1 /* 2 * Copyright (c) 1982, 1986, 1989, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * sendfile(2) and related extensions: 6 * Copyright (c) 1998, David Greenman. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 37 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 38 * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.9 2003/07/30 00:19:14 dillon Exp $ 39 */ 40 41 #include "opt_compat.h" 42 #include "opt_ktrace.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/kernel.h> 47 #include <sys/sysproto.h> 48 #include <sys/malloc.h> 49 #include <sys/filedesc.h> 50 #include <sys/event.h> 51 #include <sys/proc.h> 52 #include <sys/fcntl.h> 53 #include <sys/file.h> 54 #include <sys/filio.h> 55 #include <sys/mbuf.h> 56 #include <sys/protosw.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/signalvar.h> 60 #include <sys/uio.h> 61 #include <sys/vnode.h> 62 #include <sys/lock.h> 63 #include <sys/mount.h> 64 #ifdef KTRACE 65 #include <sys/ktrace.h> 66 #endif 67 #include <vm/vm.h> 68 #include <vm/vm_object.h> 69 #include <vm/vm_page.h> 70 #include <vm/vm_pageout.h> 71 #include <vm/vm_kern.h> 72 #include <vm/vm_extern.h> 73 #include <sys/file2.h> 74 75 static void sf_buf_init(void *arg); 76 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL) 77 78 static int sendit __P((int s, struct msghdr *mp, int flags, int *res)); 79 static int recvit __P((int s, struct msghdr *mp, caddr_t namelenp, int *res)); 80 81 static int accept1 __P((struct accept_args *uap, int compat)); 82 static int do_sendfile __P((struct sendfile_args *uap, int compat)); 83 static int getsockname1 __P((struct getsockname_args *uap, int compat)); 84 static int getpeername1 __P((struct getpeername_args *uap, int compat)); 85 86 static SLIST_HEAD(, sf_buf) sf_freelist; 87 static vm_offset_t sf_base; 88 static struct sf_buf *sf_bufs; 89 static int sf_buf_alloc_want; 90 91 /* 92 * System call interface to the socket abstraction. 93 */ 94 #if defined(COMPAT_43) || defined(COMPAT_SUNOS) 95 #define COMPAT_OLDSOCK 96 #endif 97 98 extern struct fileops socketops; 99 100 /* 101 * socket_args(int domain, int type, int protocol) 102 */ 103 int 104 socket(struct socket_args *uap) 105 { 106 struct thread *td = curthread; 107 struct proc *p = td->td_proc; 108 struct filedesc *fdp; 109 struct socket *so; 110 struct file *fp; 111 int fd, error; 112 113 KKASSERT(p); 114 fdp = p->p_fd; 115 116 error = falloc(p, &fp, &fd); 117 if (error) 118 return (error); 119 fhold(fp); 120 error = socreate(uap->domain, &so, uap->type, uap->protocol, td); 121 if (error) { 122 if (fdp->fd_ofiles[fd] == fp) { 123 fdp->fd_ofiles[fd] = NULL; 124 fdrop(fp, td); 125 } 126 } else { 127 fp->f_data = (caddr_t)so; 128 fp->f_flag = FREAD|FWRITE; 129 fp->f_ops = &socketops; 130 fp->f_type = DTYPE_SOCKET; 131 uap->sysmsg_result = fd; 132 } 133 fdrop(fp, td); 134 return (error); 135 } 136 137 /* 138 * bind_args(int s, caddr_t name, int namelen) 139 * 140 */ 141 /* ARGSUSED */ 142 int 143 bind(struct bind_args *uap) 144 { 145 struct thread *td = curthread; 146 struct proc *p = td->td_proc; 147 struct file *fp; 148 struct sockaddr *sa; 149 int error; 150 151 KKASSERT(p); 152 error = holdsock(p->p_fd, uap->s, &fp); 153 if (error) 154 return (error); 155 error = getsockaddr(&sa, uap->name, uap->namelen); 156 if (error) { 157 fdrop(fp, td); 158 return (error); 159 } 160 error = sobind((struct socket *)fp->f_data, sa, td); 161 FREE(sa, M_SONAME); 162 fdrop(fp, td); 163 return (error); 164 } 165 166 /* 167 * listen_args(int s, int backlog) 168 */ 169 /* ARGSUSED */ 170 int 171 listen(struct listen_args *uap) 172 { 173 struct thread *td = curthread; 174 struct proc *p = td->td_proc; 175 struct file *fp; 176 int error; 177 178 KKASSERT(p); 179 error = holdsock(p->p_fd, uap->s, &fp); 180 if (error) 181 return (error); 182 error = solisten((struct socket *)fp->f_data, uap->backlog, td); 183 fdrop(fp, td); 184 return(error); 185 } 186 187 /* 188 * accept_args(int s, caddr_t name, int *anamelen) 189 */ 190 static int 191 accept1(struct accept_args *uap, int compat) 192 { 193 struct thread *td = curthread; 194 struct proc *p = td->td_proc; 195 struct filedesc *fdp = p->p_fd; 196 struct file *lfp = NULL; 197 struct file *nfp = NULL; 198 struct sockaddr *sa; 199 int namelen, error, s; 200 struct socket *head, *so; 201 int fd; 202 u_int fflag; /* type must match fp->f_flag */ 203 int tmp; 204 205 if (uap->name) { 206 error = copyin((caddr_t)uap->anamelen, (caddr_t)&namelen, 207 sizeof (namelen)); 208 if(error) 209 return (error); 210 if (namelen < 0) 211 return (EINVAL); 212 } 213 error = holdsock(fdp, uap->s, &lfp); 214 if (error) 215 return (error); 216 s = splnet(); 217 head = (struct socket *)lfp->f_data; 218 if ((head->so_options & SO_ACCEPTCONN) == 0) { 219 splx(s); 220 error = EINVAL; 221 goto done; 222 } 223 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) { 224 if (head->so_state & SS_CANTRCVMORE) { 225 head->so_error = ECONNABORTED; 226 break; 227 } 228 if ((head->so_state & SS_NBIO) != 0) { 229 head->so_error = EWOULDBLOCK; 230 break; 231 } 232 error = tsleep((caddr_t)&head->so_timeo, PCATCH, "accept", 0); 233 if (error) { 234 splx(s); 235 goto done; 236 } 237 } 238 if (head->so_error) { 239 error = head->so_error; 240 head->so_error = 0; 241 splx(s); 242 goto done; 243 } 244 245 /* 246 * At this point we know that there is at least one connection 247 * ready to be accepted. Remove it from the queue prior to 248 * allocating the file descriptor for it since falloc() may 249 * block allowing another process to accept the connection 250 * instead. 251 */ 252 so = TAILQ_FIRST(&head->so_comp); 253 TAILQ_REMOVE(&head->so_comp, so, so_list); 254 head->so_qlen--; 255 256 fflag = lfp->f_flag; 257 error = falloc(p, &nfp, &fd); 258 if (error) { 259 /* 260 * Probably ran out of file descriptors. Put the 261 * unaccepted connection back onto the queue and 262 * do another wakeup so some other process might 263 * have a chance at it. 264 */ 265 TAILQ_INSERT_HEAD(&head->so_comp, so, so_list); 266 head->so_qlen++; 267 wakeup_one(&head->so_timeo); 268 splx(s); 269 goto done; 270 } 271 fhold(nfp); 272 uap->sysmsg_result = fd; 273 274 /* connection has been removed from the listen queue */ 275 KNOTE(&head->so_rcv.sb_sel.si_note, 0); 276 277 so->so_state &= ~SS_COMP; 278 so->so_head = NULL; 279 if (head->so_sigio != NULL) 280 fsetown(fgetown(head->so_sigio), &so->so_sigio); 281 282 nfp->f_data = (caddr_t)so; 283 nfp->f_flag = fflag; 284 nfp->f_ops = &socketops; 285 nfp->f_type = DTYPE_SOCKET; 286 /* Sync socket nonblocking/async state with file flags */ 287 tmp = fflag & FNONBLOCK; 288 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td); 289 tmp = fflag & FASYNC; 290 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td); 291 sa = 0; 292 error = soaccept(so, &sa); 293 if (error) { 294 /* 295 * return a namelen of zero for older code which might 296 * ignore the return value from accept. 297 */ 298 if (uap->name != NULL) { 299 namelen = 0; 300 (void) copyout((caddr_t)&namelen, 301 (caddr_t)uap->anamelen, sizeof(*uap->anamelen)); 302 } 303 goto noconnection; 304 } 305 if (sa == NULL) { 306 namelen = 0; 307 if (uap->name) 308 goto gotnoname; 309 splx(s); 310 error = 0; 311 goto done; 312 } 313 if (uap->name) { 314 /* check sa_len before it is destroyed */ 315 if (namelen > sa->sa_len) 316 namelen = sa->sa_len; 317 #ifdef COMPAT_OLDSOCK 318 if (compat) 319 ((struct osockaddr *)sa)->sa_family = 320 sa->sa_family; 321 #endif 322 error = copyout(sa, (caddr_t)uap->name, (u_int)namelen); 323 if (!error) 324 gotnoname: 325 error = copyout((caddr_t)&namelen, 326 (caddr_t)uap->anamelen, sizeof (*uap->anamelen)); 327 } 328 noconnection: 329 if (sa) 330 FREE(sa, M_SONAME); 331 332 /* 333 * close the new descriptor, assuming someone hasn't ripped it 334 * out from under us. 335 */ 336 if (error) { 337 if (fdp->fd_ofiles[fd] == nfp) { 338 fdp->fd_ofiles[fd] = NULL; 339 fdrop(nfp, td); 340 } 341 } 342 splx(s); 343 344 /* 345 * Release explicitly held references before returning. 346 */ 347 done: 348 if (nfp != NULL) 349 fdrop(nfp, td); 350 fdrop(lfp, td); 351 return (error); 352 } 353 354 int 355 accept(struct accept_args *uap) 356 { 357 return (accept1(uap, 0)); 358 } 359 360 #ifdef COMPAT_OLDSOCK 361 int 362 oaccept(struct accept_args *uap) 363 { 364 365 return (accept1(uap, 1)); 366 } 367 #endif /* COMPAT_OLDSOCK */ 368 369 /* 370 * connect_args(int s, caddr_t name, int namelen) 371 */ 372 /* ARGSUSED */ 373 int 374 connect(struct connect_args *uap) 375 { 376 struct thread *td = curthread; 377 struct proc *p = td->td_proc; 378 struct file *fp; 379 struct socket *so; 380 struct sockaddr *sa; 381 int error, s; 382 383 error = holdsock(p->p_fd, uap->s, &fp); 384 if (error) 385 return (error); 386 so = (struct socket *)fp->f_data; 387 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 388 error = EALREADY; 389 goto done; 390 } 391 error = getsockaddr(&sa, uap->name, uap->namelen); 392 if (error) 393 goto done; 394 error = soconnect(so, sa, td); 395 if (error) 396 goto bad; 397 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { 398 FREE(sa, M_SONAME); 399 error = EINPROGRESS; 400 goto done; 401 } 402 s = splnet(); 403 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 404 error = tsleep((caddr_t)&so->so_timeo, PCATCH, "connec", 0); 405 if (error) 406 break; 407 } 408 if (error == 0) { 409 error = so->so_error; 410 so->so_error = 0; 411 } 412 splx(s); 413 bad: 414 so->so_state &= ~SS_ISCONNECTING; 415 FREE(sa, M_SONAME); 416 if (error == ERESTART) 417 error = EINTR; 418 done: 419 fdrop(fp, td); 420 return (error); 421 } 422 423 /* 424 * socketpair(int domain, int type, int protocol, int *rsv) 425 */ 426 int 427 socketpair(struct socketpair_args *uap) 428 { 429 struct thread *td = curthread; 430 struct proc *p = td->td_proc; 431 struct filedesc *fdp; 432 struct file *fp1, *fp2; 433 struct socket *so1, *so2; 434 int fd, error, sv[2]; 435 436 KKASSERT(p); 437 fdp = p->p_fd; 438 error = socreate(uap->domain, &so1, uap->type, uap->protocol, td); 439 if (error) 440 return (error); 441 error = socreate(uap->domain, &so2, uap->type, uap->protocol, td); 442 if (error) 443 goto free1; 444 error = falloc(p, &fp1, &fd); 445 if (error) 446 goto free2; 447 fhold(fp1); 448 sv[0] = fd; 449 fp1->f_data = (caddr_t)so1; 450 error = falloc(p, &fp2, &fd); 451 if (error) 452 goto free3; 453 fhold(fp2); 454 fp2->f_data = (caddr_t)so2; 455 sv[1] = fd; 456 error = soconnect2(so1, so2); 457 if (error) 458 goto free4; 459 if (uap->type == SOCK_DGRAM) { 460 /* 461 * Datagram socket connection is asymmetric. 462 */ 463 error = soconnect2(so2, so1); 464 if (error) 465 goto free4; 466 } 467 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 468 fp1->f_ops = fp2->f_ops = &socketops; 469 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 470 error = copyout((caddr_t)sv, (caddr_t)uap->rsv, 2 * sizeof (int)); 471 fdrop(fp1, td); 472 fdrop(fp2, td); 473 return (error); 474 free4: 475 if (fdp->fd_ofiles[sv[1]] == fp2) { 476 fdp->fd_ofiles[sv[1]] = NULL; 477 fdrop(fp2, td); 478 } 479 fdrop(fp2, td); 480 free3: 481 if (fdp->fd_ofiles[sv[0]] == fp1) { 482 fdp->fd_ofiles[sv[0]] = NULL; 483 fdrop(fp1, td); 484 } 485 fdrop(fp1, td); 486 free2: 487 (void)soclose(so2); 488 free1: 489 (void)soclose(so1); 490 return (error); 491 } 492 493 static int 494 sendit(int s, struct msghdr *mp, int flags, int *res) 495 { 496 struct thread *td = curthread; 497 struct proc *p = td->td_proc; 498 struct file *fp; 499 struct uio auio; 500 struct iovec *iov; 501 int i; 502 struct mbuf *control; 503 struct sockaddr *to; 504 int len, error; 505 struct socket *so; 506 #ifdef KTRACE 507 struct iovec *ktriov = NULL; 508 struct uio ktruio; 509 #endif 510 511 error = holdsock(p->p_fd, s, &fp); 512 if (error) 513 return (error); 514 auio.uio_iov = mp->msg_iov; 515 auio.uio_iovcnt = mp->msg_iovlen; 516 auio.uio_segflg = UIO_USERSPACE; 517 auio.uio_rw = UIO_WRITE; 518 auio.uio_td = td; 519 auio.uio_offset = 0; /* XXX */ 520 auio.uio_resid = 0; 521 iov = mp->msg_iov; 522 for (i = 0; i < mp->msg_iovlen; i++, iov++) { 523 if ((auio.uio_resid += iov->iov_len) < 0) { 524 fdrop(fp, td); 525 return (EINVAL); 526 } 527 } 528 if (mp->msg_name) { 529 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen); 530 if (error) { 531 fdrop(fp, td); 532 return (error); 533 } 534 } else { 535 to = 0; 536 } 537 if (mp->msg_control) { 538 if (mp->msg_controllen < sizeof(struct cmsghdr) 539 #ifdef COMPAT_OLDSOCK 540 && mp->msg_flags != MSG_COMPAT 541 #endif 542 ) { 543 error = EINVAL; 544 goto bad; 545 } 546 error = sockargs(&control, mp->msg_control, 547 mp->msg_controllen, MT_CONTROL); 548 if (error) 549 goto bad; 550 #ifdef COMPAT_OLDSOCK 551 if (mp->msg_flags == MSG_COMPAT) { 552 struct cmsghdr *cm; 553 554 M_PREPEND(control, sizeof(*cm), M_WAIT); 555 if (control == 0) { 556 error = ENOBUFS; 557 goto bad; 558 } else { 559 cm = mtod(control, struct cmsghdr *); 560 cm->cmsg_len = control->m_len; 561 cm->cmsg_level = SOL_SOCKET; 562 cm->cmsg_type = SCM_RIGHTS; 563 } 564 } 565 #endif 566 } else { 567 control = 0; 568 } 569 #ifdef KTRACE 570 if (KTRPOINT(td, KTR_GENIO)) { 571 int iovlen = auio.uio_iovcnt * sizeof (struct iovec); 572 573 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 574 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen); 575 ktruio = auio; 576 } 577 #endif 578 len = auio.uio_resid; 579 so = (struct socket *)fp->f_data; 580 error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control, 581 flags, td); 582 if (error) { 583 if (auio.uio_resid != len && (error == ERESTART || 584 error == EINTR || error == EWOULDBLOCK)) 585 error = 0; 586 if (error == EPIPE) 587 psignal(p, SIGPIPE); 588 } 589 if (error == 0) 590 *res = len - auio.uio_resid; 591 #ifdef KTRACE 592 if (ktriov != NULL) { 593 if (error == 0) { 594 ktruio.uio_iov = ktriov; 595 ktruio.uio_resid = *res; 596 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error); 597 } 598 FREE(ktriov, M_TEMP); 599 } 600 #endif 601 bad: 602 fdrop(fp, td); 603 if (to) 604 FREE(to, M_SONAME); 605 return (error); 606 } 607 608 /* 609 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 610 */ 611 int 612 sendto(struct sendto_args *uap) 613 { 614 struct msghdr msg; 615 struct iovec aiov; 616 617 msg.msg_name = uap->to; 618 msg.msg_namelen = uap->tolen; 619 msg.msg_iov = &aiov; 620 msg.msg_iovlen = 1; 621 msg.msg_control = 0; 622 #ifdef COMPAT_OLDSOCK 623 msg.msg_flags = 0; 624 #endif 625 aiov.iov_base = uap->buf; 626 aiov.iov_len = uap->len; 627 return (sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result)); 628 } 629 630 #ifdef COMPAT_OLDSOCK 631 /* 632 * osend_args(int s, caddr_t buf, int len, int flags) 633 */ 634 int 635 osend(struct osend_args *uap) 636 { 637 struct msghdr msg; 638 struct iovec aiov; 639 640 msg.msg_name = 0; 641 msg.msg_namelen = 0; 642 msg.msg_iov = &aiov; 643 msg.msg_iovlen = 1; 644 aiov.iov_base = uap->buf; 645 aiov.iov_len = uap->len; 646 msg.msg_control = 0; 647 msg.msg_flags = 0; 648 return (sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result)); 649 } 650 651 /* 652 * osendmsg_args(int s, caddr_t msg, int flags) 653 */ 654 int 655 osendmsg(struct osendmsg_args *uap) 656 { 657 struct msghdr msg; 658 struct iovec aiov[UIO_SMALLIOV], *iov; 659 int error; 660 661 error = copyin(uap->msg, (caddr_t)&msg, sizeof (struct omsghdr)); 662 if (error) 663 return (error); 664 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 665 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) 666 return (EMSGSIZE); 667 MALLOC(iov, struct iovec *, 668 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 669 M_WAITOK); 670 } else 671 iov = aiov; 672 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, 673 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 674 if (error) 675 goto done; 676 msg.msg_flags = MSG_COMPAT; 677 msg.msg_iov = iov; 678 error = sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result); 679 done: 680 if (iov != aiov) 681 FREE(iov, M_IOV); 682 return (error); 683 } 684 #endif 685 686 /* 687 * sendmsg_args(int s, caddr_t msg, int flags) 688 */ 689 int 690 sendmsg(struct sendmsg_args *uap) 691 { 692 struct msghdr msg; 693 struct iovec aiov[UIO_SMALLIOV], *iov; 694 int error; 695 696 error = copyin(uap->msg, (caddr_t)&msg, sizeof (msg)); 697 if (error) 698 return (error); 699 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 700 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) 701 return (EMSGSIZE); 702 MALLOC(iov, struct iovec *, 703 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 704 M_WAITOK); 705 } else 706 iov = aiov; 707 if (msg.msg_iovlen && 708 (error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, 709 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))))) 710 goto done; 711 msg.msg_iov = iov; 712 #ifdef COMPAT_OLDSOCK 713 msg.msg_flags = 0; 714 #endif 715 error = sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result); 716 done: 717 if (iov != aiov) 718 FREE(iov, M_IOV); 719 return (error); 720 } 721 722 static int 723 recvit(int s, struct msghdr *mp, caddr_t namelenp, int *res) 724 { 725 struct thread *td = curthread; 726 struct proc *p = td->td_proc; 727 struct file *fp; 728 struct uio auio; 729 struct iovec *iov; 730 int i; 731 int len, error; 732 struct mbuf *m, *control = 0; 733 caddr_t ctlbuf; 734 struct socket *so; 735 struct sockaddr *fromsa = 0; 736 #ifdef KTRACE 737 struct iovec *ktriov = NULL; 738 struct uio ktruio; 739 #endif 740 741 error = holdsock(p->p_fd, s, &fp); 742 if (error) 743 return (error); 744 auio.uio_iov = mp->msg_iov; 745 auio.uio_iovcnt = mp->msg_iovlen; 746 auio.uio_segflg = UIO_USERSPACE; 747 auio.uio_rw = UIO_READ; 748 auio.uio_td = td; 749 auio.uio_offset = 0; /* XXX */ 750 auio.uio_resid = 0; 751 iov = mp->msg_iov; 752 for (i = 0; i < mp->msg_iovlen; i++, iov++) { 753 if ((auio.uio_resid += iov->iov_len) < 0) { 754 fdrop(fp, td); 755 return (EINVAL); 756 } 757 } 758 #ifdef KTRACE 759 if (KTRPOINT(td, KTR_GENIO)) { 760 int iovlen = auio.uio_iovcnt * sizeof (struct iovec); 761 762 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 763 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen); 764 ktruio = auio; 765 } 766 #endif 767 len = auio.uio_resid; 768 so = (struct socket *)fp->f_data; 769 error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio, 770 (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0, 771 &mp->msg_flags); 772 if (error) { 773 if (auio.uio_resid != len && (error == ERESTART || 774 error == EINTR || error == EWOULDBLOCK)) 775 error = 0; 776 } 777 #ifdef KTRACE 778 if (ktriov != NULL) { 779 if (error == 0) { 780 ktruio.uio_iov = ktriov; 781 ktruio.uio_resid = len - auio.uio_resid; 782 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error); 783 } 784 FREE(ktriov, M_TEMP); 785 } 786 #endif 787 if (error) 788 goto out; 789 *res = len - auio.uio_resid; 790 if (mp->msg_name) { 791 len = mp->msg_namelen; 792 if (len <= 0 || fromsa == 0) 793 len = 0; 794 else { 795 /* save sa_len before it is destroyed by MSG_COMPAT */ 796 len = MIN(len, fromsa->sa_len); 797 #ifdef COMPAT_OLDSOCK 798 if (mp->msg_flags & MSG_COMPAT) 799 ((struct osockaddr *)fromsa)->sa_family = 800 fromsa->sa_family; 801 #endif 802 error = copyout(fromsa, 803 (caddr_t)mp->msg_name, (unsigned)len); 804 if (error) 805 goto out; 806 } 807 mp->msg_namelen = len; 808 if (namelenp && 809 (error = copyout((caddr_t)&len, namelenp, sizeof (int)))) { 810 #ifdef COMPAT_OLDSOCK 811 if (mp->msg_flags & MSG_COMPAT) 812 error = 0; /* old recvfrom didn't check */ 813 else 814 #endif 815 goto out; 816 } 817 } 818 if (mp->msg_control) { 819 #ifdef COMPAT_OLDSOCK 820 /* 821 * We assume that old recvmsg calls won't receive access 822 * rights and other control info, esp. as control info 823 * is always optional and those options didn't exist in 4.3. 824 * If we receive rights, trim the cmsghdr; anything else 825 * is tossed. 826 */ 827 if (control && mp->msg_flags & MSG_COMPAT) { 828 if (mtod(control, struct cmsghdr *)->cmsg_level != 829 SOL_SOCKET || 830 mtod(control, struct cmsghdr *)->cmsg_type != 831 SCM_RIGHTS) { 832 mp->msg_controllen = 0; 833 goto out; 834 } 835 control->m_len -= sizeof (struct cmsghdr); 836 control->m_data += sizeof (struct cmsghdr); 837 } 838 #endif 839 len = mp->msg_controllen; 840 m = control; 841 mp->msg_controllen = 0; 842 ctlbuf = (caddr_t) mp->msg_control; 843 844 while (m && len > 0) { 845 unsigned int tocopy; 846 847 if (len >= m->m_len) 848 tocopy = m->m_len; 849 else { 850 mp->msg_flags |= MSG_CTRUNC; 851 tocopy = len; 852 } 853 854 if ((error = copyout((caddr_t)mtod(m, caddr_t), 855 ctlbuf, tocopy)) != 0) 856 goto out; 857 858 ctlbuf += tocopy; 859 len -= tocopy; 860 m = m->m_next; 861 } 862 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control; 863 } 864 out: 865 fdrop(fp, td); 866 if (fromsa) 867 FREE(fromsa, M_SONAME); 868 if (control) 869 m_freem(control); 870 return (error); 871 } 872 873 /* 874 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 875 * caddr_t from, int *fromlenaddr) 876 */ 877 int 878 recvfrom(struct recvfrom_args *uap) 879 { 880 struct msghdr msg; 881 struct iovec aiov; 882 int error; 883 884 if (uap->fromlenaddr) { 885 error = copyin((caddr_t)uap->fromlenaddr, 886 (caddr_t)&msg.msg_namelen, sizeof (msg.msg_namelen)); 887 if (error) 888 return (error); 889 } else 890 msg.msg_namelen = 0; 891 msg.msg_name = uap->from; 892 msg.msg_iov = &aiov; 893 msg.msg_iovlen = 1; 894 aiov.iov_base = uap->buf; 895 aiov.iov_len = uap->len; 896 msg.msg_control = 0; 897 msg.msg_flags = uap->flags; 898 return (recvit(uap->s, &msg, (caddr_t)uap->fromlenaddr, &uap->sysmsg_result)); 899 } 900 901 #ifdef COMPAT_OLDSOCK 902 int 903 orecvfrom(struct recvfrom_args *uap) 904 { 905 uap->flags |= MSG_COMPAT; 906 return (recvfrom(uap)); 907 } 908 #endif 909 910 #ifdef COMPAT_OLDSOCK 911 /* 912 * struct orecv_args(int s, caddr_t buf, int len, int flags) 913 */ 914 int 915 orecv(struct orecv_args *uap) 916 { 917 struct msghdr msg; 918 struct iovec aiov; 919 920 msg.msg_name = 0; 921 msg.msg_namelen = 0; 922 msg.msg_iov = &aiov; 923 msg.msg_iovlen = 1; 924 aiov.iov_base = uap->buf; 925 aiov.iov_len = uap->len; 926 msg.msg_control = 0; 927 msg.msg_flags = uap->flags; 928 return (recvit(uap->s, &msg, (caddr_t)0, &uap->sysmsg_result)); 929 } 930 931 /* 932 * Old recvmsg. This code takes advantage of the fact that the old msghdr 933 * overlays the new one, missing only the flags, and with the (old) access 934 * rights where the control fields are now. 935 * 936 * orecvmsg_args(int s, struct omsghdr *msg, int flags) 937 */ 938 int 939 orecvmsg(struct orecvmsg_args *uap) 940 { 941 struct msghdr msg; 942 struct iovec aiov[UIO_SMALLIOV], *iov; 943 int error; 944 945 error = copyin((caddr_t)uap->msg, (caddr_t)&msg, 946 sizeof (struct omsghdr)); 947 if (error) 948 return (error); 949 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 950 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) 951 return (EMSGSIZE); 952 MALLOC(iov, struct iovec *, 953 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 954 M_WAITOK); 955 } else 956 iov = aiov; 957 msg.msg_flags = uap->flags | MSG_COMPAT; 958 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov, 959 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 960 if (error) 961 goto done; 962 msg.msg_iov = iov; 963 error = recvit(uap->s, &msg, (caddr_t)&uap->msg->msg_namelen, &uap->sysmsg_result); 964 965 if (msg.msg_controllen && error == 0) 966 error = copyout((caddr_t)&msg.msg_controllen, 967 (caddr_t)&uap->msg->msg_accrightslen, sizeof (int)); 968 done: 969 if (iov != aiov) 970 FREE(iov, M_IOV); 971 return (error); 972 } 973 #endif 974 975 /* 976 * recvmsg_args(int s, struct msghdr *msg, int flags) 977 */ 978 int 979 recvmsg(struct recvmsg_args *uap) 980 { 981 struct msghdr msg; 982 struct iovec aiov[UIO_SMALLIOV], *uiov, *iov; 983 int error; 984 985 error = copyin((caddr_t)uap->msg, (caddr_t)&msg, sizeof (msg)); 986 if (error) 987 return (error); 988 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) { 989 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) 990 return (EMSGSIZE); 991 MALLOC(iov, struct iovec *, 992 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV, 993 M_WAITOK); 994 } else 995 iov = aiov; 996 #ifdef COMPAT_OLDSOCK 997 msg.msg_flags = uap->flags &~ MSG_COMPAT; 998 #else 999 msg.msg_flags = uap->flags; 1000 #endif 1001 uiov = msg.msg_iov; 1002 msg.msg_iov = iov; 1003 error = copyin((caddr_t)uiov, (caddr_t)iov, 1004 (unsigned)(msg.msg_iovlen * sizeof (struct iovec))); 1005 if (error) 1006 goto done; 1007 error = recvit(uap->s, &msg, (caddr_t)0, &uap->sysmsg_result); 1008 if (!error) { 1009 msg.msg_iov = uiov; 1010 error = copyout((caddr_t)&msg, (caddr_t)uap->msg, sizeof(msg)); 1011 } 1012 done: 1013 if (iov != aiov) 1014 FREE(iov, M_IOV); 1015 return (error); 1016 } 1017 1018 /* 1019 * shutdown_args(int s, int how) 1020 */ 1021 /* ARGSUSED */ 1022 int 1023 shutdown(struct shutdown_args *uap) 1024 { 1025 struct thread *td = curthread; 1026 struct proc *p = td->td_proc; 1027 struct file *fp; 1028 int error; 1029 1030 KKASSERT(p); 1031 error = holdsock(p->p_fd, uap->s, &fp); 1032 if (error) 1033 return (error); 1034 error = soshutdown((struct socket *)fp->f_data, uap->how); 1035 fdrop(fp, td); 1036 return(error); 1037 } 1038 1039 /* 1040 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1041 */ 1042 /* ARGSUSED */ 1043 int 1044 setsockopt(struct setsockopt_args *uap) 1045 { 1046 struct thread *td = curthread; 1047 struct proc *p = td->td_proc; 1048 struct file *fp; 1049 struct sockopt sopt; 1050 int error; 1051 1052 if (uap->val == 0 && uap->valsize != 0) 1053 return (EFAULT); 1054 if (uap->valsize < 0) 1055 return (EINVAL); 1056 1057 error = holdsock(p->p_fd, uap->s, &fp); 1058 if (error) 1059 return (error); 1060 1061 sopt.sopt_dir = SOPT_SET; 1062 sopt.sopt_level = uap->level; 1063 sopt.sopt_name = uap->name; 1064 sopt.sopt_val = uap->val; 1065 sopt.sopt_valsize = uap->valsize; 1066 sopt.sopt_td = td; 1067 error = sosetopt((struct socket *)fp->f_data, &sopt); 1068 fdrop(fp, td); 1069 return(error); 1070 } 1071 1072 /* 1073 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize) 1074 */ 1075 /* ARGSUSED */ 1076 int 1077 getsockopt(struct getsockopt_args *uap) 1078 { 1079 struct thread *td = curthread; 1080 struct proc *p = td->td_proc; 1081 int valsize, error; 1082 struct file *fp; 1083 struct sockopt sopt; 1084 1085 error = holdsock(p->p_fd, uap->s, &fp); 1086 if (error) 1087 return (error); 1088 if (uap->val) { 1089 error = copyin((caddr_t)uap->avalsize, (caddr_t)&valsize, 1090 sizeof (valsize)); 1091 if (error) { 1092 fdrop(fp, td); 1093 return (error); 1094 } 1095 if (valsize < 0) { 1096 fdrop(fp, td); 1097 return (EINVAL); 1098 } 1099 } else { 1100 valsize = 0; 1101 } 1102 1103 sopt.sopt_dir = SOPT_GET; 1104 sopt.sopt_level = uap->level; 1105 sopt.sopt_name = uap->name; 1106 sopt.sopt_val = uap->val; 1107 sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */ 1108 sopt.sopt_td = td; 1109 1110 error = sogetopt((struct socket *)fp->f_data, &sopt); 1111 if (error == 0) { 1112 valsize = sopt.sopt_valsize; 1113 error = copyout((caddr_t)&valsize, 1114 (caddr_t)uap->avalsize, sizeof (valsize)); 1115 } 1116 fdrop(fp, td); 1117 return (error); 1118 } 1119 1120 /* 1121 * getsockname_args(int fdes, caddr_t asa, int *alen) 1122 * 1123 * Get socket name. 1124 */ 1125 /* ARGSUSED */ 1126 static int 1127 getsockname1(struct getsockname_args *uap, int compat) 1128 { 1129 struct thread *td = curthread; 1130 struct proc *p = td->td_proc; 1131 struct file *fp; 1132 struct socket *so; 1133 struct sockaddr *sa; 1134 int len, error; 1135 1136 error = holdsock(p->p_fd, uap->fdes, &fp); 1137 if (error) 1138 return (error); 1139 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len)); 1140 if (error) { 1141 fdrop(fp, td); 1142 return (error); 1143 } 1144 if (len < 0) { 1145 fdrop(fp, td); 1146 return (EINVAL); 1147 } 1148 so = (struct socket *)fp->f_data; 1149 sa = 0; 1150 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa); 1151 if (error) 1152 goto bad; 1153 if (sa == 0) { 1154 len = 0; 1155 goto gotnothing; 1156 } 1157 1158 len = MIN(len, sa->sa_len); 1159 #ifdef COMPAT_OLDSOCK 1160 if (compat) 1161 ((struct osockaddr *)sa)->sa_family = sa->sa_family; 1162 #endif 1163 error = copyout(sa, (caddr_t)uap->asa, (u_int)len); 1164 if (error == 0) 1165 gotnothing: 1166 error = copyout((caddr_t)&len, (caddr_t)uap->alen, 1167 sizeof (len)); 1168 bad: 1169 if (sa) 1170 FREE(sa, M_SONAME); 1171 fdrop(fp, td); 1172 return (error); 1173 } 1174 1175 int 1176 getsockname(struct getsockname_args *uap) 1177 { 1178 1179 return (getsockname1(uap, 0)); 1180 } 1181 1182 #ifdef COMPAT_OLDSOCK 1183 int 1184 ogetsockname(struct getsockname_args *uap) 1185 { 1186 1187 return (getsockname1(uap, 1)); 1188 } 1189 #endif /* COMPAT_OLDSOCK */ 1190 1191 /* 1192 * getpeername_args(int fdes, caddr_t asa, int *alen) 1193 * 1194 * Get name of peer for connected socket. 1195 */ 1196 /* ARGSUSED */ 1197 static int 1198 getpeername1(struct getpeername_args *uap, int compat) 1199 { 1200 struct thread *td = curthread; 1201 struct proc *p = td->td_proc; 1202 struct file *fp; 1203 struct socket *so; 1204 struct sockaddr *sa; 1205 int len, error; 1206 1207 error = holdsock(p->p_fd, uap->fdes, &fp); 1208 if (error) 1209 return (error); 1210 so = (struct socket *)fp->f_data; 1211 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1212 fdrop(fp, td); 1213 return (ENOTCONN); 1214 } 1215 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len)); 1216 if (error) { 1217 fdrop(fp, td); 1218 return (error); 1219 } 1220 if (len < 0) { 1221 fdrop(fp, td); 1222 return (EINVAL); 1223 } 1224 sa = 0; 1225 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa); 1226 if (error) 1227 goto bad; 1228 if (sa == 0) { 1229 len = 0; 1230 goto gotnothing; 1231 } 1232 len = MIN(len, sa->sa_len); 1233 #ifdef COMPAT_OLDSOCK 1234 if (compat) 1235 ((struct osockaddr *)sa)->sa_family = 1236 sa->sa_family; 1237 #endif 1238 error = copyout(sa, (caddr_t)uap->asa, (u_int)len); 1239 if (error) 1240 goto bad; 1241 gotnothing: 1242 error = copyout((caddr_t)&len, (caddr_t)uap->alen, sizeof (len)); 1243 bad: 1244 if (sa) 1245 FREE(sa, M_SONAME); 1246 fdrop(fp, td); 1247 return (error); 1248 } 1249 1250 int 1251 getpeername(struct getpeername_args *uap) 1252 { 1253 return (getpeername1(uap, 0)); 1254 } 1255 1256 #ifdef COMPAT_OLDSOCK 1257 int 1258 ogetpeername(struct ogetpeername_args *uap) 1259 { 1260 /* XXX uap should have type `getpeername_args *' to begin with. */ 1261 return (getpeername1((struct getpeername_args *)uap, 1)); 1262 } 1263 #endif /* COMPAT_OLDSOCK */ 1264 1265 int 1266 sockargs(mp, buf, buflen, type) 1267 struct mbuf **mp; 1268 caddr_t buf; 1269 int buflen, type; 1270 { 1271 struct sockaddr *sa; 1272 struct mbuf *m; 1273 int error; 1274 1275 if ((u_int)buflen > MLEN) { 1276 #ifdef COMPAT_OLDSOCK 1277 if (type == MT_SONAME && (u_int)buflen <= 112) 1278 buflen = MLEN; /* unix domain compat. hack */ 1279 else 1280 #endif 1281 return (EINVAL); 1282 } 1283 m = m_get(M_WAIT, type); 1284 if (m == NULL) 1285 return (ENOBUFS); 1286 m->m_len = buflen; 1287 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen); 1288 if (error) 1289 (void) m_free(m); 1290 else { 1291 *mp = m; 1292 if (type == MT_SONAME) { 1293 sa = mtod(m, struct sockaddr *); 1294 1295 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN 1296 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1297 sa->sa_family = sa->sa_len; 1298 #endif 1299 sa->sa_len = buflen; 1300 } 1301 } 1302 return (error); 1303 } 1304 1305 int 1306 getsockaddr(namp, uaddr, len) 1307 struct sockaddr **namp; 1308 caddr_t uaddr; 1309 size_t len; 1310 { 1311 struct sockaddr *sa; 1312 int error; 1313 1314 if (len > SOCK_MAXADDRLEN) 1315 return ENAMETOOLONG; 1316 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK); 1317 error = copyin(uaddr, sa, len); 1318 if (error) { 1319 FREE(sa, M_SONAME); 1320 } else { 1321 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN 1322 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1323 sa->sa_family = sa->sa_len; 1324 #endif 1325 sa->sa_len = len; 1326 *namp = sa; 1327 } 1328 return error; 1329 } 1330 1331 /* 1332 * holdsock() - load the struct file pointer associated 1333 * with a socket into *fpp. If an error occurs, non-zero 1334 * will be returned and *fpp will be set to NULL. 1335 */ 1336 int 1337 holdsock(fdp, fdes, fpp) 1338 struct filedesc *fdp; 1339 int fdes; 1340 struct file **fpp; 1341 { 1342 struct file *fp = NULL; 1343 int error = 0; 1344 1345 if ((unsigned)fdes >= fdp->fd_nfiles || 1346 (fp = fdp->fd_ofiles[fdes]) == NULL) { 1347 error = EBADF; 1348 } else if (fp->f_type != DTYPE_SOCKET) { 1349 error = ENOTSOCK; 1350 fp = NULL; 1351 } else { 1352 fhold(fp); 1353 } 1354 *fpp = fp; 1355 return(error); 1356 } 1357 1358 /* 1359 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 1360 */ 1361 static void 1362 sf_buf_init(void *arg) 1363 { 1364 int i; 1365 1366 SLIST_INIT(&sf_freelist); 1367 sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE); 1368 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT); 1369 bzero(sf_bufs, nsfbufs * sizeof(struct sf_buf)); 1370 for (i = 0; i < nsfbufs; i++) { 1371 sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 1372 SLIST_INSERT_HEAD(&sf_freelist, &sf_bufs[i], free_list); 1373 } 1374 } 1375 1376 /* 1377 * Get an sf_buf from the freelist. Will block if none are available. 1378 */ 1379 struct sf_buf * 1380 sf_buf_alloc() 1381 { 1382 struct sf_buf *sf; 1383 int s; 1384 int error; 1385 1386 s = splimp(); 1387 while ((sf = SLIST_FIRST(&sf_freelist)) == NULL) { 1388 sf_buf_alloc_want = 1; 1389 error = tsleep(&sf_freelist, PCATCH, "sfbufa", 0); 1390 if (error) 1391 break; 1392 } 1393 if (sf != NULL) { 1394 SLIST_REMOVE_HEAD(&sf_freelist, free_list); 1395 sf->refcnt = 1; 1396 } 1397 splx(s); 1398 return (sf); 1399 } 1400 1401 #define dtosf(x) (&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT]) 1402 void 1403 sf_buf_ref(caddr_t addr, u_int size) 1404 { 1405 struct sf_buf *sf; 1406 1407 sf = dtosf(addr); 1408 if (sf->refcnt == 0) 1409 panic("sf_buf_ref: referencing a free sf_buf"); 1410 sf->refcnt++; 1411 } 1412 1413 /* 1414 * Lose a reference to an sf_buf. When none left, detach mapped page 1415 * and release resources back to the system. 1416 * 1417 * Must be called at splimp. 1418 */ 1419 void 1420 sf_buf_free(caddr_t addr, u_int size) 1421 { 1422 struct sf_buf *sf; 1423 struct vm_page *m; 1424 int s; 1425 1426 sf = dtosf(addr); 1427 if (sf->refcnt == 0) 1428 panic("sf_buf_free: freeing free sf_buf"); 1429 sf->refcnt--; 1430 if (sf->refcnt == 0) { 1431 pmap_qremove((vm_offset_t)addr, 1); 1432 m = sf->m; 1433 s = splvm(); 1434 vm_page_unwire(m, 0); 1435 /* 1436 * Check for the object going away on us. This can 1437 * happen since we don't hold a reference to it. 1438 * If so, we're responsible for freeing the page. 1439 */ 1440 if (m->wire_count == 0 && m->object == NULL) 1441 vm_page_free(m); 1442 splx(s); 1443 sf->m = NULL; 1444 SLIST_INSERT_HEAD(&sf_freelist, sf, free_list); 1445 if (sf_buf_alloc_want) { 1446 sf_buf_alloc_want = 0; 1447 wakeup(&sf_freelist); 1448 } 1449 } 1450 } 1451 1452 /* 1453 * sendfile(2). 1454 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1455 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1456 * 1457 * Send a file specified by 'fd' and starting at 'offset' to a socket 1458 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1459 * nbytes == 0. Optionally add a header and/or trailer to the socket 1460 * output. If specified, write the total number of bytes sent into *sbytes. 1461 */ 1462 int 1463 sendfile(struct sendfile_args *uap) 1464 { 1465 return (do_sendfile(uap, 0)); 1466 } 1467 1468 #ifdef COMPAT_43 1469 int 1470 osendfile(struct osendfile_args *uap) 1471 { 1472 struct sendfile_args args; 1473 1474 args.fd = uap->fd; 1475 args.s = uap->s; 1476 args.offset = uap->offset; 1477 args.nbytes = uap->nbytes; 1478 args.hdtr = uap->hdtr; 1479 args.sbytes = uap->sbytes; 1480 args.flags = uap->flags; 1481 1482 return (do_sendfile(&args, 1)); 1483 } 1484 #endif 1485 1486 int 1487 do_sendfile(struct sendfile_args *uap, int compat) 1488 { 1489 struct thread *td = curthread; 1490 struct proc *p = td->td_proc; 1491 struct file *fp; 1492 struct filedesc *fdp; 1493 struct vnode *vp; 1494 struct vm_object *obj; 1495 struct socket *so; 1496 struct mbuf *m; 1497 struct sf_buf *sf; 1498 struct vm_page *pg; 1499 struct writev_args nuap; 1500 struct sf_hdtr hdtr; 1501 off_t off, xfsize, hdtr_size, sbytes = 0; 1502 int error = 0, s; 1503 1504 KKASSERT(p); 1505 fdp = p->p_fd; 1506 1507 vp = NULL; 1508 hdtr_size = 0; 1509 /* 1510 * Do argument checking. Must be a regular file in, stream 1511 * type and connected socket out, positive offset. 1512 */ 1513 fp = holdfp(fdp, uap->fd, FREAD); 1514 if (fp == NULL) { 1515 error = EBADF; 1516 goto done; 1517 } 1518 if (fp->f_type != DTYPE_VNODE) { 1519 error = EINVAL; 1520 goto done; 1521 } 1522 vp = (struct vnode *)fp->f_data; 1523 vref(vp); 1524 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) { 1525 error = EINVAL; 1526 goto done; 1527 } 1528 fdrop(fp, td); 1529 error = holdsock(p->p_fd, uap->s, &fp); 1530 if (error) 1531 goto done; 1532 so = (struct socket *)fp->f_data; 1533 if (so->so_type != SOCK_STREAM) { 1534 error = EINVAL; 1535 goto done; 1536 } 1537 if ((so->so_state & SS_ISCONNECTED) == 0) { 1538 error = ENOTCONN; 1539 goto done; 1540 } 1541 if (uap->offset < 0) { 1542 error = EINVAL; 1543 goto done; 1544 } 1545 1546 /* 1547 * If specified, get the pointer to the sf_hdtr struct for 1548 * any headers/trailers. 1549 */ 1550 if (uap->hdtr != NULL) { 1551 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1552 if (error) 1553 goto done; 1554 /* 1555 * Send any headers. Wimp out and use writev(2). 1556 */ 1557 if (hdtr.headers != NULL) { 1558 nuap.fd = uap->s; 1559 nuap.iovp = hdtr.headers; 1560 nuap.iovcnt = hdtr.hdr_cnt; 1561 error = writev(&nuap); 1562 if (error) 1563 goto done; 1564 if (compat) 1565 sbytes += uap->sysmsg_result; 1566 else 1567 hdtr_size += uap->sysmsg_result; 1568 } 1569 } 1570 1571 /* 1572 * Protect against multiple writers to the socket. 1573 */ 1574 (void) sblock(&so->so_snd, M_WAITOK); 1575 1576 /* 1577 * Loop through the pages in the file, starting with the requested 1578 * offset. Get a file page (do I/O if necessary), map the file page 1579 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1580 * it on the socket. 1581 */ 1582 for (off = uap->offset; ; off += xfsize, sbytes += xfsize) { 1583 vm_pindex_t pindex; 1584 vm_offset_t pgoff; 1585 1586 pindex = OFF_TO_IDX(off); 1587 retry_lookup: 1588 /* 1589 * Calculate the amount to transfer. Not to exceed a page, 1590 * the EOF, or the passed in nbytes. 1591 */ 1592 xfsize = obj->un_pager.vnp.vnp_size - off; 1593 if (xfsize > PAGE_SIZE) 1594 xfsize = PAGE_SIZE; 1595 pgoff = (vm_offset_t)(off & PAGE_MASK); 1596 if (PAGE_SIZE - pgoff < xfsize) 1597 xfsize = PAGE_SIZE - pgoff; 1598 if (uap->nbytes && xfsize > (uap->nbytes - sbytes)) 1599 xfsize = uap->nbytes - sbytes; 1600 if (xfsize <= 0) 1601 break; 1602 /* 1603 * Optimize the non-blocking case by looking at the socket space 1604 * before going to the extra work of constituting the sf_buf. 1605 */ 1606 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) { 1607 if (so->so_state & SS_CANTSENDMORE) 1608 error = EPIPE; 1609 else 1610 error = EAGAIN; 1611 sbunlock(&so->so_snd); 1612 goto done; 1613 } 1614 /* 1615 * Attempt to look up the page. 1616 * 1617 * Allocate if not found 1618 * 1619 * Wait and loop if busy. 1620 */ 1621 pg = vm_page_lookup(obj, pindex); 1622 1623 if (pg == NULL) { 1624 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL); 1625 if (pg == NULL) { 1626 VM_WAIT; 1627 goto retry_lookup; 1628 } 1629 vm_page_wakeup(pg); 1630 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) { 1631 goto retry_lookup; 1632 } 1633 1634 /* 1635 * Wire the page so it does not get ripped out from under 1636 * us. 1637 */ 1638 1639 vm_page_wire(pg); 1640 1641 /* 1642 * If page is not valid for what we need, initiate I/O 1643 */ 1644 1645 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1646 struct uio auio; 1647 struct iovec aiov; 1648 int bsize; 1649 1650 /* 1651 * Ensure that our page is still around when the I/O 1652 * completes. 1653 */ 1654 vm_page_io_start(pg); 1655 1656 /* 1657 * Get the page from backing store. 1658 */ 1659 bsize = vp->v_mount->mnt_stat.f_iosize; 1660 auio.uio_iov = &aiov; 1661 auio.uio_iovcnt = 1; 1662 aiov.iov_base = 0; 1663 aiov.iov_len = MAXBSIZE; 1664 auio.uio_resid = MAXBSIZE; 1665 auio.uio_offset = trunc_page(off); 1666 auio.uio_segflg = UIO_NOCOPY; 1667 auio.uio_rw = UIO_READ; 1668 auio.uio_td = td; 1669 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td); 1670 error = VOP_READ(vp, &auio, 1671 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1672 p->p_ucred); 1673 VOP_UNLOCK(vp, 0, td); 1674 vm_page_flag_clear(pg, PG_ZERO); 1675 vm_page_io_finish(pg); 1676 if (error) { 1677 vm_page_unwire(pg, 0); 1678 /* 1679 * See if anyone else might know about this page. 1680 * If not and it is not valid, then free it. 1681 */ 1682 if (pg->wire_count == 0 && pg->valid == 0 && 1683 pg->busy == 0 && !(pg->flags & PG_BUSY) && 1684 pg->hold_count == 0) { 1685 vm_page_busy(pg); 1686 vm_page_free(pg); 1687 } 1688 sbunlock(&so->so_snd); 1689 goto done; 1690 } 1691 } 1692 1693 1694 /* 1695 * Get a sendfile buf. We usually wait as long as necessary, 1696 * but this wait can be interrupted. 1697 */ 1698 if ((sf = sf_buf_alloc()) == NULL) { 1699 s = splvm(); 1700 vm_page_unwire(pg, 0); 1701 if (pg->wire_count == 0 && pg->object == NULL) 1702 vm_page_free(pg); 1703 splx(s); 1704 sbunlock(&so->so_snd); 1705 error = EINTR; 1706 goto done; 1707 } 1708 1709 1710 /* 1711 * Allocate a kernel virtual page and insert the physical page 1712 * into it. 1713 */ 1714 1715 sf->m = pg; 1716 pmap_qenter(sf->kva, &pg, 1); 1717 /* 1718 * Get an mbuf header and set it up as having external storage. 1719 */ 1720 MGETHDR(m, M_WAIT, MT_DATA); 1721 if (m == NULL) { 1722 error = ENOBUFS; 1723 sf_buf_free((void *)sf->kva, PAGE_SIZE); 1724 sbunlock(&so->so_snd); 1725 goto done; 1726 } 1727 m->m_ext.ext_free = sf_buf_free; 1728 m->m_ext.ext_ref = sf_buf_ref; 1729 m->m_ext.ext_buf = (void *)sf->kva; 1730 m->m_ext.ext_size = PAGE_SIZE; 1731 m->m_data = (char *) sf->kva + pgoff; 1732 m->m_flags |= M_EXT; 1733 m->m_pkthdr.len = m->m_len = xfsize; 1734 /* 1735 * Add the buffer to the socket buffer chain. 1736 */ 1737 s = splnet(); 1738 retry_space: 1739 /* 1740 * Make sure that the socket is still able to take more data. 1741 * CANTSENDMORE being true usually means that the connection 1742 * was closed. so_error is true when an error was sensed after 1743 * a previous send. 1744 * The state is checked after the page mapping and buffer 1745 * allocation above since those operations may block and make 1746 * any socket checks stale. From this point forward, nothing 1747 * blocks before the pru_send (or more accurately, any blocking 1748 * results in a loop back to here to re-check). 1749 */ 1750 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1751 if (so->so_state & SS_CANTSENDMORE) { 1752 error = EPIPE; 1753 } else { 1754 error = so->so_error; 1755 so->so_error = 0; 1756 } 1757 m_freem(m); 1758 sbunlock(&so->so_snd); 1759 splx(s); 1760 goto done; 1761 } 1762 /* 1763 * Wait for socket space to become available. We do this just 1764 * after checking the connection state above in order to avoid 1765 * a race condition with sbwait(). 1766 */ 1767 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) { 1768 if (so->so_state & SS_NBIO) { 1769 m_freem(m); 1770 sbunlock(&so->so_snd); 1771 splx(s); 1772 error = EAGAIN; 1773 goto done; 1774 } 1775 error = sbwait(&so->so_snd); 1776 /* 1777 * An error from sbwait usually indicates that we've 1778 * been interrupted by a signal. If we've sent anything 1779 * then return bytes sent, otherwise return the error. 1780 */ 1781 if (error) { 1782 m_freem(m); 1783 sbunlock(&so->so_snd); 1784 splx(s); 1785 goto done; 1786 } 1787 goto retry_space; 1788 } 1789 error = 1790 (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td); 1791 splx(s); 1792 if (error) { 1793 sbunlock(&so->so_snd); 1794 goto done; 1795 } 1796 } 1797 sbunlock(&so->so_snd); 1798 1799 /* 1800 * Send trailers. Wimp out and use writev(2). 1801 */ 1802 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1803 nuap.fd = uap->s; 1804 nuap.iovp = hdtr.trailers; 1805 nuap.iovcnt = hdtr.trl_cnt; 1806 error = writev(&nuap); 1807 if (error) 1808 goto done; 1809 if (compat) 1810 sbytes += uap->sysmsg_result; 1811 else 1812 hdtr_size += uap->sysmsg_result; 1813 } 1814 1815 done: 1816 if (uap->sbytes != NULL) { 1817 if (compat == 0) 1818 sbytes += hdtr_size; 1819 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1820 } 1821 if (vp) 1822 vrele(vp); 1823 if (fp) 1824 fdrop(fp, td); 1825 return (error); 1826 } 1827