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/sysmsg.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 #include <sys/jail.h> 61 #ifdef KTRACE 62 #include <sys/ktrace.h> 63 #endif 64 #include <vm/vm.h> 65 #include <vm/vm_object.h> 66 #include <vm/vm_page.h> 67 #include <vm/vm_pageout.h> 68 #include <vm/vm_kern.h> 69 #include <vm/vm_extern.h> 70 #include <sys/file2.h> 71 #include <sys/signalvar.h> 72 #include <sys/serialize.h> 73 74 #include <sys/thread2.h> 75 #include <sys/msgport2.h> 76 #include <sys/socketvar2.h> 77 #include <net/netmsg2.h> 78 #include <vm/vm_page2.h> 79 80 extern int use_soaccept_pred_fast; 81 extern int use_sendfile_async; 82 extern int use_soconnect_async; 83 84 /* 85 * System call interface to the socket abstraction. 86 */ 87 88 extern struct fileops socketops; 89 90 /* 91 * socket_args(int domain, int type, int protocol) 92 */ 93 int 94 kern_socket(int domain, int type, int protocol, int *res) 95 { 96 struct thread *td = curthread; 97 struct filedesc *fdp = td->td_proc->p_fd; 98 struct socket *so; 99 struct file *fp; 100 int fd, error; 101 u_int fflags = 0; 102 int oflags = 0; 103 104 KKASSERT(td->td_lwp); 105 106 if (type & SOCK_NONBLOCK) { 107 type &= ~SOCK_NONBLOCK; 108 fflags |= FNONBLOCK; 109 } 110 if (type & SOCK_CLOEXEC) { 111 type &= ~SOCK_CLOEXEC; 112 oflags |= O_CLOEXEC; 113 } 114 115 error = falloc(td->td_lwp, &fp, &fd); 116 if (error) 117 return (error); 118 error = socreate(domain, &so, type, protocol, td); 119 if (error) { 120 fsetfd(fdp, NULL, fd); 121 } else { 122 fp->f_type = DTYPE_SOCKET; 123 fp->f_flag = FREAD | FWRITE | fflags; 124 fp->f_ops = &socketops; 125 fp->f_data = so; 126 if (oflags & O_CLOEXEC) 127 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 128 *res = fd; 129 fsetfd(fdp, fp, fd); 130 } 131 fdrop(fp); 132 return (error); 133 } 134 135 /* 136 * MPALMOSTSAFE 137 */ 138 int 139 sys_socket(struct sysmsg *sysmsg, const struct socket_args *uap) 140 { 141 int error; 142 143 error = kern_socket(uap->domain, uap->type, uap->protocol, 144 &sysmsg->sysmsg_iresult); 145 146 return (error); 147 } 148 149 int 150 kern_bind(int s, struct sockaddr *sa) 151 { 152 struct thread *td = curthread; 153 struct file *fp; 154 int error; 155 156 error = holdsock(td, s, &fp); 157 if (error) 158 return (error); 159 error = sobind((struct socket *)fp->f_data, sa, td); 160 dropfp(td, s, fp); 161 162 return (error); 163 } 164 165 /* 166 * bind_args(int s, caddr_t name, int namelen) 167 * 168 * MPALMOSTSAFE 169 */ 170 int 171 sys_bind(struct sysmsg *sysmsg, const struct bind_args *uap) 172 { 173 struct sockaddr *sa; 174 int error; 175 176 error = getsockaddr(&sa, uap->name, uap->namelen); 177 if (error) 178 return (error); 179 if (!prison_remote_ip(curthread, sa)) { 180 kfree(sa, M_SONAME); 181 return EAFNOSUPPORT; 182 } 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 file *fp; 194 int error; 195 196 error = holdsock(td, s, &fp); 197 if (error) 198 return (error); 199 error = solisten((struct socket *)fp->f_data, backlog, td); 200 dropfp(td, s, fp); 201 202 return (error); 203 } 204 205 /* 206 * listen_args(int s, int backlog) 207 * 208 * MPALMOSTSAFE 209 */ 210 int 211 sys_listen(struct sysmsg *sysmsg, const struct listen_args *uap) 212 { 213 int error; 214 215 error = kern_listen(uap->s, uap->backlog); 216 return (error); 217 } 218 219 /* 220 * Returns the accepted socket as well. 221 * 222 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 223 * pool token is absolutely required to avoid a sofree() race, 224 * as well as to avoid tailq handling races. 225 */ 226 static boolean_t 227 soaccept_predicate(struct netmsg_so_notify *msg) 228 { 229 struct socket *head = msg->base.nm_so; 230 struct socket *so; 231 232 if (head->so_error != 0) { 233 msg->base.lmsg.ms_error = head->so_error; 234 return (TRUE); 235 } 236 lwkt_getpooltoken(head); 237 if (!TAILQ_EMPTY(&head->so_comp)) { 238 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 239 so = TAILQ_FIRST(&head->so_comp); 240 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP); 241 TAILQ_REMOVE(&head->so_comp, so, so_list); 242 head->so_qlen--; 243 soclrstate(so, SS_COMP); 244 245 /* 246 * Keep a reference before clearing the so_head 247 * to avoid racing socket close in netisr. 248 */ 249 soreference(so); 250 so->so_head = NULL; 251 252 lwkt_relpooltoken(head); 253 254 msg->base.lmsg.ms_error = 0; 255 msg->base.nm_so = so; 256 return (TRUE); 257 } 258 lwkt_relpooltoken(head); 259 if (head->so_state & SS_CANTRCVMORE) { 260 msg->base.lmsg.ms_error = ECONNABORTED; 261 return (TRUE); 262 } 263 if (msg->nm_fflags & FNONBLOCK) { 264 msg->base.lmsg.ms_error = EWOULDBLOCK; 265 return (TRUE); 266 } 267 268 return (FALSE); 269 } 270 271 /* 272 * The second argument to kern_accept() is a handle to a struct sockaddr. 273 * This allows kern_accept() to return a pointer to an allocated struct 274 * sockaddr which must be freed later with FREE(). The caller must 275 * initialize *name to NULL. 276 */ 277 int 278 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res, 279 int sockflags) 280 { 281 struct thread *td = curthread; 282 struct filedesc *fdp = td->td_proc->p_fd; 283 struct file *lfp = NULL; 284 struct file *nfp = NULL; 285 struct sockaddr *sa; 286 struct socket *head, *so; 287 struct netmsg_so_notify msg; 288 int fd; 289 u_int fflag; /* type must match fp->f_flag */ 290 int error, tmp; 291 292 *res = -1; 293 if (name && namelen && *namelen < 0) 294 return (EINVAL); 295 296 error = holdsock(td, s, &lfp); 297 if (error) 298 return (error); 299 300 error = falloc(td->td_lwp, &nfp, &fd); 301 if (error) { /* Probably ran out of file descriptors. */ 302 fdrop(lfp); 303 return (error); 304 } 305 head = (struct socket *)lfp->f_data; 306 if ((head->so_options & SO_ACCEPTCONN) == 0) { 307 error = EINVAL; 308 goto done; 309 } 310 311 if (fflags & O_FBLOCKING) 312 fflags |= lfp->f_flag & ~FNONBLOCK; 313 else if (fflags & O_FNONBLOCKING) 314 fflags |= lfp->f_flag | FNONBLOCK; 315 else 316 fflags = lfp->f_flag; 317 318 if (use_soaccept_pred_fast) { 319 boolean_t pred; 320 321 /* Initialize necessary parts for soaccept_predicate() */ 322 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 323 msg.nm_fflags = fflags; 324 325 lwkt_getpooltoken(head); 326 pred = soaccept_predicate(&msg); 327 lwkt_relpooltoken(head); 328 329 if (pred) { 330 error = msg.base.lmsg.ms_error; 331 if (error) 332 goto done; 333 else 334 goto accepted; 335 } 336 } 337 338 /* optimize for uniprocessor case later XXX JH */ 339 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 340 0, netmsg_so_notify, netmsg_so_notify_doabort); 341 msg.nm_predicate = soaccept_predicate; 342 msg.nm_fflags = fflags; 343 msg.nm_etype = NM_REVENT; 344 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 345 if (error) 346 goto done; 347 348 accepted: 349 /* 350 * At this point we have the connection that's ready to be accepted. 351 * 352 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 353 * to eat the ref and turn it into a descriptor. 354 */ 355 so = msg.base.nm_so; 356 357 fflag = lfp->f_flag; 358 359 /* connection has been removed from the listen queue */ 360 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 361 362 if (sockflags & SOCK_KERN_NOINHERIT) { 363 fflag &= ~(FASYNC | FNONBLOCK); 364 if (sockflags & SOCK_NONBLOCK) 365 fflag |= FNONBLOCK; 366 } else { 367 if (head->so_sigio != NULL) 368 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 369 } 370 371 nfp->f_type = DTYPE_SOCKET; 372 nfp->f_flag = fflag; 373 nfp->f_ops = &socketops; 374 nfp->f_data = so; 375 /* Sync socket async state with file flags */ 376 tmp = fflag & FASYNC; 377 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 378 379 sa = NULL; 380 if (so->so_faddr != NULL) { 381 sa = so->so_faddr; 382 so->so_faddr = NULL; 383 384 soaccept_generic(so); 385 error = 0; 386 } else { 387 error = soaccept(so, &sa); 388 } 389 390 /* 391 * Set the returned name and namelen as applicable. Set the returned 392 * namelen to 0 for older code which might ignore the return value 393 * from accept. 394 */ 395 if (error == 0) { 396 if (sa && name && namelen) { 397 if (*namelen > sa->sa_len) 398 *namelen = sa->sa_len; 399 *name = sa; 400 } else { 401 if (sa) 402 kfree(sa, M_SONAME); 403 } 404 } 405 406 done: 407 /* 408 * If an error occured clear the reserved descriptor, else associate 409 * nfp with it. 410 * 411 * Note that *res is normally ignored if an error is returned but 412 * a syscall message will still have access to the result code. 413 */ 414 if (error) { 415 fsetfd(fdp, NULL, fd); 416 } else { 417 if (sockflags & SOCK_CLOEXEC) 418 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 419 *res = fd; 420 fsetfd(fdp, nfp, fd); 421 } 422 fdrop(nfp); 423 dropfp(td, s, lfp); 424 425 return (error); 426 } 427 428 /* 429 * accept(int s, caddr_t name, int *anamelen) 430 * 431 * MPALMOSTSAFE 432 */ 433 int 434 sys_accept(struct sysmsg *sysmsg, const struct accept_args *uap) 435 { 436 struct sockaddr *sa = NULL; 437 int sa_len; 438 int error; 439 440 if (uap->name) { 441 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 442 if (error) 443 return (error); 444 445 error = kern_accept(uap->s, 0, &sa, &sa_len, 446 &sysmsg->sysmsg_iresult, 0); 447 448 if (error == 0) { 449 prison_local_ip(curthread, sa); 450 error = copyout(sa, uap->name, sa_len); 451 } 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 &sysmsg->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 sysmsg *sysmsg, const 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 &sysmsg->sysmsg_iresult, 0); 485 486 if (error == 0) { 487 prison_local_ip(curthread, sa); 488 error = copyout(sa, uap->name, sa_len); 489 } 490 if (error == 0) { 491 error = copyout(&sa_len, uap->anamelen, 492 sizeof(*uap->anamelen)); 493 } 494 if (sa) 495 kfree(sa, M_SONAME); 496 } else { 497 error = kern_accept(uap->s, fflags, NULL, 0, 498 &sysmsg->sysmsg_iresult, 0); 499 } 500 return (error); 501 } 502 503 /* 504 * accept4(int s, caddr_t name, int *anamelen, int flags) 505 * 506 * MPALMOSTSAFE 507 */ 508 int 509 sys_accept4(struct sysmsg *sysmsg, const struct accept4_args *uap) 510 { 511 struct sockaddr *sa = NULL; 512 int sa_len; 513 int error; 514 int sockflags; 515 516 if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC)) 517 return (EINVAL); 518 sockflags = uap->flags | SOCK_KERN_NOINHERIT; 519 520 if (uap->name) { 521 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 522 if (error) 523 return (error); 524 525 error = kern_accept(uap->s, 0, &sa, &sa_len, 526 &sysmsg->sysmsg_iresult, sockflags); 527 528 if (error == 0) { 529 prison_local_ip(curthread, sa); 530 error = copyout(sa, uap->name, sa_len); 531 } 532 if (error == 0) { 533 error = copyout(&sa_len, uap->anamelen, 534 sizeof(*uap->anamelen)); 535 } 536 if (sa) 537 kfree(sa, M_SONAME); 538 } else { 539 error = kern_accept(uap->s, 0, NULL, 0, 540 &sysmsg->sysmsg_iresult, sockflags); 541 } 542 return (error); 543 } 544 545 /* 546 * Returns TRUE if predicate satisfied. 547 */ 548 static boolean_t 549 soconnected_predicate(struct netmsg_so_notify *msg) 550 { 551 struct socket *so = msg->base.nm_so; 552 553 /* check predicate */ 554 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 555 msg->base.lmsg.ms_error = so->so_error; 556 return (TRUE); 557 } 558 559 return (FALSE); 560 } 561 562 int 563 kern_connect(int s, int fflags, struct sockaddr *sa) 564 { 565 struct thread *td = curthread; 566 struct file *fp; 567 struct socket *so; 568 int error, interrupted = 0; 569 570 error = holdsock(td, s, &fp); 571 if (error) 572 return (error); 573 so = (struct socket *)fp->f_data; 574 575 if (fflags & O_FBLOCKING) 576 /* fflags &= ~FNONBLOCK; */; 577 else if (fflags & O_FNONBLOCKING) 578 fflags |= FNONBLOCK; 579 else 580 fflags = fp->f_flag; 581 582 if (so->so_state & SS_ISCONNECTING) { 583 error = EALREADY; 584 goto done; 585 } 586 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE); 587 if (error) 588 goto bad; 589 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 590 error = EINPROGRESS; 591 goto done; 592 } 593 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 594 struct netmsg_so_notify msg; 595 596 netmsg_init_abortable(&msg.base, so, 597 &curthread->td_msgport, 598 0, 599 netmsg_so_notify, 600 netmsg_so_notify_doabort); 601 msg.nm_predicate = soconnected_predicate; 602 msg.nm_etype = NM_REVENT; 603 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 604 if (error == EINTR || error == ERESTART) 605 interrupted = 1; 606 } 607 if (error == 0) { 608 error = so->so_error; 609 so->so_error = 0; 610 } 611 bad: 612 if (!interrupted) 613 soclrstate(so, SS_ISCONNECTING); 614 if (error == ERESTART) 615 error = EINTR; 616 done: 617 dropfp(td, s, fp); 618 619 return (error); 620 } 621 622 /* 623 * connect_args(int s, caddr_t name, int namelen) 624 * 625 * MPALMOSTSAFE 626 */ 627 int 628 sys_connect(struct sysmsg *sysmsg, const struct connect_args *uap) 629 { 630 struct sockaddr *sa; 631 int error; 632 633 error = getsockaddr(&sa, uap->name, uap->namelen); 634 if (error) 635 return (error); 636 if (!prison_remote_ip(curthread, sa)) { 637 kfree(sa, M_SONAME); 638 return EAFNOSUPPORT; 639 } 640 error = kern_connect(uap->s, 0, sa); 641 kfree(sa, M_SONAME); 642 643 return (error); 644 } 645 646 /* 647 * connect_args(int s, int fflags, caddr_t name, int namelen) 648 * 649 * MPALMOSTSAFE 650 */ 651 int 652 sys_extconnect(struct sysmsg *sysmsg, const struct extconnect_args *uap) 653 { 654 struct sockaddr *sa; 655 int error; 656 int fflags = uap->flags & O_FMASK; 657 658 error = getsockaddr(&sa, uap->name, uap->namelen); 659 if (error) 660 return (error); 661 if (!prison_remote_ip(curthread, sa)) { 662 kfree(sa, M_SONAME); 663 return EAFNOSUPPORT; 664 } 665 error = kern_connect(uap->s, fflags, sa); 666 kfree(sa, M_SONAME); 667 668 return (error); 669 } 670 671 int 672 kern_socketpair(int domain, int type, int protocol, int *sv) 673 { 674 struct thread *td = curthread; 675 struct filedesc *fdp; 676 struct file *fp1, *fp2; 677 struct socket *so1, *so2; 678 int fd1, fd2, error; 679 u_int fflags = 0; 680 int oflags = 0; 681 682 if (type & SOCK_NONBLOCK) { 683 type &= ~SOCK_NONBLOCK; 684 fflags |= FNONBLOCK; 685 } 686 if (type & SOCK_CLOEXEC) { 687 type &= ~SOCK_CLOEXEC; 688 oflags |= O_CLOEXEC; 689 } 690 691 fdp = td->td_proc->p_fd; 692 error = socreate(domain, &so1, type, protocol, td); 693 if (error) 694 return (error); 695 error = socreate(domain, &so2, type, protocol, td); 696 if (error) 697 goto free1; 698 error = falloc(td->td_lwp, &fp1, &fd1); 699 if (error) 700 goto free2; 701 sv[0] = fd1; 702 fp1->f_data = so1; 703 error = falloc(td->td_lwp, &fp2, &fd2); 704 if (error) 705 goto free3; 706 fp2->f_data = so2; 707 sv[1] = fd2; 708 error = soconnect2(so1, so2); 709 if (error) 710 goto free4; 711 if (type == SOCK_DGRAM) { 712 /* 713 * Datagram socket connection is asymmetric. 714 */ 715 error = soconnect2(so2, so1); 716 if (error) 717 goto free4; 718 } 719 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 720 fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags; 721 fp1->f_ops = fp2->f_ops = &socketops; 722 if (oflags & O_CLOEXEC) { 723 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE; 724 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE; 725 } 726 fsetfd(fdp, fp1, fd1); 727 fsetfd(fdp, fp2, fd2); 728 fdrop(fp1); 729 fdrop(fp2); 730 return (error); 731 free4: 732 fsetfd(fdp, NULL, fd2); 733 fdrop(fp2); 734 free3: 735 fsetfd(fdp, NULL, fd1); 736 fdrop(fp1); 737 free2: 738 (void)soclose(so2, 0); 739 free1: 740 (void)soclose(so1, 0); 741 return (error); 742 } 743 744 /* 745 * socketpair(int domain, int type, int protocol, int *rsv) 746 */ 747 int 748 sys_socketpair(struct sysmsg *sysmsg, const struct socketpair_args *uap) 749 { 750 int error, sockv[2]; 751 752 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 753 754 if (error == 0) { 755 error = copyout(sockv, uap->rsv, sizeof(sockv)); 756 757 if (error != 0) { 758 kern_close(sockv[0]); 759 kern_close(sockv[1]); 760 } 761 } 762 763 return (error); 764 } 765 766 int 767 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 768 struct mbuf *control, int flags, size_t *res) 769 { 770 struct thread *td = curthread; 771 struct lwp *lp = td->td_lwp; 772 struct proc *p = td->td_proc; 773 struct file *fp; 774 size_t len; 775 int error; 776 struct socket *so; 777 #ifdef KTRACE 778 struct iovec *ktriov = NULL; 779 struct uio ktruio; 780 #endif 781 782 error = holdsock(td, s, &fp); 783 if (error) 784 return (error); 785 #ifdef KTRACE 786 if (KTRPOINT(td, KTR_GENIO)) { 787 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 788 789 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 790 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 791 ktruio = *auio; 792 } 793 #endif 794 len = auio->uio_resid; 795 so = (struct socket *)fp->f_data; 796 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 797 if (fp->f_flag & FNONBLOCK) 798 flags |= MSG_FNONBLOCKING; 799 } 800 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 801 if (error) { 802 if (auio->uio_resid != len && (error == ERESTART || 803 error == EINTR || error == EWOULDBLOCK)) 804 error = 0; 805 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 806 !(so->so_options & SO_NOSIGPIPE)) 807 lwpsignal(p, lp, SIGPIPE); 808 } 809 #ifdef KTRACE 810 if (ktriov != NULL) { 811 if (error == 0) { 812 ktruio.uio_iov = ktriov; 813 ktruio.uio_resid = len - auio->uio_resid; 814 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 815 } 816 kfree(ktriov, M_TEMP); 817 } 818 #endif 819 if (error == 0) 820 *res = len - auio->uio_resid; 821 dropfp(td, s, fp); 822 823 return (error); 824 } 825 826 /* 827 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 828 * 829 * MPALMOSTSAFE 830 */ 831 int 832 sys_sendto(struct sysmsg *sysmsg, const struct sendto_args *uap) 833 { 834 struct thread *td = curthread; 835 struct uio auio; 836 struct iovec aiov; 837 struct sockaddr *sa = NULL; 838 int error; 839 840 if (uap->to) { 841 error = getsockaddr(&sa, uap->to, uap->tolen); 842 if (error) 843 return (error); 844 if (!prison_remote_ip(curthread, sa)) { 845 kfree(sa, M_SONAME); 846 return EAFNOSUPPORT; 847 } 848 } 849 aiov.iov_base = uap->buf; 850 aiov.iov_len = uap->len; 851 auio.uio_iov = &aiov; 852 auio.uio_iovcnt = 1; 853 auio.uio_offset = 0; 854 auio.uio_resid = uap->len; 855 auio.uio_segflg = UIO_USERSPACE; 856 auio.uio_rw = UIO_WRITE; 857 auio.uio_td = td; 858 859 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 860 &sysmsg->sysmsg_szresult); 861 862 if (sa) 863 kfree(sa, M_SONAME); 864 return (error); 865 } 866 867 /* 868 * sendmsg_args(int s, caddr_t msg, int flags) 869 * 870 * MPALMOSTSAFE 871 */ 872 int 873 sys_sendmsg(struct sysmsg *sysmsg, const struct sendmsg_args *uap) 874 { 875 struct thread *td = curthread; 876 struct msghdr msg; 877 struct uio auio; 878 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 879 struct sockaddr *sa = NULL; 880 struct mbuf *control = NULL; 881 int error; 882 883 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 884 if (error) 885 return (error); 886 887 /* 888 * Conditionally copyin msg.msg_name. 889 */ 890 if (msg.msg_name) { 891 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 892 if (error) 893 return (error); 894 if (!prison_remote_ip(curthread, sa)) { 895 kfree(sa, M_SONAME); 896 return EAFNOSUPPORT; 897 } 898 } 899 900 /* 901 * Populate auio. 902 */ 903 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 904 &auio.uio_resid); 905 if (error) 906 goto cleanup2; 907 auio.uio_iov = iov; 908 auio.uio_iovcnt = msg.msg_iovlen; 909 auio.uio_offset = 0; 910 auio.uio_segflg = UIO_USERSPACE; 911 auio.uio_rw = UIO_WRITE; 912 auio.uio_td = td; 913 914 /* 915 * Conditionally copyin msg.msg_control. 916 */ 917 if (msg.msg_control) { 918 if (msg.msg_controllen < sizeof(struct cmsghdr) || 919 msg.msg_controllen > MLEN) { 920 error = EINVAL; 921 goto cleanup; 922 } 923 control = m_get(M_WAITOK, MT_CONTROL); 924 if (control == NULL) { 925 error = ENOBUFS; 926 goto cleanup; 927 } 928 control->m_len = msg.msg_controllen; 929 error = copyin(msg.msg_control, mtod(control, caddr_t), 930 msg.msg_controllen); 931 if (error) { 932 m_free(control); 933 goto cleanup; 934 } 935 } 936 937 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 938 &sysmsg->sysmsg_szresult); 939 940 cleanup: 941 iovec_free(&iov, aiov); 942 cleanup2: 943 if (sa) 944 kfree(sa, M_SONAME); 945 return (error); 946 } 947 948 /* 949 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 950 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 951 * Don't forget to FREE() and m_free() these if they are returned. 952 */ 953 int 954 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 955 struct mbuf **control, int *flags, size_t *res) 956 { 957 struct thread *td = curthread; 958 struct file *fp; 959 size_t len; 960 int error; 961 int lflags; 962 struct socket *so; 963 #ifdef KTRACE 964 struct iovec *ktriov = NULL; 965 struct uio ktruio; 966 #endif 967 968 error = holdsock(td, s, &fp); 969 if (error) 970 return (error); 971 #ifdef KTRACE 972 if (KTRPOINT(td, KTR_GENIO)) { 973 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 974 975 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 976 bcopy(auio->uio_iov, ktriov, iovlen); 977 ktruio = *auio; 978 } 979 #endif 980 len = auio->uio_resid; 981 so = (struct socket *)fp->f_data; 982 983 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 984 if (fp->f_flag & FNONBLOCK) { 985 if (flags) { 986 *flags |= MSG_FNONBLOCKING; 987 } else { 988 lflags = MSG_FNONBLOCKING; 989 flags = &lflags; 990 } 991 } 992 } 993 994 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 995 if (error) { 996 if (auio->uio_resid != len && (error == ERESTART || 997 error == EINTR || error == EWOULDBLOCK)) 998 error = 0; 999 } 1000 #ifdef KTRACE 1001 if (ktriov != NULL) { 1002 if (error == 0) { 1003 ktruio.uio_iov = ktriov; 1004 ktruio.uio_resid = len - auio->uio_resid; 1005 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 1006 } 1007 kfree(ktriov, M_TEMP); 1008 } 1009 #endif 1010 if (error == 0) 1011 *res = len - auio->uio_resid; 1012 dropfp(td, s, fp); 1013 1014 return (error); 1015 } 1016 1017 /* 1018 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 1019 * caddr_t from, int *fromlenaddr) 1020 * 1021 * MPALMOSTSAFE 1022 */ 1023 int 1024 sys_recvfrom(struct sysmsg *sysmsg, const struct recvfrom_args *uap) 1025 { 1026 struct thread *td = curthread; 1027 struct uio auio; 1028 struct iovec aiov; 1029 struct sockaddr *sa = NULL; 1030 int error, fromlen; 1031 int flags; 1032 1033 if (uap->from && uap->fromlenaddr) { 1034 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 1035 if (error) 1036 return (error); 1037 if (fromlen < 0) 1038 return (EINVAL); 1039 } else { 1040 fromlen = 0; 1041 } 1042 aiov.iov_base = uap->buf; 1043 aiov.iov_len = uap->len; 1044 auio.uio_iov = &aiov; 1045 auio.uio_iovcnt = 1; 1046 auio.uio_offset = 0; 1047 auio.uio_resid = uap->len; 1048 auio.uio_segflg = UIO_USERSPACE; 1049 auio.uio_rw = UIO_READ; 1050 auio.uio_td = td; 1051 flags = uap->flags; 1052 1053 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 1054 &flags, &sysmsg->sysmsg_szresult); 1055 1056 if (error == 0 && uap->from) { 1057 /* note: sa may still be NULL */ 1058 if (sa) { 1059 fromlen = MIN(fromlen, sa->sa_len); 1060 prison_local_ip(curthread, sa); 1061 error = copyout(sa, uap->from, fromlen); 1062 } else { 1063 fromlen = 0; 1064 } 1065 if (error == 0) { 1066 error = copyout(&fromlen, uap->fromlenaddr, 1067 sizeof(fromlen)); 1068 } 1069 } 1070 if (sa) 1071 kfree(sa, M_SONAME); 1072 1073 return (error); 1074 } 1075 1076 /* 1077 * recvmsg_args(int s, struct msghdr *msg, int flags) 1078 * 1079 * MPALMOSTSAFE 1080 */ 1081 int 1082 sys_recvmsg(struct sysmsg *sysmsg, const struct recvmsg_args *uap) 1083 { 1084 struct thread *td = curthread; 1085 struct msghdr msg; 1086 struct uio auio; 1087 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1088 struct mbuf *m, *control = NULL; 1089 struct sockaddr *sa = NULL; 1090 caddr_t ctlbuf; 1091 socklen_t *ufromlenp, *ucontrollenp; 1092 int error, fromlen, controllen, len, flags, *uflagsp; 1093 1094 /* 1095 * This copyin handles everything except the iovec. 1096 */ 1097 error = copyin(uap->msg, &msg, sizeof(msg)); 1098 if (error) 1099 return (error); 1100 1101 if (msg.msg_name && msg.msg_namelen < 0) 1102 return (EINVAL); 1103 if (msg.msg_control && msg.msg_controllen < 0) 1104 return (EINVAL); 1105 1106 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1107 msg_namelen)); 1108 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1109 msg_controllen)); 1110 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1111 msg_flags)); 1112 1113 /* 1114 * Populate auio. 1115 */ 1116 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1117 &auio.uio_resid); 1118 if (error) 1119 return (error); 1120 auio.uio_iov = iov; 1121 auio.uio_iovcnt = msg.msg_iovlen; 1122 auio.uio_offset = 0; 1123 auio.uio_segflg = UIO_USERSPACE; 1124 auio.uio_rw = UIO_READ; 1125 auio.uio_td = td; 1126 1127 flags = uap->flags; 1128 1129 error = kern_recvmsg(uap->s, 1130 (msg.msg_name ? &sa : NULL), &auio, 1131 (msg.msg_control ? &control : NULL), &flags, 1132 &sysmsg->sysmsg_szresult); 1133 1134 /* 1135 * Conditionally copyout the name and populate the namelen field. 1136 */ 1137 if (error == 0 && msg.msg_name) { 1138 /* note: sa may still be NULL */ 1139 if (sa != NULL) { 1140 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1141 prison_local_ip(curthread, sa); 1142 error = copyout(sa, msg.msg_name, fromlen); 1143 } else { 1144 fromlen = 0; 1145 } 1146 if (error == 0) 1147 error = copyout(&fromlen, ufromlenp, 1148 sizeof(*ufromlenp)); 1149 } 1150 1151 /* 1152 * Copyout msg.msg_control and msg.msg_controllen. 1153 */ 1154 if (error == 0 && msg.msg_control) { 1155 len = msg.msg_controllen; 1156 m = control; 1157 ctlbuf = (caddr_t)msg.msg_control; 1158 1159 while(m && len > 0) { 1160 unsigned int tocopy; 1161 1162 if (len >= m->m_len) { 1163 tocopy = m->m_len; 1164 } else { 1165 msg.msg_flags |= MSG_CTRUNC; 1166 tocopy = len; 1167 } 1168 1169 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1170 if (error) 1171 goto cleanup; 1172 1173 ctlbuf += tocopy; 1174 len -= tocopy; 1175 m = m->m_next; 1176 } 1177 controllen = ctlbuf - (caddr_t)msg.msg_control; 1178 error = copyout(&controllen, ucontrollenp, 1179 sizeof(*ucontrollenp)); 1180 } 1181 1182 if (error == 0) 1183 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1184 1185 cleanup: 1186 if (sa) 1187 kfree(sa, M_SONAME); 1188 iovec_free(&iov, aiov); 1189 if (control) 1190 m_freem(control); 1191 return (error); 1192 } 1193 1194 /* 1195 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1196 * in kernel pointer instead of a userland pointer. This allows us 1197 * to manipulate socket options in the emulation code. 1198 */ 1199 int 1200 kern_setsockopt(int s, struct sockopt *sopt) 1201 { 1202 struct thread *td = curthread; 1203 struct file *fp; 1204 int error; 1205 1206 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1207 return (EFAULT); 1208 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1209 return (EINVAL); 1210 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1211 return (EINVAL); 1212 1213 error = holdsock(td, s, &fp); 1214 if (error) 1215 return (error); 1216 1217 error = sosetopt((struct socket *)fp->f_data, sopt); 1218 dropfp(td, s, fp); 1219 1220 return (error); 1221 } 1222 1223 /* 1224 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1225 * 1226 * MPALMOSTSAFE 1227 */ 1228 int 1229 sys_setsockopt(struct sysmsg *sysmsg, const struct setsockopt_args *uap) 1230 { 1231 struct thread *td = curthread; 1232 struct sockopt sopt; 1233 int error; 1234 1235 sopt.sopt_level = uap->level; 1236 sopt.sopt_name = uap->name; 1237 sopt.sopt_valsize = uap->valsize; 1238 sopt.sopt_td = td; 1239 sopt.sopt_val = NULL; 1240 1241 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1242 return (EINVAL); 1243 if (uap->val) { 1244 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1245 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1246 if (error) 1247 goto out; 1248 } 1249 1250 error = kern_setsockopt(uap->s, &sopt); 1251 out: 1252 if (uap->val) 1253 kfree(sopt.sopt_val, M_TEMP); 1254 return(error); 1255 } 1256 1257 /* 1258 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1259 * in kernel pointer instead of a userland pointer. This allows us 1260 * to manipulate socket options in the emulation code. 1261 */ 1262 int 1263 kern_getsockopt(int s, struct sockopt *sopt) 1264 { 1265 struct thread *td = curthread; 1266 struct file *fp; 1267 int error; 1268 1269 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1270 return (EFAULT); 1271 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1272 return (EINVAL); 1273 1274 error = holdsock(td, s, &fp); 1275 if (error) 1276 return (error); 1277 1278 error = sogetopt((struct socket *)fp->f_data, sopt); 1279 dropfp(td, s, fp); 1280 1281 return (error); 1282 } 1283 1284 /* 1285 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1286 * 1287 * MPALMOSTSAFE 1288 */ 1289 int 1290 sys_getsockopt(struct sysmsg *sysmsg, const struct getsockopt_args *uap) 1291 { 1292 struct thread *td = curthread; 1293 struct sockopt sopt; 1294 int error, valsize, valszmax, mflag = 0; 1295 1296 if (uap->val) { 1297 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1298 if (error) 1299 return (error); 1300 } else { 1301 valsize = 0; 1302 } 1303 1304 sopt.sopt_level = uap->level; 1305 sopt.sopt_name = uap->name; 1306 sopt.sopt_valsize = valsize; 1307 sopt.sopt_td = td; 1308 sopt.sopt_val = NULL; 1309 1310 if (td->td_proc->p_ucred->cr_uid == 0) { 1311 valszmax = SOMAXOPT_SIZE0; 1312 mflag = M_NULLOK; 1313 } else { 1314 valszmax = SOMAXOPT_SIZE; 1315 } 1316 if (sopt.sopt_valsize > valszmax) /* unsigned */ 1317 return (EINVAL); 1318 if (uap->val) { 1319 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, 1320 M_WAITOK | mflag); 1321 if (sopt.sopt_val == NULL) 1322 return (ENOBUFS); 1323 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1324 if (error) 1325 goto out; 1326 } 1327 1328 error = kern_getsockopt(uap->s, &sopt); 1329 if (error) 1330 goto out; 1331 valsize = sopt.sopt_valsize; 1332 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1333 if (error) 1334 goto out; 1335 if (uap->val) 1336 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1337 out: 1338 if (uap->val) 1339 kfree(sopt.sopt_val, M_TEMP); 1340 return (error); 1341 } 1342 1343 /* 1344 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1345 * This allows kern_getsockname() to return a pointer to an allocated struct 1346 * sockaddr which must be freed later with FREE(). The caller must 1347 * initialize *name to NULL. 1348 */ 1349 int 1350 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1351 { 1352 struct thread *td = curthread; 1353 struct file *fp; 1354 struct socket *so; 1355 struct sockaddr *sa = NULL; 1356 int error; 1357 1358 error = holdsock(td, s, &fp); 1359 if (error) 1360 return (error); 1361 if (*namelen < 0) { 1362 fdrop(fp); 1363 return (EINVAL); 1364 } 1365 so = (struct socket *)fp->f_data; 1366 error = so_pru_sockaddr(so, &sa); 1367 if (error == 0) { 1368 if (sa == NULL) { 1369 *namelen = 0; 1370 } else { 1371 *namelen = MIN(*namelen, sa->sa_len); 1372 *name = sa; 1373 } 1374 } 1375 dropfp(td, s, fp); 1376 1377 return (error); 1378 } 1379 1380 /* 1381 * getsockname_args(int fdes, caddr_t asa, int *alen) 1382 * 1383 * Get socket name. 1384 * 1385 * MPALMOSTSAFE 1386 */ 1387 int 1388 sys_getsockname(struct sysmsg *sysmsg, const struct getsockname_args *uap) 1389 { 1390 struct sockaddr *sa = NULL; 1391 struct sockaddr satmp; 1392 int error, sa_len_in, sa_len_out; 1393 1394 error = copyin(uap->alen, &sa_len_in, sizeof(sa_len_in)); 1395 if (error) 1396 return (error); 1397 1398 sa_len_out = sa_len_in; 1399 error = kern_getsockname(uap->fdes, &sa, &sa_len_out); 1400 1401 if (error == 0) { 1402 if (sa) { 1403 prison_local_ip(curthread, sa); 1404 error = copyout(sa, uap->asa, sa_len_out); 1405 } else { 1406 /* 1407 * unnamed uipc sockets don't bother storing 1408 * sockaddr, simulate an AF_LOCAL sockaddr. 1409 */ 1410 sa_len_out = sizeof(satmp); 1411 if (sa_len_out > sa_len_in) 1412 sa_len_out = sa_len_in; 1413 if (sa_len_out < 0) 1414 sa_len_out = 0; 1415 bzero(&satmp, sizeof(satmp)); 1416 satmp.sa_len = sa_len_out; 1417 satmp.sa_family = AF_LOCAL; 1418 error = copyout(&satmp, uap->asa, sa_len_out); 1419 } 1420 } 1421 if (error == 0 && sa_len_out != sa_len_in) 1422 error = copyout(&sa_len_out, uap->alen, sizeof(*uap->alen)); 1423 if (sa) 1424 kfree(sa, M_SONAME); 1425 return (error); 1426 } 1427 1428 /* 1429 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1430 * This allows kern_getpeername() to return a pointer to an allocated struct 1431 * sockaddr which must be freed later with FREE(). The caller must 1432 * initialize *name to NULL. 1433 */ 1434 int 1435 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1436 { 1437 struct thread *td = curthread; 1438 struct file *fp; 1439 struct socket *so; 1440 struct sockaddr *sa = NULL; 1441 int error; 1442 1443 error = holdsock(td, s, &fp); 1444 if (error) 1445 return (error); 1446 if (*namelen < 0) { 1447 fdrop(fp); 1448 return (EINVAL); 1449 } 1450 so = (struct socket *)fp->f_data; 1451 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1452 fdrop(fp); 1453 return (ENOTCONN); 1454 } 1455 error = so_pru_peeraddr(so, &sa); 1456 if (error == 0) { 1457 if (sa == NULL) { 1458 *namelen = 0; 1459 } else { 1460 *namelen = MIN(*namelen, sa->sa_len); 1461 *name = sa; 1462 } 1463 } 1464 dropfp(td, s, fp); 1465 1466 return (error); 1467 } 1468 1469 /* 1470 * getpeername_args(int fdes, caddr_t asa, int *alen) 1471 * 1472 * Get name of peer for connected socket. 1473 * 1474 * MPALMOSTSAFE 1475 */ 1476 int 1477 sys_getpeername(struct sysmsg *sysmsg, const struct getpeername_args *uap) 1478 { 1479 struct sockaddr *sa = NULL; 1480 int error, sa_len; 1481 1482 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1483 if (error) 1484 return (error); 1485 1486 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1487 1488 if (error == 0) { 1489 prison_local_ip(curthread, sa); 1490 error = copyout(sa, uap->asa, sa_len); 1491 } 1492 if (error == 0) 1493 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1494 if (sa) 1495 kfree(sa, M_SONAME); 1496 return (error); 1497 } 1498 1499 int 1500 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1501 { 1502 struct sockaddr *sa; 1503 int error; 1504 1505 *namp = NULL; 1506 if (len > SOCK_MAXADDRLEN) 1507 return ENAMETOOLONG; 1508 if (len < offsetof(struct sockaddr, sa_data[0])) 1509 return EDOM; 1510 sa = kmalloc(len, M_SONAME, M_WAITOK); 1511 error = copyin(uaddr, sa, len); 1512 if (error) { 1513 kfree(sa, M_SONAME); 1514 } else { 1515 sa->sa_len = len; 1516 *namp = sa; 1517 } 1518 return error; 1519 } 1520 1521 /* 1522 * Detach a mapped page and release resources back to the system. 1523 * We must release our wiring and if the object is ripped out 1524 * from under the vm_page we become responsible for freeing the 1525 * page. 1526 * 1527 * MPSAFE 1528 */ 1529 static void 1530 sf_buf_mfree(void *arg) 1531 { 1532 struct sf_buf *sf = arg; 1533 vm_page_t m; 1534 1535 m = sf_buf_page(sf); 1536 if (sf_buf_free(sf)) { 1537 /* sf invalid now */ 1538 vm_page_sbusy_drop(m); 1539 #if 0 1540 if (m->object == NULL && 1541 m->wire_count == 0 && 1542 (m->flags & PG_NEED_COMMIT) == 0) { 1543 vm_page_free(m); 1544 } else { 1545 vm_page_wakeup(m); 1546 } 1547 #endif 1548 } 1549 } 1550 1551 /* 1552 * sendfile(2). 1553 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1554 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1555 * 1556 * Send a file specified by 'fd' and starting at 'offset' to a socket 1557 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1558 * nbytes == 0. Optionally add a header and/or trailer to the socket 1559 * output. If specified, write the total number of bytes sent into *sbytes. 1560 * 1561 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1562 * the headers to count against the remaining bytes to be sent from 1563 * the file descriptor. We may wish to implement a compatibility syscall 1564 * in the future. 1565 * 1566 * MPALMOSTSAFE 1567 */ 1568 int 1569 sys_sendfile(struct sysmsg *sysmsg, const struct sendfile_args *uap) 1570 { 1571 struct thread *td = curthread; 1572 struct file *fp; 1573 struct vnode *vp = NULL; 1574 struct sf_hdtr hdtr; 1575 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1576 struct uio auio; 1577 struct mbuf *mheader = NULL; 1578 size_t hbytes = 0; 1579 size_t tbytes; 1580 off_t hdtr_size = 0; 1581 off_t sbytes; 1582 int error; 1583 1584 /* 1585 * Do argument checking. Must be a regular file in, stream 1586 * type and connected socket out, positive offset. 1587 */ 1588 fp = holdfp(td, uap->fd, FREAD); 1589 if (fp == NULL) { 1590 return (EBADF); 1591 } 1592 if (fp->f_type != DTYPE_VNODE) { 1593 fdrop(fp); 1594 return (EINVAL); 1595 } 1596 vp = (struct vnode *)fp->f_data; 1597 vref(vp); 1598 dropfp(td, uap->fd, fp); 1599 1600 /* 1601 * If specified, get the pointer to the sf_hdtr struct for 1602 * any headers/trailers. 1603 */ 1604 if (uap->hdtr) { 1605 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1606 if (error) 1607 goto done; 1608 /* 1609 * Send any headers. 1610 */ 1611 if (hdtr.headers) { 1612 error = iovec_copyin(hdtr.headers, &iov, aiov, 1613 hdtr.hdr_cnt, &hbytes); 1614 if (error) 1615 goto done; 1616 auio.uio_iov = iov; 1617 auio.uio_iovcnt = hdtr.hdr_cnt; 1618 auio.uio_offset = 0; 1619 auio.uio_segflg = UIO_USERSPACE; 1620 auio.uio_rw = UIO_WRITE; 1621 auio.uio_td = td; 1622 auio.uio_resid = hbytes; 1623 1624 mheader = m_uiomove(&auio); 1625 1626 iovec_free(&iov, aiov); 1627 if (mheader == NULL) 1628 goto done; 1629 } 1630 } 1631 1632 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1633 &sbytes, uap->flags); 1634 if (error) 1635 goto done; 1636 1637 /* 1638 * Send trailers. Wimp out and use writev(2). 1639 */ 1640 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1641 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1642 hdtr.trl_cnt, &auio.uio_resid); 1643 if (error) 1644 goto done; 1645 auio.uio_iov = iov; 1646 auio.uio_iovcnt = hdtr.trl_cnt; 1647 auio.uio_offset = 0; 1648 auio.uio_segflg = UIO_USERSPACE; 1649 auio.uio_rw = UIO_WRITE; 1650 auio.uio_td = td; 1651 1652 tbytes = 0; /* avoid gcc warnings */ 1653 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1654 1655 iovec_free(&iov, aiov); 1656 if (error) 1657 goto done; 1658 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1659 } 1660 1661 done: 1662 if (vp) 1663 vrele(vp); 1664 if (uap->sbytes != NULL) { 1665 sbytes += hdtr_size; 1666 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1667 } 1668 return (error); 1669 } 1670 1671 int 1672 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1673 struct mbuf *mheader, off_t *sbytes, int flags) 1674 { 1675 struct thread *td = curthread; 1676 struct vm_object *obj; 1677 struct socket *so; 1678 struct file *fp; 1679 struct mbuf *m, *mp; 1680 struct sf_buf *sf; 1681 struct vm_page *pg; 1682 off_t off, xfsize, xbytes; 1683 off_t hbytes = 0; 1684 int error = 0; 1685 1686 if (vp->v_type != VREG) { 1687 error = EINVAL; 1688 goto done0; 1689 } 1690 if ((obj = vp->v_object) == NULL) { 1691 error = EINVAL; 1692 goto done0; 1693 } 1694 error = holdsock(td, sfd, &fp); 1695 if (error) 1696 goto done0; 1697 so = (struct socket *)fp->f_data; 1698 if (so->so_type != SOCK_STREAM) { 1699 error = EINVAL; 1700 goto done1; 1701 } 1702 if ((so->so_state & SS_ISCONNECTED) == 0) { 1703 error = ENOTCONN; 1704 goto done1; 1705 } 1706 if (offset < 0) { 1707 error = EINVAL; 1708 goto done1; 1709 } 1710 1711 /* 1712 * preallocation is required for asynchronous passing of mbufs, 1713 * otherwise we can wind up building up an infinite number of 1714 * mbufs during the asynchronous latency. 1715 */ 1716 if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) { 1717 error = EINVAL; 1718 goto done1; 1719 } 1720 1721 *sbytes = 0; 1722 xbytes = 0; 1723 1724 /* 1725 * Protect against multiple writers to the socket. 1726 * We need at least a shared lock on the VM object 1727 */ 1728 ssb_lock(&so->so_snd, M_WAITOK); 1729 vm_object_hold_shared(obj); 1730 1731 /* 1732 * Loop through the pages in the file, starting with the requested 1733 * offset. Get a file page (do I/O if necessary), map the file page 1734 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1735 * it on the socket. 1736 */ 1737 for (off = offset; ; 1738 off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) { 1739 vm_pindex_t pindex; 1740 vm_offset_t pgoff; 1741 long space; 1742 int loops; 1743 1744 pindex = OFF_TO_IDX(off); 1745 loops = 0; 1746 1747 retry_lookup: 1748 /* 1749 * Calculate the amount to transfer. Not to exceed a page, 1750 * the EOF, or the passed in nbytes. 1751 */ 1752 xfsize = vp->v_filesize - off; 1753 if (xfsize > PAGE_SIZE) 1754 xfsize = PAGE_SIZE; 1755 pgoff = (vm_offset_t)(off & PAGE_MASK); 1756 if (PAGE_SIZE - pgoff < xfsize) 1757 xfsize = PAGE_SIZE - pgoff; 1758 if (nbytes && xfsize > (nbytes - xbytes)) 1759 xfsize = nbytes - xbytes; 1760 if (xfsize <= 0) 1761 break; 1762 /* 1763 * Optimize the non-blocking case by looking at the socket space 1764 * before going to the extra work of constituting the sf_buf. 1765 */ 1766 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1767 space = ssb_space_prealloc(&so->so_snd); 1768 else 1769 space = ssb_space(&so->so_snd); 1770 1771 if ((fp->f_flag & FNONBLOCK) && space <= 0) { 1772 if (so->so_state & SS_CANTSENDMORE) 1773 error = EPIPE; 1774 else 1775 error = EAGAIN; 1776 goto done; 1777 } 1778 1779 /* 1780 * Attempt to look up the page. 1781 * 1782 * Try to find the data using a shared vm_object token and 1783 * vm_page_lookup_sbusy_try() first. 1784 * 1785 * If data is missing, use a UIO_NOCOPY VOP_READ to load 1786 * the missing data and loop back up. We avoid all sorts 1787 * of problems by not trying to hold onto the page during 1788 * the I/O. 1789 * 1790 * NOTE: The soft-busy will temporary block filesystem 1791 * truncation operations when a file is removed 1792 * while the sendfile is running. 1793 */ 1794 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize); 1795 if (pg == NULL) { 1796 struct uio auio; 1797 struct iovec aiov; 1798 int bsize; 1799 1800 if (++loops > 100000) { 1801 kprintf("sendfile: VOP operation failed " 1802 "to retain page\n"); 1803 error = EIO; 1804 goto done; 1805 } 1806 1807 vm_object_drop(obj); 1808 bsize = vp->v_mount->mnt_stat.f_iosize; 1809 auio.uio_iov = &aiov; 1810 auio.uio_iovcnt = 1; 1811 aiov.iov_base = 0; 1812 aiov.iov_len = MAXBSIZE; 1813 auio.uio_resid = MAXBSIZE; 1814 auio.uio_offset = trunc_page(off); 1815 auio.uio_segflg = UIO_NOCOPY; 1816 auio.uio_rw = UIO_READ; 1817 auio.uio_td = td; 1818 1819 vn_lock(vp, LK_SHARED | LK_RETRY); 1820 error = VOP_READ_FP(vp, &auio, 1821 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1822 td->td_ucred, fp); 1823 vn_unlock(vp); 1824 vm_object_hold_shared(obj); 1825 1826 if (error) 1827 goto done; 1828 goto retry_lookup; 1829 } 1830 1831 /* 1832 * Get a sendfile buf. We usually wait as long as necessary, 1833 * but this wait can be interrupted. 1834 */ 1835 if ((sf = sf_buf_alloc(pg)) == NULL) { 1836 vm_page_sbusy_drop(pg); 1837 /* vm_page_try_to_free(pg); */ 1838 error = EINTR; 1839 goto done; 1840 } 1841 1842 /* 1843 * Get an mbuf header and set it up as having external storage. 1844 */ 1845 MGETHDR(m, M_WAITOK, MT_DATA); 1846 if (m == NULL) { 1847 error = ENOBUFS; 1848 vm_page_sbusy_drop(pg); 1849 /* vm_page_try_to_free(pg); */ 1850 sf_buf_free(sf); 1851 goto done; 1852 } 1853 1854 m->m_ext.ext_free = sf_buf_mfree; 1855 m->m_ext.ext_ref = sf_buf_ref; 1856 m->m_ext.ext_arg = sf; 1857 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1858 m->m_ext.ext_size = PAGE_SIZE; 1859 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1860 m->m_flags |= M_EXT; 1861 m->m_pkthdr.len = m->m_len = xfsize; 1862 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1863 1864 if (mheader != NULL) { 1865 hbytes = mheader->m_pkthdr.len; 1866 mheader->m_pkthdr.len += m->m_pkthdr.len; 1867 m_cat(mheader, m); 1868 m = mheader; 1869 mheader = NULL; 1870 } else { 1871 hbytes = 0; 1872 } 1873 1874 /* 1875 * Add the buffer to the socket buffer chain. 1876 */ 1877 crit_enter(); 1878 retry_space: 1879 /* 1880 * Make sure that the socket is still able to take more data. 1881 * CANTSENDMORE being true usually means that the connection 1882 * was closed. so_error is true when an error was sensed after 1883 * a previous send. 1884 * The state is checked after the page mapping and buffer 1885 * allocation above since those operations may block and make 1886 * any socket checks stale. From this point forward, nothing 1887 * blocks before the pru_send (or more accurately, any blocking 1888 * results in a loop back to here to re-check). 1889 */ 1890 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1891 if (so->so_state & SS_CANTSENDMORE) { 1892 error = EPIPE; 1893 } else { 1894 error = so->so_error; 1895 so->so_error = 0; 1896 } 1897 m_freem(m); 1898 crit_exit(); 1899 goto done; 1900 } 1901 /* 1902 * Wait for socket space to become available. We do this just 1903 * after checking the connection state above in order to avoid 1904 * a race condition with ssb_wait(). 1905 */ 1906 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1907 space = ssb_space_prealloc(&so->so_snd); 1908 else 1909 space = ssb_space(&so->so_snd); 1910 1911 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1912 if (fp->f_flag & FNONBLOCK) { 1913 m_freem(m); 1914 crit_exit(); 1915 error = EAGAIN; 1916 goto done; 1917 } 1918 error = ssb_wait(&so->so_snd); 1919 /* 1920 * An error from ssb_wait usually indicates that we've 1921 * been interrupted by a signal. If we've sent anything 1922 * then return bytes sent, otherwise return the error. 1923 */ 1924 if (error) { 1925 m_freem(m); 1926 crit_exit(); 1927 goto done; 1928 } 1929 goto retry_space; 1930 } 1931 1932 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1933 for (mp = m; mp != NULL; mp = mp->m_next) 1934 ssb_preallocstream(&so->so_snd, mp); 1935 } 1936 if (use_sendfile_async) 1937 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1938 else 1939 error = so_pru_send(so, 0, m, NULL, NULL, td); 1940 1941 crit_exit(); 1942 if (error) 1943 goto done; 1944 } 1945 if (mheader != NULL) { 1946 *sbytes += mheader->m_pkthdr.len; 1947 1948 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1949 for (mp = mheader; 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, mheader, NULL, NULL, td); 1954 else 1955 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1956 1957 mheader = NULL; 1958 } 1959 done: 1960 vm_object_drop(obj); 1961 ssb_unlock(&so->so_snd); 1962 done1: 1963 dropfp(td, sfd, fp); 1964 done0: 1965 if (mheader != NULL) 1966 m_freem(mheader); 1967 return (error); 1968 } 1969