kern_sendfile.c revision 101842
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: head/sys/kern/uipc_syscalls.c 101842 2002-08-13 19:03:19Z dg $ 38 */ 39 40#include "opt_compat.h" 41#include "opt_ktrace.h" 42#include "opt_mac.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/kernel.h> 47#include <sys/lock.h> 48#include <sys/mac.h> 49#include <sys/mutex.h> 50#include <sys/sysproto.h> 51#include <sys/malloc.h> 52#include <sys/filedesc.h> 53#include <sys/event.h> 54#include <sys/proc.h> 55#include <sys/fcntl.h> 56#include <sys/file.h> 57#include <sys/lock.h> 58#include <sys/mount.h> 59#include <sys/mbuf.h> 60#include <sys/protosw.h> 61#include <sys/socket.h> 62#include <sys/socketvar.h> 63#include <sys/signalvar.h> 64#include <sys/uio.h> 65#include <sys/vnode.h> 66#ifdef KTRACE 67#include <sys/ktrace.h> 68#endif 69 70#include <vm/vm.h> 71#include <vm/vm_object.h> 72#include <vm/vm_page.h> 73#include <vm/vm_pageout.h> 74#include <vm/vm_kern.h> 75#include <vm/vm_extern.h> 76 77static void sf_buf_init(void *arg); 78SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL) 79 80static int sendit(struct thread *td, int s, struct msghdr *mp, int flags); 81static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp); 82 83static int accept1(struct thread *td, struct accept_args *uap, int compat); 84static int do_sendfile(struct thread *td, struct sendfile_args *uap, int compat); 85static int getsockname1(struct thread *td, struct getsockname_args *uap, 86 int compat); 87static int getpeername1(struct thread *td, struct getpeername_args *uap, 88 int compat); 89 90/* 91 * Expanded sf_freelist head. Really an SLIST_HEAD() in disguise, with the 92 * sf_freelist head with the sf_lock mutex. 93 */ 94static struct { 95 SLIST_HEAD(, sf_buf) sf_head; 96 struct mtx sf_lock; 97} sf_freelist; 98 99vm_offset_t sf_base; 100struct sf_buf *sf_bufs; 101u_int sf_buf_alloc_want; 102 103/* 104 * System call interface to the socket abstraction. 105 */ 106#if defined(COMPAT_43) || defined(COMPAT_SUNOS) 107#define COMPAT_OLDSOCK 108#endif 109 110extern struct fileops socketops; 111 112/* 113 * MPSAFE 114 */ 115int 116socket(td, uap) 117 struct thread *td; 118 register struct socket_args /* { 119 int domain; 120 int type; 121 int protocol; 122 } */ *uap; 123{ 124 struct filedesc *fdp; 125 struct socket *so; 126 struct file *fp; 127 int fd, error; 128 129 mtx_lock(&Giant); 130 fdp = td->td_proc->p_fd; 131 error = falloc(td, &fp, &fd); 132 if (error) 133 goto done2; 134 fhold(fp); 135 error = socreate(uap->domain, &so, uap->type, uap->protocol, 136 td->td_ucred, td); 137 FILEDESC_LOCK(fdp); 138 if (error) { 139 if (fdp->fd_ofiles[fd] == fp) { 140 fdp->fd_ofiles[fd] = NULL; 141 FILEDESC_UNLOCK(fdp); 142 fdrop(fp, td); 143 } else 144 FILEDESC_UNLOCK(fdp); 145 } else { 146 fp->f_data = so; /* already has ref count */ 147 fp->f_flag = FREAD|FWRITE; 148 fp->f_ops = &socketops; 149 fp->f_type = DTYPE_SOCKET; 150 FILEDESC_UNLOCK(fdp); 151 td->td_retval[0] = fd; 152 } 153 fdrop(fp, td); 154done2: 155 mtx_unlock(&Giant); 156 return (error); 157} 158 159/* 160 * MPSAFE 161 */ 162/* ARGSUSED */ 163int 164bind(td, uap) 165 struct thread *td; 166 register struct bind_args /* { 167 int s; 168 caddr_t name; 169 int namelen; 170 } */ *uap; 171{ 172 struct socket *so; 173 struct sockaddr *sa; 174 int error; 175 176 mtx_lock(&Giant); 177 if ((error = fgetsock(td, uap->s, &so, NULL)) != 0) 178 goto done2; 179 if ((error = getsockaddr(&sa, uap->name, uap->namelen)) != 0) 180 goto done1; 181#ifdef MAC 182 error = mac_check_socket_bind(td->td_ucred, so, sa); 183 if (error) { 184 FREE(sa, M_SONAME); 185 goto done1; 186 } 187#endif 188 error = sobind(so, sa, td); 189 FREE(sa, M_SONAME); 190done1: 191 fputsock(so); 192done2: 193 mtx_unlock(&Giant); 194 return (error); 195} 196 197/* 198 * MPSAFE 199 */ 200/* ARGSUSED */ 201int 202listen(td, uap) 203 struct thread *td; 204 register struct listen_args /* { 205 int s; 206 int backlog; 207 } */ *uap; 208{ 209 struct socket *so; 210 int error; 211 212 mtx_lock(&Giant); 213 if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) { 214#ifdef MAC 215 error = mac_check_socket_listen(td->td_ucred, so); 216 if (error) 217 goto done; 218#endif 219 error = solisten(so, uap->backlog, td); 220#ifdef MAC 221done: 222#endif 223 fputsock(so); 224 } 225 mtx_unlock(&Giant); 226 return(error); 227} 228 229/* 230 * accept1() 231 * MPSAFE 232 */ 233static int 234accept1(td, uap, compat) 235 struct thread *td; 236 register struct accept_args /* { 237 int s; 238 caddr_t name; 239 int *anamelen; 240 } */ *uap; 241 int compat; 242{ 243 struct filedesc *fdp; 244 struct file *nfp = NULL; 245 struct sockaddr *sa; 246 int namelen, error, s; 247 struct socket *head, *so; 248 int fd; 249 u_int fflag; 250 251 mtx_lock(&Giant); 252 fdp = td->td_proc->p_fd; 253 if (uap->name) { 254 error = copyin(uap->anamelen, &namelen, sizeof (namelen)); 255 if(error) 256 goto done2; 257 if (namelen < 0) 258 return (EINVAL); 259 } 260 error = fgetsock(td, uap->s, &head, &fflag); 261 if (error) 262 goto done2; 263 s = splnet(); 264 if ((head->so_options & SO_ACCEPTCONN) == 0) { 265 splx(s); 266 error = EINVAL; 267 goto done; 268 } 269 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) { 270 splx(s); 271 error = EWOULDBLOCK; 272 goto done; 273 } 274 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) { 275 if (head->so_state & SS_CANTRCVMORE) { 276 head->so_error = ECONNABORTED; 277 break; 278 } 279 error = tsleep(&head->so_timeo, PSOCK | PCATCH, 280 "accept", 0); 281 if (error) { 282 splx(s); 283 goto done; 284 } 285 } 286 if (head->so_error) { 287 error = head->so_error; 288 head->so_error = 0; 289 splx(s); 290 goto done; 291 } 292 293 /* 294 * At this point we know that there is at least one connection 295 * ready to be accepted. Remove it from the queue prior to 296 * allocating the file descriptor for it since falloc() may 297 * block allowing another process to accept the connection 298 * instead. 299 */ 300 so = TAILQ_FIRST(&head->so_comp); 301 TAILQ_REMOVE(&head->so_comp, so, so_list); 302 head->so_qlen--; 303 304 error = falloc(td, &nfp, &fd); 305 if (error) { 306 /* 307 * Probably ran out of file descriptors. Put the 308 * unaccepted connection back onto the queue and 309 * do another wakeup so some other process might 310 * have a chance at it. 311 */ 312 TAILQ_INSERT_HEAD(&head->so_comp, so, so_list); 313 head->so_qlen++; 314 wakeup_one(&head->so_timeo); 315 splx(s); 316 goto done; 317 } 318 fhold(nfp); 319 td->td_retval[0] = fd; 320 321 /* connection has been removed from the listen queue */ 322 KNOTE(&head->so_rcv.sb_sel.si_note, 0); 323 324 so->so_state &= ~SS_COMP; 325 so->so_head = NULL; 326 if (head->so_sigio != NULL) 327 fsetown(fgetown(head->so_sigio), &so->so_sigio); 328 329 FILE_LOCK(nfp); 330 soref(so); /* file descriptor reference */ 331 nfp->f_data = so; /* nfp has ref count from falloc */ 332 nfp->f_flag = fflag; 333 nfp->f_ops = &socketops; 334 nfp->f_type = DTYPE_SOCKET; 335 FILE_UNLOCK(nfp); 336 sa = 0; 337 error = soaccept(so, &sa); 338 if (error) { 339 /* 340 * return a namelen of zero for older code which might 341 * ignore the return value from accept. 342 */ 343 if (uap->name != NULL) { 344 namelen = 0; 345 (void) copyout(&namelen, 346 uap->anamelen, sizeof(*uap->anamelen)); 347 } 348 goto noconnection; 349 } 350 if (sa == NULL) { 351 namelen = 0; 352 if (uap->name) 353 goto gotnoname; 354 splx(s); 355 error = 0; 356 goto done; 357 } 358 if (uap->name) { 359 /* check sa_len before it is destroyed */ 360 if (namelen > sa->sa_len) 361 namelen = sa->sa_len; 362#ifdef COMPAT_OLDSOCK 363 if (compat) 364 ((struct osockaddr *)sa)->sa_family = 365 sa->sa_family; 366#endif 367 error = copyout(sa, uap->name, (u_int)namelen); 368 if (!error) 369gotnoname: 370 error = copyout(&namelen, 371 uap->anamelen, sizeof (*uap->anamelen)); 372 } 373noconnection: 374 if (sa) 375 FREE(sa, M_SONAME); 376 377 /* 378 * close the new descriptor, assuming someone hasn't ripped it 379 * out from under us. 380 */ 381 if (error) { 382 FILEDESC_LOCK(fdp); 383 if (fdp->fd_ofiles[fd] == nfp) { 384 fdp->fd_ofiles[fd] = NULL; 385 FILEDESC_UNLOCK(fdp); 386 fdrop(nfp, td); 387 } else { 388 FILEDESC_UNLOCK(fdp); 389 } 390 } 391 splx(s); 392 393 /* 394 * Release explicitly held references before returning. 395 */ 396done: 397 if (nfp != NULL) 398 fdrop(nfp, td); 399 fputsock(head); 400done2: 401 mtx_unlock(&Giant); 402 return (error); 403} 404 405/* 406 * MPSAFE (accept1() is MPSAFE) 407 */ 408int 409accept(td, uap) 410 struct thread *td; 411 struct accept_args *uap; 412{ 413 414 return (accept1(td, uap, 0)); 415} 416 417#ifdef COMPAT_OLDSOCK 418/* 419 * MPSAFE (accept1() is MPSAFE) 420 */ 421int 422oaccept(td, uap) 423 struct thread *td; 424 struct accept_args *uap; 425{ 426 427 return (accept1(td, uap, 1)); 428} 429#endif /* COMPAT_OLDSOCK */ 430 431/* 432 * MPSAFE 433 */ 434/* ARGSUSED */ 435int 436connect(td, uap) 437 struct thread *td; 438 register struct connect_args /* { 439 int s; 440 caddr_t name; 441 int namelen; 442 } */ *uap; 443{ 444 struct socket *so; 445 struct sockaddr *sa; 446 int error, s; 447 448 mtx_lock(&Giant); 449 if ((error = fgetsock(td, uap->s, &so, NULL)) != 0) 450 goto done2; 451 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 452 error = EALREADY; 453 goto done1; 454 } 455 error = getsockaddr(&sa, uap->name, uap->namelen); 456 if (error) 457 goto done1; 458#ifdef MAC 459 error = mac_check_socket_connect(td->td_ucred, so, sa); 460 if (error) 461 goto bad; 462#endif 463 error = soconnect(so, sa, td); 464 if (error) 465 goto bad; 466 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 467 FREE(sa, M_SONAME); 468 error = EINPROGRESS; 469 goto done1; 470 } 471 s = splnet(); 472 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 473 error = tsleep(&so->so_timeo, PSOCK | PCATCH, "connec", 0); 474 if (error) 475 break; 476 } 477 if (error == 0) { 478 error = so->so_error; 479 so->so_error = 0; 480 } 481 splx(s); 482bad: 483 so->so_state &= ~SS_ISCONNECTING; 484 FREE(sa, M_SONAME); 485 if (error == ERESTART) 486 error = EINTR; 487done1: 488 fputsock(so); 489done2: 490 mtx_unlock(&Giant); 491 return (error); 492} 493 494/* 495 * MPSAFE 496 */ 497int 498socketpair(td, uap) 499 struct thread *td; 500 register struct socketpair_args /* { 501 int domain; 502 int type; 503 int protocol; 504 int *rsv; 505 } */ *uap; 506{ 507 register struct filedesc *fdp = td->td_proc->p_fd; 508 struct file *fp1, *fp2; 509 struct socket *so1, *so2; 510 int fd, error, sv[2]; 511 512 mtx_lock(&Giant); 513 error = socreate(uap->domain, &so1, uap->type, uap->protocol, 514 td->td_ucred, td); 515 if (error) 516 goto done2; 517 error = socreate(uap->domain, &so2, uap->type, uap->protocol, 518 td->td_ucred, td); 519 if (error) 520 goto free1; 521 error = falloc(td, &fp1, &fd); 522 if (error) 523 goto free2; 524 fhold(fp1); 525 sv[0] = fd; 526 fp1->f_data = so1; /* so1 already has ref count */ 527 error = falloc(td, &fp2, &fd); 528 if (error) 529 goto free3; 530 fhold(fp2); 531 fp2->f_data = so2; /* so2 already has ref count */ 532 sv[1] = fd; 533 error = soconnect2(so1, so2); 534 if (error) 535 goto free4; 536 if (uap->type == SOCK_DGRAM) { 537 /* 538 * Datagram socket connection is asymmetric. 539 */ 540 error = soconnect2(so2, so1); 541 if (error) 542 goto free4; 543 } 544 FILE_LOCK(fp1); 545 fp1->f_flag = FREAD|FWRITE; 546 fp1->f_ops = &socketops; 547 fp1->f_type = DTYPE_SOCKET; 548 FILE_UNLOCK(fp1); 549 FILE_LOCK(fp2); 550 fp2->f_flag = FREAD|FWRITE; 551 fp2->f_ops = &socketops; 552 fp2->f_type = DTYPE_SOCKET; 553 FILE_UNLOCK(fp2); 554 error = copyout(sv, uap->rsv, 2 * sizeof (int)); 555 fdrop(fp1, td); 556 fdrop(fp2, td); 557 goto done2; 558free4: 559 FILEDESC_LOCK(fdp); 560 if (fdp->fd_ofiles[sv[1]] == fp2) { 561 fdp->fd_ofiles[sv[1]] = NULL; 562 FILEDESC_UNLOCK(fdp); 563 fdrop(fp2, td); 564 } else 565 FILEDESC_UNLOCK(fdp); 566 fdrop(fp2, td); 567free3: 568 FILEDESC_LOCK(fdp); 569 if (fdp->fd_ofiles[sv[0]] == fp1) { 570 fdp->fd_ofiles[sv[0]] = NULL; 571 FILEDESC_UNLOCK(fdp); 572 fdrop(fp1, td); 573 } else 574 FILEDESC_UNLOCK(fdp); 575 fdrop(fp1, td); 576free2: 577 (void)soclose(so2); 578free1: 579 (void)soclose(so1); 580done2: 581 mtx_unlock(&Giant); 582 return (error); 583} 584 585static int 586sendit(td, s, mp, flags) 587 register struct thread *td; 588 int s; 589 register struct msghdr *mp; 590 int flags; 591{ 592 struct uio auio; 593 register struct iovec *iov; 594 register int i; 595 struct mbuf *control; 596 struct sockaddr *to = NULL; 597 int len, error; 598 struct socket *so; 599#ifdef KTRACE 600 struct iovec *ktriov = NULL; 601 struct uio ktruio; 602 int iovlen; 603#endif 604 605 if ((error = fgetsock(td, s, &so, NULL)) != 0) 606 return (error); 607 auio.uio_iov = mp->msg_iov; 608 auio.uio_iovcnt = mp->msg_iovlen; 609 auio.uio_segflg = UIO_USERSPACE; 610 auio.uio_rw = UIO_WRITE; 611 auio.uio_td = td; 612 auio.uio_offset = 0; /* XXX */ 613 auio.uio_resid = 0; 614 iov = mp->msg_iov; 615 for (i = 0; i < mp->msg_iovlen; i++, iov++) { 616 if ((auio.uio_resid += iov->iov_len) < 0) { 617 error = EINVAL; 618 goto bad; 619 } 620 } 621 if (mp->msg_name) { 622 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen); 623 if (error) 624 goto bad; 625 } 626 if (mp->msg_control) { 627 if (mp->msg_controllen < sizeof(struct cmsghdr) 628#ifdef COMPAT_OLDSOCK 629 && mp->msg_flags != MSG_COMPAT 630#endif 631 ) { 632 error = EINVAL; 633 goto bad; 634 } 635 error = sockargs(&control, mp->msg_control, 636 mp->msg_controllen, MT_CONTROL); 637 if (error) 638 goto bad; 639#ifdef COMPAT_OLDSOCK 640 if (mp->msg_flags == MSG_COMPAT) { 641 register struct cmsghdr *cm; 642 643 M_PREPEND(control, sizeof(*cm), M_TRYWAIT); 644 if (control == 0) { 645 error = ENOBUFS; 646 goto bad; 647 } else { 648 cm = mtod(control, struct cmsghdr *); 649 cm->cmsg_len = control->m_len; 650 cm->cmsg_level = SOL_SOCKET; 651 cm->cmsg_type = SCM_RIGHTS; 652 } 653 } 654#endif 655 } else { 656 control = 0; 657 } 658#ifdef KTRACE 659 if (KTRPOINT(td, KTR_GENIO)) { 660 iovlen = auio.uio_iovcnt * sizeof (struct iovec); 661 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 662 bcopy(auio.uio_iov, ktriov, iovlen); 663 ktruio = auio; 664 } 665#endif 666 len = auio.uio_resid; 667 error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control, 668 flags, td); 669 if (error) { 670 if (auio.uio_resid != len && (error == ERESTART || 671 error == EINTR || error == EWOULDBLOCK)) 672 error = 0; 673 /* Generation of SIGPIPE can be controlled per socket */ 674 if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE)) { 675 PROC_LOCK(td->td_proc); 676 psignal(td->td_proc, SIGPIPE); 677 PROC_UNLOCK(td->td_proc); 678 } 679 } 680 if (error == 0) 681 td->td_retval[0] = len - auio.uio_resid; 682#ifdef KTRACE 683 if (ktriov != NULL) { 684 if (error == 0) { 685 ktruio.uio_iov = ktriov; 686 ktruio.uio_resid = td->td_retval[0]; 687 ktrgenio(s, UIO_WRITE, &ktruio, error); 688 } 689 FREE(ktriov, M_TEMP); 690 } 691#endif 692bad: 693 fputsock(so); 694 if (to) 695 FREE(to, M_SONAME); 696 return (error); 697} 698 699/* 700 * MPSAFE 701 */ 702int 703sendto(td, uap) 704 struct thread *td; 705 register struct sendto_args /* { 706 int s; 707 caddr_t buf; 708 size_t len; 709 int flags; 710 caddr_t to; 711 int tolen; 712 } */ *uap; 713{ 714 struct msghdr msg; 715 struct iovec aiov; 716 int error; 717 718 msg.msg_name = uap->to; 719 msg.msg_namelen = uap->tolen; 720 msg.msg_iov = &aiov; 721 msg.msg_iovlen = 1; 722 msg.msg_control = 0; 723#ifdef COMPAT_OLDSOCK 724 msg.msg_flags = 0; 725#endif 726 aiov.iov_base = uap->buf; 727 aiov.iov_len = uap->len; 728 mtx_lock(&Giant); 729 error = sendit(td, uap->s, &msg, uap->flags); 730 mtx_unlock(&Giant); 731 return (error); 732} 733 734#ifdef COMPAT_OLDSOCK 735/* 736 * MPSAFE 737 */ 738int 739osend(td, uap) 740 struct thread *td; 741 register struct osend_args /* { 742 int s; 743 caddr_t buf; 744 int len; 745 int flags; 746 } */ *uap; 747{ 748 struct msghdr msg; 749 struct iovec aiov; 750 int error; 751 752 msg.msg_name = 0; 753 msg.msg_namelen = 0; 754 msg.msg_iov = &aiov; 755 msg.msg_iovlen = 1; 756 aiov.iov_base = uap->buf; 757 aiov.iov_len = uap->len; 758 msg.msg_control = 0; 759 msg.msg_flags = 0; 760 mtx_lock(&Giant); 761 error = sendit(td, uap->s, &msg, uap->flags); 762 mtx_unlock(&Giant); 763 return (error); 764} 765 766/* 767 * MPSAFE 768 */ 769int 770osendmsg(td, uap) 771 struct thread *td; 772 register struct osendmsg_args /* { 773 int s; 774 caddr_t msg; 775 int flags; 776 } */ *uap; 777{ 778 struct msghdr msg; 779 struct iovec aiov[UIO_SMALLIOV], *iov; 780 int error; 781 782 mtx_lock(&Giant); 783 error = copyin(uap->msg, &msg, sizeof (struct omsghdr)); 784 if (error) 785 goto done2; 786 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 787 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { 788 error = EMSGSIZE; 789 goto done2; 790 } 791 MALLOC(iov, struct iovec *, 792 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 793 M_WAITOK); 794 } else { 795 iov = aiov; 796 } 797 error = copyin(msg.msg_iov, iov, 798 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 799 if (error) 800 goto done; 801 msg.msg_flags = MSG_COMPAT; 802 msg.msg_iov = iov; 803 error = sendit(td, uap->s, &msg, uap->flags); 804done: 805 if (iov != aiov) 806 FREE(iov, M_IOV); 807done2: 808 mtx_unlock(&Giant); 809 return (error); 810} 811#endif 812 813/* 814 * MPSAFE 815 */ 816int 817sendmsg(td, uap) 818 struct thread *td; 819 register struct sendmsg_args /* { 820 int s; 821 caddr_t msg; 822 int flags; 823 } */ *uap; 824{ 825 struct msghdr msg; 826 struct iovec aiov[UIO_SMALLIOV], *iov; 827 int error; 828 829 mtx_lock(&Giant); 830 error = copyin(uap->msg, &msg, sizeof (msg)); 831 if (error) 832 goto done2; 833 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 834 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { 835 error = EMSGSIZE; 836 goto done2; 837 } 838 MALLOC(iov, struct iovec *, 839 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 840 M_WAITOK); 841 } else { 842 iov = aiov; 843 } 844 if (msg.msg_iovlen && 845 (error = copyin(msg.msg_iov, iov, 846 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))))) 847 goto done; 848 msg.msg_iov = iov; 849#ifdef COMPAT_OLDSOCK 850 msg.msg_flags = 0; 851#endif 852 error = sendit(td, uap->s, &msg, uap->flags); 853done: 854 if (iov != aiov) 855 FREE(iov, M_IOV); 856done2: 857 mtx_unlock(&Giant); 858 return (error); 859} 860 861static int 862recvit(td, s, mp, namelenp) 863 register struct thread *td; 864 int s; 865 register struct msghdr *mp; 866 void *namelenp; 867{ 868 struct uio auio; 869 register struct iovec *iov; 870 register int i; 871 int len, error; 872 struct mbuf *m, *control = 0; 873 caddr_t ctlbuf; 874 struct socket *so; 875 struct sockaddr *fromsa = 0; 876#ifdef KTRACE 877 struct iovec *ktriov = NULL; 878 struct uio ktruio; 879 int iovlen; 880#endif 881 882 if ((error = fgetsock(td, s, &so, NULL)) != 0) 883 return (error); 884 auio.uio_iov = mp->msg_iov; 885 auio.uio_iovcnt = mp->msg_iovlen; 886 auio.uio_segflg = UIO_USERSPACE; 887 auio.uio_rw = UIO_READ; 888 auio.uio_td = td; 889 auio.uio_offset = 0; /* XXX */ 890 auio.uio_resid = 0; 891 iov = mp->msg_iov; 892 for (i = 0; i < mp->msg_iovlen; i++, iov++) { 893 if ((auio.uio_resid += iov->iov_len) < 0) { 894 fputsock(so); 895 return (EINVAL); 896 } 897 } 898#ifdef KTRACE 899 if (KTRPOINT(td, KTR_GENIO)) { 900 iovlen = auio.uio_iovcnt * sizeof (struct iovec); 901 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 902 bcopy(auio.uio_iov, ktriov, iovlen); 903 ktruio = auio; 904 } 905#endif 906 len = auio.uio_resid; 907 error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio, 908 (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0, 909 &mp->msg_flags); 910 if (error) { 911 if (auio.uio_resid != len && (error == ERESTART || 912 error == EINTR || error == EWOULDBLOCK)) 913 error = 0; 914 } 915#ifdef KTRACE 916 if (ktriov != NULL) { 917 if (error == 0) { 918 ktruio.uio_iov = ktriov; 919 ktruio.uio_resid = len - auio.uio_resid; 920 ktrgenio(s, UIO_READ, &ktruio, error); 921 } 922 FREE(ktriov, M_TEMP); 923 } 924#endif 925 if (error) 926 goto out; 927 td->td_retval[0] = len - auio.uio_resid; 928 if (mp->msg_name) { 929 len = mp->msg_namelen; 930 if (len <= 0 || fromsa == 0) 931 len = 0; 932 else { 933#ifndef MIN 934#define MIN(a,b) ((a)>(b)?(b):(a)) 935#endif 936 /* save sa_len before it is destroyed by MSG_COMPAT */ 937 len = MIN(len, fromsa->sa_len); 938#ifdef COMPAT_OLDSOCK 939 if (mp->msg_flags & MSG_COMPAT) 940 ((struct osockaddr *)fromsa)->sa_family = 941 fromsa->sa_family; 942#endif 943 error = copyout(fromsa, mp->msg_name, (unsigned)len); 944 if (error) 945 goto out; 946 } 947 mp->msg_namelen = len; 948 if (namelenp && 949 (error = copyout(&len, namelenp, sizeof (int)))) { 950#ifdef COMPAT_OLDSOCK 951 if (mp->msg_flags & MSG_COMPAT) 952 error = 0; /* old recvfrom didn't check */ 953 else 954#endif 955 goto out; 956 } 957 } 958 if (mp->msg_control) { 959#ifdef COMPAT_OLDSOCK 960 /* 961 * We assume that old recvmsg calls won't receive access 962 * rights and other control info, esp. as control info 963 * is always optional and those options didn't exist in 4.3. 964 * If we receive rights, trim the cmsghdr; anything else 965 * is tossed. 966 */ 967 if (control && mp->msg_flags & MSG_COMPAT) { 968 if (mtod(control, struct cmsghdr *)->cmsg_level != 969 SOL_SOCKET || 970 mtod(control, struct cmsghdr *)->cmsg_type != 971 SCM_RIGHTS) { 972 mp->msg_controllen = 0; 973 goto out; 974 } 975 control->m_len -= sizeof (struct cmsghdr); 976 control->m_data += sizeof (struct cmsghdr); 977 } 978#endif 979 len = mp->msg_controllen; 980 m = control; 981 mp->msg_controllen = 0; 982 ctlbuf = mp->msg_control; 983 984 while (m && len > 0) { 985 unsigned int tocopy; 986 987 if (len >= m->m_len) 988 tocopy = m->m_len; 989 else { 990 mp->msg_flags |= MSG_CTRUNC; 991 tocopy = len; 992 } 993 994 if ((error = copyout(mtod(m, caddr_t), 995 ctlbuf, tocopy)) != 0) 996 goto out; 997 998 ctlbuf += tocopy; 999 len -= tocopy; 1000 m = m->m_next; 1001 } 1002 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control; 1003 } 1004out: 1005 fputsock(so); 1006 if (fromsa) 1007 FREE(fromsa, M_SONAME); 1008 if (control) 1009 m_freem(control); 1010 return (error); 1011} 1012 1013/* 1014 * MPSAFE 1015 */ 1016int 1017recvfrom(td, uap) 1018 struct thread *td; 1019 register struct recvfrom_args /* { 1020 int s; 1021 caddr_t buf; 1022 size_t len; 1023 int flags; 1024 caddr_t from; 1025 int *fromlenaddr; 1026 } */ *uap; 1027{ 1028 struct msghdr msg; 1029 struct iovec aiov; 1030 int error; 1031 1032 mtx_lock(&Giant); 1033 if (uap->fromlenaddr) { 1034 error = copyin(uap->fromlenaddr, 1035 &msg.msg_namelen, sizeof (msg.msg_namelen)); 1036 if (error) 1037 goto done2; 1038 } else { 1039 msg.msg_namelen = 0; 1040 } 1041 msg.msg_name = uap->from; 1042 msg.msg_iov = &aiov; 1043 msg.msg_iovlen = 1; 1044 aiov.iov_base = uap->buf; 1045 aiov.iov_len = uap->len; 1046 msg.msg_control = 0; 1047 msg.msg_flags = uap->flags; 1048 error = recvit(td, uap->s, &msg, uap->fromlenaddr); 1049done2: 1050 mtx_unlock(&Giant); 1051 return(error); 1052} 1053 1054#ifdef COMPAT_OLDSOCK 1055/* 1056 * MPSAFE 1057 */ 1058int 1059orecvfrom(td, uap) 1060 struct thread *td; 1061 struct recvfrom_args *uap; 1062{ 1063 1064 uap->flags |= MSG_COMPAT; 1065 return (recvfrom(td, uap)); 1066} 1067#endif 1068 1069 1070#ifdef COMPAT_OLDSOCK 1071/* 1072 * MPSAFE 1073 */ 1074int 1075orecv(td, uap) 1076 struct thread *td; 1077 register struct orecv_args /* { 1078 int s; 1079 caddr_t buf; 1080 int len; 1081 int flags; 1082 } */ *uap; 1083{ 1084 struct msghdr msg; 1085 struct iovec aiov; 1086 int error; 1087 1088 mtx_lock(&Giant); 1089 msg.msg_name = 0; 1090 msg.msg_namelen = 0; 1091 msg.msg_iov = &aiov; 1092 msg.msg_iovlen = 1; 1093 aiov.iov_base = uap->buf; 1094 aiov.iov_len = uap->len; 1095 msg.msg_control = 0; 1096 msg.msg_flags = uap->flags; 1097 error = recvit(td, uap->s, &msg, NULL); 1098 mtx_unlock(&Giant); 1099 return (error); 1100} 1101 1102/* 1103 * Old recvmsg. This code takes advantage of the fact that the old msghdr 1104 * overlays the new one, missing only the flags, and with the (old) access 1105 * rights where the control fields are now. 1106 * 1107 * MPSAFE 1108 */ 1109int 1110orecvmsg(td, uap) 1111 struct thread *td; 1112 register struct orecvmsg_args /* { 1113 int s; 1114 struct omsghdr *msg; 1115 int flags; 1116 } */ *uap; 1117{ 1118 struct msghdr msg; 1119 struct iovec aiov[UIO_SMALLIOV], *iov; 1120 int error; 1121 1122 error = copyin(uap->msg, &msg, sizeof (struct omsghdr)); 1123 if (error) 1124 return (error); 1125 1126 mtx_lock(&Giant); 1127 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 1128 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { 1129 error = EMSGSIZE; 1130 goto done2; 1131 } 1132 MALLOC(iov, struct iovec *, 1133 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 1134 M_WAITOK); 1135 } else { 1136 iov = aiov; 1137 } 1138 msg.msg_flags = uap->flags | MSG_COMPAT; 1139 error = copyin(msg.msg_iov, iov, 1140 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 1141 if (error) 1142 goto done; 1143 msg.msg_iov = iov; 1144 error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen); 1145 1146 if (msg.msg_controllen && error == 0) 1147 error = copyout(&msg.msg_controllen, 1148 &uap->msg->msg_accrightslen, sizeof (int)); 1149done: 1150 if (iov != aiov) 1151 FREE(iov, M_IOV); 1152done2: 1153 mtx_unlock(&Giant); 1154 return (error); 1155} 1156#endif 1157 1158/* 1159 * MPSAFE 1160 */ 1161int 1162recvmsg(td, uap) 1163 struct thread *td; 1164 register struct recvmsg_args /* { 1165 int s; 1166 struct msghdr *msg; 1167 int flags; 1168 } */ *uap; 1169{ 1170 struct msghdr msg; 1171 struct iovec aiov[UIO_SMALLIOV], *uiov, *iov; 1172 register int error; 1173 1174 mtx_lock(&Giant); 1175 error = copyin(uap->msg, &msg, sizeof (msg)); 1176 if (error) 1177 goto done2; 1178 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 1179 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) { 1180 error = EMSGSIZE; 1181 goto done2; 1182 } 1183 MALLOC(iov, struct iovec *, 1184 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 1185 M_WAITOK); 1186 } else { 1187 iov = aiov; 1188 } 1189#ifdef COMPAT_OLDSOCK 1190 msg.msg_flags = uap->flags &~ MSG_COMPAT; 1191#else 1192 msg.msg_flags = uap->flags; 1193#endif 1194 uiov = msg.msg_iov; 1195 msg.msg_iov = iov; 1196 error = copyin(uiov, iov, 1197 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 1198 if (error) 1199 goto done; 1200 error = recvit(td, uap->s, &msg, NULL); 1201 if (!error) { 1202 msg.msg_iov = uiov; 1203 error = copyout(&msg, uap->msg, sizeof(msg)); 1204 } 1205done: 1206 if (iov != aiov) 1207 FREE(iov, M_IOV); 1208done2: 1209 mtx_unlock(&Giant); 1210 return (error); 1211} 1212 1213/* 1214 * MPSAFE 1215 */ 1216/* ARGSUSED */ 1217int 1218shutdown(td, uap) 1219 struct thread *td; 1220 register struct shutdown_args /* { 1221 int s; 1222 int how; 1223 } */ *uap; 1224{ 1225 struct socket *so; 1226 int error; 1227 1228 mtx_lock(&Giant); 1229 if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) { 1230 error = soshutdown(so, uap->how); 1231 fputsock(so); 1232 } 1233 mtx_unlock(&Giant); 1234 return(error); 1235} 1236 1237/* 1238 * MPSAFE 1239 */ 1240/* ARGSUSED */ 1241int 1242setsockopt(td, uap) 1243 struct thread *td; 1244 register struct setsockopt_args /* { 1245 int s; 1246 int level; 1247 int name; 1248 caddr_t val; 1249 int valsize; 1250 } */ *uap; 1251{ 1252 struct socket *so; 1253 struct sockopt sopt; 1254 int error; 1255 1256 if (uap->val == 0 && uap->valsize != 0) 1257 return (EFAULT); 1258 if (uap->valsize < 0) 1259 return (EINVAL); 1260 1261 mtx_lock(&Giant); 1262 if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) { 1263 sopt.sopt_dir = SOPT_SET; 1264 sopt.sopt_level = uap->level; 1265 sopt.sopt_name = uap->name; 1266 sopt.sopt_val = uap->val; 1267 sopt.sopt_valsize = uap->valsize; 1268 sopt.sopt_td = td; 1269 error = sosetopt(so, &sopt); 1270 fputsock(so); 1271 } 1272 mtx_unlock(&Giant); 1273 return(error); 1274} 1275 1276/* 1277 * MPSAFE 1278 */ 1279/* ARGSUSED */ 1280int 1281getsockopt(td, uap) 1282 struct thread *td; 1283 register struct getsockopt_args /* { 1284 int s; 1285 int level; 1286 int name; 1287 caddr_t val; 1288 int *avalsize; 1289 } */ *uap; 1290{ 1291 int valsize, error; 1292 struct socket *so; 1293 struct sockopt sopt; 1294 1295 mtx_lock(&Giant); 1296 if ((error = fgetsock(td, uap->s, &so, NULL)) != 0) 1297 goto done2; 1298 if (uap->val) { 1299 error = copyin(uap->avalsize, &valsize, sizeof (valsize)); 1300 if (error) 1301 goto done1; 1302 if (valsize < 0) { 1303 error = EINVAL; 1304 goto done1; 1305 } 1306 } else { 1307 valsize = 0; 1308 } 1309 1310 sopt.sopt_dir = SOPT_GET; 1311 sopt.sopt_level = uap->level; 1312 sopt.sopt_name = uap->name; 1313 sopt.sopt_val = uap->val; 1314 sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */ 1315 sopt.sopt_td = td; 1316 1317 error = sogetopt(so, &sopt); 1318 if (error == 0) { 1319 valsize = sopt.sopt_valsize; 1320 error = copyout(&valsize, uap->avalsize, sizeof (valsize)); 1321 } 1322done1: 1323 fputsock(so); 1324done2: 1325 mtx_unlock(&Giant); 1326 return (error); 1327} 1328 1329/* 1330 * getsockname1() - Get socket name. 1331 * 1332 * MPSAFE 1333 */ 1334/* ARGSUSED */ 1335static int 1336getsockname1(td, uap, compat) 1337 struct thread *td; 1338 register struct getsockname_args /* { 1339 int fdes; 1340 caddr_t asa; 1341 int *alen; 1342 } */ *uap; 1343 int compat; 1344{ 1345 struct socket *so; 1346 struct sockaddr *sa; 1347 int len, error; 1348 1349 mtx_lock(&Giant); 1350 if ((error = fgetsock(td, uap->fdes, &so, NULL)) != 0) 1351 goto done2; 1352 error = copyin(uap->alen, &len, sizeof (len)); 1353 if (error) 1354 goto done1; 1355 if (len < 0) { 1356 error = EINVAL; 1357 goto done1; 1358 } 1359 sa = 0; 1360 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa); 1361 if (error) 1362 goto bad; 1363 if (sa == 0) { 1364 len = 0; 1365 goto gotnothing; 1366 } 1367 1368 len = MIN(len, sa->sa_len); 1369#ifdef COMPAT_OLDSOCK 1370 if (compat) 1371 ((struct osockaddr *)sa)->sa_family = sa->sa_family; 1372#endif 1373 error = copyout(sa, uap->asa, (u_int)len); 1374 if (error == 0) 1375gotnothing: 1376 error = copyout(&len, uap->alen, sizeof (len)); 1377bad: 1378 if (sa) 1379 FREE(sa, M_SONAME); 1380done1: 1381 fputsock(so); 1382done2: 1383 mtx_unlock(&Giant); 1384 return (error); 1385} 1386 1387/* 1388 * MPSAFE 1389 */ 1390int 1391getsockname(td, uap) 1392 struct thread *td; 1393 struct getsockname_args *uap; 1394{ 1395 1396 return (getsockname1(td, uap, 0)); 1397} 1398 1399#ifdef COMPAT_OLDSOCK 1400/* 1401 * MPSAFE 1402 */ 1403int 1404ogetsockname(td, uap) 1405 struct thread *td; 1406 struct getsockname_args *uap; 1407{ 1408 1409 return (getsockname1(td, uap, 1)); 1410} 1411#endif /* COMPAT_OLDSOCK */ 1412 1413/* 1414 * getpeername1() - Get name of peer for connected socket. 1415 * 1416 * MPSAFE 1417 */ 1418/* ARGSUSED */ 1419static int 1420getpeername1(td, uap, compat) 1421 struct thread *td; 1422 register struct getpeername_args /* { 1423 int fdes; 1424 caddr_t asa; 1425 int *alen; 1426 } */ *uap; 1427 int compat; 1428{ 1429 struct socket *so; 1430 struct sockaddr *sa; 1431 int len, error; 1432 1433 mtx_lock(&Giant); 1434 if ((error = fgetsock(td, uap->fdes, &so, NULL)) != 0) 1435 goto done2; 1436 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1437 error = ENOTCONN; 1438 goto done1; 1439 } 1440 error = copyin(uap->alen, &len, sizeof (len)); 1441 if (error) 1442 goto done1; 1443 if (len < 0) { 1444 error = EINVAL; 1445 goto done1; 1446 } 1447 sa = 0; 1448 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa); 1449 if (error) 1450 goto bad; 1451 if (sa == 0) { 1452 len = 0; 1453 goto gotnothing; 1454 } 1455 len = MIN(len, sa->sa_len); 1456#ifdef COMPAT_OLDSOCK 1457 if (compat) 1458 ((struct osockaddr *)sa)->sa_family = 1459 sa->sa_family; 1460#endif 1461 error = copyout(sa, uap->asa, (u_int)len); 1462 if (error) 1463 goto bad; 1464gotnothing: 1465 error = copyout(&len, uap->alen, sizeof (len)); 1466bad: 1467 if (sa) 1468 FREE(sa, M_SONAME); 1469done1: 1470 fputsock(so); 1471done2: 1472 mtx_unlock(&Giant); 1473 return (error); 1474} 1475 1476/* 1477 * MPSAFE 1478 */ 1479int 1480getpeername(td, uap) 1481 struct thread *td; 1482 struct getpeername_args *uap; 1483{ 1484 1485 return (getpeername1(td, uap, 0)); 1486} 1487 1488#ifdef COMPAT_OLDSOCK 1489/* 1490 * MPSAFE 1491 */ 1492int 1493ogetpeername(td, uap) 1494 struct thread *td; 1495 struct ogetpeername_args *uap; 1496{ 1497 1498 /* XXX uap should have type `getpeername_args *' to begin with. */ 1499 return (getpeername1(td, (struct getpeername_args *)uap, 1)); 1500} 1501#endif /* COMPAT_OLDSOCK */ 1502 1503int 1504sockargs(mp, buf, buflen, type) 1505 struct mbuf **mp; 1506 caddr_t buf; 1507 int buflen, type; 1508{ 1509 register struct sockaddr *sa; 1510 register struct mbuf *m; 1511 int error; 1512 1513 if ((u_int)buflen > MLEN) { 1514#ifdef COMPAT_OLDSOCK 1515 if (type == MT_SONAME && (u_int)buflen <= 112) 1516 buflen = MLEN; /* unix domain compat. hack */ 1517 else 1518#endif 1519 return (EINVAL); 1520 } 1521 m = m_get(M_TRYWAIT, type); 1522 if (m == NULL) 1523 return (ENOBUFS); 1524 m->m_len = buflen; 1525 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen); 1526 if (error) 1527 (void) m_free(m); 1528 else { 1529 *mp = m; 1530 if (type == MT_SONAME) { 1531 sa = mtod(m, struct sockaddr *); 1532 1533#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN 1534 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1535 sa->sa_family = sa->sa_len; 1536#endif 1537 sa->sa_len = buflen; 1538 } 1539 } 1540 return (error); 1541} 1542 1543int 1544getsockaddr(namp, uaddr, len) 1545 struct sockaddr **namp; 1546 caddr_t uaddr; 1547 size_t len; 1548{ 1549 struct sockaddr *sa; 1550 int error; 1551 1552 if (len > SOCK_MAXADDRLEN) 1553 return ENAMETOOLONG; 1554 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1555 error = copyin(uaddr, sa, len); 1556 if (error) { 1557 FREE(sa, M_SONAME); 1558 } else { 1559#if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN 1560 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1561 sa->sa_family = sa->sa_len; 1562#endif 1563 sa->sa_len = len; 1564 *namp = sa; 1565 } 1566 return error; 1567} 1568 1569/* 1570 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 1571 */ 1572static void 1573sf_buf_init(void *arg) 1574{ 1575 int i; 1576 1577 mtx_init(&sf_freelist.sf_lock, "sf_bufs list lock", NULL, MTX_DEF); 1578 mtx_lock(&sf_freelist.sf_lock); 1579 SLIST_INIT(&sf_freelist.sf_head); 1580 sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE); 1581 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, 1582 M_NOWAIT | M_ZERO); 1583 for (i = 0; i < nsfbufs; i++) { 1584 sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 1585 SLIST_INSERT_HEAD(&sf_freelist.sf_head, &sf_bufs[i], free_list); 1586 } 1587 sf_buf_alloc_want = 0; 1588 mtx_unlock(&sf_freelist.sf_lock); 1589} 1590 1591/* 1592 * Get an sf_buf from the freelist. Will block if none are available. 1593 */ 1594struct sf_buf * 1595sf_buf_alloc() 1596{ 1597 struct sf_buf *sf; 1598 int error; 1599 1600 mtx_lock(&sf_freelist.sf_lock); 1601 while ((sf = SLIST_FIRST(&sf_freelist.sf_head)) == NULL) { 1602 sf_buf_alloc_want++; 1603 error = msleep(&sf_freelist, &sf_freelist.sf_lock, PVM|PCATCH, 1604 "sfbufa", 0); 1605 sf_buf_alloc_want--; 1606 1607 /* 1608 * If we got a signal, don't risk going back to sleep. 1609 */ 1610 if (error) 1611 break; 1612 } 1613 if (sf != NULL) 1614 SLIST_REMOVE_HEAD(&sf_freelist.sf_head, free_list); 1615 mtx_unlock(&sf_freelist.sf_lock); 1616 return (sf); 1617} 1618 1619#define dtosf(x) (&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT]) 1620 1621/* 1622 * Detatch mapped page and release resources back to the system. 1623 */ 1624void 1625sf_buf_free(void *addr, void *args) 1626{ 1627 struct sf_buf *sf; 1628 struct vm_page *m; 1629 1630 GIANT_REQUIRED; 1631 1632 sf = dtosf(addr); 1633 pmap_qremove((vm_offset_t)addr, 1); 1634 m = sf->m; 1635 vm_page_lock_queues(); 1636 vm_page_unwire(m, 0); 1637 /* 1638 * Check for the object going away on us. This can 1639 * happen since we don't hold a reference to it. 1640 * If so, we're responsible for freeing the page. 1641 */ 1642 if (m->wire_count == 0 && m->object == NULL) 1643 vm_page_free(m); 1644 vm_page_unlock_queues(); 1645 sf->m = NULL; 1646 mtx_lock(&sf_freelist.sf_lock); 1647 SLIST_INSERT_HEAD(&sf_freelist.sf_head, sf, free_list); 1648 if (sf_buf_alloc_want > 0) 1649 wakeup_one(&sf_freelist); 1650 mtx_unlock(&sf_freelist.sf_lock); 1651} 1652 1653/* 1654 * sendfile(2) 1655 * 1656 * MPSAFE 1657 * 1658 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1659 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1660 * 1661 * Send a file specified by 'fd' and starting at 'offset' to a socket 1662 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1663 * nbytes == 0. Optionally add a header and/or trailer to the socket 1664 * output. If specified, write the total number of bytes sent into *sbytes. 1665 * 1666 */ 1667int 1668sendfile(struct thread *td, struct sendfile_args *uap) 1669{ 1670 1671 return (do_sendfile(td, uap, 0)); 1672} 1673 1674#ifdef COMPAT_FREEBSD4 1675int 1676freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap) 1677{ 1678 struct sendfile_args args; 1679 1680 args.fd = uap->fd; 1681 args.s = uap->s; 1682 args.offset = uap->offset; 1683 args.nbytes = uap->nbytes; 1684 args.hdtr = uap->hdtr; 1685 args.sbytes = uap->sbytes; 1686 args.flags = uap->flags; 1687 1688 return (do_sendfile(td, &args, 1)); 1689} 1690#endif /* COMPAT_FREEBSD4 */ 1691 1692static int 1693do_sendfile(struct thread *td, struct sendfile_args *uap, int compat) 1694{ 1695 struct vnode *vp; 1696 struct vm_object *obj; 1697 struct socket *so = NULL; 1698 struct mbuf *m; 1699 struct sf_buf *sf; 1700 struct vm_page *pg; 1701 struct writev_args nuap; 1702 struct sf_hdtr hdtr; 1703 off_t off, xfsize, hdtr_size, sbytes = 0; 1704 int error, s; 1705 1706 mtx_lock(&Giant); 1707 1708 hdtr_size = 0; 1709 1710 /* 1711 * The descriptor must be a regular file and have a backing VM object. 1712 */ 1713 if ((error = fgetvp_read(td, uap->fd, &vp)) != 0) 1714 goto done; 1715 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) { 1716 error = EINVAL; 1717 goto done; 1718 } 1719 if ((error = fgetsock(td, uap->s, &so, NULL)) != 0) 1720 goto done; 1721 if (so->so_type != SOCK_STREAM) { 1722 error = EINVAL; 1723 goto done; 1724 } 1725 if ((so->so_state & SS_ISCONNECTED) == 0) { 1726 error = ENOTCONN; 1727 goto done; 1728 } 1729 if (uap->offset < 0) { 1730 error = EINVAL; 1731 goto done; 1732 } 1733 1734 /* 1735 * If specified, get the pointer to the sf_hdtr struct for 1736 * any headers/trailers. 1737 */ 1738 if (uap->hdtr != NULL) { 1739 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1740 if (error) 1741 goto done; 1742 /* 1743 * Send any headers. Wimp out and use writev(2). 1744 */ 1745 if (hdtr.headers != NULL) { 1746 nuap.fd = uap->s; 1747 nuap.iovp = hdtr.headers; 1748 nuap.iovcnt = hdtr.hdr_cnt; 1749 error = writev(td, &nuap); 1750 if (error) 1751 goto done; 1752 if (compat) 1753 sbytes += td->td_retval[0]; 1754 else 1755 hdtr_size += td->td_retval[0]; 1756 } 1757 } 1758 1759 /* 1760 * Protect against multiple writers to the socket. 1761 */ 1762 (void) sblock(&so->so_snd, M_WAITOK); 1763 1764 /* 1765 * Loop through the pages in the file, starting with the requested 1766 * offset. Get a file page (do I/O if necessary), map the file page 1767 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1768 * it on the socket. 1769 */ 1770 for (off = uap->offset; ; off += xfsize, sbytes += xfsize) { 1771 vm_pindex_t pindex; 1772 vm_offset_t pgoff; 1773 1774 pindex = OFF_TO_IDX(off); 1775retry_lookup: 1776 /* 1777 * Calculate the amount to transfer. Not to exceed a page, 1778 * the EOF, or the passed in nbytes. 1779 */ 1780 xfsize = obj->un_pager.vnp.vnp_size - off; 1781 if (xfsize > PAGE_SIZE) 1782 xfsize = PAGE_SIZE; 1783 pgoff = (vm_offset_t)(off & PAGE_MASK); 1784 if (PAGE_SIZE - pgoff < xfsize) 1785 xfsize = PAGE_SIZE - pgoff; 1786 if (uap->nbytes && xfsize > (uap->nbytes - sbytes)) 1787 xfsize = uap->nbytes - sbytes; 1788 if (xfsize <= 0) 1789 break; 1790 /* 1791 * Optimize the non-blocking case by looking at the socket space 1792 * before going to the extra work of constituting the sf_buf. 1793 */ 1794 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) { 1795 if (so->so_state & SS_CANTSENDMORE) 1796 error = EPIPE; 1797 else 1798 error = EAGAIN; 1799 sbunlock(&so->so_snd); 1800 goto done; 1801 } 1802 /* 1803 * Attempt to look up the page. 1804 * 1805 * Allocate if not found 1806 * 1807 * Wait and loop if busy. 1808 */ 1809 pg = vm_page_lookup(obj, pindex); 1810 1811 if (pg == NULL) { 1812 pg = vm_page_alloc(obj, pindex, 1813 VM_ALLOC_NORMAL | VM_ALLOC_WIRED); 1814 if (pg == NULL) { 1815 VM_WAIT; 1816 goto retry_lookup; 1817 } 1818 vm_page_lock_queues(); 1819 vm_page_wakeup(pg); 1820 } else { 1821 vm_page_lock_queues(); 1822 if (vm_page_sleep_if_busy(pg, TRUE, "sfpbsy")) 1823 goto retry_lookup; 1824 /* 1825 * Wire the page so it does not get ripped out from 1826 * under us. 1827 */ 1828 vm_page_wire(pg); 1829 } 1830 1831 /* 1832 * If page is not valid for what we need, initiate I/O 1833 */ 1834 1835 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1836 int bsize, resid; 1837 1838 /* 1839 * Ensure that our page is still around when the I/O 1840 * completes. 1841 */ 1842 vm_page_io_start(pg); 1843 vm_page_unlock_queues(); 1844 1845 /* 1846 * Get the page from backing store. 1847 */ 1848 bsize = vp->v_mount->mnt_stat.f_iosize; 1849 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td); 1850 error = vn_rdwr(UIO_READ, vp, NULL, MAXBSIZE, 1851 trunc_page(off), UIO_NOCOPY, IO_NODELOCKED | 1852 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1853 td->td_ucred, &resid, td); 1854 VOP_UNLOCK(vp, 0, td); 1855 vm_page_lock_queues(); 1856 vm_page_flag_clear(pg, PG_ZERO); 1857 vm_page_io_finish(pg); 1858 if (error) { 1859 vm_page_unwire(pg, 0); 1860 /* 1861 * See if anyone else might know about this page. 1862 * If not and it is not valid, then free it. 1863 */ 1864 if (pg->wire_count == 0 && pg->valid == 0 && 1865 pg->busy == 0 && !(pg->flags & PG_BUSY) && 1866 pg->hold_count == 0) { 1867 vm_page_busy(pg); 1868 vm_page_free(pg); 1869 } 1870 vm_page_unlock_queues(); 1871 sbunlock(&so->so_snd); 1872 goto done; 1873 } 1874 } 1875 vm_page_unlock_queues(); 1876 1877 /* 1878 * Get a sendfile buf. We usually wait as long as necessary, 1879 * but this wait can be interrupted. 1880 */ 1881 if ((sf = sf_buf_alloc()) == NULL) { 1882 vm_page_lock_queues(); 1883 vm_page_unwire(pg, 0); 1884 if (pg->wire_count == 0 && pg->object == NULL) 1885 vm_page_free(pg); 1886 vm_page_unlock_queues(); 1887 sbunlock(&so->so_snd); 1888 error = EINTR; 1889 goto done; 1890 } 1891 1892 /* 1893 * Allocate a kernel virtual page and insert the physical page 1894 * into it. 1895 */ 1896 sf->m = pg; 1897 pmap_qenter(sf->kva, &pg, 1); 1898 /* 1899 * Get an mbuf header and set it up as having external storage. 1900 */ 1901 MGETHDR(m, M_TRYWAIT, MT_DATA); 1902 if (m == NULL) { 1903 error = ENOBUFS; 1904 sf_buf_free((void *)sf->kva, NULL); 1905 sbunlock(&so->so_snd); 1906 goto done; 1907 } 1908 /* 1909 * Setup external storage for mbuf. 1910 */ 1911 MEXTADD(m, sf->kva, PAGE_SIZE, sf_buf_free, NULL, M_RDONLY, 1912 EXT_SFBUF); 1913 m->m_data = (char *) sf->kva + pgoff; 1914 m->m_pkthdr.len = m->m_len = xfsize; 1915 /* 1916 * Add the buffer to the socket buffer chain. 1917 */ 1918 s = splnet(); 1919retry_space: 1920 /* 1921 * Make sure that the socket is still able to take more data. 1922 * CANTSENDMORE being true usually means that the connection 1923 * was closed. so_error is true when an error was sensed after 1924 * a previous send. 1925 * The state is checked after the page mapping and buffer 1926 * allocation above since those operations may block and make 1927 * any socket checks stale. From this point forward, nothing 1928 * blocks before the pru_send (or more accurately, any blocking 1929 * results in a loop back to here to re-check). 1930 */ 1931 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1932 if (so->so_state & SS_CANTSENDMORE) { 1933 error = EPIPE; 1934 } else { 1935 error = so->so_error; 1936 so->so_error = 0; 1937 } 1938 m_freem(m); 1939 sbunlock(&so->so_snd); 1940 splx(s); 1941 goto done; 1942 } 1943 /* 1944 * Wait for socket space to become available. We do this just 1945 * after checking the connection state above in order to avoid 1946 * a race condition with sbwait(). 1947 */ 1948 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { 1949 if (so->so_state & SS_NBIO) { 1950 m_freem(m); 1951 sbunlock(&so->so_snd); 1952 splx(s); 1953 error = EAGAIN; 1954 goto done; 1955 } 1956 error = sbwait(&so->so_snd); 1957 /* 1958 * An error from sbwait usually indicates that we've 1959 * been interrupted by a signal. If we've sent anything 1960 * then return bytes sent, otherwise return the error. 1961 */ 1962 if (error) { 1963 m_freem(m); 1964 sbunlock(&so->so_snd); 1965 splx(s); 1966 goto done; 1967 } 1968 goto retry_space; 1969 } 1970 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td); 1971 splx(s); 1972 if (error) { 1973 sbunlock(&so->so_snd); 1974 goto done; 1975 } 1976 } 1977 sbunlock(&so->so_snd); 1978 1979 /* 1980 * Send trailers. Wimp out and use writev(2). 1981 */ 1982 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1983 nuap.fd = uap->s; 1984 nuap.iovp = hdtr.trailers; 1985 nuap.iovcnt = hdtr.trl_cnt; 1986 error = writev(td, &nuap); 1987 if (error) 1988 goto done; 1989 if (compat) 1990 sbytes += td->td_retval[0]; 1991 else 1992 hdtr_size += td->td_retval[0]; 1993 } 1994 1995done: 1996 /* 1997 * If there was no error we have to clear td->td_retval[0] 1998 * because it may have been set by writev. 1999 */ 2000 if (error == 0) { 2001 td->td_retval[0] = 0; 2002 } 2003 if (uap->sbytes != NULL) { 2004 if (!compat) 2005 sbytes += hdtr_size; 2006 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 2007 } 2008 if (vp) 2009 vrele(vp); 2010 if (so) 2011 fputsock(so); 2012 mtx_unlock(&Giant); 2013 return (error); 2014} 2015