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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 33 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 34 */ 35 36 #include "opt_ktrace.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/sysproto.h> 42 #include <sys/malloc.h> 43 #include <sys/filedesc.h> 44 #include <sys/event.h> 45 #include <sys/proc.h> 46 #include <sys/fcntl.h> 47 #include <sys/file.h> 48 #include <sys/filio.h> 49 #include <sys/kern_syscall.h> 50 #include <sys/mbuf.h> 51 #include <sys/protosw.h> 52 #include <sys/sfbuf.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/socketops.h> 56 #include <sys/uio.h> 57 #include <sys/vnode.h> 58 #include <sys/lock.h> 59 #include <sys/mount.h> 60 #ifdef KTRACE 61 #include <sys/ktrace.h> 62 #endif 63 #include <vm/vm.h> 64 #include <vm/vm_object.h> 65 #include <vm/vm_page.h> 66 #include <vm/vm_pageout.h> 67 #include <vm/vm_kern.h> 68 #include <vm/vm_extern.h> 69 #include <sys/file2.h> 70 #include <sys/signalvar.h> 71 #include <sys/serialize.h> 72 73 #include <sys/thread2.h> 74 #include <sys/msgport2.h> 75 #include <sys/socketvar2.h> 76 #include <net/netmsg2.h> 77 #include <vm/vm_page2.h> 78 79 extern int use_soaccept_pred_fast; 80 extern int use_sendfile_async; 81 extern int use_soconnect_async; 82 83 /* 84 * System call interface to the socket abstraction. 85 */ 86 87 extern struct fileops socketops; 88 89 /* 90 * socket_args(int domain, int type, int protocol) 91 */ 92 int 93 kern_socket(int domain, int type, int protocol, int *res) 94 { 95 struct thread *td = curthread; 96 struct filedesc *fdp = td->td_proc->p_fd; 97 struct socket *so; 98 struct file *fp; 99 int fd, error; 100 101 KKASSERT(td->td_lwp); 102 103 error = falloc(td->td_lwp, &fp, &fd); 104 if (error) 105 return (error); 106 error = socreate(domain, &so, type, protocol, td); 107 if (error) { 108 fsetfd(fdp, NULL, fd); 109 } else { 110 fp->f_type = DTYPE_SOCKET; 111 fp->f_flag = FREAD | FWRITE; 112 fp->f_ops = &socketops; 113 fp->f_data = so; 114 *res = fd; 115 fsetfd(fdp, fp, fd); 116 } 117 fdrop(fp); 118 return (error); 119 } 120 121 /* 122 * MPALMOSTSAFE 123 */ 124 int 125 sys_socket(struct socket_args *uap) 126 { 127 int error; 128 129 error = kern_socket(uap->domain, uap->type, uap->protocol, 130 &uap->sysmsg_iresult); 131 132 return (error); 133 } 134 135 int 136 kern_bind(int s, struct sockaddr *sa) 137 { 138 struct thread *td = curthread; 139 struct proc *p = td->td_proc; 140 struct file *fp; 141 int error; 142 143 KKASSERT(p); 144 error = holdsock(p->p_fd, s, &fp); 145 if (error) 146 return (error); 147 error = sobind((struct socket *)fp->f_data, sa, td); 148 fdrop(fp); 149 return (error); 150 } 151 152 /* 153 * bind_args(int s, caddr_t name, int namelen) 154 * 155 * MPALMOSTSAFE 156 */ 157 int 158 sys_bind(struct bind_args *uap) 159 { 160 struct sockaddr *sa; 161 int error; 162 163 error = getsockaddr(&sa, uap->name, uap->namelen); 164 if (error) 165 return (error); 166 error = kern_bind(uap->s, sa); 167 kfree(sa, M_SONAME); 168 169 return (error); 170 } 171 172 int 173 kern_listen(int s, int backlog) 174 { 175 struct thread *td = curthread; 176 struct proc *p = td->td_proc; 177 struct file *fp; 178 int error; 179 180 KKASSERT(p); 181 error = holdsock(p->p_fd, s, &fp); 182 if (error) 183 return (error); 184 error = solisten((struct socket *)fp->f_data, backlog, td); 185 fdrop(fp); 186 return(error); 187 } 188 189 /* 190 * listen_args(int s, int backlog) 191 * 192 * MPALMOSTSAFE 193 */ 194 int 195 sys_listen(struct listen_args *uap) 196 { 197 int error; 198 199 error = kern_listen(uap->s, uap->backlog); 200 return (error); 201 } 202 203 /* 204 * Returns the accepted socket as well. 205 * 206 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 207 * pool token is absolutely required to avoid a sofree() race, 208 * as well as to avoid tailq handling races. 209 */ 210 static boolean_t 211 soaccept_predicate(struct netmsg_so_notify *msg) 212 { 213 struct socket *head = msg->base.nm_so; 214 struct socket *so; 215 216 if (head->so_error != 0) { 217 msg->base.lmsg.ms_error = head->so_error; 218 return (TRUE); 219 } 220 lwkt_getpooltoken(head); 221 if (!TAILQ_EMPTY(&head->so_comp)) { 222 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 223 so = TAILQ_FIRST(&head->so_comp); 224 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP); 225 TAILQ_REMOVE(&head->so_comp, so, so_list); 226 head->so_qlen--; 227 soclrstate(so, SS_COMP); 228 229 /* 230 * Keep a reference before clearing the so_head 231 * to avoid racing socket close in netisr. 232 */ 233 soreference(so); 234 so->so_head = NULL; 235 236 lwkt_relpooltoken(head); 237 238 msg->base.lmsg.ms_error = 0; 239 msg->base.nm_so = so; 240 return (TRUE); 241 } 242 lwkt_relpooltoken(head); 243 if (head->so_state & SS_CANTRCVMORE) { 244 msg->base.lmsg.ms_error = ECONNABORTED; 245 return (TRUE); 246 } 247 if (msg->nm_fflags & FNONBLOCK) { 248 msg->base.lmsg.ms_error = EWOULDBLOCK; 249 return (TRUE); 250 } 251 252 return (FALSE); 253 } 254 255 /* 256 * The second argument to kern_accept() is a handle to a struct sockaddr. 257 * This allows kern_accept() to return a pointer to an allocated struct 258 * sockaddr which must be freed later with FREE(). The caller must 259 * initialize *name to NULL. 260 */ 261 int 262 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 263 { 264 struct thread *td = curthread; 265 struct filedesc *fdp = td->td_proc->p_fd; 266 struct file *lfp = NULL; 267 struct file *nfp = NULL; 268 struct sockaddr *sa; 269 struct socket *head, *so; 270 struct netmsg_so_notify msg; 271 int fd; 272 u_int fflag; /* type must match fp->f_flag */ 273 int error, tmp; 274 275 *res = -1; 276 if (name && namelen && *namelen < 0) 277 return (EINVAL); 278 279 error = holdsock(td->td_proc->p_fd, s, &lfp); 280 if (error) 281 return (error); 282 283 error = falloc(td->td_lwp, &nfp, &fd); 284 if (error) { /* Probably ran out of file descriptors. */ 285 fdrop(lfp); 286 return (error); 287 } 288 head = (struct socket *)lfp->f_data; 289 if ((head->so_options & SO_ACCEPTCONN) == 0) { 290 error = EINVAL; 291 goto done; 292 } 293 294 if (fflags & O_FBLOCKING) 295 fflags |= lfp->f_flag & ~FNONBLOCK; 296 else if (fflags & O_FNONBLOCKING) 297 fflags |= lfp->f_flag | FNONBLOCK; 298 else 299 fflags = lfp->f_flag; 300 301 if (use_soaccept_pred_fast) { 302 boolean_t pred; 303 304 /* Initialize necessary parts for soaccept_predicate() */ 305 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 306 msg.nm_fflags = fflags; 307 308 lwkt_getpooltoken(head); 309 pred = soaccept_predicate(&msg); 310 lwkt_relpooltoken(head); 311 312 if (pred) { 313 error = msg.base.lmsg.ms_error; 314 if (error) 315 goto done; 316 else 317 goto accepted; 318 } 319 } 320 321 /* optimize for uniprocessor case later XXX JH */ 322 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 323 0, netmsg_so_notify, netmsg_so_notify_doabort); 324 msg.nm_predicate = soaccept_predicate; 325 msg.nm_fflags = fflags; 326 msg.nm_etype = NM_REVENT; 327 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 328 if (error) 329 goto done; 330 331 accepted: 332 /* 333 * At this point we have the connection that's ready to be accepted. 334 * 335 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 336 * to eat the ref and turn it into a descriptor. 337 */ 338 so = msg.base.nm_so; 339 340 fflag = lfp->f_flag; 341 342 /* connection has been removed from the listen queue */ 343 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 344 345 if (head->so_sigio != NULL) 346 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 347 348 nfp->f_type = DTYPE_SOCKET; 349 nfp->f_flag = fflag; 350 nfp->f_ops = &socketops; 351 nfp->f_data = so; 352 /* Sync socket nonblocking/async state with file flags */ 353 tmp = fflag & FNONBLOCK; 354 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 355 tmp = fflag & FASYNC; 356 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 357 358 sa = NULL; 359 if (so->so_faddr != NULL) { 360 sa = so->so_faddr; 361 so->so_faddr = NULL; 362 363 soaccept_generic(so); 364 error = 0; 365 } else { 366 error = soaccept(so, &sa); 367 } 368 369 /* 370 * Set the returned name and namelen as applicable. Set the returned 371 * namelen to 0 for older code which might ignore the return value 372 * from accept. 373 */ 374 if (error == 0) { 375 if (sa && name && namelen) { 376 if (*namelen > sa->sa_len) 377 *namelen = sa->sa_len; 378 *name = sa; 379 } else { 380 if (sa) 381 kfree(sa, M_SONAME); 382 } 383 } 384 385 done: 386 /* 387 * If an error occured clear the reserved descriptor, else associate 388 * nfp with it. 389 * 390 * Note that *res is normally ignored if an error is returned but 391 * a syscall message will still have access to the result code. 392 */ 393 if (error) { 394 fsetfd(fdp, NULL, fd); 395 } else { 396 *res = fd; 397 fsetfd(fdp, nfp, fd); 398 } 399 fdrop(nfp); 400 fdrop(lfp); 401 return (error); 402 } 403 404 /* 405 * accept(int s, caddr_t name, int *anamelen) 406 * 407 * MPALMOSTSAFE 408 */ 409 int 410 sys_accept(struct accept_args *uap) 411 { 412 struct sockaddr *sa = NULL; 413 int sa_len; 414 int error; 415 416 if (uap->name) { 417 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 418 if (error) 419 return (error); 420 421 error = kern_accept(uap->s, 0, &sa, &sa_len, 422 &uap->sysmsg_iresult); 423 424 if (error == 0) 425 error = copyout(sa, uap->name, sa_len); 426 if (error == 0) { 427 error = copyout(&sa_len, uap->anamelen, 428 sizeof(*uap->anamelen)); 429 } 430 if (sa) 431 kfree(sa, M_SONAME); 432 } else { 433 error = kern_accept(uap->s, 0, NULL, 0, 434 &uap->sysmsg_iresult); 435 } 436 return (error); 437 } 438 439 /* 440 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 441 * 442 * MPALMOSTSAFE 443 */ 444 int 445 sys_extaccept(struct extaccept_args *uap) 446 { 447 struct sockaddr *sa = NULL; 448 int sa_len; 449 int error; 450 int fflags = uap->flags & O_FMASK; 451 452 if (uap->name) { 453 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 454 if (error) 455 return (error); 456 457 error = kern_accept(uap->s, fflags, &sa, &sa_len, 458 &uap->sysmsg_iresult); 459 460 if (error == 0) 461 error = copyout(sa, uap->name, sa_len); 462 if (error == 0) { 463 error = copyout(&sa_len, uap->anamelen, 464 sizeof(*uap->anamelen)); 465 } 466 if (sa) 467 kfree(sa, M_SONAME); 468 } else { 469 error = kern_accept(uap->s, fflags, NULL, 0, 470 &uap->sysmsg_iresult); 471 } 472 return (error); 473 } 474 475 476 /* 477 * Returns TRUE if predicate satisfied. 478 */ 479 static boolean_t 480 soconnected_predicate(struct netmsg_so_notify *msg) 481 { 482 struct socket *so = msg->base.nm_so; 483 484 /* check predicate */ 485 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 486 msg->base.lmsg.ms_error = so->so_error; 487 return (TRUE); 488 } 489 490 return (FALSE); 491 } 492 493 int 494 kern_connect(int s, int fflags, struct sockaddr *sa) 495 { 496 struct thread *td = curthread; 497 struct proc *p = td->td_proc; 498 struct file *fp; 499 struct socket *so; 500 int error, interrupted = 0; 501 502 error = holdsock(p->p_fd, s, &fp); 503 if (error) 504 return (error); 505 so = (struct socket *)fp->f_data; 506 507 if (fflags & O_FBLOCKING) 508 /* fflags &= ~FNONBLOCK; */; 509 else if (fflags & O_FNONBLOCKING) 510 fflags |= FNONBLOCK; 511 else 512 fflags = fp->f_flag; 513 514 if (so->so_state & SS_ISCONNECTING) { 515 error = EALREADY; 516 goto done; 517 } 518 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE); 519 if (error) 520 goto bad; 521 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 522 error = EINPROGRESS; 523 goto done; 524 } 525 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 526 struct netmsg_so_notify msg; 527 528 netmsg_init_abortable(&msg.base, so, 529 &curthread->td_msgport, 530 0, 531 netmsg_so_notify, 532 netmsg_so_notify_doabort); 533 msg.nm_predicate = soconnected_predicate; 534 msg.nm_etype = NM_REVENT; 535 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 536 if (error == EINTR || error == ERESTART) 537 interrupted = 1; 538 } 539 if (error == 0) { 540 error = so->so_error; 541 so->so_error = 0; 542 } 543 bad: 544 if (!interrupted) 545 soclrstate(so, SS_ISCONNECTING); 546 if (error == ERESTART) 547 error = EINTR; 548 done: 549 fdrop(fp); 550 return (error); 551 } 552 553 /* 554 * connect_args(int s, caddr_t name, int namelen) 555 * 556 * MPALMOSTSAFE 557 */ 558 int 559 sys_connect(struct connect_args *uap) 560 { 561 struct sockaddr *sa; 562 int error; 563 564 error = getsockaddr(&sa, uap->name, uap->namelen); 565 if (error) 566 return (error); 567 error = kern_connect(uap->s, 0, sa); 568 kfree(sa, M_SONAME); 569 570 return (error); 571 } 572 573 /* 574 * connect_args(int s, int fflags, caddr_t name, int namelen) 575 * 576 * MPALMOSTSAFE 577 */ 578 int 579 sys_extconnect(struct extconnect_args *uap) 580 { 581 struct sockaddr *sa; 582 int error; 583 int fflags = uap->flags & O_FMASK; 584 585 error = getsockaddr(&sa, uap->name, uap->namelen); 586 if (error) 587 return (error); 588 error = kern_connect(uap->s, fflags, sa); 589 kfree(sa, M_SONAME); 590 591 return (error); 592 } 593 594 int 595 kern_socketpair(int domain, int type, int protocol, int *sv) 596 { 597 struct thread *td = curthread; 598 struct filedesc *fdp; 599 struct file *fp1, *fp2; 600 struct socket *so1, *so2; 601 int fd1, fd2, error; 602 603 fdp = td->td_proc->p_fd; 604 error = socreate(domain, &so1, type, protocol, td); 605 if (error) 606 return (error); 607 error = socreate(domain, &so2, type, protocol, td); 608 if (error) 609 goto free1; 610 error = falloc(td->td_lwp, &fp1, &fd1); 611 if (error) 612 goto free2; 613 sv[0] = fd1; 614 fp1->f_data = so1; 615 error = falloc(td->td_lwp, &fp2, &fd2); 616 if (error) 617 goto free3; 618 fp2->f_data = so2; 619 sv[1] = fd2; 620 error = soconnect2(so1, so2); 621 if (error) 622 goto free4; 623 if (type == SOCK_DGRAM) { 624 /* 625 * Datagram socket connection is asymmetric. 626 */ 627 error = soconnect2(so2, so1); 628 if (error) 629 goto free4; 630 } 631 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 632 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 633 fp1->f_ops = fp2->f_ops = &socketops; 634 fsetfd(fdp, fp1, fd1); 635 fsetfd(fdp, fp2, fd2); 636 fdrop(fp1); 637 fdrop(fp2); 638 return (error); 639 free4: 640 fsetfd(fdp, NULL, fd2); 641 fdrop(fp2); 642 free3: 643 fsetfd(fdp, NULL, fd1); 644 fdrop(fp1); 645 free2: 646 (void)soclose(so2, 0); 647 free1: 648 (void)soclose(so1, 0); 649 return (error); 650 } 651 652 /* 653 * socketpair(int domain, int type, int protocol, int *rsv) 654 */ 655 int 656 sys_socketpair(struct socketpair_args *uap) 657 { 658 int error, sockv[2]; 659 660 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 661 662 if (error == 0) { 663 error = copyout(sockv, uap->rsv, sizeof(sockv)); 664 665 if (error != 0) { 666 kern_close(sockv[0]); 667 kern_close(sockv[1]); 668 } 669 } 670 671 return (error); 672 } 673 674 int 675 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 676 struct mbuf *control, int flags, size_t *res) 677 { 678 struct thread *td = curthread; 679 struct lwp *lp = td->td_lwp; 680 struct proc *p = td->td_proc; 681 struct file *fp; 682 size_t len; 683 int error; 684 struct socket *so; 685 #ifdef KTRACE 686 struct iovec *ktriov = NULL; 687 struct uio ktruio; 688 #endif 689 690 error = holdsock(p->p_fd, s, &fp); 691 if (error) 692 return (error); 693 #ifdef KTRACE 694 if (KTRPOINT(td, KTR_GENIO)) { 695 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 696 697 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 698 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 699 ktruio = *auio; 700 } 701 #endif 702 len = auio->uio_resid; 703 so = (struct socket *)fp->f_data; 704 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 705 if (fp->f_flag & FNONBLOCK) 706 flags |= MSG_FNONBLOCKING; 707 } 708 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 709 if (error) { 710 if (auio->uio_resid != len && (error == ERESTART || 711 error == EINTR || error == EWOULDBLOCK)) 712 error = 0; 713 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 714 !(so->so_options & SO_NOSIGPIPE)) 715 lwpsignal(p, lp, SIGPIPE); 716 } 717 #ifdef KTRACE 718 if (ktriov != NULL) { 719 if (error == 0) { 720 ktruio.uio_iov = ktriov; 721 ktruio.uio_resid = len - auio->uio_resid; 722 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 723 } 724 kfree(ktriov, M_TEMP); 725 } 726 #endif 727 if (error == 0) 728 *res = len - auio->uio_resid; 729 fdrop(fp); 730 return (error); 731 } 732 733 /* 734 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 735 * 736 * MPALMOSTSAFE 737 */ 738 int 739 sys_sendto(struct sendto_args *uap) 740 { 741 struct thread *td = curthread; 742 struct uio auio; 743 struct iovec aiov; 744 struct sockaddr *sa = NULL; 745 int error; 746 747 if (uap->to) { 748 error = getsockaddr(&sa, uap->to, uap->tolen); 749 if (error) 750 return (error); 751 } 752 aiov.iov_base = uap->buf; 753 aiov.iov_len = uap->len; 754 auio.uio_iov = &aiov; 755 auio.uio_iovcnt = 1; 756 auio.uio_offset = 0; 757 auio.uio_resid = uap->len; 758 auio.uio_segflg = UIO_USERSPACE; 759 auio.uio_rw = UIO_WRITE; 760 auio.uio_td = td; 761 762 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 763 &uap->sysmsg_szresult); 764 765 if (sa) 766 kfree(sa, M_SONAME); 767 return (error); 768 } 769 770 /* 771 * sendmsg_args(int s, caddr_t msg, int flags) 772 * 773 * MPALMOSTSAFE 774 */ 775 int 776 sys_sendmsg(struct sendmsg_args *uap) 777 { 778 struct thread *td = curthread; 779 struct msghdr msg; 780 struct uio auio; 781 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 782 struct sockaddr *sa = NULL; 783 struct mbuf *control = NULL; 784 int error; 785 786 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 787 if (error) 788 return (error); 789 790 /* 791 * Conditionally copyin msg.msg_name. 792 */ 793 if (msg.msg_name) { 794 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 795 if (error) 796 return (error); 797 } 798 799 /* 800 * Populate auio. 801 */ 802 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 803 &auio.uio_resid); 804 if (error) 805 goto cleanup2; 806 auio.uio_iov = iov; 807 auio.uio_iovcnt = msg.msg_iovlen; 808 auio.uio_offset = 0; 809 auio.uio_segflg = UIO_USERSPACE; 810 auio.uio_rw = UIO_WRITE; 811 auio.uio_td = td; 812 813 /* 814 * Conditionally copyin msg.msg_control. 815 */ 816 if (msg.msg_control) { 817 if (msg.msg_controllen < sizeof(struct cmsghdr) || 818 msg.msg_controllen > MLEN) { 819 error = EINVAL; 820 goto cleanup; 821 } 822 control = m_get(M_WAITOK, MT_CONTROL); 823 if (control == NULL) { 824 error = ENOBUFS; 825 goto cleanup; 826 } 827 control->m_len = msg.msg_controllen; 828 error = copyin(msg.msg_control, mtod(control, caddr_t), 829 msg.msg_controllen); 830 if (error) { 831 m_free(control); 832 goto cleanup; 833 } 834 } 835 836 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 837 &uap->sysmsg_szresult); 838 839 cleanup: 840 iovec_free(&iov, aiov); 841 cleanup2: 842 if (sa) 843 kfree(sa, M_SONAME); 844 return (error); 845 } 846 847 /* 848 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 849 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 850 * Don't forget to FREE() and m_free() these if they are returned. 851 */ 852 int 853 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 854 struct mbuf **control, int *flags, size_t *res) 855 { 856 struct thread *td = curthread; 857 struct proc *p = td->td_proc; 858 struct file *fp; 859 size_t len; 860 int error; 861 int lflags; 862 struct socket *so; 863 #ifdef KTRACE 864 struct iovec *ktriov = NULL; 865 struct uio ktruio; 866 #endif 867 868 error = holdsock(p->p_fd, s, &fp); 869 if (error) 870 return (error); 871 #ifdef KTRACE 872 if (KTRPOINT(td, KTR_GENIO)) { 873 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 874 875 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 876 bcopy(auio->uio_iov, ktriov, iovlen); 877 ktruio = *auio; 878 } 879 #endif 880 len = auio->uio_resid; 881 so = (struct socket *)fp->f_data; 882 883 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 884 if (fp->f_flag & FNONBLOCK) { 885 if (flags) { 886 *flags |= MSG_FNONBLOCKING; 887 } else { 888 lflags = MSG_FNONBLOCKING; 889 flags = &lflags; 890 } 891 } 892 } 893 894 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 895 if (error) { 896 if (auio->uio_resid != len && (error == ERESTART || 897 error == EINTR || error == EWOULDBLOCK)) 898 error = 0; 899 } 900 #ifdef KTRACE 901 if (ktriov != NULL) { 902 if (error == 0) { 903 ktruio.uio_iov = ktriov; 904 ktruio.uio_resid = len - auio->uio_resid; 905 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 906 } 907 kfree(ktriov, M_TEMP); 908 } 909 #endif 910 if (error == 0) 911 *res = len - auio->uio_resid; 912 fdrop(fp); 913 return (error); 914 } 915 916 /* 917 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 918 * caddr_t from, int *fromlenaddr) 919 * 920 * MPALMOSTSAFE 921 */ 922 int 923 sys_recvfrom(struct recvfrom_args *uap) 924 { 925 struct thread *td = curthread; 926 struct uio auio; 927 struct iovec aiov; 928 struct sockaddr *sa = NULL; 929 int error, fromlen; 930 931 if (uap->from && uap->fromlenaddr) { 932 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 933 if (error) 934 return (error); 935 if (fromlen < 0) 936 return (EINVAL); 937 } else { 938 fromlen = 0; 939 } 940 aiov.iov_base = uap->buf; 941 aiov.iov_len = uap->len; 942 auio.uio_iov = &aiov; 943 auio.uio_iovcnt = 1; 944 auio.uio_offset = 0; 945 auio.uio_resid = uap->len; 946 auio.uio_segflg = UIO_USERSPACE; 947 auio.uio_rw = UIO_READ; 948 auio.uio_td = td; 949 950 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 951 &uap->flags, &uap->sysmsg_szresult); 952 953 if (error == 0 && uap->from) { 954 /* note: sa may still be NULL */ 955 if (sa) { 956 fromlen = MIN(fromlen, sa->sa_len); 957 error = copyout(sa, uap->from, fromlen); 958 } else { 959 fromlen = 0; 960 } 961 if (error == 0) { 962 error = copyout(&fromlen, uap->fromlenaddr, 963 sizeof(fromlen)); 964 } 965 } 966 if (sa) 967 kfree(sa, M_SONAME); 968 969 return (error); 970 } 971 972 /* 973 * recvmsg_args(int s, struct msghdr *msg, int flags) 974 * 975 * MPALMOSTSAFE 976 */ 977 int 978 sys_recvmsg(struct recvmsg_args *uap) 979 { 980 struct thread *td = curthread; 981 struct msghdr msg; 982 struct uio auio; 983 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 984 struct mbuf *m, *control = NULL; 985 struct sockaddr *sa = NULL; 986 caddr_t ctlbuf; 987 socklen_t *ufromlenp, *ucontrollenp; 988 int error, fromlen, controllen, len, flags, *uflagsp; 989 990 /* 991 * This copyin handles everything except the iovec. 992 */ 993 error = copyin(uap->msg, &msg, sizeof(msg)); 994 if (error) 995 return (error); 996 997 if (msg.msg_name && msg.msg_namelen < 0) 998 return (EINVAL); 999 if (msg.msg_control && msg.msg_controllen < 0) 1000 return (EINVAL); 1001 1002 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1003 msg_namelen)); 1004 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1005 msg_controllen)); 1006 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1007 msg_flags)); 1008 1009 /* 1010 * Populate auio. 1011 */ 1012 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1013 &auio.uio_resid); 1014 if (error) 1015 return (error); 1016 auio.uio_iov = iov; 1017 auio.uio_iovcnt = msg.msg_iovlen; 1018 auio.uio_offset = 0; 1019 auio.uio_segflg = UIO_USERSPACE; 1020 auio.uio_rw = UIO_READ; 1021 auio.uio_td = td; 1022 1023 flags = uap->flags; 1024 1025 error = kern_recvmsg(uap->s, 1026 (msg.msg_name ? &sa : NULL), &auio, 1027 (msg.msg_control ? &control : NULL), &flags, 1028 &uap->sysmsg_szresult); 1029 1030 /* 1031 * Conditionally copyout the name and populate the namelen field. 1032 */ 1033 if (error == 0 && msg.msg_name) { 1034 /* note: sa may still be NULL */ 1035 if (sa != NULL) { 1036 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1037 error = copyout(sa, msg.msg_name, fromlen); 1038 } else { 1039 fromlen = 0; 1040 } 1041 if (error == 0) 1042 error = copyout(&fromlen, ufromlenp, 1043 sizeof(*ufromlenp)); 1044 } 1045 1046 /* 1047 * Copyout msg.msg_control and msg.msg_controllen. 1048 */ 1049 if (error == 0 && msg.msg_control) { 1050 len = msg.msg_controllen; 1051 m = control; 1052 ctlbuf = (caddr_t)msg.msg_control; 1053 1054 while(m && len > 0) { 1055 unsigned int tocopy; 1056 1057 if (len >= m->m_len) { 1058 tocopy = m->m_len; 1059 } else { 1060 msg.msg_flags |= MSG_CTRUNC; 1061 tocopy = len; 1062 } 1063 1064 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1065 if (error) 1066 goto cleanup; 1067 1068 ctlbuf += tocopy; 1069 len -= tocopy; 1070 m = m->m_next; 1071 } 1072 controllen = ctlbuf - (caddr_t)msg.msg_control; 1073 error = copyout(&controllen, ucontrollenp, 1074 sizeof(*ucontrollenp)); 1075 } 1076 1077 if (error == 0) 1078 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1079 1080 cleanup: 1081 if (sa) 1082 kfree(sa, M_SONAME); 1083 iovec_free(&iov, aiov); 1084 if (control) 1085 m_freem(control); 1086 return (error); 1087 } 1088 1089 /* 1090 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1091 * in kernel pointer instead of a userland pointer. This allows us 1092 * to manipulate socket options in the emulation code. 1093 */ 1094 int 1095 kern_setsockopt(int s, struct sockopt *sopt) 1096 { 1097 struct thread *td = curthread; 1098 struct proc *p = td->td_proc; 1099 struct file *fp; 1100 int error; 1101 1102 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1103 return (EFAULT); 1104 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1105 return (EINVAL); 1106 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1107 return (EINVAL); 1108 1109 error = holdsock(p->p_fd, s, &fp); 1110 if (error) 1111 return (error); 1112 1113 error = sosetopt((struct socket *)fp->f_data, sopt); 1114 fdrop(fp); 1115 return (error); 1116 } 1117 1118 /* 1119 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1120 * 1121 * MPALMOSTSAFE 1122 */ 1123 int 1124 sys_setsockopt(struct setsockopt_args *uap) 1125 { 1126 struct thread *td = curthread; 1127 struct sockopt sopt; 1128 int error; 1129 1130 sopt.sopt_level = uap->level; 1131 sopt.sopt_name = uap->name; 1132 sopt.sopt_valsize = uap->valsize; 1133 sopt.sopt_td = td; 1134 sopt.sopt_val = NULL; 1135 1136 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1137 return (EINVAL); 1138 if (uap->val) { 1139 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1140 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1141 if (error) 1142 goto out; 1143 } 1144 1145 error = kern_setsockopt(uap->s, &sopt); 1146 out: 1147 if (uap->val) 1148 kfree(sopt.sopt_val, M_TEMP); 1149 return(error); 1150 } 1151 1152 /* 1153 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1154 * in kernel pointer instead of a userland pointer. This allows us 1155 * to manipulate socket options in the emulation code. 1156 */ 1157 int 1158 kern_getsockopt(int s, struct sockopt *sopt) 1159 { 1160 struct thread *td = curthread; 1161 struct proc *p = td->td_proc; 1162 struct file *fp; 1163 int error; 1164 1165 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1166 return (EFAULT); 1167 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1168 return (EINVAL); 1169 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1170 return (EINVAL); 1171 1172 error = holdsock(p->p_fd, s, &fp); 1173 if (error) 1174 return (error); 1175 1176 error = sogetopt((struct socket *)fp->f_data, sopt); 1177 fdrop(fp); 1178 return (error); 1179 } 1180 1181 /* 1182 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1183 * 1184 * MPALMOSTSAFE 1185 */ 1186 int 1187 sys_getsockopt(struct getsockopt_args *uap) 1188 { 1189 struct thread *td = curthread; 1190 struct sockopt sopt; 1191 int error, valsize; 1192 1193 if (uap->val) { 1194 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1195 if (error) 1196 return (error); 1197 } else { 1198 valsize = 0; 1199 } 1200 1201 sopt.sopt_level = uap->level; 1202 sopt.sopt_name = uap->name; 1203 sopt.sopt_valsize = valsize; 1204 sopt.sopt_td = td; 1205 sopt.sopt_val = NULL; 1206 1207 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1208 return (EINVAL); 1209 if (uap->val) { 1210 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1211 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1212 if (error) 1213 goto out; 1214 } 1215 1216 error = kern_getsockopt(uap->s, &sopt); 1217 if (error) 1218 goto out; 1219 valsize = sopt.sopt_valsize; 1220 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1221 if (error) 1222 goto out; 1223 if (uap->val) 1224 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1225 out: 1226 if (uap->val) 1227 kfree(sopt.sopt_val, M_TEMP); 1228 return (error); 1229 } 1230 1231 /* 1232 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1233 * This allows kern_getsockname() to return a pointer to an allocated struct 1234 * sockaddr which must be freed later with FREE(). The caller must 1235 * initialize *name to NULL. 1236 */ 1237 int 1238 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1239 { 1240 struct thread *td = curthread; 1241 struct proc *p = td->td_proc; 1242 struct file *fp; 1243 struct socket *so; 1244 struct sockaddr *sa = NULL; 1245 int error; 1246 1247 error = holdsock(p->p_fd, s, &fp); 1248 if (error) 1249 return (error); 1250 if (*namelen < 0) { 1251 fdrop(fp); 1252 return (EINVAL); 1253 } 1254 so = (struct socket *)fp->f_data; 1255 error = so_pru_sockaddr(so, &sa); 1256 if (error == 0) { 1257 if (sa == NULL) { 1258 *namelen = 0; 1259 } else { 1260 *namelen = MIN(*namelen, sa->sa_len); 1261 *name = sa; 1262 } 1263 } 1264 1265 fdrop(fp); 1266 return (error); 1267 } 1268 1269 /* 1270 * getsockname_args(int fdes, caddr_t asa, int *alen) 1271 * 1272 * Get socket name. 1273 * 1274 * MPALMOSTSAFE 1275 */ 1276 int 1277 sys_getsockname(struct getsockname_args *uap) 1278 { 1279 struct sockaddr *sa = NULL; 1280 int error, sa_len; 1281 1282 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1283 if (error) 1284 return (error); 1285 1286 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1287 1288 if (error == 0) 1289 error = copyout(sa, uap->asa, sa_len); 1290 if (error == 0) 1291 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1292 if (sa) 1293 kfree(sa, M_SONAME); 1294 return (error); 1295 } 1296 1297 /* 1298 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1299 * This allows kern_getpeername() to return a pointer to an allocated struct 1300 * sockaddr which must be freed later with FREE(). The caller must 1301 * initialize *name to NULL. 1302 */ 1303 int 1304 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1305 { 1306 struct thread *td = curthread; 1307 struct proc *p = td->td_proc; 1308 struct file *fp; 1309 struct socket *so; 1310 struct sockaddr *sa = NULL; 1311 int error; 1312 1313 error = holdsock(p->p_fd, s, &fp); 1314 if (error) 1315 return (error); 1316 if (*namelen < 0) { 1317 fdrop(fp); 1318 return (EINVAL); 1319 } 1320 so = (struct socket *)fp->f_data; 1321 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1322 fdrop(fp); 1323 return (ENOTCONN); 1324 } 1325 error = so_pru_peeraddr(so, &sa); 1326 if (error == 0) { 1327 if (sa == NULL) { 1328 *namelen = 0; 1329 } else { 1330 *namelen = MIN(*namelen, sa->sa_len); 1331 *name = sa; 1332 } 1333 } 1334 1335 fdrop(fp); 1336 return (error); 1337 } 1338 1339 /* 1340 * getpeername_args(int fdes, caddr_t asa, int *alen) 1341 * 1342 * Get name of peer for connected socket. 1343 * 1344 * MPALMOSTSAFE 1345 */ 1346 int 1347 sys_getpeername(struct getpeername_args *uap) 1348 { 1349 struct sockaddr *sa = NULL; 1350 int error, sa_len; 1351 1352 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1353 if (error) 1354 return (error); 1355 1356 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1357 1358 if (error == 0) 1359 error = copyout(sa, uap->asa, sa_len); 1360 if (error == 0) 1361 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1362 if (sa) 1363 kfree(sa, M_SONAME); 1364 return (error); 1365 } 1366 1367 int 1368 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1369 { 1370 struct sockaddr *sa; 1371 int error; 1372 1373 *namp = NULL; 1374 if (len > SOCK_MAXADDRLEN) 1375 return ENAMETOOLONG; 1376 if (len < offsetof(struct sockaddr, sa_data[0])) 1377 return EDOM; 1378 sa = kmalloc(len, M_SONAME, M_WAITOK); 1379 error = copyin(uaddr, sa, len); 1380 if (error) { 1381 kfree(sa, M_SONAME); 1382 } else { 1383 #if BYTE_ORDER != BIG_ENDIAN 1384 /* 1385 * The bind(), connect(), and sendto() syscalls were not 1386 * versioned for COMPAT_43. Thus, this check must stay. 1387 */ 1388 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1389 sa->sa_family = sa->sa_len; 1390 #endif 1391 sa->sa_len = len; 1392 *namp = sa; 1393 } 1394 return error; 1395 } 1396 1397 /* 1398 * Detach a mapped page and release resources back to the system. 1399 * We must release our wiring and if the object is ripped out 1400 * from under the vm_page we become responsible for freeing the 1401 * page. 1402 * 1403 * MPSAFE 1404 */ 1405 static void 1406 sf_buf_mfree(void *arg) 1407 { 1408 struct sf_buf *sf = arg; 1409 vm_page_t m; 1410 1411 m = sf_buf_page(sf); 1412 if (sf_buf_free(sf)) { 1413 /* sf invalid now */ 1414 /* 1415 vm_page_busy_wait(m, FALSE, "sockpgf"); 1416 vm_page_wakeup(m); 1417 */ 1418 vm_page_unhold(m); 1419 #if 0 1420 if (m->object == NULL && 1421 m->wire_count == 0 && 1422 (m->flags & PG_NEED_COMMIT) == 0) { 1423 vm_page_free(m); 1424 } else { 1425 vm_page_wakeup(m); 1426 } 1427 #endif 1428 } 1429 } 1430 1431 /* 1432 * sendfile(2). 1433 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1434 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1435 * 1436 * Send a file specified by 'fd' and starting at 'offset' to a socket 1437 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1438 * nbytes == 0. Optionally add a header and/or trailer to the socket 1439 * output. If specified, write the total number of bytes sent into *sbytes. 1440 * 1441 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1442 * the headers to count against the remaining bytes to be sent from 1443 * the file descriptor. We may wish to implement a compatibility syscall 1444 * in the future. 1445 * 1446 * MPALMOSTSAFE 1447 */ 1448 int 1449 sys_sendfile(struct sendfile_args *uap) 1450 { 1451 struct thread *td = curthread; 1452 struct proc *p = td->td_proc; 1453 struct file *fp; 1454 struct vnode *vp = NULL; 1455 struct sf_hdtr hdtr; 1456 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1457 struct uio auio; 1458 struct mbuf *mheader = NULL; 1459 size_t hbytes = 0; 1460 size_t tbytes; 1461 off_t hdtr_size = 0; 1462 off_t sbytes; 1463 int error; 1464 1465 KKASSERT(p); 1466 1467 /* 1468 * Do argument checking. Must be a regular file in, stream 1469 * type and connected socket out, positive offset. 1470 */ 1471 fp = holdfp(p->p_fd, uap->fd, FREAD); 1472 if (fp == NULL) { 1473 return (EBADF); 1474 } 1475 if (fp->f_type != DTYPE_VNODE) { 1476 fdrop(fp); 1477 return (EINVAL); 1478 } 1479 vp = (struct vnode *)fp->f_data; 1480 vref(vp); 1481 fdrop(fp); 1482 1483 /* 1484 * If specified, get the pointer to the sf_hdtr struct for 1485 * any headers/trailers. 1486 */ 1487 if (uap->hdtr) { 1488 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1489 if (error) 1490 goto done; 1491 /* 1492 * Send any headers. 1493 */ 1494 if (hdtr.headers) { 1495 error = iovec_copyin(hdtr.headers, &iov, aiov, 1496 hdtr.hdr_cnt, &hbytes); 1497 if (error) 1498 goto done; 1499 auio.uio_iov = iov; 1500 auio.uio_iovcnt = hdtr.hdr_cnt; 1501 auio.uio_offset = 0; 1502 auio.uio_segflg = UIO_USERSPACE; 1503 auio.uio_rw = UIO_WRITE; 1504 auio.uio_td = td; 1505 auio.uio_resid = hbytes; 1506 1507 mheader = m_uiomove(&auio); 1508 1509 iovec_free(&iov, aiov); 1510 if (mheader == NULL) 1511 goto done; 1512 } 1513 } 1514 1515 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1516 &sbytes, uap->flags); 1517 if (error) 1518 goto done; 1519 1520 /* 1521 * Send trailers. Wimp out and use writev(2). 1522 */ 1523 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1524 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1525 hdtr.trl_cnt, &auio.uio_resid); 1526 if (error) 1527 goto done; 1528 auio.uio_iov = iov; 1529 auio.uio_iovcnt = hdtr.trl_cnt; 1530 auio.uio_offset = 0; 1531 auio.uio_segflg = UIO_USERSPACE; 1532 auio.uio_rw = UIO_WRITE; 1533 auio.uio_td = td; 1534 1535 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1536 1537 iovec_free(&iov, aiov); 1538 if (error) 1539 goto done; 1540 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1541 } 1542 1543 done: 1544 if (vp) 1545 vrele(vp); 1546 if (uap->sbytes != NULL) { 1547 sbytes += hdtr_size; 1548 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1549 } 1550 return (error); 1551 } 1552 1553 int 1554 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1555 struct mbuf *mheader, off_t *sbytes, int flags) 1556 { 1557 struct thread *td = curthread; 1558 struct proc *p = td->td_proc; 1559 struct vm_object *obj; 1560 struct socket *so; 1561 struct file *fp; 1562 struct mbuf *m, *mp; 1563 struct sf_buf *sf; 1564 struct vm_page *pg; 1565 off_t off, xfsize, xbytes; 1566 off_t hbytes = 0; 1567 int error = 0; 1568 1569 if (vp->v_type != VREG) { 1570 error = EINVAL; 1571 goto done0; 1572 } 1573 if ((obj = vp->v_object) == NULL) { 1574 error = EINVAL; 1575 goto done0; 1576 } 1577 error = holdsock(p->p_fd, sfd, &fp); 1578 if (error) 1579 goto done0; 1580 so = (struct socket *)fp->f_data; 1581 if (so->so_type != SOCK_STREAM) { 1582 error = EINVAL; 1583 goto done; 1584 } 1585 if ((so->so_state & SS_ISCONNECTED) == 0) { 1586 error = ENOTCONN; 1587 goto done; 1588 } 1589 if (offset < 0) { 1590 error = EINVAL; 1591 goto done; 1592 } 1593 1594 /* 1595 * preallocation is required for asynchronous passing of mbufs, 1596 * otherwise we can wind up building up an infinite number of 1597 * mbufs during the asynchronous latency. 1598 */ 1599 if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) { 1600 error = EINVAL; 1601 goto done; 1602 } 1603 1604 *sbytes = 0; 1605 xbytes = 0; 1606 /* 1607 * Protect against multiple writers to the socket. 1608 */ 1609 ssb_lock(&so->so_snd, M_WAITOK); 1610 1611 /* 1612 * Loop through the pages in the file, starting with the requested 1613 * offset. Get a file page (do I/O if necessary), map the file page 1614 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1615 * it on the socket. 1616 */ 1617 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) { 1618 vm_pindex_t pindex; 1619 vm_offset_t pgoff; 1620 long space; 1621 1622 pindex = OFF_TO_IDX(off); 1623 retry_lookup: 1624 /* 1625 * Calculate the amount to transfer. Not to exceed a page, 1626 * the EOF, or the passed in nbytes. 1627 */ 1628 xfsize = vp->v_filesize - off; 1629 if (xfsize > PAGE_SIZE) 1630 xfsize = PAGE_SIZE; 1631 pgoff = (vm_offset_t)(off & PAGE_MASK); 1632 if (PAGE_SIZE - pgoff < xfsize) 1633 xfsize = PAGE_SIZE - pgoff; 1634 if (nbytes && xfsize > (nbytes - xbytes)) 1635 xfsize = nbytes - xbytes; 1636 if (xfsize <= 0) 1637 break; 1638 /* 1639 * Optimize the non-blocking case by looking at the socket space 1640 * before going to the extra work of constituting the sf_buf. 1641 */ 1642 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1643 space = ssb_space_prealloc(&so->so_snd); 1644 else 1645 space = ssb_space(&so->so_snd); 1646 1647 if ((fp->f_flag & FNONBLOCK) && space <= 0) { 1648 if (so->so_state & SS_CANTSENDMORE) 1649 error = EPIPE; 1650 else 1651 error = EAGAIN; 1652 ssb_unlock(&so->so_snd); 1653 goto done; 1654 } 1655 /* 1656 * Attempt to look up the page. 1657 * 1658 * Allocate if not found, wait and loop if busy, then hold the page. 1659 * We hold rather than wire the page because we do not want to prevent 1660 * filesystem truncation operations from occuring on the file. This 1661 * can happen even under normal operation if the file being sent is 1662 * remove()d after the sendfile() call completes, because the socket buffer 1663 * may still be draining. tmpfs will crash if we try to use wire. 1664 */ 1665 vm_object_hold(obj); 1666 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error); 1667 if (error) { 1668 vm_page_sleep_busy(pg, TRUE, "sfpbsy"); 1669 vm_object_drop(obj); 1670 goto retry_lookup; 1671 } 1672 if (pg == NULL) { 1673 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL | 1674 VM_ALLOC_NULL_OK); 1675 if (pg == NULL) { 1676 vm_wait(0); 1677 vm_object_drop(obj); 1678 goto retry_lookup; 1679 } 1680 } 1681 vm_page_hold(pg); 1682 vm_object_drop(obj); 1683 1684 /* 1685 * If page is not valid for what we need, initiate I/O 1686 */ 1687 1688 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1689 struct uio auio; 1690 struct iovec aiov; 1691 int bsize; 1692 1693 /* 1694 * Ensure that our page is still around when the I/O 1695 * completes. 1696 * 1697 * Ensure that our page is not modified while part of 1698 * a mbuf as this could mess up tcp checksums, DMA, 1699 * etc (XXX NEEDS WORK). The softbusy is supposed to 1700 * help here but it actually doesn't. 1701 * 1702 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying 1703 * VM pages are not protected by the soft-busy 1704 * unless we vm_page_protect... READ them, and 1705 * they STILL aren't protected against 1706 * modification via the buffer cache (VOP_WRITE). 1707 * 1708 * Fixing the second issue is particularly 1709 * difficult. 1710 * 1711 * XXX We also can't soft-busy anyway because it can 1712 * deadlock against the syncer doing a vfs_msync(), 1713 * vfs_msync->vmntvnodesca->vfs_msync_scan2-> 1714 * vm_object_page_clean->(scan)-> ... page 1715 * busy-wait. 1716 */ 1717 /*vm_page_io_start(pg);*/ 1718 vm_page_wakeup(pg); 1719 1720 /* 1721 * Get the page from backing store. 1722 */ 1723 bsize = vp->v_mount->mnt_stat.f_iosize; 1724 auio.uio_iov = &aiov; 1725 auio.uio_iovcnt = 1; 1726 aiov.iov_base = 0; 1727 aiov.iov_len = MAXBSIZE; 1728 auio.uio_resid = MAXBSIZE; 1729 auio.uio_offset = trunc_page(off); 1730 auio.uio_segflg = UIO_NOCOPY; 1731 auio.uio_rw = UIO_READ; 1732 auio.uio_td = td; 1733 vn_lock(vp, LK_SHARED | LK_RETRY); 1734 error = VOP_READ(vp, &auio, 1735 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1736 td->td_ucred); 1737 vn_unlock(vp); 1738 vm_page_flag_clear(pg, PG_ZERO); 1739 vm_page_busy_wait(pg, FALSE, "sockpg"); 1740 /*vm_page_io_finish(pg);*/ 1741 if (error) { 1742 vm_page_wakeup(pg); 1743 vm_page_unhold(pg); 1744 /* vm_page_try_to_free(pg); */ 1745 ssb_unlock(&so->so_snd); 1746 goto done; 1747 } 1748 } 1749 1750 1751 /* 1752 * Get a sendfile buf. We usually wait as long as necessary, 1753 * but this wait can be interrupted. 1754 */ 1755 if ((sf = sf_buf_alloc(pg)) == NULL) { 1756 vm_page_wakeup(pg); 1757 vm_page_unhold(pg); 1758 /* vm_page_try_to_free(pg); */ 1759 ssb_unlock(&so->so_snd); 1760 error = EINTR; 1761 goto done; 1762 } 1763 1764 /* 1765 * Get an mbuf header and set it up as having external storage. 1766 */ 1767 MGETHDR(m, M_WAITOK, MT_DATA); 1768 if (m == NULL) { 1769 error = ENOBUFS; 1770 vm_page_wakeup(pg); 1771 vm_page_unhold(pg); 1772 /* vm_page_try_to_free(pg); */ 1773 sf_buf_free(sf); 1774 ssb_unlock(&so->so_snd); 1775 goto done; 1776 } 1777 1778 vm_page_wakeup(pg); 1779 1780 m->m_ext.ext_free = sf_buf_mfree; 1781 m->m_ext.ext_ref = sf_buf_ref; 1782 m->m_ext.ext_arg = sf; 1783 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1784 m->m_ext.ext_size = PAGE_SIZE; 1785 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1786 m->m_flags |= M_EXT; 1787 m->m_pkthdr.len = m->m_len = xfsize; 1788 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1789 1790 if (mheader != NULL) { 1791 hbytes = mheader->m_pkthdr.len; 1792 mheader->m_pkthdr.len += m->m_pkthdr.len; 1793 m_cat(mheader, m); 1794 m = mheader; 1795 mheader = NULL; 1796 } else 1797 hbytes = 0; 1798 1799 /* 1800 * Add the buffer to the socket buffer chain. 1801 */ 1802 crit_enter(); 1803 retry_space: 1804 /* 1805 * Make sure that the socket is still able to take more data. 1806 * CANTSENDMORE being true usually means that the connection 1807 * was closed. so_error is true when an error was sensed after 1808 * a previous send. 1809 * The state is checked after the page mapping and buffer 1810 * allocation above since those operations may block and make 1811 * any socket checks stale. From this point forward, nothing 1812 * blocks before the pru_send (or more accurately, any blocking 1813 * results in a loop back to here to re-check). 1814 */ 1815 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1816 if (so->so_state & SS_CANTSENDMORE) { 1817 error = EPIPE; 1818 } else { 1819 error = so->so_error; 1820 so->so_error = 0; 1821 } 1822 m_freem(m); 1823 ssb_unlock(&so->so_snd); 1824 crit_exit(); 1825 goto done; 1826 } 1827 /* 1828 * Wait for socket space to become available. We do this just 1829 * after checking the connection state above in order to avoid 1830 * a race condition with ssb_wait(). 1831 */ 1832 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1833 space = ssb_space_prealloc(&so->so_snd); 1834 else 1835 space = ssb_space(&so->so_snd); 1836 1837 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1838 if (fp->f_flag & FNONBLOCK) { 1839 m_freem(m); 1840 ssb_unlock(&so->so_snd); 1841 crit_exit(); 1842 error = EAGAIN; 1843 goto done; 1844 } 1845 error = ssb_wait(&so->so_snd); 1846 /* 1847 * An error from ssb_wait usually indicates that we've 1848 * been interrupted by a signal. If we've sent anything 1849 * then return bytes sent, otherwise return the error. 1850 */ 1851 if (error) { 1852 m_freem(m); 1853 ssb_unlock(&so->so_snd); 1854 crit_exit(); 1855 goto done; 1856 } 1857 goto retry_space; 1858 } 1859 1860 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1861 for (mp = m; mp != NULL; mp = mp->m_next) 1862 ssb_preallocstream(&so->so_snd, mp); 1863 } 1864 if (use_sendfile_async) 1865 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1866 else 1867 error = so_pru_send(so, 0, m, NULL, NULL, td); 1868 1869 crit_exit(); 1870 if (error) { 1871 ssb_unlock(&so->so_snd); 1872 goto done; 1873 } 1874 } 1875 if (mheader != NULL) { 1876 *sbytes += mheader->m_pkthdr.len; 1877 1878 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1879 for (mp = mheader; mp != NULL; mp = mp->m_next) 1880 ssb_preallocstream(&so->so_snd, mp); 1881 } 1882 if (use_sendfile_async) 1883 error = so_pru_senda(so, 0, mheader, NULL, NULL, td); 1884 else 1885 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1886 1887 mheader = NULL; 1888 } 1889 ssb_unlock(&so->so_snd); 1890 1891 done: 1892 fdrop(fp); 1893 done0: 1894 if (mheader != NULL) 1895 m_freem(mheader); 1896 return (error); 1897 } 1898