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 u_int fflags = 0; 101 int oflags = 0; 102 103 KKASSERT(td->td_lwp); 104 105 if (type & SOCK_NONBLOCK) { 106 type &= ~SOCK_NONBLOCK; 107 fflags |= FNONBLOCK; 108 } 109 if (type & SOCK_CLOEXEC) { 110 type &= ~SOCK_CLOEXEC; 111 oflags |= O_CLOEXEC; 112 } 113 114 error = falloc(td->td_lwp, &fp, &fd); 115 if (error) 116 return (error); 117 error = socreate(domain, &so, type, protocol, td); 118 if (error) { 119 fsetfd(fdp, NULL, fd); 120 } else { 121 fp->f_type = DTYPE_SOCKET; 122 fp->f_flag = FREAD | FWRITE | fflags; 123 fp->f_ops = &socketops; 124 fp->f_data = so; 125 if (oflags & O_CLOEXEC) 126 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 127 if (fflags & FNONBLOCK) { 128 int tmp = 1; 129 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 130 } 131 *res = fd; 132 fsetfd(fdp, fp, fd); 133 } 134 fdrop(fp); 135 return (error); 136 } 137 138 /* 139 * MPALMOSTSAFE 140 */ 141 int 142 sys_socket(struct socket_args *uap) 143 { 144 int error; 145 146 error = kern_socket(uap->domain, uap->type, uap->protocol, 147 &uap->sysmsg_iresult); 148 149 return (error); 150 } 151 152 int 153 kern_bind(int s, struct sockaddr *sa) 154 { 155 struct thread *td = curthread; 156 struct proc *p = td->td_proc; 157 struct file *fp; 158 int error; 159 160 KKASSERT(p); 161 error = holdsock(p->p_fd, s, &fp); 162 if (error) 163 return (error); 164 error = sobind((struct socket *)fp->f_data, sa, td); 165 fdrop(fp); 166 return (error); 167 } 168 169 /* 170 * bind_args(int s, caddr_t name, int namelen) 171 * 172 * MPALMOSTSAFE 173 */ 174 int 175 sys_bind(struct bind_args *uap) 176 { 177 struct sockaddr *sa; 178 int error; 179 180 error = getsockaddr(&sa, uap->name, uap->namelen); 181 if (error) 182 return (error); 183 error = kern_bind(uap->s, sa); 184 kfree(sa, M_SONAME); 185 186 return (error); 187 } 188 189 int 190 kern_listen(int s, int backlog) 191 { 192 struct thread *td = curthread; 193 struct proc *p = td->td_proc; 194 struct file *fp; 195 int error; 196 197 KKASSERT(p); 198 error = holdsock(p->p_fd, s, &fp); 199 if (error) 200 return (error); 201 error = solisten((struct socket *)fp->f_data, backlog, td); 202 fdrop(fp); 203 return(error); 204 } 205 206 /* 207 * listen_args(int s, int backlog) 208 * 209 * MPALMOSTSAFE 210 */ 211 int 212 sys_listen(struct listen_args *uap) 213 { 214 int error; 215 216 error = kern_listen(uap->s, uap->backlog); 217 return (error); 218 } 219 220 /* 221 * Returns the accepted socket as well. 222 * 223 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 224 * pool token is absolutely required to avoid a sofree() race, 225 * as well as to avoid tailq handling races. 226 */ 227 static boolean_t 228 soaccept_predicate(struct netmsg_so_notify *msg) 229 { 230 struct socket *head = msg->base.nm_so; 231 struct socket *so; 232 233 if (head->so_error != 0) { 234 msg->base.lmsg.ms_error = head->so_error; 235 return (TRUE); 236 } 237 lwkt_getpooltoken(head); 238 if (!TAILQ_EMPTY(&head->so_comp)) { 239 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 240 so = TAILQ_FIRST(&head->so_comp); 241 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP); 242 TAILQ_REMOVE(&head->so_comp, so, so_list); 243 head->so_qlen--; 244 soclrstate(so, SS_COMP); 245 246 /* 247 * Keep a reference before clearing the so_head 248 * to avoid racing socket close in netisr. 249 */ 250 soreference(so); 251 so->so_head = NULL; 252 253 lwkt_relpooltoken(head); 254 255 msg->base.lmsg.ms_error = 0; 256 msg->base.nm_so = so; 257 return (TRUE); 258 } 259 lwkt_relpooltoken(head); 260 if (head->so_state & SS_CANTRCVMORE) { 261 msg->base.lmsg.ms_error = ECONNABORTED; 262 return (TRUE); 263 } 264 if (msg->nm_fflags & FNONBLOCK) { 265 msg->base.lmsg.ms_error = EWOULDBLOCK; 266 return (TRUE); 267 } 268 269 return (FALSE); 270 } 271 272 /* 273 * The second argument to kern_accept() is a handle to a struct sockaddr. 274 * This allows kern_accept() to return a pointer to an allocated struct 275 * sockaddr which must be freed later with FREE(). The caller must 276 * initialize *name to NULL. 277 */ 278 int 279 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res, 280 int sockflags) 281 { 282 struct thread *td = curthread; 283 struct filedesc *fdp = td->td_proc->p_fd; 284 struct file *lfp = NULL; 285 struct file *nfp = NULL; 286 struct sockaddr *sa; 287 struct socket *head, *so; 288 struct netmsg_so_notify msg; 289 int fd; 290 u_int fflag; /* type must match fp->f_flag */ 291 int error, tmp; 292 293 *res = -1; 294 if (name && namelen && *namelen < 0) 295 return (EINVAL); 296 297 error = holdsock(td->td_proc->p_fd, s, &lfp); 298 if (error) 299 return (error); 300 301 error = falloc(td->td_lwp, &nfp, &fd); 302 if (error) { /* Probably ran out of file descriptors. */ 303 fdrop(lfp); 304 return (error); 305 } 306 head = (struct socket *)lfp->f_data; 307 if ((head->so_options & SO_ACCEPTCONN) == 0) { 308 error = EINVAL; 309 goto done; 310 } 311 312 if (fflags & O_FBLOCKING) 313 fflags |= lfp->f_flag & ~FNONBLOCK; 314 else if (fflags & O_FNONBLOCKING) 315 fflags |= lfp->f_flag | FNONBLOCK; 316 else 317 fflags = lfp->f_flag; 318 319 if (use_soaccept_pred_fast) { 320 boolean_t pred; 321 322 /* Initialize necessary parts for soaccept_predicate() */ 323 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 324 msg.nm_fflags = fflags; 325 326 lwkt_getpooltoken(head); 327 pred = soaccept_predicate(&msg); 328 lwkt_relpooltoken(head); 329 330 if (pred) { 331 error = msg.base.lmsg.ms_error; 332 if (error) 333 goto done; 334 else 335 goto accepted; 336 } 337 } 338 339 /* optimize for uniprocessor case later XXX JH */ 340 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 341 0, netmsg_so_notify, netmsg_so_notify_doabort); 342 msg.nm_predicate = soaccept_predicate; 343 msg.nm_fflags = fflags; 344 msg.nm_etype = NM_REVENT; 345 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 346 if (error) 347 goto done; 348 349 accepted: 350 /* 351 * At this point we have the connection that's ready to be accepted. 352 * 353 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 354 * to eat the ref and turn it into a descriptor. 355 */ 356 so = msg.base.nm_so; 357 358 fflag = lfp->f_flag; 359 360 /* connection has been removed from the listen queue */ 361 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 362 363 if (sockflags & SOCK_KERN_NOINHERIT) { 364 fflag &= ~(FASYNC | FNONBLOCK); 365 if (sockflags & SOCK_NONBLOCK) 366 fflag |= FNONBLOCK; 367 } else { 368 if (head->so_sigio != NULL) 369 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 370 } 371 372 nfp->f_type = DTYPE_SOCKET; 373 nfp->f_flag = fflag; 374 nfp->f_ops = &socketops; 375 nfp->f_data = so; 376 /* Sync socket nonblocking/async state with file flags */ 377 tmp = fflag & FNONBLOCK; 378 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 379 tmp = fflag & FASYNC; 380 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 381 382 sa = NULL; 383 if (so->so_faddr != NULL) { 384 sa = so->so_faddr; 385 so->so_faddr = NULL; 386 387 soaccept_generic(so); 388 error = 0; 389 } else { 390 error = soaccept(so, &sa); 391 } 392 393 /* 394 * Set the returned name and namelen as applicable. Set the returned 395 * namelen to 0 for older code which might ignore the return value 396 * from accept. 397 */ 398 if (error == 0) { 399 if (sa && name && namelen) { 400 if (*namelen > sa->sa_len) 401 *namelen = sa->sa_len; 402 *name = sa; 403 } else { 404 if (sa) 405 kfree(sa, M_SONAME); 406 } 407 } 408 409 done: 410 /* 411 * If an error occured clear the reserved descriptor, else associate 412 * nfp with it. 413 * 414 * Note that *res is normally ignored if an error is returned but 415 * a syscall message will still have access to the result code. 416 */ 417 if (error) { 418 fsetfd(fdp, NULL, fd); 419 } else { 420 if (sockflags & SOCK_CLOEXEC) 421 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 422 *res = fd; 423 fsetfd(fdp, nfp, fd); 424 } 425 fdrop(nfp); 426 fdrop(lfp); 427 return (error); 428 } 429 430 /* 431 * accept(int s, caddr_t name, int *anamelen) 432 * 433 * MPALMOSTSAFE 434 */ 435 int 436 sys_accept(struct accept_args *uap) 437 { 438 struct sockaddr *sa = NULL; 439 int sa_len; 440 int error; 441 442 if (uap->name) { 443 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 444 if (error) 445 return (error); 446 447 error = kern_accept(uap->s, 0, &sa, &sa_len, 448 &uap->sysmsg_iresult, 0); 449 450 if (error == 0) 451 error = copyout(sa, uap->name, sa_len); 452 if (error == 0) { 453 error = copyout(&sa_len, uap->anamelen, 454 sizeof(*uap->anamelen)); 455 } 456 if (sa) 457 kfree(sa, M_SONAME); 458 } else { 459 error = kern_accept(uap->s, 0, NULL, 0, 460 &uap->sysmsg_iresult, 0); 461 } 462 return (error); 463 } 464 465 /* 466 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 467 * 468 * MPALMOSTSAFE 469 */ 470 int 471 sys_extaccept(struct extaccept_args *uap) 472 { 473 struct sockaddr *sa = NULL; 474 int sa_len; 475 int error; 476 int fflags = uap->flags & O_FMASK; 477 478 if (uap->name) { 479 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 480 if (error) 481 return (error); 482 483 error = kern_accept(uap->s, fflags, &sa, &sa_len, 484 &uap->sysmsg_iresult, 0); 485 486 if (error == 0) 487 error = copyout(sa, uap->name, sa_len); 488 if (error == 0) { 489 error = copyout(&sa_len, uap->anamelen, 490 sizeof(*uap->anamelen)); 491 } 492 if (sa) 493 kfree(sa, M_SONAME); 494 } else { 495 error = kern_accept(uap->s, fflags, NULL, 0, 496 &uap->sysmsg_iresult, 0); 497 } 498 return (error); 499 } 500 501 /* 502 * accept4(int s, caddr_t name, int *anamelen, int flags) 503 * 504 * MPALMOSTSAFE 505 */ 506 int 507 sys_accept4(struct accept4_args *uap) 508 { 509 struct sockaddr *sa = NULL; 510 int sa_len; 511 int error; 512 int sockflags; 513 514 if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC)) 515 return (EINVAL); 516 sockflags = uap->flags | SOCK_KERN_NOINHERIT; 517 518 if (uap->name) { 519 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 520 if (error) 521 return (error); 522 523 error = kern_accept(uap->s, 0, &sa, &sa_len, 524 &uap->sysmsg_iresult, sockflags); 525 526 if (error == 0) 527 error = copyout(sa, uap->name, sa_len); 528 if (error == 0) { 529 error = copyout(&sa_len, uap->anamelen, 530 sizeof(*uap->anamelen)); 531 } 532 if (sa) 533 kfree(sa, M_SONAME); 534 } else { 535 error = kern_accept(uap->s, 0, NULL, 0, 536 &uap->sysmsg_iresult, sockflags); 537 } 538 return (error); 539 } 540 541 /* 542 * Returns TRUE if predicate satisfied. 543 */ 544 static boolean_t 545 soconnected_predicate(struct netmsg_so_notify *msg) 546 { 547 struct socket *so = msg->base.nm_so; 548 549 /* check predicate */ 550 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 551 msg->base.lmsg.ms_error = so->so_error; 552 return (TRUE); 553 } 554 555 return (FALSE); 556 } 557 558 int 559 kern_connect(int s, int fflags, struct sockaddr *sa) 560 { 561 struct thread *td = curthread; 562 struct proc *p = td->td_proc; 563 struct file *fp; 564 struct socket *so; 565 int error, interrupted = 0; 566 567 error = holdsock(p->p_fd, s, &fp); 568 if (error) 569 return (error); 570 so = (struct socket *)fp->f_data; 571 572 if (fflags & O_FBLOCKING) 573 /* fflags &= ~FNONBLOCK; */; 574 else if (fflags & O_FNONBLOCKING) 575 fflags |= FNONBLOCK; 576 else 577 fflags = fp->f_flag; 578 579 if (so->so_state & SS_ISCONNECTING) { 580 error = EALREADY; 581 goto done; 582 } 583 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE); 584 if (error) 585 goto bad; 586 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 587 error = EINPROGRESS; 588 goto done; 589 } 590 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 591 struct netmsg_so_notify msg; 592 593 netmsg_init_abortable(&msg.base, so, 594 &curthread->td_msgport, 595 0, 596 netmsg_so_notify, 597 netmsg_so_notify_doabort); 598 msg.nm_predicate = soconnected_predicate; 599 msg.nm_etype = NM_REVENT; 600 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 601 if (error == EINTR || error == ERESTART) 602 interrupted = 1; 603 } 604 if (error == 0) { 605 error = so->so_error; 606 so->so_error = 0; 607 } 608 bad: 609 if (!interrupted) 610 soclrstate(so, SS_ISCONNECTING); 611 if (error == ERESTART) 612 error = EINTR; 613 done: 614 fdrop(fp); 615 return (error); 616 } 617 618 /* 619 * connect_args(int s, caddr_t name, int namelen) 620 * 621 * MPALMOSTSAFE 622 */ 623 int 624 sys_connect(struct connect_args *uap) 625 { 626 struct sockaddr *sa; 627 int error; 628 629 error = getsockaddr(&sa, uap->name, uap->namelen); 630 if (error) 631 return (error); 632 error = kern_connect(uap->s, 0, sa); 633 kfree(sa, M_SONAME); 634 635 return (error); 636 } 637 638 /* 639 * connect_args(int s, int fflags, caddr_t name, int namelen) 640 * 641 * MPALMOSTSAFE 642 */ 643 int 644 sys_extconnect(struct extconnect_args *uap) 645 { 646 struct sockaddr *sa; 647 int error; 648 int fflags = uap->flags & O_FMASK; 649 650 error = getsockaddr(&sa, uap->name, uap->namelen); 651 if (error) 652 return (error); 653 error = kern_connect(uap->s, fflags, sa); 654 kfree(sa, M_SONAME); 655 656 return (error); 657 } 658 659 int 660 kern_socketpair(int domain, int type, int protocol, int *sv) 661 { 662 struct thread *td = curthread; 663 struct filedesc *fdp; 664 struct file *fp1, *fp2; 665 struct socket *so1, *so2; 666 int fd1, fd2, error; 667 u_int fflags = 0; 668 int oflags = 0; 669 670 if (type & SOCK_NONBLOCK) { 671 type &= ~SOCK_NONBLOCK; 672 fflags |= FNONBLOCK; 673 } 674 if (type & SOCK_CLOEXEC) { 675 type &= ~SOCK_CLOEXEC; 676 oflags |= O_CLOEXEC; 677 } 678 679 fdp = td->td_proc->p_fd; 680 error = socreate(domain, &so1, type, protocol, td); 681 if (error) 682 return (error); 683 error = socreate(domain, &so2, type, protocol, td); 684 if (error) 685 goto free1; 686 error = falloc(td->td_lwp, &fp1, &fd1); 687 if (error) 688 goto free2; 689 sv[0] = fd1; 690 fp1->f_data = so1; 691 error = falloc(td->td_lwp, &fp2, &fd2); 692 if (error) 693 goto free3; 694 fp2->f_data = so2; 695 sv[1] = fd2; 696 error = soconnect2(so1, so2); 697 if (error) 698 goto free4; 699 if (type == SOCK_DGRAM) { 700 /* 701 * Datagram socket connection is asymmetric. 702 */ 703 error = soconnect2(so2, so1); 704 if (error) 705 goto free4; 706 } 707 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 708 fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags; 709 fp1->f_ops = fp2->f_ops = &socketops; 710 if (oflags & O_CLOEXEC) { 711 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE; 712 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE; 713 } 714 if (fflags & FNONBLOCK) { 715 int tmp; 716 717 tmp = 1; 718 fo_ioctl(fp1, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 719 tmp = 1; 720 fo_ioctl(fp2, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 721 } 722 fsetfd(fdp, fp1, fd1); 723 fsetfd(fdp, fp2, fd2); 724 fdrop(fp1); 725 fdrop(fp2); 726 return (error); 727 free4: 728 fsetfd(fdp, NULL, fd2); 729 fdrop(fp2); 730 free3: 731 fsetfd(fdp, NULL, fd1); 732 fdrop(fp1); 733 free2: 734 (void)soclose(so2, 0); 735 free1: 736 (void)soclose(so1, 0); 737 return (error); 738 } 739 740 /* 741 * socketpair(int domain, int type, int protocol, int *rsv) 742 */ 743 int 744 sys_socketpair(struct socketpair_args *uap) 745 { 746 int error, sockv[2]; 747 748 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 749 750 if (error == 0) { 751 error = copyout(sockv, uap->rsv, sizeof(sockv)); 752 753 if (error != 0) { 754 kern_close(sockv[0]); 755 kern_close(sockv[1]); 756 } 757 } 758 759 return (error); 760 } 761 762 int 763 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 764 struct mbuf *control, int flags, size_t *res) 765 { 766 struct thread *td = curthread; 767 struct lwp *lp = td->td_lwp; 768 struct proc *p = td->td_proc; 769 struct file *fp; 770 size_t len; 771 int error; 772 struct socket *so; 773 #ifdef KTRACE 774 struct iovec *ktriov = NULL; 775 struct uio ktruio; 776 #endif 777 778 error = holdsock(p->p_fd, s, &fp); 779 if (error) 780 return (error); 781 #ifdef KTRACE 782 if (KTRPOINT(td, KTR_GENIO)) { 783 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 784 785 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 786 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 787 ktruio = *auio; 788 } 789 #endif 790 len = auio->uio_resid; 791 so = (struct socket *)fp->f_data; 792 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 793 if (fp->f_flag & FNONBLOCK) 794 flags |= MSG_FNONBLOCKING; 795 } 796 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 797 if (error) { 798 if (auio->uio_resid != len && (error == ERESTART || 799 error == EINTR || error == EWOULDBLOCK)) 800 error = 0; 801 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 802 !(so->so_options & SO_NOSIGPIPE)) 803 lwpsignal(p, lp, SIGPIPE); 804 } 805 #ifdef KTRACE 806 if (ktriov != NULL) { 807 if (error == 0) { 808 ktruio.uio_iov = ktriov; 809 ktruio.uio_resid = len - auio->uio_resid; 810 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 811 } 812 kfree(ktriov, M_TEMP); 813 } 814 #endif 815 if (error == 0) 816 *res = len - auio->uio_resid; 817 fdrop(fp); 818 return (error); 819 } 820 821 /* 822 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 823 * 824 * MPALMOSTSAFE 825 */ 826 int 827 sys_sendto(struct sendto_args *uap) 828 { 829 struct thread *td = curthread; 830 struct uio auio; 831 struct iovec aiov; 832 struct sockaddr *sa = NULL; 833 int error; 834 835 if (uap->to) { 836 error = getsockaddr(&sa, uap->to, uap->tolen); 837 if (error) 838 return (error); 839 } 840 aiov.iov_base = uap->buf; 841 aiov.iov_len = uap->len; 842 auio.uio_iov = &aiov; 843 auio.uio_iovcnt = 1; 844 auio.uio_offset = 0; 845 auio.uio_resid = uap->len; 846 auio.uio_segflg = UIO_USERSPACE; 847 auio.uio_rw = UIO_WRITE; 848 auio.uio_td = td; 849 850 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 851 &uap->sysmsg_szresult); 852 853 if (sa) 854 kfree(sa, M_SONAME); 855 return (error); 856 } 857 858 /* 859 * sendmsg_args(int s, caddr_t msg, int flags) 860 * 861 * MPALMOSTSAFE 862 */ 863 int 864 sys_sendmsg(struct sendmsg_args *uap) 865 { 866 struct thread *td = curthread; 867 struct msghdr msg; 868 struct uio auio; 869 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 870 struct sockaddr *sa = NULL; 871 struct mbuf *control = NULL; 872 int error; 873 874 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 875 if (error) 876 return (error); 877 878 /* 879 * Conditionally copyin msg.msg_name. 880 */ 881 if (msg.msg_name) { 882 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 883 if (error) 884 return (error); 885 } 886 887 /* 888 * Populate auio. 889 */ 890 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 891 &auio.uio_resid); 892 if (error) 893 goto cleanup2; 894 auio.uio_iov = iov; 895 auio.uio_iovcnt = msg.msg_iovlen; 896 auio.uio_offset = 0; 897 auio.uio_segflg = UIO_USERSPACE; 898 auio.uio_rw = UIO_WRITE; 899 auio.uio_td = td; 900 901 /* 902 * Conditionally copyin msg.msg_control. 903 */ 904 if (msg.msg_control) { 905 if (msg.msg_controllen < sizeof(struct cmsghdr) || 906 msg.msg_controllen > MLEN) { 907 error = EINVAL; 908 goto cleanup; 909 } 910 control = m_get(M_WAITOK, MT_CONTROL); 911 if (control == NULL) { 912 error = ENOBUFS; 913 goto cleanup; 914 } 915 control->m_len = msg.msg_controllen; 916 error = copyin(msg.msg_control, mtod(control, caddr_t), 917 msg.msg_controllen); 918 if (error) { 919 m_free(control); 920 goto cleanup; 921 } 922 } 923 924 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 925 &uap->sysmsg_szresult); 926 927 cleanup: 928 iovec_free(&iov, aiov); 929 cleanup2: 930 if (sa) 931 kfree(sa, M_SONAME); 932 return (error); 933 } 934 935 /* 936 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 937 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 938 * Don't forget to FREE() and m_free() these if they are returned. 939 */ 940 int 941 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 942 struct mbuf **control, int *flags, size_t *res) 943 { 944 struct thread *td = curthread; 945 struct proc *p = td->td_proc; 946 struct file *fp; 947 size_t len; 948 int error; 949 int lflags; 950 struct socket *so; 951 #ifdef KTRACE 952 struct iovec *ktriov = NULL; 953 struct uio ktruio; 954 #endif 955 956 error = holdsock(p->p_fd, s, &fp); 957 if (error) 958 return (error); 959 #ifdef KTRACE 960 if (KTRPOINT(td, KTR_GENIO)) { 961 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 962 963 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 964 bcopy(auio->uio_iov, ktriov, iovlen); 965 ktruio = *auio; 966 } 967 #endif 968 len = auio->uio_resid; 969 so = (struct socket *)fp->f_data; 970 971 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 972 if (fp->f_flag & FNONBLOCK) { 973 if (flags) { 974 *flags |= MSG_FNONBLOCKING; 975 } else { 976 lflags = MSG_FNONBLOCKING; 977 flags = &lflags; 978 } 979 } 980 } 981 982 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 983 if (error) { 984 if (auio->uio_resid != len && (error == ERESTART || 985 error == EINTR || error == EWOULDBLOCK)) 986 error = 0; 987 } 988 #ifdef KTRACE 989 if (ktriov != NULL) { 990 if (error == 0) { 991 ktruio.uio_iov = ktriov; 992 ktruio.uio_resid = len - auio->uio_resid; 993 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 994 } 995 kfree(ktriov, M_TEMP); 996 } 997 #endif 998 if (error == 0) 999 *res = len - auio->uio_resid; 1000 fdrop(fp); 1001 return (error); 1002 } 1003 1004 /* 1005 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 1006 * caddr_t from, int *fromlenaddr) 1007 * 1008 * MPALMOSTSAFE 1009 */ 1010 int 1011 sys_recvfrom(struct recvfrom_args *uap) 1012 { 1013 struct thread *td = curthread; 1014 struct uio auio; 1015 struct iovec aiov; 1016 struct sockaddr *sa = NULL; 1017 int error, fromlen; 1018 1019 if (uap->from && uap->fromlenaddr) { 1020 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 1021 if (error) 1022 return (error); 1023 if (fromlen < 0) 1024 return (EINVAL); 1025 } else { 1026 fromlen = 0; 1027 } 1028 aiov.iov_base = uap->buf; 1029 aiov.iov_len = uap->len; 1030 auio.uio_iov = &aiov; 1031 auio.uio_iovcnt = 1; 1032 auio.uio_offset = 0; 1033 auio.uio_resid = uap->len; 1034 auio.uio_segflg = UIO_USERSPACE; 1035 auio.uio_rw = UIO_READ; 1036 auio.uio_td = td; 1037 1038 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 1039 &uap->flags, &uap->sysmsg_szresult); 1040 1041 if (error == 0 && uap->from) { 1042 /* note: sa may still be NULL */ 1043 if (sa) { 1044 fromlen = MIN(fromlen, sa->sa_len); 1045 error = copyout(sa, uap->from, fromlen); 1046 } else { 1047 fromlen = 0; 1048 } 1049 if (error == 0) { 1050 error = copyout(&fromlen, uap->fromlenaddr, 1051 sizeof(fromlen)); 1052 } 1053 } 1054 if (sa) 1055 kfree(sa, M_SONAME); 1056 1057 return (error); 1058 } 1059 1060 /* 1061 * recvmsg_args(int s, struct msghdr *msg, int flags) 1062 * 1063 * MPALMOSTSAFE 1064 */ 1065 int 1066 sys_recvmsg(struct recvmsg_args *uap) 1067 { 1068 struct thread *td = curthread; 1069 struct msghdr msg; 1070 struct uio auio; 1071 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1072 struct mbuf *m, *control = NULL; 1073 struct sockaddr *sa = NULL; 1074 caddr_t ctlbuf; 1075 socklen_t *ufromlenp, *ucontrollenp; 1076 int error, fromlen, controllen, len, flags, *uflagsp; 1077 1078 /* 1079 * This copyin handles everything except the iovec. 1080 */ 1081 error = copyin(uap->msg, &msg, sizeof(msg)); 1082 if (error) 1083 return (error); 1084 1085 if (msg.msg_name && msg.msg_namelen < 0) 1086 return (EINVAL); 1087 if (msg.msg_control && msg.msg_controllen < 0) 1088 return (EINVAL); 1089 1090 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1091 msg_namelen)); 1092 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1093 msg_controllen)); 1094 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1095 msg_flags)); 1096 1097 /* 1098 * Populate auio. 1099 */ 1100 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1101 &auio.uio_resid); 1102 if (error) 1103 return (error); 1104 auio.uio_iov = iov; 1105 auio.uio_iovcnt = msg.msg_iovlen; 1106 auio.uio_offset = 0; 1107 auio.uio_segflg = UIO_USERSPACE; 1108 auio.uio_rw = UIO_READ; 1109 auio.uio_td = td; 1110 1111 flags = uap->flags; 1112 1113 error = kern_recvmsg(uap->s, 1114 (msg.msg_name ? &sa : NULL), &auio, 1115 (msg.msg_control ? &control : NULL), &flags, 1116 &uap->sysmsg_szresult); 1117 1118 /* 1119 * Conditionally copyout the name and populate the namelen field. 1120 */ 1121 if (error == 0 && msg.msg_name) { 1122 /* note: sa may still be NULL */ 1123 if (sa != NULL) { 1124 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1125 error = copyout(sa, msg.msg_name, fromlen); 1126 } else { 1127 fromlen = 0; 1128 } 1129 if (error == 0) 1130 error = copyout(&fromlen, ufromlenp, 1131 sizeof(*ufromlenp)); 1132 } 1133 1134 /* 1135 * Copyout msg.msg_control and msg.msg_controllen. 1136 */ 1137 if (error == 0 && msg.msg_control) { 1138 len = msg.msg_controllen; 1139 m = control; 1140 ctlbuf = (caddr_t)msg.msg_control; 1141 1142 while(m && len > 0) { 1143 unsigned int tocopy; 1144 1145 if (len >= m->m_len) { 1146 tocopy = m->m_len; 1147 } else { 1148 msg.msg_flags |= MSG_CTRUNC; 1149 tocopy = len; 1150 } 1151 1152 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1153 if (error) 1154 goto cleanup; 1155 1156 ctlbuf += tocopy; 1157 len -= tocopy; 1158 m = m->m_next; 1159 } 1160 controllen = ctlbuf - (caddr_t)msg.msg_control; 1161 error = copyout(&controllen, ucontrollenp, 1162 sizeof(*ucontrollenp)); 1163 } 1164 1165 if (error == 0) 1166 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1167 1168 cleanup: 1169 if (sa) 1170 kfree(sa, M_SONAME); 1171 iovec_free(&iov, aiov); 1172 if (control) 1173 m_freem(control); 1174 return (error); 1175 } 1176 1177 /* 1178 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1179 * in kernel pointer instead of a userland pointer. This allows us 1180 * to manipulate socket options in the emulation code. 1181 */ 1182 int 1183 kern_setsockopt(int s, struct sockopt *sopt) 1184 { 1185 struct thread *td = curthread; 1186 struct proc *p = td->td_proc; 1187 struct file *fp; 1188 int error; 1189 1190 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1191 return (EFAULT); 1192 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1193 return (EINVAL); 1194 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1195 return (EINVAL); 1196 1197 error = holdsock(p->p_fd, s, &fp); 1198 if (error) 1199 return (error); 1200 1201 error = sosetopt((struct socket *)fp->f_data, sopt); 1202 fdrop(fp); 1203 return (error); 1204 } 1205 1206 /* 1207 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1208 * 1209 * MPALMOSTSAFE 1210 */ 1211 int 1212 sys_setsockopt(struct setsockopt_args *uap) 1213 { 1214 struct thread *td = curthread; 1215 struct sockopt sopt; 1216 int error; 1217 1218 sopt.sopt_level = uap->level; 1219 sopt.sopt_name = uap->name; 1220 sopt.sopt_valsize = uap->valsize; 1221 sopt.sopt_td = td; 1222 sopt.sopt_val = NULL; 1223 1224 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1225 return (EINVAL); 1226 if (uap->val) { 1227 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1228 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1229 if (error) 1230 goto out; 1231 } 1232 1233 error = kern_setsockopt(uap->s, &sopt); 1234 out: 1235 if (uap->val) 1236 kfree(sopt.sopt_val, M_TEMP); 1237 return(error); 1238 } 1239 1240 /* 1241 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1242 * in kernel pointer instead of a userland pointer. This allows us 1243 * to manipulate socket options in the emulation code. 1244 */ 1245 int 1246 kern_getsockopt(int s, struct sockopt *sopt) 1247 { 1248 struct thread *td = curthread; 1249 struct proc *p = td->td_proc; 1250 struct file *fp; 1251 int error; 1252 1253 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1254 return (EFAULT); 1255 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1256 return (EINVAL); 1257 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1258 return (EINVAL); 1259 1260 error = holdsock(p->p_fd, s, &fp); 1261 if (error) 1262 return (error); 1263 1264 error = sogetopt((struct socket *)fp->f_data, sopt); 1265 fdrop(fp); 1266 return (error); 1267 } 1268 1269 /* 1270 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1271 * 1272 * MPALMOSTSAFE 1273 */ 1274 int 1275 sys_getsockopt(struct getsockopt_args *uap) 1276 { 1277 struct thread *td = curthread; 1278 struct sockopt sopt; 1279 int error, valsize; 1280 1281 if (uap->val) { 1282 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1283 if (error) 1284 return (error); 1285 } else { 1286 valsize = 0; 1287 } 1288 1289 sopt.sopt_level = uap->level; 1290 sopt.sopt_name = uap->name; 1291 sopt.sopt_valsize = valsize; 1292 sopt.sopt_td = td; 1293 sopt.sopt_val = NULL; 1294 1295 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1296 return (EINVAL); 1297 if (uap->val) { 1298 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1299 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1300 if (error) 1301 goto out; 1302 } 1303 1304 error = kern_getsockopt(uap->s, &sopt); 1305 if (error) 1306 goto out; 1307 valsize = sopt.sopt_valsize; 1308 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1309 if (error) 1310 goto out; 1311 if (uap->val) 1312 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1313 out: 1314 if (uap->val) 1315 kfree(sopt.sopt_val, M_TEMP); 1316 return (error); 1317 } 1318 1319 /* 1320 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1321 * This allows kern_getsockname() to return a pointer to an allocated struct 1322 * sockaddr which must be freed later with FREE(). The caller must 1323 * initialize *name to NULL. 1324 */ 1325 int 1326 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1327 { 1328 struct thread *td = curthread; 1329 struct proc *p = td->td_proc; 1330 struct file *fp; 1331 struct socket *so; 1332 struct sockaddr *sa = NULL; 1333 int error; 1334 1335 error = holdsock(p->p_fd, s, &fp); 1336 if (error) 1337 return (error); 1338 if (*namelen < 0) { 1339 fdrop(fp); 1340 return (EINVAL); 1341 } 1342 so = (struct socket *)fp->f_data; 1343 error = so_pru_sockaddr(so, &sa); 1344 if (error == 0) { 1345 if (sa == NULL) { 1346 *namelen = 0; 1347 } else { 1348 *namelen = MIN(*namelen, sa->sa_len); 1349 *name = sa; 1350 } 1351 } 1352 1353 fdrop(fp); 1354 return (error); 1355 } 1356 1357 /* 1358 * getsockname_args(int fdes, caddr_t asa, int *alen) 1359 * 1360 * Get socket name. 1361 * 1362 * MPALMOSTSAFE 1363 */ 1364 int 1365 sys_getsockname(struct getsockname_args *uap) 1366 { 1367 struct sockaddr *sa = NULL; 1368 int error, sa_len; 1369 1370 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1371 if (error) 1372 return (error); 1373 1374 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1375 1376 if (error == 0) 1377 error = copyout(sa, uap->asa, sa_len); 1378 if (error == 0) 1379 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1380 if (sa) 1381 kfree(sa, M_SONAME); 1382 return (error); 1383 } 1384 1385 /* 1386 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1387 * This allows kern_getpeername() to return a pointer to an allocated struct 1388 * sockaddr which must be freed later with FREE(). The caller must 1389 * initialize *name to NULL. 1390 */ 1391 int 1392 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1393 { 1394 struct thread *td = curthread; 1395 struct proc *p = td->td_proc; 1396 struct file *fp; 1397 struct socket *so; 1398 struct sockaddr *sa = NULL; 1399 int error; 1400 1401 error = holdsock(p->p_fd, s, &fp); 1402 if (error) 1403 return (error); 1404 if (*namelen < 0) { 1405 fdrop(fp); 1406 return (EINVAL); 1407 } 1408 so = (struct socket *)fp->f_data; 1409 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1410 fdrop(fp); 1411 return (ENOTCONN); 1412 } 1413 error = so_pru_peeraddr(so, &sa); 1414 if (error == 0) { 1415 if (sa == NULL) { 1416 *namelen = 0; 1417 } else { 1418 *namelen = MIN(*namelen, sa->sa_len); 1419 *name = sa; 1420 } 1421 } 1422 1423 fdrop(fp); 1424 return (error); 1425 } 1426 1427 /* 1428 * getpeername_args(int fdes, caddr_t asa, int *alen) 1429 * 1430 * Get name of peer for connected socket. 1431 * 1432 * MPALMOSTSAFE 1433 */ 1434 int 1435 sys_getpeername(struct getpeername_args *uap) 1436 { 1437 struct sockaddr *sa = NULL; 1438 int error, sa_len; 1439 1440 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1441 if (error) 1442 return (error); 1443 1444 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1445 1446 if (error == 0) 1447 error = copyout(sa, uap->asa, sa_len); 1448 if (error == 0) 1449 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1450 if (sa) 1451 kfree(sa, M_SONAME); 1452 return (error); 1453 } 1454 1455 int 1456 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1457 { 1458 struct sockaddr *sa; 1459 int error; 1460 1461 *namp = NULL; 1462 if (len > SOCK_MAXADDRLEN) 1463 return ENAMETOOLONG; 1464 if (len < offsetof(struct sockaddr, sa_data[0])) 1465 return EDOM; 1466 sa = kmalloc(len, M_SONAME, M_WAITOK); 1467 error = copyin(uaddr, sa, len); 1468 if (error) { 1469 kfree(sa, M_SONAME); 1470 } else { 1471 #if BYTE_ORDER != BIG_ENDIAN 1472 /* 1473 * The bind(), connect(), and sendto() syscalls were not 1474 * versioned for COMPAT_43. Thus, this check must stay. 1475 */ 1476 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1477 sa->sa_family = sa->sa_len; 1478 #endif 1479 sa->sa_len = len; 1480 *namp = sa; 1481 } 1482 return error; 1483 } 1484 1485 /* 1486 * Detach a mapped page and release resources back to the system. 1487 * We must release our wiring and if the object is ripped out 1488 * from under the vm_page we become responsible for freeing the 1489 * page. 1490 * 1491 * MPSAFE 1492 */ 1493 static void 1494 sf_buf_mfree(void *arg) 1495 { 1496 struct sf_buf *sf = arg; 1497 vm_page_t m; 1498 1499 m = sf_buf_page(sf); 1500 if (sf_buf_free(sf)) { 1501 /* sf invalid now */ 1502 /* 1503 vm_page_busy_wait(m, FALSE, "sockpgf"); 1504 vm_page_wakeup(m); 1505 */ 1506 vm_page_unhold(m); 1507 #if 0 1508 if (m->object == NULL && 1509 m->wire_count == 0 && 1510 (m->flags & PG_NEED_COMMIT) == 0) { 1511 vm_page_free(m); 1512 } else { 1513 vm_page_wakeup(m); 1514 } 1515 #endif 1516 } 1517 } 1518 1519 /* 1520 * sendfile(2). 1521 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1522 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1523 * 1524 * Send a file specified by 'fd' and starting at 'offset' to a socket 1525 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1526 * nbytes == 0. Optionally add a header and/or trailer to the socket 1527 * output. If specified, write the total number of bytes sent into *sbytes. 1528 * 1529 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1530 * the headers to count against the remaining bytes to be sent from 1531 * the file descriptor. We may wish to implement a compatibility syscall 1532 * in the future. 1533 * 1534 * MPALMOSTSAFE 1535 */ 1536 int 1537 sys_sendfile(struct sendfile_args *uap) 1538 { 1539 struct thread *td = curthread; 1540 struct proc *p = td->td_proc; 1541 struct file *fp; 1542 struct vnode *vp = NULL; 1543 struct sf_hdtr hdtr; 1544 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1545 struct uio auio; 1546 struct mbuf *mheader = NULL; 1547 size_t hbytes = 0; 1548 size_t tbytes; 1549 off_t hdtr_size = 0; 1550 off_t sbytes; 1551 int error; 1552 1553 KKASSERT(p); 1554 1555 /* 1556 * Do argument checking. Must be a regular file in, stream 1557 * type and connected socket out, positive offset. 1558 */ 1559 fp = holdfp(p->p_fd, uap->fd, FREAD); 1560 if (fp == NULL) { 1561 return (EBADF); 1562 } 1563 if (fp->f_type != DTYPE_VNODE) { 1564 fdrop(fp); 1565 return (EINVAL); 1566 } 1567 vp = (struct vnode *)fp->f_data; 1568 vref(vp); 1569 fdrop(fp); 1570 1571 /* 1572 * If specified, get the pointer to the sf_hdtr struct for 1573 * any headers/trailers. 1574 */ 1575 if (uap->hdtr) { 1576 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1577 if (error) 1578 goto done; 1579 /* 1580 * Send any headers. 1581 */ 1582 if (hdtr.headers) { 1583 error = iovec_copyin(hdtr.headers, &iov, aiov, 1584 hdtr.hdr_cnt, &hbytes); 1585 if (error) 1586 goto done; 1587 auio.uio_iov = iov; 1588 auio.uio_iovcnt = hdtr.hdr_cnt; 1589 auio.uio_offset = 0; 1590 auio.uio_segflg = UIO_USERSPACE; 1591 auio.uio_rw = UIO_WRITE; 1592 auio.uio_td = td; 1593 auio.uio_resid = hbytes; 1594 1595 mheader = m_uiomove(&auio); 1596 1597 iovec_free(&iov, aiov); 1598 if (mheader == NULL) 1599 goto done; 1600 } 1601 } 1602 1603 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1604 &sbytes, uap->flags); 1605 if (error) 1606 goto done; 1607 1608 /* 1609 * Send trailers. Wimp out and use writev(2). 1610 */ 1611 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1612 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1613 hdtr.trl_cnt, &auio.uio_resid); 1614 if (error) 1615 goto done; 1616 auio.uio_iov = iov; 1617 auio.uio_iovcnt = hdtr.trl_cnt; 1618 auio.uio_offset = 0; 1619 auio.uio_segflg = UIO_USERSPACE; 1620 auio.uio_rw = UIO_WRITE; 1621 auio.uio_td = td; 1622 1623 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1624 1625 iovec_free(&iov, aiov); 1626 if (error) 1627 goto done; 1628 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1629 } 1630 1631 done: 1632 if (vp) 1633 vrele(vp); 1634 if (uap->sbytes != NULL) { 1635 sbytes += hdtr_size; 1636 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1637 } 1638 return (error); 1639 } 1640 1641 int 1642 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1643 struct mbuf *mheader, off_t *sbytes, int flags) 1644 { 1645 struct thread *td = curthread; 1646 struct proc *p = td->td_proc; 1647 struct vm_object *obj; 1648 struct socket *so; 1649 struct file *fp; 1650 struct mbuf *m, *mp; 1651 struct sf_buf *sf; 1652 struct vm_page *pg; 1653 off_t off, xfsize, xbytes; 1654 off_t hbytes = 0; 1655 int error = 0; 1656 1657 if (vp->v_type != VREG) { 1658 error = EINVAL; 1659 goto done0; 1660 } 1661 if ((obj = vp->v_object) == NULL) { 1662 error = EINVAL; 1663 goto done0; 1664 } 1665 error = holdsock(p->p_fd, sfd, &fp); 1666 if (error) 1667 goto done0; 1668 so = (struct socket *)fp->f_data; 1669 if (so->so_type != SOCK_STREAM) { 1670 error = EINVAL; 1671 goto done; 1672 } 1673 if ((so->so_state & SS_ISCONNECTED) == 0) { 1674 error = ENOTCONN; 1675 goto done; 1676 } 1677 if (offset < 0) { 1678 error = EINVAL; 1679 goto done; 1680 } 1681 1682 /* 1683 * preallocation is required for asynchronous passing of mbufs, 1684 * otherwise we can wind up building up an infinite number of 1685 * mbufs during the asynchronous latency. 1686 */ 1687 if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) { 1688 error = EINVAL; 1689 goto done; 1690 } 1691 1692 *sbytes = 0; 1693 xbytes = 0; 1694 /* 1695 * Protect against multiple writers to the socket. 1696 */ 1697 ssb_lock(&so->so_snd, M_WAITOK); 1698 1699 /* 1700 * Loop through the pages in the file, starting with the requested 1701 * offset. Get a file page (do I/O if necessary), map the file page 1702 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1703 * it on the socket. 1704 */ 1705 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) { 1706 vm_pindex_t pindex; 1707 vm_offset_t pgoff; 1708 long space; 1709 1710 pindex = OFF_TO_IDX(off); 1711 retry_lookup: 1712 /* 1713 * Calculate the amount to transfer. Not to exceed a page, 1714 * the EOF, or the passed in nbytes. 1715 */ 1716 xfsize = vp->v_filesize - off; 1717 if (xfsize > PAGE_SIZE) 1718 xfsize = PAGE_SIZE; 1719 pgoff = (vm_offset_t)(off & PAGE_MASK); 1720 if (PAGE_SIZE - pgoff < xfsize) 1721 xfsize = PAGE_SIZE - pgoff; 1722 if (nbytes && xfsize > (nbytes - xbytes)) 1723 xfsize = nbytes - xbytes; 1724 if (xfsize <= 0) 1725 break; 1726 /* 1727 * Optimize the non-blocking case by looking at the socket space 1728 * before going to the extra work of constituting the sf_buf. 1729 */ 1730 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1731 space = ssb_space_prealloc(&so->so_snd); 1732 else 1733 space = ssb_space(&so->so_snd); 1734 1735 if ((fp->f_flag & FNONBLOCK) && space <= 0) { 1736 if (so->so_state & SS_CANTSENDMORE) 1737 error = EPIPE; 1738 else 1739 error = EAGAIN; 1740 ssb_unlock(&so->so_snd); 1741 goto done; 1742 } 1743 /* 1744 * Attempt to look up the page. 1745 * 1746 * Allocate if not found, wait and loop if busy, then hold the page. 1747 * We hold rather than wire the page because we do not want to prevent 1748 * filesystem truncation operations from occuring on the file. This 1749 * can happen even under normal operation if the file being sent is 1750 * remove()d after the sendfile() call completes, because the socket buffer 1751 * may still be draining. tmpfs will crash if we try to use wire. 1752 */ 1753 vm_object_hold(obj); 1754 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error); 1755 if (error) { 1756 vm_page_sleep_busy(pg, TRUE, "sfpbsy"); 1757 vm_object_drop(obj); 1758 goto retry_lookup; 1759 } 1760 if (pg == NULL) { 1761 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL | 1762 VM_ALLOC_NULL_OK); 1763 if (pg == NULL) { 1764 vm_wait(0); 1765 vm_object_drop(obj); 1766 goto retry_lookup; 1767 } 1768 } 1769 vm_page_hold(pg); 1770 vm_object_drop(obj); 1771 1772 /* 1773 * If page is not valid for what we need, initiate I/O 1774 */ 1775 1776 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1777 struct uio auio; 1778 struct iovec aiov; 1779 int bsize; 1780 1781 /* 1782 * Ensure that our page is still around when the I/O 1783 * completes. 1784 * 1785 * Ensure that our page is not modified while part of 1786 * a mbuf as this could mess up tcp checksums, DMA, 1787 * etc (XXX NEEDS WORK). The softbusy is supposed to 1788 * help here but it actually doesn't. 1789 * 1790 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying 1791 * VM pages are not protected by the soft-busy 1792 * unless we vm_page_protect... READ them, and 1793 * they STILL aren't protected against 1794 * modification via the buffer cache (VOP_WRITE). 1795 * 1796 * Fixing the second issue is particularly 1797 * difficult. 1798 * 1799 * XXX We also can't soft-busy anyway because it can 1800 * deadlock against the syncer doing a vfs_msync(), 1801 * vfs_msync->vmntvnodesca->vfs_msync_scan2-> 1802 * vm_object_page_clean->(scan)-> ... page 1803 * busy-wait. 1804 */ 1805 /*vm_page_io_start(pg);*/ 1806 vm_page_wakeup(pg); 1807 1808 /* 1809 * Get the page from backing store. 1810 */ 1811 bsize = vp->v_mount->mnt_stat.f_iosize; 1812 auio.uio_iov = &aiov; 1813 auio.uio_iovcnt = 1; 1814 aiov.iov_base = 0; 1815 aiov.iov_len = MAXBSIZE; 1816 auio.uio_resid = MAXBSIZE; 1817 auio.uio_offset = trunc_page(off); 1818 auio.uio_segflg = UIO_NOCOPY; 1819 auio.uio_rw = UIO_READ; 1820 auio.uio_td = td; 1821 vn_lock(vp, LK_SHARED | LK_RETRY); 1822 error = VOP_READ(vp, &auio, 1823 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1824 td->td_ucred); 1825 vn_unlock(vp); 1826 vm_page_flag_clear(pg, PG_ZERO); 1827 vm_page_busy_wait(pg, FALSE, "sockpg"); 1828 /*vm_page_io_finish(pg);*/ 1829 if (error) { 1830 vm_page_wakeup(pg); 1831 vm_page_unhold(pg); 1832 /* vm_page_try_to_free(pg); */ 1833 ssb_unlock(&so->so_snd); 1834 goto done; 1835 } 1836 } 1837 1838 1839 /* 1840 * Get a sendfile buf. We usually wait as long as necessary, 1841 * but this wait can be interrupted. 1842 */ 1843 if ((sf = sf_buf_alloc(pg)) == NULL) { 1844 vm_page_wakeup(pg); 1845 vm_page_unhold(pg); 1846 /* vm_page_try_to_free(pg); */ 1847 ssb_unlock(&so->so_snd); 1848 error = EINTR; 1849 goto done; 1850 } 1851 1852 /* 1853 * Get an mbuf header and set it up as having external storage. 1854 */ 1855 MGETHDR(m, M_WAITOK, MT_DATA); 1856 if (m == NULL) { 1857 error = ENOBUFS; 1858 vm_page_wakeup(pg); 1859 vm_page_unhold(pg); 1860 /* vm_page_try_to_free(pg); */ 1861 sf_buf_free(sf); 1862 ssb_unlock(&so->so_snd); 1863 goto done; 1864 } 1865 1866 vm_page_wakeup(pg); 1867 1868 m->m_ext.ext_free = sf_buf_mfree; 1869 m->m_ext.ext_ref = sf_buf_ref; 1870 m->m_ext.ext_arg = sf; 1871 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1872 m->m_ext.ext_size = PAGE_SIZE; 1873 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1874 m->m_flags |= M_EXT; 1875 m->m_pkthdr.len = m->m_len = xfsize; 1876 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1877 1878 if (mheader != NULL) { 1879 hbytes = mheader->m_pkthdr.len; 1880 mheader->m_pkthdr.len += m->m_pkthdr.len; 1881 m_cat(mheader, m); 1882 m = mheader; 1883 mheader = NULL; 1884 } else 1885 hbytes = 0; 1886 1887 /* 1888 * Add the buffer to the socket buffer chain. 1889 */ 1890 crit_enter(); 1891 retry_space: 1892 /* 1893 * Make sure that the socket is still able to take more data. 1894 * CANTSENDMORE being true usually means that the connection 1895 * was closed. so_error is true when an error was sensed after 1896 * a previous send. 1897 * The state is checked after the page mapping and buffer 1898 * allocation above since those operations may block and make 1899 * any socket checks stale. From this point forward, nothing 1900 * blocks before the pru_send (or more accurately, any blocking 1901 * results in a loop back to here to re-check). 1902 */ 1903 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1904 if (so->so_state & SS_CANTSENDMORE) { 1905 error = EPIPE; 1906 } else { 1907 error = so->so_error; 1908 so->so_error = 0; 1909 } 1910 m_freem(m); 1911 ssb_unlock(&so->so_snd); 1912 crit_exit(); 1913 goto done; 1914 } 1915 /* 1916 * Wait for socket space to become available. We do this just 1917 * after checking the connection state above in order to avoid 1918 * a race condition with ssb_wait(). 1919 */ 1920 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1921 space = ssb_space_prealloc(&so->so_snd); 1922 else 1923 space = ssb_space(&so->so_snd); 1924 1925 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1926 if (fp->f_flag & FNONBLOCK) { 1927 m_freem(m); 1928 ssb_unlock(&so->so_snd); 1929 crit_exit(); 1930 error = EAGAIN; 1931 goto done; 1932 } 1933 error = ssb_wait(&so->so_snd); 1934 /* 1935 * An error from ssb_wait usually indicates that we've 1936 * been interrupted by a signal. If we've sent anything 1937 * then return bytes sent, otherwise return the error. 1938 */ 1939 if (error) { 1940 m_freem(m); 1941 ssb_unlock(&so->so_snd); 1942 crit_exit(); 1943 goto done; 1944 } 1945 goto retry_space; 1946 } 1947 1948 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1949 for (mp = m; mp != NULL; mp = mp->m_next) 1950 ssb_preallocstream(&so->so_snd, mp); 1951 } 1952 if (use_sendfile_async) 1953 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1954 else 1955 error = so_pru_send(so, 0, m, NULL, NULL, td); 1956 1957 crit_exit(); 1958 if (error) { 1959 ssb_unlock(&so->so_snd); 1960 goto done; 1961 } 1962 } 1963 if (mheader != NULL) { 1964 *sbytes += mheader->m_pkthdr.len; 1965 1966 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1967 for (mp = mheader; mp != NULL; mp = mp->m_next) 1968 ssb_preallocstream(&so->so_snd, mp); 1969 } 1970 if (use_sendfile_async) 1971 error = so_pru_senda(so, 0, mheader, NULL, NULL, td); 1972 else 1973 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1974 1975 mheader = NULL; 1976 } 1977 ssb_unlock(&so->so_snd); 1978 1979 done: 1980 fdrop(fp); 1981 done0: 1982 if (mheader != NULL) 1983 m_freem(mheader); 1984 return (error); 1985 } 1986