf *sf; 1492 struct sfbuf_mref *sfm; 1493 struct vm_page *pg; 1494 off_t off, xfsize; 1495 off_t hbytes = 0; 1496 int error = 0; 1497 1498 if (vp->v_type != VREG) { 1499 error = EINVAL; 1500 goto done0; 1501 } 1502 if ((obj = vp->v_object) == NULL) { 1503 error = EINVAL; 1504 goto done0; 1505 } 1506 error = holdsock(p->p_fd, sfd, &fp); 1507 if (error) 1508 goto done0; 1509 so = (struct socket *)fp->f_data; 1510 if (so->so_type != SOCK_STREAM) { 1511 error = EINVAL; 1512 goto done; 1513 } 1514 if ((so->so_state & SS_ISCONNECTED) == 0) { 1515 error = ENOTCONN; 1516 goto done; 1517 } 1518 if (offset < 0) { 1519 error = EINVAL; 1520 goto done; 1521 } 1522 1523 *sbytes = 0; 1524 /* 1525 * Protect against multiple writers to the socket. 1526 */ 1527 ssb_lock(&so->so_snd, M_WAITOK); 1528 1529 /* 1530 * Loop through the pages in the file, starting with the requested 1531 * offset. Get a file page (do I/O if necessary), map the file page 1532 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1533 * it on the socket. 1534 */ 1535 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1536 vm_pindex_t pindex; 1537 vm_offset_t pgoff; 1538 1539 pindex = OFF_TO_IDX(off); 1540 retry_lookup: 1541 /* 1542 * Calculate the amount to transfer. Not to exceed a page, 1543 * the EOF, or the passed in nbytes. 1544 */ 1545 xfsize = vp->v_filesize - off; 1546 if (xfsize > PAGE_SIZE) 1547 xfsize = PAGE_SIZE; 1548 pgoff = (vm_offset_t)(off & PAGE_MASK); 1549 if (PAGE_SIZE - pgoff < xfsize) 1550 xfsize = PAGE_SIZE - pgoff; 1551 if (nbytes && xfsize > (nbytes - *sbytes)) 1552 xfsize = nbytes - *sbytes; 1553 if (xfsize <= 0) 1554 break; 1555 /* 1556 * Optimize the non-blocking case by looking at the socket space 1557 * before going to the extra work of constituting the sf_buf. 1558 */ 1559 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) { 1560 if (so->so_state & SS_CANTSENDMORE) 1561 error = EPIPE; 1562 else 1563 error = EAGAIN; 1564 ssb_unlock(&so->so_snd); 1565 goto done; 1566 } 1567 /* 1568 * Attempt to look up the page. 1569 * 1570 * Allocate if not found, wait and loop if busy, then 1571 * wire the page. critical section protection is 1572 * required to maintain the object association (an 1573 * interrupt can free the page) through to the 1574 * vm_page_wire() call. 1575 */ 1576 crit_enter(); 1577 pg = vm_page_lookup(obj, pindex); 1578 if (pg == NULL) { 1579 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1580 if (pg == NULL) { 1581 vm_wait(); 1582 crit_exit(); 1583 goto retry_lookup; 1584 } 1585 vm_page_wakeup(pg); 1586 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1587 crit_exit(); 1588 goto retry_lookup; 1589 } 1590 vm_page_wire(pg); 1591 crit_exit(); 1592 1593 /* 1594 * If page is not valid for what we need, initiate I/O 1595 */ 1596 1597 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1598 struct uio auio; 1599 struct iovec aiov; 1600 int bsize; 1601 1602 /* 1603 * Ensure that our page is still around when the I/O 1604 * completes. 1605 */ 1606 vm_page_io_start(pg); 1607 1608 /* 1609 * Get the page from backing store. 1610 */ 1611 bsize = vp->v_mount->mnt_stat.f_iosize; 1612 auio.uio_iov = &aiov; 1613 auio.uio_iovcnt = 1; 1614 aiov.iov_base = 0; 1615 aiov.iov_len = MAXBSIZE; 1616 auio.uio_resid = MAXBSIZE; 1617 auio.uio_offset = trunc_page(off); 1618 auio.uio_segflg = UIO_NOCOPY; 1619 auio.uio_rw = UIO_READ; 1620 auio.uio_td = td; 1621 vn_lock(vp, LK_SHARED | LK_RETRY); 1622 error = VOP_READ(vp, &auio, 1623 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1624 p->p_ucred); 1625 vn_unlock(vp); 1626 vm_page_flag_clear(pg, PG_ZERO); 1627 vm_page_io_finish(pg); 1628 if (error) { 1629 crit_enter(); 1630 vm_page_unwire(pg, 0); 1631 vm_page_try_to_free(pg); 1632 crit_exit(); 1633 ssb_unlock(&so->so_snd); 1634 goto done; 1635 } 1636 } 1637 1638 1639 /* 1640 * Get a sendfile buf. We usually wait as long as necessary, 1641 * but this wait can be interrupted. 1642 */ 1643 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) { 1644 crit_enter(); 1645 vm_page_unwire(pg, 0); 1646 vm_page_try_to_free(pg); 1647 crit_exit(); 1648 ssb_unlock(&so->so_snd); 1649 error = EINTR; 1650 goto done; 1651 } 1652 1653 /* 1654 * Get an mbuf header and set it up as having external storage. 1655 */ 1656 MGETHDR(m, MB_WAIT, MT_DATA); 1657 if (m == NULL) { 1658 error = ENOBUFS; 1659 sf_buf_free(sf); 1660 ssb_unlock(&so->so_snd); 1661 goto done; 1662 } 1663 1664 /* 1665 * sfm is a temporary hack, use a per-cpu cache for this. 1666 */ 1667 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK); 1668 sfm->sf = sf; 1669 sfm->mref_count = 1; 1670 lwkt_serialize_init(&sfm->serializer); 1671 1672 m->m_ext.ext_free = sf_buf_mfree; 1673 m->m_ext.ext_ref = sf_buf_mref; 1674 m->m_ext.ext_arg = sfm; 1675 m->m_ext.ext_buf = (void *)sf->kva; 1676 m->m_ext.ext_size = PAGE_SIZE; 1677 m->m_data = (char *) sf->kva + pgoff; 1678 m->m_flags |= M_EXT; 1679 m->m_pkthdr.len = m->m_len = xfsize; 1680 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1681 1682 if (mheader != NULL) { 1683 hbytes = mheader->m_pkthdr.len; 1684 mheader->m_pkthdr.len += m->m_pkthdr.len; 1685 m_cat(mheader, m); 1686 m = mheader; 1687 mheader = NULL; 1688 } else 1689 hbytes = 0; 1690 1691 /* 1692 * Add the buffer to the socket buffer chain. 1693 */ 1694 crit_enter(); 1695 retry_space: 1696 /* 1697 * Make sure that the socket is still able to take more data. 1698 * CANTSENDMORE being true usually means that the connection 1699 * was closed. so_error is true when an error was sensed after 1700 * a previous send. 1701 * The state is checked after the page mapping and buffer 1702 * allocation above since those operations may block and make 1703 * any socket checks stale. From this point forward, nothing 1704 * blocks before the pru_send (or more accurately, any blocking 1705 * results in a loop back to here to re-check). 1706 */ 1707 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1708 if (so->so_state & SS_CANTSENDMORE) { 1709 error = EPIPE; 1710 } else { 1711 error = so->so_error; 1712 so->so_error = 0; 1713 } 1714 m_freem(m); 1715 ssb_unlock(&so->so_snd); 1716 crit_exit(); 1717 goto done; 1718 } 1719 /* 1720 * Wait for socket space to become available. We do this just 1721 * after checking the connection state above in order to avoid 1722 * a race condition with ssb_wait(). 1723 */ 1724 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) { 1725 if (fp->f_flag & FNONBLOCK) { 1726 m_freem(m); 1727 ssb_unlock(&so->so_snd); 1728 crit_exit(); 1729 error = EAGAIN; 1730 goto done; 1731 } 1732 error = ssb_wait(&so->so_snd); 1733 /* 1734 * An error from ssb_wait usually indicates that we've 1735 * been interrupted by a signal. If we've sent anything 1736 * then return bytes sent, otherwise return the error. 1737 */ 1738 if (error) { 1739 m_freem(m); 1740 ssb_unlock(&so->so_snd); 1741 crit_exit(); 1742 goto done; 1743 } 1744 goto retry_space; 1745 } 1746 error = so_pru_send(so, 0, m, NULL, NULL, td); 1747 crit_exit(); 1748 if (error) { 1749 ssb_unlock(&so->so_snd); 1750 goto done; 1751 } 1752 } 1753 if (mheader != NULL) { 1754 *sbytes += mheader->m_pkthdr.len; 1755 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1756 mheader = NULL; 1757 } 1758 ssb_unlock(&so->so_snd); 1759 1760 done: 1761 fdrop(fp); 1762 done0: 1763 if (mheader != NULL) 1764 m_freem(mheader); 1765 return (error); 1766 } 1767 1768 int 1769 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1770 { 1771 #ifdef SCTP 1772 struct thread *td = curthread; 1773 struct proc *p = td->td_proc; 1774 struct file *lfp = NULL; 1775 struct file *nfp = NULL; 1776 int error; 1777 struct socket *head, *so; 1778 caddr_t assoc_id; 1779 int fd; 1780 short fflag; /* type must match fp->f_flag */ 1781 1782 assoc_id = uap->name; 1783 error = holdsock(p->p_fd, uap->sd, &lfp); 1784 if (error) { 1785 return (error); 1786 } 1787 crit_enter(); 1788 head = (struct socket *)lfp->f_data; 1789 error = sctp_can_peel_off(head, assoc_id); 1790 if (error) { 1791 crit_exit(); 1792 goto done; 1793 } 1794 /* 1795 * At this point we know we do have a assoc to pull 1796 * we proceed to get the fd setup. This may block 1797 * but that is ok. 1798 */ 1799 1800 fflag = lfp->f_flag; 1801 error = falloc(p, &nfp, &fd); 1802 if (error) { 1803 /* 1804 * Probably ran out of file descriptors. Put the 1805 * unaccepted connection back onto the queue and 1806 * do another wakeup so some other process might 1807 * have a chance at it. 1808 */ 1809 crit_exit(); 1810 goto done; 1811 } 1812 uap->sysmsg_result = fd; 1813 1814 so = sctp_get_peeloff(head, assoc_id, &error); 1815 if (so == NULL) { 1816 /* 1817 * Either someone else peeled it off OR 1818 * we can't get a socket. 1819 */ 1820 goto noconnection; 1821 } 1822 so->so_state &= ~SS_COMP; 1823 so->so_state &= ~SS_NOFDREF; 1824 so->so_head = NULL; 1825 if (head->so_sigio != NULL) 1826 fsetown(fgetown(head->so_sigio), &so->so_sigio); 1827 1828 nfp->f_type = DTYPE_SOCKET; 1829 nfp->f_flag = fflag; 1830 nfp->f_ops = &socketops; 1831 nfp->f_data = so; 1832 1833 noconnection: 1834 /* 1835 * Assign the file pointer to the reserved descriptor, or clear 1836 * the reserved descriptor if an error occured. 1837 */ 1838 if (error) 1839 fsetfd(p, NULL, fd); 1840 else 1841 fsetfd(p, nfp, fd); 1842 crit_exit(); 1843 /* 1844 * Release explicitly held references before returning. 1845 */ 1846 done: 1847 if (nfp != NULL) 1848 fdrop(nfp); 1849 fdrop(lfp); 1850 return (error); 1851 #else /* SCTP */ 1852 return(EOPNOTSUPP); 1853 #endif /* SCTP */ 1854 } 1855