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 1.41 2004/08/24 21:53:38 dillon Exp $ 39 */ 40 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/sysproto.h> 47 #include <sys/malloc.h> 48 #include <sys/filedesc.h> 49 #include <sys/event.h> 50 #include <sys/proc.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filio.h> 54 #include <sys/kern_syscall.h> 55 #include <sys/mbuf.h> 56 #include <sys/protosw.h> 57 #include <sys/sfbuf.h> 58 #include <sys/socket.h> 59 #include <sys/socketvar.h> 60 #include <sys/socketops.h> 61 #include <sys/uio.h> 62 #include <sys/vnode.h> 63 #include <sys/lock.h> 64 #include <sys/mount.h> 65 #ifdef KTRACE 66 #include <sys/ktrace.h> 67 #endif 68 #include <vm/vm.h> 69 #include <vm/vm_object.h> 70 #include <vm/vm_page.h> 71 #include <vm/vm_pageout.h> 72 #include <vm/vm_kern.h> 73 #include <vm/vm_extern.h> 74 #include <sys/file2.h> 75 #include <sys/signalvar.h> 76 77 #include <sys/thread2.h> 78 #include <sys/msgport2.h> 79 80 struct sfbuf_mref { 81 struct sf_buf *sf; 82 int mref_count; 83 }; 84 85 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures"); 86 87 /* 88 * System call interface to the socket abstraction. 89 */ 90 91 extern struct fileops socketops; 92 93 /* 94 * socket_args(int domain, int type, int protocol) 95 */ 96 int 97 kern_socket(int domain, int type, int protocol, int *res) 98 { 99 struct thread *td = curthread; 100 struct proc *p = td->td_proc; 101 struct filedesc *fdp; 102 struct socket *so; 103 struct file *fp; 104 int fd, error; 105 106 KKASSERT(p); 107 fdp = p->p_fd; 108 109 error = falloc(p, &fp, &fd); 110 if (error) 111 return (error); 112 fhold(fp); 113 error = socreate(domain, &so, type, protocol, td); 114 if (error) { 115 if (fdp->fd_ofiles[fd] == fp) { 116 fdp->fd_ofiles[fd] = NULL; 117 fdrop(fp, td); 118 } 119 } else { 120 fp->f_data = (caddr_t)so; 121 fp->f_flag = FREAD|FWRITE; 122 fp->f_ops = &socketops; 123 fp->f_type = DTYPE_SOCKET; 124 *res = fd; 125 } 126 fdrop(fp, td); 127 return (error); 128 } 129 130 int 131 socket(struct socket_args *uap) 132 { 133 int error; 134 135 error = kern_socket(uap->domain, uap->type, uap->protocol, 136 &uap->sysmsg_result); 137 138 return (error); 139 } 140 141 int 142 kern_bind(int s, struct sockaddr *sa) 143 { 144 struct thread *td = curthread; 145 struct proc *p = td->td_proc; 146 struct file *fp; 147 int error; 148 149 KKASSERT(p); 150 error = holdsock(p->p_fd, s, &fp); 151 if (error) 152 return (error); 153 error = sobind((struct socket *)fp->f_data, sa, td); 154 fdrop(fp, td); 155 return (error); 156 } 157 158 /* 159 * bind_args(int s, caddr_t name, int namelen) 160 */ 161 int 162 bind(struct bind_args *uap) 163 { 164 struct sockaddr *sa; 165 int error; 166 167 error = getsockaddr(&sa, uap->name, uap->namelen); 168 if (error) 169 return (error); 170 error = kern_bind(uap->s, sa); 171 FREE(sa, M_SONAME); 172 173 return (error); 174 } 175 176 int 177 kern_listen(int s, int backlog) 178 { 179 struct thread *td = curthread; 180 struct proc *p = td->td_proc; 181 struct file *fp; 182 int error; 183 184 KKASSERT(p); 185 error = holdsock(p->p_fd, s, &fp); 186 if (error) 187 return (error); 188 error = solisten((struct socket *)fp->f_data, backlog, td); 189 fdrop(fp, td); 190 return(error); 191 } 192 193 /* 194 * listen_args(int s, int backlog) 195 */ 196 int 197 listen(struct listen_args *uap) 198 { 199 int error; 200 201 error = kern_listen(uap->s, uap->backlog); 202 return (error); 203 } 204 205 /* 206 * Returns the accepted socket as well. 207 */ 208 static boolean_t 209 soaccept_predicate(struct netmsg *msg0) 210 { 211 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 212 struct socket *head = msg->nm_so; 213 214 if (head->so_error != 0) { 215 msg->nm_lmsg.ms_error = head->so_error; 216 return (TRUE); 217 } 218 if (!TAILQ_EMPTY(&head->so_comp)) { 219 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 220 msg->nm_so = TAILQ_FIRST(&head->so_comp); 221 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list); 222 head->so_qlen--; 223 224 msg->nm_lmsg.ms_error = 0; 225 return (TRUE); 226 } 227 if (head->so_state & SS_CANTRCVMORE) { 228 msg->nm_lmsg.ms_error = ECONNABORTED; 229 return (TRUE); 230 } 231 if (head->so_state & SS_NBIO) { 232 msg->nm_lmsg.ms_error = EWOULDBLOCK; 233 return (TRUE); 234 } 235 236 return (FALSE); 237 } 238 239 /* 240 * The second argument to kern_accept() is a handle to a struct sockaddr. 241 * This allows kern_accept() to return a pointer to an allocated struct 242 * sockaddr which must be freed later with FREE(). The caller must 243 * initialize *name to NULL. 244 */ 245 int 246 kern_accept(int s, struct sockaddr **name, int *namelen, int *res) 247 { 248 struct thread *td = curthread; 249 struct proc *p = td->td_proc; 250 struct filedesc *fdp = p->p_fd; 251 struct file *lfp = NULL; 252 struct file *nfp = NULL; 253 struct sockaddr *sa; 254 struct socket *head, *so; 255 struct netmsg_so_notify msg; 256 lwkt_port_t port; 257 int fd; 258 u_int fflag; /* type must match fp->f_flag */ 259 int error, tmp; 260 261 if (name && namelen && *namelen < 0) 262 return (EINVAL); 263 264 error = holdsock(fdp, s, &lfp); 265 if (error) 266 return (error); 267 268 error = falloc(p, &nfp, &fd); 269 if (error) { /* Probably ran out of file descriptors. */ 270 *res = -1; 271 fdrop(lfp, td); 272 return (error); 273 } 274 fhold(nfp); 275 *res = fd; 276 277 head = (struct socket *)lfp->f_data; 278 if ((head->so_options & SO_ACCEPTCONN) == 0) { 279 error = EINVAL; 280 goto done; 281 } 282 283 /* optimize for uniprocessor case later XXX JH */ 284 port = head->so_proto->pr_mport(head, NULL, PRU_PRED); 285 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport, 286 MSGF_PCATCH | MSGF_ABORTABLE, 287 lwkt_cmd_func(netmsg_so_notify), 288 lwkt_cmd_func(netmsg_so_notify_abort)); 289 msg.nm_predicate = soaccept_predicate; 290 msg.nm_so = head; 291 msg.nm_etype = NM_REVENT; 292 error = lwkt_domsg(port, &msg.nm_lmsg); 293 if (error) 294 goto done; 295 296 /* 297 * At this point we have the connection that's ready to be accepted. 298 */ 299 so = msg.nm_so; 300 301 fflag = lfp->f_flag; 302 303 /* connection has been removed from the listen queue */ 304 KNOTE(&head->so_rcv.sb_sel.si_note, 0); 305 306 so->so_state &= ~SS_COMP; 307 so->so_head = NULL; 308 if (head->so_sigio != NULL) 309 fsetown(fgetown(head->so_sigio), &so->so_sigio); 310 311 nfp->f_data = (caddr_t)so; 312 nfp->f_flag = fflag; 313 nfp->f_ops = &socketops; 314 nfp->f_type = DTYPE_SOCKET; 315 /* Sync socket nonblocking/async state with file flags */ 316 tmp = fflag & FNONBLOCK; 317 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td); 318 tmp = fflag & FASYNC; 319 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td); 320 321 sa = NULL; 322 error = soaccept(so, &sa); 323 324 /* 325 * Set the returned name and namelen as applicable. Set the returned 326 * namelen to 0 for older code which might ignore the return value 327 * from accept. 328 */ 329 if (error == 0) { 330 if (sa && name && namelen) { 331 if (*namelen > sa->sa_len) 332 *namelen = sa->sa_len; 333 *name = sa; 334 } else { 335 if (sa) 336 FREE(sa, M_SONAME); 337 } 338 } 339 340 done: 341 /* 342 * close the new descriptor, assuming someone hasn't ripped it 343 * out from under us. Note that *res is normally ignored if an 344 * error is returned but a syscall message will still have access 345 * to the result code. 346 */ 347 if (error) { 348 *res = -1; 349 if (fdp->fd_ofiles[fd] == nfp) { 350 fdp->fd_ofiles[fd] = NULL; 351 fdrop(nfp, td); 352 } 353 } 354 355 /* 356 * Release explicitly held references before returning. 357 */ 358 if (nfp != NULL) 359 fdrop(nfp, td); 360 fdrop(lfp, td); 361 return (error); 362 } 363 364 /* 365 * accept_args(int s, caddr_t name, int *anamelen) 366 */ 367 int 368 accept(struct accept_args *uap) 369 { 370 struct sockaddr *sa = NULL; 371 int sa_len; 372 int error; 373 374 if (uap->name) { 375 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 376 if (error) 377 return (error); 378 379 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result); 380 381 if (error == 0) 382 error = copyout(sa, uap->name, sa_len); 383 if (error == 0) { 384 error = copyout(&sa_len, uap->anamelen, 385 sizeof(*uap->anamelen)); 386 } 387 if (sa) 388 FREE(sa, M_SONAME); 389 } else { 390 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result); 391 } 392 return (error); 393 } 394 395 /* 396 * Returns TRUE if predicate satisfied. 397 */ 398 static boolean_t 399 soconnected_predicate(struct netmsg *msg0) 400 { 401 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0; 402 struct socket *so = msg->nm_so; 403 404 /* check predicate */ 405 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 406 msg->nm_lmsg.ms_error = so->so_error; 407 return (TRUE); 408 } 409 410 return (FALSE); 411 } 412 413 int 414 kern_connect(int s, struct sockaddr *sa) 415 { 416 struct thread *td = curthread; 417 struct proc *p = td->td_proc; 418 struct file *fp; 419 struct socket *so; 420 int error; 421 422 error = holdsock(p->p_fd, s, &fp); 423 if (error) 424 return (error); 425 so = (struct socket *)fp->f_data; 426 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 427 error = EALREADY; 428 goto done; 429 } 430 error = soconnect(so, sa, td); 431 if (error) 432 goto bad; 433 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 434 error = EINPROGRESS; 435 goto done; 436 } 437 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 438 struct netmsg_so_notify msg; 439 lwkt_port_t port; 440 441 port = so->so_proto->pr_mport(so, sa, PRU_PRED); 442 lwkt_initmsg(&msg.nm_lmsg, 443 &curthread->td_msgport, 444 MSGF_PCATCH | MSGF_ABORTABLE, 445 lwkt_cmd_func(netmsg_so_notify), 446 lwkt_cmd_func(netmsg_so_notify_abort)); 447 msg.nm_predicate = soconnected_predicate; 448 msg.nm_so = so; 449 msg.nm_etype = NM_REVENT; 450 error = lwkt_domsg(port, &msg.nm_lmsg); 451 } 452 if (error == 0) { 453 error = so->so_error; 454 so->so_error = 0; 455 } 456 bad: 457 so->so_state &= ~SS_ISCONNECTING; 458 if (error == ERESTART) 459 error = EINTR; 460 done: 461 fdrop(fp, td); 462 return (error); 463 } 464 465 /* 466 * connect_args(int s, caddr_t name, int namelen) 467 */ 468 int 469 connect(struct connect_args *uap) 470 { 471 struct sockaddr *sa; 472 int error; 473 474 error = getsockaddr(&sa, uap->name, uap->namelen); 475 if (error) 476 return (error); 477 error = kern_connect(uap->s, sa); 478 FREE(sa, M_SONAME); 479 480 return (error); 481 } 482 483 int 484 kern_socketpair(int domain, int type, int protocol, int *sv) 485 { 486 struct thread *td = curthread; 487 struct proc *p = td->td_proc; 488 struct filedesc *fdp; 489 struct file *fp1, *fp2; 490 struct socket *so1, *so2; 491 int fd, error; 492 493 KKASSERT(p); 494 fdp = p->p_fd; 495 error = socreate(domain, &so1, type, protocol, td); 496 if (error) 497 return (error); 498 error = socreate(domain, &so2, type, protocol, td); 499 if (error) 500 goto free1; 501 error = falloc(p, &fp1, &fd); 502 if (error) 503 goto free2; 504 fhold(fp1); 505 sv[0] = fd; 506 fp1->f_data = (caddr_t)so1; 507 error = falloc(p, &fp2, &fd); 508 if (error) 509 goto free3; 510 fhold(fp2); 511 fp2->f_data = (caddr_t)so2; 512 sv[1] = fd; 513 error = soconnect2(so1, so2); 514 if (error) 515 goto free4; 516 if (type == SOCK_DGRAM) { 517 /* 518 * Datagram socket connection is asymmetric. 519 */ 520 error = soconnect2(so2, so1); 521 if (error) 522 goto free4; 523 } 524 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 525 fp1->f_ops = fp2->f_ops = &socketops; 526 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 527 fdrop(fp1, td); 528 fdrop(fp2, td); 529 return (error); 530 free4: 531 if (fdp->fd_ofiles[sv[1]] == fp2) { 532 fdp->fd_ofiles[sv[1]] = NULL; 533 fdrop(fp2, td); 534 } 535 fdrop(fp2, td); 536 free3: 537 if (fdp->fd_ofiles[sv[0]] == fp1) { 538 fdp->fd_ofiles[sv[0]] = NULL; 539 fdrop(fp1, td); 540 } 541 fdrop(fp1, td); 542 free2: 543 (void)soclose(so2); 544 free1: 545 (void)soclose(so1); 546 return (error); 547 } 548 549 /* 550 * socketpair(int domain, int type, int protocol, int *rsv) 551 */ 552 int 553 socketpair(struct socketpair_args *uap) 554 { 555 int error, sockv[2]; 556 557 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 558 559 if (error == 0) 560 error = copyout(sockv, uap->rsv, sizeof(sockv)); 561 return (error); 562 } 563 564 int 565 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 566 struct mbuf *control, int flags, int *res) 567 { 568 struct thread *td = curthread; 569 struct proc *p = td->td_proc; 570 struct file *fp; 571 int len, error; 572 struct socket *so; 573 #ifdef KTRACE 574 struct iovec *ktriov = NULL; 575 struct uio ktruio; 576 #endif 577 578 error = holdsock(p->p_fd, s, &fp); 579 if (error) 580 return (error); 581 if (auio->uio_resid < 0) { 582 error = EINVAL; 583 goto done; 584 } 585 #ifdef KTRACE 586 if (KTRPOINT(td, KTR_GENIO)) { 587 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 588 589 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 590 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 591 ktruio = *auio; 592 } 593 #endif 594 len = auio->uio_resid; 595 so = (struct socket *)fp->f_data; 596 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 597 if (error) { 598 if (auio->uio_resid != len && (error == ERESTART || 599 error == EINTR || error == EWOULDBLOCK)) 600 error = 0; 601 if (error == EPIPE) 602 psignal(p, SIGPIPE); 603 } 604 #ifdef KTRACE 605 if (ktriov != NULL) { 606 if (error == 0) { 607 ktruio.uio_iov = ktriov; 608 ktruio.uio_resid = len - auio->uio_resid; 609 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error); 610 } 611 FREE(ktriov, M_TEMP); 612 } 613 #endif 614 if (error == 0) 615 *res = len - auio->uio_resid; 616 done: 617 fdrop(fp, td); 618 return (error); 619 } 620 621 /* 622 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 623 */ 624 int 625 sendto(struct sendto_args *uap) 626 { 627 struct thread *td = curthread; 628 struct uio auio; 629 struct iovec aiov; 630 struct sockaddr *sa = NULL; 631 int error; 632 633 if (uap->to) { 634 error = getsockaddr(&sa, uap->to, uap->tolen); 635 if (error) 636 return (error); 637 } 638 aiov.iov_base = uap->buf; 639 aiov.iov_len = uap->len; 640 auio.uio_iov = &aiov; 641 auio.uio_iovcnt = 1; 642 auio.uio_offset = 0; 643 auio.uio_resid = uap->len; 644 auio.uio_segflg = UIO_USERSPACE; 645 auio.uio_rw = UIO_WRITE; 646 auio.uio_td = td; 647 648 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 649 &uap->sysmsg_result); 650 651 if (sa) 652 FREE(sa, M_SONAME); 653 return (error); 654 } 655 656 /* 657 * sendmsg_args(int s, caddr_t msg, int flags) 658 */ 659 int 660 sendmsg(struct sendmsg_args *uap) 661 { 662 struct thread *td = curthread; 663 struct msghdr msg; 664 struct uio auio; 665 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 666 struct sockaddr *sa = NULL; 667 struct mbuf *control = NULL; 668 int error; 669 670 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 671 if (error) 672 return (error); 673 674 /* 675 * Conditionally copyin msg.msg_name. 676 */ 677 if (msg.msg_name) { 678 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 679 if (error) 680 return (error); 681 } 682 683 /* 684 * Populate auio. 685 */ 686 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 687 &auio.uio_resid); 688 if (error) 689 goto cleanup; 690 auio.uio_iov = iov; 691 auio.uio_iovcnt = msg.msg_iovlen; 692 auio.uio_offset = 0; 693 auio.uio_segflg = UIO_USERSPACE; 694 auio.uio_rw = UIO_WRITE; 695 auio.uio_td = td; 696 697 /* 698 * Conditionally copyin msg.msg_control. 699 */ 700 if (msg.msg_control) { 701 if (msg.msg_controllen < sizeof(struct cmsghdr) || 702 msg.msg_controllen > MLEN) { 703 error = EINVAL; 704 goto cleanup; 705 } 706 control = m_get(MB_WAIT, MT_CONTROL); 707 if (control == NULL) { 708 error = ENOBUFS; 709 goto cleanup; 710 } 711 control->m_len = msg.msg_controllen; 712 error = copyin(msg.msg_control, mtod(control, caddr_t), 713 msg.msg_controllen); 714 if (error) { 715 m_free(control); 716 goto cleanup; 717 } 718 } 719 720 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 721 &uap->sysmsg_result); 722 723 cleanup: 724 if (sa) 725 FREE(sa, M_SONAME); 726 iovec_free(&iov, aiov); 727 return (error); 728 } 729 730 /* 731 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 732 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 733 * Don't forget to FREE() and m_free() these if they are returned. 734 */ 735 int 736 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 737 struct mbuf **control, int *flags, int *res) 738 { 739 struct thread *td = curthread; 740 struct proc *p = td->td_proc; 741 struct file *fp; 742 int len, error; 743 struct socket *so; 744 #ifdef KTRACE 745 struct iovec *ktriov = NULL; 746 struct uio ktruio; 747 #endif 748 749 error = holdsock(p->p_fd, s, &fp); 750 if (error) 751 return (error); 752 if (auio->uio_resid < 0) { 753 error = EINVAL; 754 goto done; 755 } 756 #ifdef KTRACE 757 if (KTRPOINT(td, KTR_GENIO)) { 758 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 759 760 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 761 bcopy(auio->uio_iov, ktriov, iovlen); 762 ktruio = *auio; 763 } 764 #endif 765 len = auio->uio_resid; 766 so = (struct socket *)fp->f_data; 767 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 768 if (error) { 769 if (auio->uio_resid != len && (error == ERESTART || 770 error == EINTR || error == EWOULDBLOCK)) 771 error = 0; 772 } 773 #ifdef KTRACE 774 if (ktriov != NULL) { 775 if (error == 0) { 776 ktruio.uio_iov = ktriov; 777 ktruio.uio_resid = len - auio->uio_resid; 778 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error); 779 } 780 FREE(ktriov, M_TEMP); 781 } 782 #endif 783 if (error == 0) 784 *res = len - auio->uio_resid; 785 done: 786 fdrop(fp, td); 787 return (error); 788 } 789 790 /* 791 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 792 * caddr_t from, int *fromlenaddr) 793 */ 794 int 795 recvfrom(struct recvfrom_args *uap) 796 { 797 struct thread *td = curthread; 798 struct uio auio; 799 struct iovec aiov; 800 struct sockaddr *sa = NULL; 801 int error, fromlen; 802 803 if (uap->from && uap->fromlenaddr) { 804 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 805 if (error) 806 return (error); 807 if (fromlen < 0) 808 return (EINVAL); 809 } else { 810 fromlen = 0; 811 } 812 aiov.iov_base = uap->buf; 813 aiov.iov_len = uap->len; 814 auio.uio_iov = &aiov; 815 auio.uio_iovcnt = 1; 816 auio.uio_offset = 0; 817 auio.uio_resid = uap->len; 818 auio.uio_segflg = UIO_USERSPACE; 819 auio.uio_rw = UIO_READ; 820 auio.uio_td = td; 821 822 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 823 &uap->flags, &uap->sysmsg_result); 824 825 if (error == 0 && uap->from) { 826 /* note: sa may still be NULL */ 827 if (sa) { 828 fromlen = MIN(fromlen, sa->sa_len); 829 error = copyout(sa, uap->from, fromlen); 830 } else { 831 fromlen = 0; 832 } 833 if (error == 0) { 834 error = copyout(&fromlen, uap->fromlenaddr, 835 sizeof(fromlen)); 836 } 837 } 838 if (sa) 839 FREE(sa, M_SONAME); 840 841 return (error); 842 } 843 844 /* 845 * recvmsg_args(int s, struct msghdr *msg, int flags) 846 */ 847 int 848 recvmsg(struct recvmsg_args *uap) 849 { 850 struct thread *td = curthread; 851 struct msghdr msg; 852 struct uio auio; 853 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 854 struct mbuf *m, *control = NULL; 855 struct sockaddr *sa = NULL; 856 caddr_t ctlbuf; 857 socklen_t *ufromlenp, *ucontrollenp; 858 int error, fromlen, controllen, len, flags, *uflagsp; 859 860 /* 861 * This copyin handles everything except the iovec. 862 */ 863 error = copyin(uap->msg, &msg, sizeof(msg)); 864 if (error) 865 return (error); 866 867 if (msg.msg_name && msg.msg_namelen < 0) 868 return (EINVAL); 869 if (msg.msg_control && msg.msg_controllen < 0) 870 return (EINVAL); 871 872 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 873 msg_namelen)); 874 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 875 msg_controllen)); 876 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 877 msg_flags)); 878 879 /* 880 * Populate auio. 881 */ 882 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 883 &auio.uio_resid); 884 if (error) 885 return (error); 886 auio.uio_iov = iov; 887 auio.uio_iovcnt = msg.msg_iovlen; 888 auio.uio_offset = 0; 889 auio.uio_segflg = UIO_USERSPACE; 890 auio.uio_rw = UIO_READ; 891 auio.uio_td = td; 892 893 flags = msg.msg_flags; 894 895 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio, 896 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result); 897 898 /* 899 * Conditionally copyout the name and populate the namelen field. 900 */ 901 if (error == 0 && msg.msg_name) { 902 fromlen = MIN(msg.msg_namelen, sa->sa_len); 903 error = copyout(sa, msg.msg_name, fromlen); 904 if (error == 0) 905 error = copyout(&fromlen, ufromlenp, 906 sizeof(*ufromlenp)); 907 } 908 909 /* 910 * Copyout msg.msg_control and msg.msg_controllen. 911 */ 912 if (error == 0 && msg.msg_control) { 913 len = msg.msg_controllen; 914 m = control; 915 ctlbuf = (caddr_t)msg.msg_control; 916 917 while(m && len > 0) { 918 unsigned int tocopy; 919 920 if (len >= m->m_len) { 921 tocopy = m->m_len; 922 } else { 923 msg.msg_flags |= MSG_CTRUNC; 924 tocopy = len; 925 } 926 927 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 928 if (error) 929 goto cleanup; 930 931 ctlbuf += tocopy; 932 len -= tocopy; 933 m = m->m_next; 934 } 935 controllen = ctlbuf - (caddr_t)msg.msg_control; 936 error = copyout(&controllen, ucontrollenp, 937 sizeof(*ucontrollenp)); 938 } 939 940 if (error == 0) 941 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 942 943 cleanup: 944 if (sa) 945 FREE(sa, M_SONAME); 946 iovec_free(&iov, aiov); 947 if (control) 948 m_freem(control); 949 return (error); 950 } 951 952 /* 953 * shutdown_args(int s, int how) 954 */ 955 int 956 kern_shutdown(int s, int how) 957 { 958 struct thread *td = curthread; 959 struct proc *p = td->td_proc; 960 struct file *fp; 961 int error; 962 963 KKASSERT(p); 964 error = holdsock(p->p_fd, s, &fp); 965 if (error) 966 return (error); 967 error = soshutdown((struct socket *)fp->f_data, how); 968 fdrop(fp, td); 969 return(error); 970 } 971 972 int 973 shutdown(struct shutdown_args *uap) 974 { 975 int error; 976 977 error = kern_shutdown(uap->s, uap->how); 978 979 return (error); 980 } 981 982 /* 983 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 984 * in kernel pointer instead of a userland pointer. This allows us 985 * to manipulate socket options in the emulation code. 986 */ 987 int 988 kern_setsockopt(int s, struct sockopt *sopt) 989 { 990 struct thread *td = curthread; 991 struct proc *p = td->td_proc; 992 struct file *fp; 993 int error; 994 995 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 996 return (EFAULT); 997 if (sopt->sopt_valsize < 0) 998 return (EINVAL); 999 1000 error = holdsock(p->p_fd, s, &fp); 1001 if (error) 1002 return (error); 1003 1004 error = sosetopt((struct socket *)fp->f_data, sopt); 1005 fdrop(fp, td); 1006 return (error); 1007 } 1008 1009 /* 1010 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1011 */ 1012 int 1013 setsockopt(struct setsockopt_args *uap) 1014 { 1015 struct thread *td = curthread; 1016 struct sockopt sopt; 1017 int error; 1018 1019 sopt.sopt_dir = SOPT_SET; 1020 sopt.sopt_level = uap->level; 1021 sopt.sopt_name = uap->name; 1022 sopt.sopt_val = uap->val; 1023 sopt.sopt_valsize = uap->valsize; 1024 sopt.sopt_td = td; 1025 1026 error = kern_setsockopt(uap->s, &sopt); 1027 return(error); 1028 } 1029 1030 /* 1031 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1032 * in kernel pointer instead of a userland pointer. This allows us 1033 * to manipulate socket options in the emulation code. 1034 */ 1035 int 1036 kern_getsockopt(int s, struct sockopt *sopt) 1037 { 1038 struct thread *td = curthread; 1039 struct proc *p = td->td_proc; 1040 struct file *fp; 1041 int error; 1042 1043 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0) 1044 return (EFAULT); 1045 if (sopt->sopt_valsize < 0) 1046 return (EINVAL); 1047 1048 error = holdsock(p->p_fd, s, &fp); 1049 if (error) 1050 return (error); 1051 1052 error = sogetopt((struct socket *)fp->f_data, sopt); 1053 fdrop(fp, td); 1054 return (error); 1055 } 1056 1057 /* 1058 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1059 */ 1060 int 1061 getsockopt(struct getsockopt_args *uap) 1062 { 1063 struct thread *td = curthread; 1064 struct sockopt sopt; 1065 int error, valsize; 1066 1067 if (uap->val) { 1068 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1069 if (error) 1070 return (error); 1071 if (valsize < 0) 1072 return (EINVAL); 1073 } else { 1074 valsize = 0; 1075 } 1076 1077 sopt.sopt_dir = SOPT_GET; 1078 sopt.sopt_level = uap->level; 1079 sopt.sopt_name = uap->name; 1080 sopt.sopt_val = uap->val; 1081 sopt.sopt_valsize = valsize; 1082 sopt.sopt_td = td; 1083 1084 error = kern_getsockopt(uap->s, &sopt); 1085 if (error == 0) { 1086 valsize = sopt.sopt_valsize; 1087 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1088 } 1089 return (error); 1090 } 1091 1092 /* 1093 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1094 * This allows kern_getsockname() to return a pointer to an allocated struct 1095 * sockaddr which must be freed later with FREE(). The caller must 1096 * initialize *name to NULL. 1097 */ 1098 int 1099 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1100 { 1101 struct thread *td = curthread; 1102 struct proc *p = td->td_proc; 1103 struct file *fp; 1104 struct socket *so; 1105 struct sockaddr *sa = NULL; 1106 int error; 1107 1108 error = holdsock(p->p_fd, s, &fp); 1109 if (error) 1110 return (error); 1111 if (*namelen < 0) { 1112 fdrop(fp, td); 1113 return (EINVAL); 1114 } 1115 so = (struct socket *)fp->f_data; 1116 error = so_pru_sockaddr(so, &sa); 1117 if (error == 0) { 1118 if (sa == 0) { 1119 *namelen = 0; 1120 } else { 1121 *namelen = MIN(*namelen, sa->sa_len); 1122 *name = sa; 1123 } 1124 } 1125 1126 fdrop(fp, td); 1127 return (error); 1128 } 1129 1130 /* 1131 * getsockname_args(int fdes, caddr_t asa, int *alen) 1132 * 1133 * Get socket name. 1134 */ 1135 int 1136 getsockname(struct getsockname_args *uap) 1137 { 1138 struct sockaddr *sa = NULL; 1139 int error, sa_len; 1140 1141 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1142 if (error) 1143 return (error); 1144 1145 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1146 1147 if (error == 0) 1148 error = copyout(sa, uap->asa, sa_len); 1149 if (error == 0) 1150 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1151 if (sa) 1152 FREE(sa, M_SONAME); 1153 return (error); 1154 } 1155 1156 /* 1157 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1158 * This allows kern_getpeername() to return a pointer to an allocated struct 1159 * sockaddr which must be freed later with FREE(). The caller must 1160 * initialize *name to NULL. 1161 */ 1162 int 1163 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1164 { 1165 struct thread *td = curthread; 1166 struct proc *p = td->td_proc; 1167 struct file *fp; 1168 struct socket *so; 1169 struct sockaddr *sa = NULL; 1170 int error; 1171 1172 error = holdsock(p->p_fd, s, &fp); 1173 if (error) 1174 return (error); 1175 if (*namelen < 0) { 1176 fdrop(fp, td); 1177 return (EINVAL); 1178 } 1179 so = (struct socket *)fp->f_data; 1180 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1181 fdrop(fp, td); 1182 return (ENOTCONN); 1183 } 1184 error = so_pru_peeraddr(so, &sa); 1185 if (error == 0) { 1186 if (sa == 0) { 1187 *namelen = 0; 1188 } else { 1189 *namelen = MIN(*namelen, sa->sa_len); 1190 *name = sa; 1191 } 1192 } 1193 1194 fdrop(fp, td); 1195 return (error); 1196 } 1197 1198 /* 1199 * getpeername_args(int fdes, caddr_t asa, int *alen) 1200 * 1201 * Get name of peer for connected socket. 1202 */ 1203 int 1204 getpeername(struct getpeername_args *uap) 1205 { 1206 struct sockaddr *sa = NULL; 1207 int error, sa_len; 1208 1209 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1210 if (error) 1211 return (error); 1212 1213 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1214 1215 if (error == 0) 1216 error = copyout(sa, uap->asa, sa_len); 1217 if (error == 0) 1218 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1219 if (sa) 1220 FREE(sa, M_SONAME); 1221 return (error); 1222 } 1223 1224 int 1225 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1226 { 1227 struct sockaddr *sa; 1228 int error; 1229 1230 *namp = NULL; 1231 if (len > SOCK_MAXADDRLEN) 1232 return ENAMETOOLONG; 1233 if (len < offsetof(struct sockaddr, sa_data[0])) 1234 return EDOM; 1235 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1236 error = copyin(uaddr, sa, len); 1237 if (error) { 1238 FREE(sa, M_SONAME); 1239 } else { 1240 #if BYTE_ORDER != BIG_ENDIAN 1241 /* 1242 * The bind(), connect(), and sendto() syscalls were not 1243 * versioned for COMPAT_43. Thus, this check must stay. 1244 */ 1245 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1246 sa->sa_family = sa->sa_len; 1247 #endif 1248 sa->sa_len = len; 1249 *namp = sa; 1250 } 1251 return error; 1252 } 1253 1254 /* 1255 * holdsock() - load the struct file pointer associated 1256 * with a socket into *fpp. If an error occurs, non-zero 1257 * will be returned and *fpp will be set to NULL. 1258 */ 1259 int 1260 holdsock(fdp, fdes, fpp) 1261 struct filedesc *fdp; 1262 int fdes; 1263 struct file **fpp; 1264 { 1265 struct file *fp = NULL; 1266 int error = 0; 1267 1268 if ((unsigned)fdes >= fdp->fd_nfiles || 1269 (fp = fdp->fd_ofiles[fdes]) == NULL) { 1270 error = EBADF; 1271 } else if (fp->f_type != DTYPE_SOCKET) { 1272 error = ENOTSOCK; 1273 fp = NULL; 1274 } else { 1275 fhold(fp); 1276 } 1277 *fpp = fp; 1278 return(error); 1279 } 1280 1281 /* 1282 * Detach a mapped page and release resources back to the system. 1283 * We must release our wiring and if the object is ripped out 1284 * from under the vm_page we become responsible for freeing the 1285 * page. 1286 * 1287 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING 1288 */ 1289 static void 1290 sf_buf_mref(void *arg) 1291 { 1292 struct sfbuf_mref *sfm = arg; 1293 1294 ++sfm->mref_count; 1295 } 1296 1297 static void 1298 sf_buf_mfree(void *arg) 1299 { 1300 struct sfbuf_mref *sfm = arg; 1301 vm_page_t m; 1302 int s; 1303 1304 KKASSERT(sfm->mref_count > 0); 1305 if (--sfm->mref_count == 0) { 1306 m = sf_buf_page(sfm->sf); 1307 sf_buf_free(sfm->sf); 1308 s = splvm(); 1309 vm_page_unwire(m, 0); 1310 if (m->wire_count == 0 && m->object == NULL) 1311 vm_page_free(m); 1312 splx(s); 1313 free(sfm, M_SENDFILE); 1314 } 1315 } 1316 1317 /* 1318 * sendfile(2). 1319 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1320 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1321 * 1322 * Send a file specified by 'fd' and starting at 'offset' to a socket 1323 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1324 * nbytes == 0. Optionally add a header and/or trailer to the socket 1325 * output. If specified, write the total number of bytes sent into *sbytes. 1326 * 1327 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1328 * the headers to count against the remaining bytes to be sent from 1329 * the file descriptor. We may wish to implement a compatibility syscall 1330 * in the future. 1331 */ 1332 int 1333 sendfile(struct sendfile_args *uap) 1334 { 1335 struct thread *td = curthread; 1336 struct proc *p = td->td_proc; 1337 struct file *fp; 1338 struct filedesc *fdp; 1339 struct vnode *vp = NULL; 1340 struct sf_hdtr hdtr; 1341 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1342 struct uio auio; 1343 struct mbuf *mheader = NULL; 1344 off_t hdtr_size = 0, sbytes; 1345 int error, hbytes = 0, tbytes; 1346 1347 KKASSERT(p); 1348 fdp = p->p_fd; 1349 1350 /* 1351 * Do argument checking. Must be a regular file in, stream 1352 * type and connected socket out, positive offset. 1353 */ 1354 fp = holdfp(fdp, uap->fd, FREAD); 1355 if (fp == NULL) { 1356 return (EBADF); 1357 } 1358 if (fp->f_type != DTYPE_VNODE) { 1359 fdrop(fp, td); 1360 return (EINVAL); 1361 } 1362 vp = (struct vnode *)fp->f_data; 1363 vref(vp); 1364 fdrop(fp, td); 1365 1366 /* 1367 * If specified, get the pointer to the sf_hdtr struct for 1368 * any headers/trailers. 1369 */ 1370 if (uap->hdtr) { 1371 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1372 if (error) 1373 goto done; 1374 /* 1375 * Send any headers. 1376 */ 1377 if (hdtr.headers) { 1378 error = iovec_copyin(hdtr.headers, &iov, aiov, 1379 hdtr.hdr_cnt, &hbytes); 1380 if (error) 1381 goto done; 1382 auio.uio_iov = iov; 1383 auio.uio_iovcnt = hdtr.hdr_cnt; 1384 auio.uio_offset = 0; 1385 auio.uio_segflg = UIO_USERSPACE; 1386 auio.uio_rw = UIO_WRITE; 1387 auio.uio_td = td; 1388 auio.uio_resid = hbytes; 1389 1390 mheader = m_uiomove(&auio, MB_WAIT, 0); 1391 1392 iovec_free(&iov, aiov); 1393 if (mheader == NULL) 1394 goto done; 1395 } 1396 } 1397 1398 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1399 &sbytes, uap->flags); 1400 if (error) 1401 goto done; 1402 1403 /* 1404 * Send trailers. Wimp out and use writev(2). 1405 */ 1406 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1407 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1408 hdtr.trl_cnt, &auio.uio_resid); 1409 if (error) 1410 goto done; 1411 auio.uio_iov = iov; 1412 auio.uio_iovcnt = hdtr.trl_cnt; 1413 auio.uio_offset = 0; 1414 auio.uio_segflg = UIO_USERSPACE; 1415 auio.uio_rw = UIO_WRITE; 1416 auio.uio_td = td; 1417 1418 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1419 1420 iovec_free(&iov, aiov); 1421 if (error) 1422 goto done; 1423 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1424 } 1425 1426 done: 1427 if (uap->sbytes != NULL) { 1428 sbytes += hdtr_size; 1429 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1430 } 1431 if (vp) 1432 vrele(vp); 1433 return (error); 1434 } 1435 1436 int 1437 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1438 struct mbuf *mheader, off_t *sbytes, int flags) 1439 { 1440 struct thread *td = curthread; 1441 struct proc *p = td->td_proc; 1442 struct vm_object *obj; 1443 struct socket *so; 1444 struct file *fp; 1445 struct mbuf *m; 1446 struct sf_buf *sf; 1447 struct sfbuf_mref *sfm; 1448 struct vm_page *pg; 1449 off_t off, xfsize; 1450 off_t hbytes = 0; 1451 int error = 0; 1452 int s; 1453 1454 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) { 1455 error = EINVAL; 1456 goto done; 1457 } 1458 error = holdsock(p->p_fd, sfd, &fp); 1459 if (error) 1460 goto done; 1461 so = (struct socket *)fp->f_data; 1462 if (so->so_type != SOCK_STREAM) { 1463 error = EINVAL; 1464 goto done; 1465 } 1466 if ((so->so_state & SS_ISCONNECTED) == 0) { 1467 error = ENOTCONN; 1468 goto done; 1469 } 1470 if (offset < 0) { 1471 error = EINVAL; 1472 goto done; 1473 } 1474 1475 *sbytes = 0; 1476 /* 1477 * Protect against multiple writers to the socket. 1478 */ 1479 (void) sblock(&so->so_snd, M_WAITOK); 1480 1481 /* 1482 * Loop through the pages in the file, starting with the requested 1483 * offset. Get a file page (do I/O if necessary), map the file page 1484 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1485 * it on the socket. 1486 */ 1487 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) { 1488 vm_pindex_t pindex; 1489 vm_offset_t pgoff; 1490 1491 pindex = OFF_TO_IDX(off); 1492 retry_lookup: 1493 /* 1494 * Calculate the amount to transfer. Not to exceed a page, 1495 * the EOF, or the passed in nbytes. 1496 */ 1497 xfsize = obj->un_pager.vnp.vnp_size - off; 1498 if (xfsize > PAGE_SIZE) 1499 xfsize = PAGE_SIZE; 1500 pgoff = (vm_offset_t)(off & PAGE_MASK); 1501 if (PAGE_SIZE - pgoff < xfsize) 1502 xfsize = PAGE_SIZE - pgoff; 1503 if (nbytes && xfsize > (nbytes - *sbytes)) 1504 xfsize = nbytes - *sbytes; 1505 if (xfsize <= 0) 1506 break; 1507 /* 1508 * Optimize the non-blocking case by looking at the socket space 1509 * before going to the extra work of constituting the sf_buf. 1510 */ 1511 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) { 1512 if (so->so_state & SS_CANTSENDMORE) 1513 error = EPIPE; 1514 else 1515 error = EAGAIN; 1516 sbunlock(&so->so_snd); 1517 goto done; 1518 } 1519 /* 1520 * Attempt to look up the page. 1521 * 1522 * Allocate if not found, wait and loop if busy, then 1523 * wire the page. splvm() protection is required to 1524 * maintain the object association (an interrupt can 1525 * free the page) through to the vm_page_wire() call. 1526 */ 1527 s = splvm(); 1528 pg = vm_page_lookup(obj, pindex); 1529 if (pg == NULL) { 1530 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1531 if (pg == NULL) { 1532 vm_wait(); 1533 splx(s); 1534 goto retry_lookup; 1535 } 1536 vm_page_wakeup(pg); 1537 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1538 splx(s); 1539 goto retry_lookup; 1540 } 1541 vm_page_wire(pg); 1542 splx(s); 1543 1544 /* 1545 * If page is not valid for what we need, initiate I/O 1546 */ 1547 1548 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1549 struct uio auio; 1550 struct iovec aiov; 1551 int bsize; 1552 1553 /* 1554 * Ensure that our page is still around when the I/O 1555 * completes. 1556 */ 1557 vm_page_io_start(pg); 1558 1559 /* 1560 * Get the page from backing store. 1561 */ 1562 bsize = vp->v_mount->mnt_stat.f_iosize; 1563 auio.uio_iov = &aiov; 1564 auio.uio_iovcnt = 1; 1565 aiov.iov_base = 0; 1566 aiov.iov_len = MAXBSIZE; 1567 auio.uio_resid = MAXBSIZE; 1568 auio.uio_offset = trunc_page(off); 1569 auio.uio_segflg = UIO_NOCOPY; 1570 auio.uio_rw = UIO_READ; 1571 auio.uio_td = td; 1572 vn_lock(vp, NULL, LK_SHARED | LK_NOPAUSE | LK_RETRY, td); 1573 error = VOP_READ(vp, &auio, 1574 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1575 p->p_ucred); 1576 VOP_UNLOCK(vp, NULL, 0, td); 1577 vm_page_flag_clear(pg, PG_ZERO); 1578 vm_page_io_finish(pg); 1579 if (error) { 1580 vm_page_unwire(pg, 0); 1581 /* 1582 * See if anyone else might know about this page. 1583 * If not and it is not valid, then free it. 1584 */ 1585 if (pg->wire_count == 0 && pg->valid == 0 && 1586 pg->busy == 0 && !(pg->flags & PG_BUSY) && 1587 pg->hold_count == 0) { 1588 vm_page_busy(pg); 1589 vm_page_free(pg); 1590 } 1591 sbunlock(&so->so_snd); 1592 goto done; 1593 } 1594 } 1595 1596 1597 /* 1598 * Get a sendfile buf. We usually wait as long as necessary, 1599 * but this wait can be interrupted. 1600 */ 1601 if ((sf = sf_buf_alloc(pg, SFBA_PCATCH)) == NULL) { 1602 s = splvm(); 1603 vm_page_unwire(pg, 0); 1604 if (pg->wire_count == 0 && pg->object == NULL) 1605 vm_page_free(pg); 1606 splx(s); 1607 sbunlock(&so->so_snd); 1608 error = EINTR; 1609 goto done; 1610 } 1611 1612 /* 1613 * Get an mbuf header and set it up as having external storage. 1614 */ 1615 MGETHDR(m, MB_WAIT, MT_DATA); 1616 if (m == NULL) { 1617 error = ENOBUFS; 1618 sf_buf_free(sf); 1619 sbunlock(&so->so_snd); 1620 goto done; 1621 } 1622 1623 /* 1624 * sfm is a temporary hack, use a per-cpu cache for this. 1625 */ 1626 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK); 1627 sfm->sf = sf; 1628 sfm->mref_count = 1; 1629 1630 m->m_ext.ext_nfree.new = sf_buf_mfree; 1631 m->m_ext.ext_nref.new = sf_buf_mref; 1632 m->m_ext.ext_arg = sfm; 1633 m->m_ext.ext_buf = (void *)sf->kva; 1634 m->m_ext.ext_size = PAGE_SIZE; 1635 m->m_data = (char *) sf->kva + pgoff; 1636 m->m_flags |= M_EXT; 1637 m->m_pkthdr.len = m->m_len = xfsize; 1638 KKASSERT((m->m_flags & (M_EXT_OLD|M_EXT_CLUSTER)) == 0); 1639 1640 if (mheader != NULL) { 1641 hbytes = mheader->m_pkthdr.len; 1642 mheader->m_pkthdr.len += m->m_pkthdr.len; 1643 m_cat(mheader, m); 1644 m = mheader; 1645 mheader = NULL; 1646 } else 1647 hbytes = 0; 1648 1649 /* 1650 * Add the buffer to the socket buffer chain. 1651 */ 1652 s = splnet(); 1653 retry_space: 1654 /* 1655 * Make sure that the socket is still able to take more data. 1656 * CANTSENDMORE being true usually means that the connection 1657 * was closed. so_error is true when an error was sensed after 1658 * a previous send. 1659 * The state is checked after the page mapping and buffer 1660 * allocation above since those operations may block and make 1661 * any socket checks stale. From this point forward, nothing 1662 * blocks before the pru_send (or more accurately, any blocking 1663 * results in a loop back to here to re-check). 1664 */ 1665 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1666 if (so->so_state & SS_CANTSENDMORE) { 1667 error = EPIPE; 1668 } else { 1669 error = so->so_error; 1670 so->so_error = 0; 1671 } 1672 m_freem(m); 1673 sbunlock(&so->so_snd); 1674 splx(s); 1675 goto done; 1676 } 1677 /* 1678 * Wait for socket space to become available. We do this just 1679 * after checking the connection state above in order to avoid 1680 * a race condition with sbwait(). 1681 */ 1682 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { 1683 if (so->so_state & SS_NBIO) { 1684 m_freem(m); 1685 sbunlock(&so->so_snd); 1686 splx(s); 1687 error = EAGAIN; 1688 goto done; 1689 } 1690 error = sbwait(&so->so_snd); 1691 /* 1692 * An error from sbwait usually indicates that we've 1693 * been interrupted by a signal. If we've sent anything 1694 * then return bytes sent, otherwise return the error. 1695 */ 1696 if (error) { 1697 m_freem(m); 1698 sbunlock(&so->so_snd); 1699 splx(s); 1700 goto done; 1701 } 1702 goto retry_space; 1703 } 1704 error = so_pru_send(so, 0, m, NULL, NULL, td); 1705 splx(s); 1706 if (error) { 1707 sbunlock(&so->so_snd); 1708 goto done; 1709 } 1710 } 1711 if (mheader != NULL) { 1712 *sbytes += mheader->m_pkthdr.len; 1713 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1714 mheader = NULL; 1715 } 1716 sbunlock(&so->so_snd); 1717 1718 done: 1719 if (fp) 1720 fdrop(fp, td); 1721 if (mheader != NULL) 1722 m_freem(mheader); 1723 return (error); 1724 } 1725