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 1.77 2007/01/08 21:41:56 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 * 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 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 ksignal(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 cleanup2; 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 iovec_free(&iov, aiov); 782 cleanup2: 783 if (sa) 784 FREE(sa, M_SONAME); 785 return (error); 786 } 787 788 /* 789 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 790 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 791 * Don't forget to FREE() and m_free() these if they are returned. 792 */ 793 int 794 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 795 struct mbuf **control, int *flags, int *res) 796 { 797 struct thread *td = curthread; 798 struct proc *p = td->td_proc; 799 struct file *fp; 800 int len, error; 801 int lflags; 802 struct socket *so; 803 #ifdef KTRACE 804 struct iovec *ktriov = NULL; 805 struct uio ktruio; 806 #endif 807 808 error = holdsock(p->p_fd, s, &fp); 809 if (error) 810 return (error); 811 if (auio->uio_resid < 0) { 812 error = EINVAL; 813 goto done; 814 } 815 #ifdef KTRACE 816 if (KTRPOINT(td, KTR_GENIO)) { 817 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 818 819 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 820 bcopy(auio->uio_iov, ktriov, iovlen); 821 ktruio = *auio; 822 } 823 #endif 824 len = auio->uio_resid; 825 so = (struct socket *)fp->f_data; 826 827 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 828 if (fp->f_flag & FNONBLOCK) { 829 if (flags) { 830 *flags |= MSG_FNONBLOCKING; 831 } else { 832 lflags = MSG_FNONBLOCKING; 833 flags = &lflags; 834 } 835 } 836 } 837 838 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 839 if (error) { 840 if (auio->uio_resid != len && (error == ERESTART || 841 error == EINTR || error == EWOULDBLOCK)) 842 error = 0; 843 } 844 #ifdef KTRACE 845 if (ktriov != NULL) { 846 if (error == 0) { 847 ktruio.uio_iov = ktriov; 848 ktruio.uio_resid = len - auio->uio_resid; 849 ktrgenio(p, s, UIO_READ, &ktruio, error); 850 } 851 FREE(ktriov, M_TEMP); 852 } 853 #endif 854 if (error == 0) 855 *res = len - auio->uio_resid; 856 done: 857 fdrop(fp); 858 return (error); 859 } 860 861 /* 862 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 863 * caddr_t from, int *fromlenaddr) 864 */ 865 int 866 sys_recvfrom(struct recvfrom_args *uap) 867 { 868 struct thread *td = curthread; 869 struct uio auio; 870 struct iovec aiov; 871 struct sockaddr *sa = NULL; 872 int error, fromlen; 873 874 if (uap->from && uap->fromlenaddr) { 875 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 876 if (error) 877 return (error); 878 if (fromlen < 0) 879 return (EINVAL); 880 } else { 881 fromlen = 0; 882 } 883 aiov.iov_base = uap->buf; 884 aiov.iov_len = uap->len; 885 auio.uio_iov = &aiov; 886 auio.uio_iovcnt = 1; 887 auio.uio_offset = 0; 888 auio.uio_resid = uap->len; 889 auio.uio_segflg = UIO_USERSPACE; 890 auio.uio_rw = UIO_READ; 891 auio.uio_td = td; 892 893 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 894 &uap->flags, &uap->sysmsg_result); 895 896 if (error == 0 && uap->from) { 897 /* note: sa may still be NULL */ 898 if (sa) { 899 fromlen = MIN(fromlen, sa->sa_len); 900 error = copyout(sa, uap->from, fromlen); 901 } else { 902 fromlen = 0; 903 } 904 if (error == 0) { 905 error = copyout(&fromlen, uap->fromlenaddr, 906 sizeof(fromlen)); 907 } 908 } 909 if (sa) 910 FREE(sa, M_SONAME); 911 912 return (error); 913 } 914 915 /* 916 * recvmsg_args(int s, struct msghdr *msg, int flags) 917 */ 918 int 919 sys_recvmsg(struct recvmsg_args *uap) 920 { 921 struct thread *td = curthread; 922 struct msghdr msg; 923 struct uio auio; 924 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 925 struct mbuf *m, *control = NULL; 926 struct sockaddr *sa = NULL; 927 caddr_t ctlbuf; 928 socklen_t *ufromlenp, *ucontrollenp; 929 int error, fromlen, controllen, len, flags, *uflagsp; 930 931 /* 932 * This copyin handles everything except the iovec. 933 */ 934 error = copyin(uap->msg, &msg, sizeof(msg)); 935 if (error) 936 return (error); 937 938 if (msg.msg_name && msg.msg_namelen < 0) 939 return (EINVAL); 940 if (msg.msg_control && msg.msg_controllen < 0) 941 return (EINVAL); 942 943 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 944 msg_namelen)); 945 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 946 msg_controllen)); 947 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 948 msg_flags)); 949 950 /* 951 * Populate auio. 952 */ 953 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 954 &auio.uio_resid); 955 if (error) 956 return (error); 957 auio.uio_iov = iov; 958 auio.uio_iovcnt = msg.msg_iovlen; 959 auio.uio_offset = 0; 960 auio.uio_segflg = UIO_USERSPACE; 961 auio.uio_rw = UIO_READ; 962 auio.uio_td = td; 963 964 flags = uap->flags; 965 966 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio, 967 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result); 968 969 /* 970 * Conditionally copyout the name and populate the namelen field. 971 */ 972 if (error == 0 && msg.msg_name) { 973 fromlen = MIN(msg.msg_namelen, sa->sa_len); 974 error = copyout(sa, msg.msg_name, fromlen); 975 if (error == 0) 976 error = copyout(&fromlen, ufromlenp, 977 sizeof(*ufromlenp)); 978 } 979 980 /* 981 * Copyout msg.msg_control and msg.msg_controllen. 982 */ 983 if (error == 0 && msg.msg_control) { 984 len = msg.msg_controllen; 985 m = control; 986 ctlbuf = (caddr_t)msg.msg_control; 987 988 while(m && len > 0) { 989 unsigned int tocopy; 990 991 if (len >= m->m_len) { 992 tocopy = m->m_len; 993 } else { 994 msg.msg_flags |= MSG_CTRUNC; 995 tocopy = len; 996 } 997 998 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 999 if (error) 1000 goto cleanup; 1001 1002 ctlbuf += tocopy; 1003 len -= tocopy; 1004 m = m->m_next; 1005 } 1006 controllen = ctlbuf - (caddr_t)msg.msg_control; 1007 error = copyout(&controllen, ucontrollenp, 1008 sizeof(*ucontrollenp)); 1009 } 1010 1011 if (error == 0) 1012 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1013 1014 cleanup: 1015 if (sa) 1016 FREE(sa, M_SONAME); 1017 iovec_free(&iov, aiov); 1018 if (control) 1019 m_freem(control); 1020 return (error); 1021 } 1022 1023 /* 1024 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1025 * in kernel pointer instead of a userland pointer. This allows us 1026 * to manipulate socket options in the emulation code. 1027 */ 1028 int 1029 kern_setsockopt(int s, struct sockopt *sopt) 1030 { 1031 struct thread *td = curthread; 1032 struct proc *p = td->td_proc; 1033 struct file *fp; 1034 int error; 1035 1036 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1037 return (EFAULT); 1038 if (sopt->sopt_valsize < 0) 1039 return (EINVAL); 1040 1041 error = holdsock(p->p_fd, s, &fp); 1042 if (error) 1043 return (error); 1044 1045 error = sosetopt((struct socket *)fp->f_data, sopt); 1046 fdrop(fp); 1047 return (error); 1048 } 1049 1050 /* 1051 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1052 */ 1053 int 1054 sys_setsockopt(struct setsockopt_args *uap) 1055 { 1056 struct thread *td = curthread; 1057 struct sockopt sopt; 1058 int error; 1059 1060 sopt.sopt_level = uap->level; 1061 sopt.sopt_name = uap->name; 1062 sopt.sopt_val = uap->val; 1063 sopt.sopt_valsize = uap->valsize; 1064 sopt.sopt_td = td; 1065 1066 error = kern_setsockopt(uap->s, &sopt); 1067 return(error); 1068 } 1069 1070 /* 1071 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1072 * in kernel pointer instead of a userland pointer. This allows us 1073 * to manipulate socket options in the emulation code. 1074 */ 1075 int 1076 kern_getsockopt(int s, struct sockopt *sopt) 1077 { 1078 struct thread *td = curthread; 1079 struct proc *p = td->td_proc; 1080 struct file *fp; 1081 int error; 1082 1083 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1084 return (EFAULT); 1085 if (sopt->sopt_valsize < 0) 1086 return (EINVAL); 1087 1088 error = holdsock(p->p_fd, s, &fp); 1089 if (error) 1090 return (error); 1091 1092 error = sogetopt((struct socket *)fp->f_data, sopt); 1093 fdrop(fp); 1094 return (error); 1095 } 1096 1097 /* 1098 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1099 */ 1100 int 1101 sys_getsockopt(struct getsockopt_args *uap) 1102 { 1103 struct thread *td = curthread; 1104 struct sockopt sopt; 1105 int error, valsize; 1106 1107 if (uap->val) { 1108 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1109 if (error) 1110 return (error); 1111 if (valsize < 0) 1112 return (EINVAL); 1113 } else { 1114 valsize = 0; 1115 } 1116 1117 sopt.sopt_level = uap->level; 1118 sopt.sopt_name = uap->name; 1119 sopt.sopt_val = uap->val; 1120 sopt.sopt_valsize = valsize; 1121 sopt.sopt_td = td; 1122 1123 error = kern_getsockopt(uap->s, &sopt); 1124 if (error == 0) { 1125 valsize = sopt.sopt_valsize; 1126 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1127 } 1128 return (error); 1129 } 1130 1131 /* 1132 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1133 * This allows kern_getsockname() to return a pointer to an allocated struct 1134 * sockaddr which must be freed later with FREE(). The caller must 1135 * initialize *name to NULL. 1136 */ 1137 int 1138 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1139 { 1140 struct thread *td = curthread; 1141 struct proc *p = td->td_proc; 1142 struct file *fp; 1143 struct socket *so; 1144 struct sockaddr *sa = NULL; 1145 int error; 1146 1147 error = holdsock(p->p_fd, s, &fp); 1148 if (error) 1149 return (error); 1150 if (*namelen < 0) { 1151 fdrop(fp); 1152 return (EINVAL); 1153 } 1154 so = (struct socket *)fp->f_data; 1155 error = so_pru_sockaddr(so, &sa); 1156 if (error == 0) { 1157 if (sa == 0) { 1158 *namelen = 0; 1159 } else { 1160 *namelen = MIN(*namelen, sa->sa_len); 1161 *name = sa; 1162 } 1163 } 1164 1165 fdrop(fp); 1166 return (error); 1167 } 1168 1169 /* 1170 * getsockname_args(int fdes, caddr_t asa, int *alen) 1171 * 1172 * Get socket name. 1173 */ 1174 int 1175 sys_getsockname(struct getsockname_args *uap) 1176 { 1177 struct sockaddr *sa = NULL; 1178 int error, sa_len; 1179 1180 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1181 if (error) 1182 return (error); 1183 1184 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1185 1186 if (error == 0) 1187 error = copyout(sa, uap->asa, sa_len); 1188 if (error == 0) 1189 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1190 if (sa) 1191 FREE(sa, M_SONAME); 1192 return (error); 1193 } 1194 1195 /* 1196 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1197 * This allows kern_getpeername() to return a pointer to an allocated struct 1198 * sockaddr which must be freed later with FREE(). The caller must 1199 * initialize *name to NULL. 1200 */ 1201 int 1202 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1203 { 1204 struct thread *td = curthread; 1205 struct proc *p = td->td_proc; 1206 struct file *fp; 1207 struct socket *so; 1208 struct sockaddr *sa = NULL; 1209 int error; 1210 1211 error = holdsock(p->p_fd, s, &fp); 1212 if (error) 1213 return (error); 1214 if (*namelen < 0) { 1215 fdrop(fp); 1216 return (EINVAL); 1217 } 1218 so = (struct socket *)fp->f_data; 1219 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1220 fdrop(fp); 1221 return (ENOTCONN); 1222 } 1223 error = so_pru_peeraddr(so, &sa); 1224 if (error == 0) { 1225 if (sa == 0) { 1226 *namelen = 0; 1227 } else { 1228 *namelen = MIN(*namelen, sa->sa_len); 1229 *name = sa; 1230 } 1231 } 1232 1233 fdrop(fp); 1234 return (error); 1235 } 1236 1237 /* 1238 * getpeername_args(int fdes, caddr_t asa, int *alen) 1239 * 1240 * Get name of peer for connected socket. 1241 */ 1242 int 1243 sys_getpeername(struct getpeername_args *uap) 1244 { 1245 struct sockaddr *sa = NULL; 1246 int error, sa_len; 1247 1248 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1249 if (error) 1250 return (error); 1251 1252 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1253 1254 if (error == 0) 1255 error = copyout(sa, uap->asa, sa_len); 1256 if (error == 0) 1257 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1258 if (sa) 1259 FREE(sa, M_SONAME); 1260 return (error); 1261 } 1262 1263 int 1264 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1265 { 1266 struct sockaddr *sa; 1267 int error; 1268 1269 *namp = NULL; 1270 if (len > SOCK_MAXADDRLEN) 1271 return ENAMETOOLONG; 1272 if (len < offsetof(struct sockaddr, sa_data[0])) 1273 return EDOM; 1274 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1275 error = copyin(uaddr, sa, len); 1276 if (error) { 1277 FREE(sa, M_SONAME); 1278 } else { 1279 #if BYTE_ORDER != BIG_ENDIAN 1280 /* 1281 * The bind(), connect(), and sendto() syscalls were not 1282 * versioned for COMPAT_43. Thus, this check must stay. 1283 */ 1284 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1285 sa->sa_family = sa->sa_len; 1286 #endif 1287 sa->sa_len = len; 1288 *namp = sa; 1289 } 1290 return error; 1291 } 1292 1293 /* 1294 * Detach a mapped page and release resources back to the system. 1295 * We must release our wiring and if the object is ripped out 1296 * from under the vm_page we become responsible for freeing the 1297 * page. These routines must be MPSAFE. 1298 * 1299 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1300 * 1301 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required. 1302 */ 1303 static void 1304 sf_buf_mref(void *arg) 1305 { 1306 struct sfbuf_mref *sfm = arg; 1307 1308 /* 1309 * We must already hold a ref so there is no race to 0, just 1310 * atomically increment the count. 1311 */ 1312 atomic_add_int(&sfm->mref_count, 1); 1313 } 1314 1315 static void 1316 sf_buf_mfree(void *arg) 1317 { 1318 struct sfbuf_mref *sfm = arg; 1319 vm_page_t m; 1320 1321 KKASSERT(sfm->mref_count > 0); 1322 if (sfm->mref_count == 1) { 1323 /* 1324 * We are the only holder so no further locking is required, 1325 * the sfbuf can simply be freed. 1326 */ 1327 sfm->mref_count = 0; 1328 goto freeit; 1329 } else { 1330 /* 1331 * There may be other holders, we must obtain the serializer 1332 * to protect against a sf_buf_mfree() race to 0. An atomic 1333 * operation is still required for races against 1334 * sf_buf_mref(). 1335 * 1336 * XXX vm_page_*() and SFBUF routines not MPSAFE yet. 1337 */ 1338 lwkt_serialize_enter(&sfm->serializer); 1339 atomic_subtract_int(&sfm->mref_count, 1); 1340 if (sfm->mref_count == 0) { 1341 lwkt_serialize_exit(&sfm->serializer); 1342 freeit: 1343 get_mplock(); 1344 crit_enter(); 1345 m = sf_buf_page(sfm->sf); 1346 sf_buf_free(sfm->sf); 1347 vm_page_unwire(m, 0); 1348 if (m->wire_count == 0 && m->object == NULL) 1349 vm_page_try_to_free(m); 1350 crit_exit(); 1351 rel_mplock(); 1352 kfree(sfm, M_SENDFILE); 1353 } else { 1354 lwkt_serialize_exit(&sfm->serializer); 1355 } 1356 } 1357 } 1358 1359 /* 1360 * sendfile(2). 1361 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1362 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1363 * 1364 * Send a file specified by 'fd' and starting at 'offset' to a socket 1365 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1366 * nbytes == 0. Optionally add a header and/or trailer to the socket 1367 * output. If specified, write the total number of bytes sent into *sbytes. 1368 * 1369 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1370 * the headers to count against the remaining bytes to be sent from 1371 * the file descriptor. We may wish to implement a compatibility syscall 1372 * in the future. 1373 */ 1374 int 1375 sys_sendfile(struct sendfile_args *uap) 1376 { 1377 struct thread *td = curthread; 1378 struct proc *p = td->td_proc; 1379 struct file *fp; 1380 struct vnode *vp = NULL; 1381 struct sf_hdtr hdtr; 1382 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1383 struct uio auio; 1384 struct mbuf *mheader = NULL; 1385 off_t hdtr_size = 0, sbytes; 1386 int error, hbytes = 0, tbytes; 1387 1388 KKASSERT(p); 1389 1390 /* 1391 * Do argument checking. Must be a regular file in, stream 1392 * type and connected socket out, positive offset. 1393 */ 1394 fp = holdfp(p->p_fd, uap->fd, FREAD); 1395 if (fp == NULL) { 1396 return (EBADF); 1397 } 1398 if (fp->f_type != DTYPE_VNODE) { 1399 fdrop(fp); 1400 return (EINVAL); 1401 } 1402 vp = (struct vnode *)fp->f_data; 1403 vref(vp); 1404 fdrop(fp); 1405 1406 /* 1407 * If specified, get the pointer to the sf_hdtr struct for 1408 * any headers/trailers. 1409 */ 1410 if (uap->hdtr) { 1411 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1412 if (error) 1413 goto done; 1414 /* 1415 * Send any headers. 1416 */ 1417 if (hdtr.headers) { 1418 error = iovec_copyin(hdtr.headers, &iov, aiov, 1419 hdtr.hdr_cnt, &hbytes); 1420 if (error) 1421 goto done; 1422 auio.uio_iov = iov; 1423 auio.uio_iovcnt = hdtr.hdr_cnt; 1424 auio.uio_offset = 0; 1425 auio.uio_segflg = UIO_USERSPACE; 1426 auio.uio_rw = UIO_WRITE; 1427 auio.uio_td = td; 1428 auio.uio_resid = hbytes; 1429 1430 mheader = m_uiomove(&auio); 1431 1432 iovec_free(&iov, aiov); 1433 if (mheader == NULL) 1434 goto done; 1435 } 1436 } 1437 1438 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1439 &sbytes, uap->flags); 1440 if (error) 1441 goto done; 1442 1443 /* 1444 * Send trailers. Wimp out and use writev(2). 1445 */ 1446 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1447 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1448 hdtr.trl_cnt, &auio.uio_resid); 1449 if (error) 1450 goto done; 1451 auio.uio_iov = iov; 1452 auio.uio_iovcnt = hdtr.trl_cnt; 1453 auio.uio_offset = 0; 1454 auio.uio_segflg = UIO_USERSPACE; 1455 auio.uio_rw = UIO_WRITE; 1456 auio.uio_td = td; 1457 1458 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1459 1460 iovec_free(&iov, aiov); 1461 if (error) 1462 goto done; 1463 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1464 } 1465 1466 done: 1467 if (uap->sbytes != NULL) { 1468 sbytes += hdtr_size; 1469 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1470 } 1471 if (vp) 1472 vrele(vp); 1473 return (error); 1474 } 1475 1476 int 1477 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1478 struct mbuf *mheader, off_t *sbytes, int flags) 1479 { 1480 struct thread *td = curthread; 1481 struct proc *p = td->td_proc; 1482 struct vm_object *obj; 1483 struct socket *so; 1484 struct file *fp; 1485 struct mbuf *m; 1486 struct sf_buf *sf; 1487 struct sfbuf_mref *sfm; 1488 struct vm_page *pg; 1489 off_t off, xfsize; 1490 off_t hbytes = 0; 1491 int error = 0; 1492 1493 if (vp->v_type != VREG) { 1494 error = EINVAL; 1495 goto done0; 1496 } 1497 if ((obj = vp->v_object) == NULL) { 1498 error = EINVAL; 1499 goto done0; 1500 } 1501 error = holdsock(p->p_fd, sfd, &fp); 1502 if (error) 1503 goto done0; 1504 so = (struct socket *)fp->f_data; 1505 if (so->so_type != SOCK_STREAM) { 1506 error = EINVAL; 1507 goto done; 1508 } 1509 if ((so->so_state & SS_ISCONNECTED) == 0) { 1510 error = ENOTCONN; 1511 goto done; 1512 } 1513 if (offset < 0) { 1514 error = EINVAL; 1515 goto done; 1516 } 1517 1518 *sbytes = 0; 1519 /* 1520 * Protect against multiple writers to the socket. 1521 */ 1522 (void) sblock(&so->so_snd, M_WAITOK); 1523 1524 /* 1525 * Loop through the pages in the file, starting with the requested 1526 * offset. Get a file page (do I/O if necessary), map the file page 1527 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1528 * it on the socket. 1529 */ 1530 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1531 vm_pindex_t pindex; 1532 vm_offset_t pgoff; 1533 1534 pindex = OFF_TO_IDX(off); 1535 retry_lookup: 1536 /* 1537 * Calculate the amount to transfer. Not to exceed a page, 1538 * the EOF, or the passed in nbytes. 1539 */ 1540 xfsize = vp->v_filesize - off; 1541 if (xfsize > PAGE_SIZE) 1542 xfsize = PAGE_SIZE; 1543 pgoff = (vm_offset_t)(off & PAGE_MASK); 1544 if (PAGE_SIZE - pgoff < xfsize) 1545 xfsize = PAGE_SIZE - pgoff; 1546 if (nbytes && xfsize > (nbytes - *sbytes)) 1547 xfsize = nbytes - *sbytes; 1548 if (xfsize <= 0) 1549 break; 1550 /* 1551 * Optimize the non-blocking case by looking at the socket space 1552 * before going to the extra work of constituting the sf_buf. 1553 */ 1554 if ((fp->f_flag & FNONBLOCK) && sbspace(&so->so_snd) <= 0) { 1555 if (so->so_state & SS_CANTSENDMORE) 1556 error = EPIPE; 1557 else 1558 error = EAGAIN; 1559 sbunlock(&so->so_snd); 1560 goto done; 1561 } 1562 /* 1563 * Attempt to look up the page. 1564 * 1565 * Allocate if not found, wait and loop if busy, then 1566 * wire the page. critical section protection is 1567 * required to maintain the object association (an 1568 * interrupt can free the page) through to the 1569 * vm_page_wire() call. 1570 */ 1571 crit_enter(); 1572 pg = vm_page_lookup(obj, pindex); 1573 if (pg == NULL) { 1574 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1575 if (pg == NULL) { 1576 vm_wait(); 1577 crit_exit(); 1578 goto retry_lookup; 1579 } 1580 vm_page_wakeup(pg); 1581 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1582 crit_exit(); 1583 goto retry_lookup; 1584 } 1585 vm_page_wire(pg); 1586 crit_exit(); 1587 1588 /* 1589 * If page is not valid for what we need, initiate I/O 1590 */ 1591 1592 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1593 struct uio auio; 1594 struct iovec aiov; 1595 int bsize; 1596 1597 /* 1598 * Ensure that our page is still around when the I/O 1599 * completes. 1600 */ 1601 vm_page_io_start(pg); 1602 1603 /* 1604 * Get the page from backing store. 1605 */ 1606 bsize = vp->v_mount->mnt_stat.f_iosize; 1607 auio.uio_iov = &aiov; 1608 auio.uio_iovcnt = 1; 1609 aiov.iov_base = 0; 1610 aiov.iov_len = MAXBSIZE; 1611 auio.uio_resid = MAXBSIZE; 1612 auio.uio_offset = trunc_page(off); 1613 auio.uio_segflg = UIO_NOCOPY; 1614 auio.uio_rw = UIO_READ; 1615 auio.uio_td = td; 1616 vn_lock(vp, LK_SHARED | LK_RETRY); 1617 error = VOP_READ(vp, &auio, 1618 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1619 p->p_ucred); 1620 vn_unlock(vp); 1621 vm_page_flag_clear(pg, PG_ZERO); 1622 vm_page_io_finish(pg); 1623 if (error) { 1624 crit_enter(); 1625 vm_page_unwire(pg, 0); 1626 vm_page_try_to_free(pg); 1627 crit_exit(); 1628 sbunlock(&so->so_snd); 1629 goto done; 1630 } 1631 } 1632 1633 1634 /* 1635 * Get a sendfile buf. We usually wait as long as necessary, 1636 * but this wait can be interrupted. 1637 */ 1638 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) { 1639 crit_enter(); 1640 vm_page_unwire(pg, 0); 1641 vm_page_try_to_free(pg); 1642 crit_exit(); 1643 sbunlock(&so->so_snd); 1644 error = EINTR; 1645 goto done; 1646 } 1647 1648 /* 1649 * Get an mbuf header and set it up as having external storage. 1650 */ 1651 MGETHDR(m, MB_WAIT, MT_DATA); 1652 if (m == NULL) { 1653 error = ENOBUFS; 1654 sf_buf_free(sf); 1655 sbunlock(&so->so_snd); 1656 goto done; 1657 } 1658 1659 /* 1660 * sfm is a temporary hack, use a per-cpu cache for this. 1661 */ 1662 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK); 1663 sfm->sf = sf; 1664 sfm->mref_count = 1; 1665 lwkt_serialize_init(&sfm->serializer); 1666 1667 m->m_ext.ext_free = sf_buf_mfree; 1668 m->m_ext.ext_ref = sf_buf_mref; 1669 m->m_ext.ext_arg = sfm; 1670 m->m_ext.ext_buf = (void *)sf->kva; 1671 m->m_ext.ext_size = PAGE_SIZE; 1672 m->m_data = (char *) sf->kva + pgoff; 1673 m->m_flags |= M_EXT; 1674 m->m_pkthdr.len = m->m_len = xfsize; 1675 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1676 1677 if (mheader != NULL) { 1678 hbytes = mheader->m_pkthdr.len; 1679 mheader->m_pkthdr.len += m->m_pkthdr.len; 1680 m_cat(mheader, m); 1681 m = mheader; 1682 mheader = NULL; 1683 } else 1684 hbytes = 0; 1685 1686 /* 1687 * Add the buffer to the socket buffer chain. 1688 */ 1689 crit_enter(); 1690 retry_space: 1691 /* 1692 * Make sure that the socket is still able to take more data. 1693 * CANTSENDMORE being true usually means that the connection 1694 * was closed. so_error is true when an error was sensed after 1695 * a previous send. 1696 * The state is checked after the page mapping and buffer 1697 * allocation above since those operations may block and make 1698 * any socket checks stale. From this point forward, nothing 1699 * blocks before the pru_send (or more accurately, any blocking 1700 * results in a loop back to here to re-check). 1701 */ 1702 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1703 if (so->so_state & SS_CANTSENDMORE) { 1704 error = EPIPE; 1705 } else { 1706 error = so->so_error; 1707 so->so_error = 0; 1708 } 1709 m_freem(m); 1710 sbunlock(&so->so_snd); 1711 crit_exit(); 1712 goto done; 1713 } 1714 /* 1715 * Wait for socket space to become available. We do this just 1716 * after checking the connection state above in order to avoid 1717 * a race condition with sbwait(). 1718 */ 1719 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { 1720 if (fp->f_flag & FNONBLOCK) { 1721 m_freem(m); 1722 sbunlock(&so->so_snd); 1723 crit_exit(); 1724 error = EAGAIN; 1725 goto done; 1726 } 1727 error = sbwait(&so->so_snd); 1728 /* 1729 * An error from sbwait usually indicates that we've 1730 * been interrupted by a signal. If we've sent anything 1731 * then return bytes sent, otherwise return the error. 1732 */ 1733 if (error) { 1734 m_freem(m); 1735 sbunlock(&so->so_snd); 1736 crit_exit(); 1737 goto done; 1738 } 1739 goto retry_space; 1740 } 1741 error = so_pru_send(so, 0, m, NULL, NULL, td); 1742 crit_exit(); 1743 if (error) { 1744 sbunlock(&so->so_snd); 1745 goto done; 1746 } 1747 } 1748 if (mheader != NULL) { 1749 *sbytes += mheader->m_pkthdr.len; 1750 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1751 mheader = NULL; 1752 } 1753 sbunlock(&so->so_snd); 1754 1755 done: 1756 fdrop(fp); 1757 done0: 1758 if (mheader != NULL) 1759 m_freem(mheader); 1760 return (error); 1761 } 1762 1763 int 1764 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1765 { 1766 #ifdef SCTP 1767 struct thread *td = curthread; 1768 struct proc *p = td->td_proc; 1769 struct file *lfp = NULL; 1770 struct file *nfp = NULL; 1771 int error; 1772 struct socket *head, *so; 1773 caddr_t assoc_id; 1774 int fd; 1775 short fflag; /* type must match fp->f_flag */ 1776 1777 assoc_id = uap->name; 1778 error = holdsock(p->p_fd, uap->sd, &lfp); 1779 if (error) { 1780 return (error); 1781 } 1782 crit_enter(); 1783 head = (struct socket *)lfp->f_data; 1784 error = sctp_can_peel_off(head, assoc_id); 1785 if (error) { 1786 crit_exit(); 1787 goto done; 1788 } 1789 /* 1790 * At this point we know we do have a assoc to pull 1791 * we proceed to get the fd setup. This may block 1792 * but that is ok. 1793 */ 1794 1795 fflag = lfp->f_flag; 1796 error = falloc(p, &nfp, &fd); 1797 if (error) { 1798 /* 1799 * Probably ran out of file descriptors. Put the 1800 * unaccepted connection back onto the queue and 1801 * do another wakeup so some other process might 1802 * have a chance at it. 1803 */ 1804 crit_exit(); 1805 goto done; 1806 } 1807 uap->sysmsg_result = fd; 1808 1809 so = sctp_get_peeloff(head, assoc_id, &error); 1810 if (so == NULL) { 1811 /* 1812 * Either someone else peeled it off OR 1813 * we can't get a socket. 1814 */ 1815 goto noconnection; 1816 } 1817 so->so_state &= ~SS_COMP; 1818 so->so_state &= ~SS_NOFDREF; 1819 so->so_head = NULL; 1820 if (head->so_sigio != NULL) 1821 fsetown(fgetown(head->so_sigio), &so->so_sigio); 1822 1823 nfp->f_type = DTYPE_SOCKET; 1824 nfp->f_flag = fflag; 1825 nfp->f_ops = &socketops; 1826 nfp->f_data = so; 1827 1828 noconnection: 1829 /* 1830 * Assign the file pointer to the reserved descriptor, or clear 1831 * the reserved descriptor if an error occured. 1832 */ 1833 if (error) 1834 fsetfd(p, NULL, fd); 1835 else 1836 fsetfd(p, nfp, fd); 1837 crit_exit(); 1838 /* 1839 * Release explicitly held references before returning. 1840 */ 1841 done: 1842 if (nfp != NULL) 1843 fdrop(nfp); 1844 fdrop(lfp); 1845 return (error); 1846 #else /* SCTP */ 1847 return(EOPNOTSUPP); 1848 #endif /* SCTP */ 1849 } 1850