1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/filedesc.h> 47 #include <sys/filio.h> 48 #include <sys/fcntl.h> 49 #include <sys/file.h> 50 #include <sys/proc.h> 51 #include <sys/signalvar.h> 52 #include <sys/socketvar.h> 53 #include <sys/uio.h> 54 #include <sys/kernel.h> 55 #include <sys/limits.h> 56 #include <sys/malloc.h> 57 #include <sys/poll.h> 58 #include <sys/resourcevar.h> 59 #include <sys/selinfo.h> 60 #include <sys/sleepqueue.h> 61 #include <sys/syscallsubr.h> 62 #include <sys/sysctl.h> 63 #include <sys/sysent.h> 64 #include <sys/vnode.h> 65 #include <sys/bio.h> 66 #include <sys/buf.h> 67 #include <sys/condvar.h> 68 #ifdef KTRACE 69 #include <sys/ktrace.h> 70 #endif 71 #include <vm/vm.h> 72 #include <vm/vm_page.h> 73 74 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); 75 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); 76 MALLOC_DEFINE(M_IOV, "iov", "large iov's"); 77 78 static int pollscan(struct thread *, struct pollfd *, u_int); 79 static int selscan(struct thread *, fd_mask **, fd_mask **, int); 80 static int dofileread(struct thread *, int, struct file *, struct uio *, 81 off_t, int); 82 static int dofilewrite(struct thread *, int, struct file *, struct uio *, 83 off_t, int); 84 static void doselwakeup(struct selinfo *, int); 85 86 /* 87 * Read system call. 88 */ 89 #ifndef _SYS_SYSPROTO_H_ 90 struct read_args { 91 int fd; 92 void *buf; 93 size_t nbyte; 94 }; 95 #endif 96 /* 97 * MPSAFE 98 */ 99 int 100 read(td, uap) 101 struct thread *td; 102 struct read_args *uap; 103 { 104 struct uio auio; 105 struct iovec aiov; 106 int error; 107 108 if (uap->nbyte > INT_MAX) 109 return (EINVAL); 110 aiov.iov_base = uap->buf; 111 aiov.iov_len = uap->nbyte; 112 auio.uio_iov = &aiov; 113 auio.uio_iovcnt = 1; 114 auio.uio_resid = uap->nbyte; 115 auio.uio_segflg = UIO_USERSPACE; 116 error = kern_readv(td, uap->fd, &auio); 117 return(error); 118 } 119 120 /* 121 * Positioned read system call 122 */ 123 #ifndef _SYS_SYSPROTO_H_ 124 struct pread_args { 125 int fd; 126 void *buf; 127 size_t nbyte; 128 int pad; 129 off_t offset; 130 }; 131 #endif 132 /* 133 * MPSAFE 134 */ 135 int 136 pread(td, uap) 137 struct thread *td; 138 struct pread_args *uap; 139 { 140 struct uio auio; 141 struct iovec aiov; 142 int error; 143 144 if (uap->nbyte > INT_MAX) 145 return (EINVAL); 146 aiov.iov_base = uap->buf; 147 aiov.iov_len = uap->nbyte; 148 auio.uio_iov = &aiov; 149 auio.uio_iovcnt = 1; 150 auio.uio_resid = uap->nbyte; 151 auio.uio_segflg = UIO_USERSPACE; 152 error = kern_preadv(td, uap->fd, &auio, uap->offset); 153 return(error); 154 } 155 156 /* 157 * Scatter read system call. 158 */ 159 #ifndef _SYS_SYSPROTO_H_ 160 struct readv_args { 161 int fd; 162 struct iovec *iovp; 163 u_int iovcnt; 164 }; 165 #endif 166 /* 167 * MPSAFE 168 */ 169 int 170 readv(struct thread *td, struct readv_args *uap) 171 { 172 struct uio *auio; 173 int error; 174 175 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 176 if (error) 177 return (error); 178 error = kern_readv(td, uap->fd, auio); 179 free(auio, M_IOV); 180 return (error); 181 } 182 183 int 184 kern_readv(struct thread *td, int fd, struct uio *auio) 185 { 186 struct file *fp; 187 int error; 188 189 error = fget_read(td, fd, &fp); 190 if (error) 191 return (error); 192 error = dofileread(td, fd, fp, auio, (off_t)-1, 0); 193 fdrop(fp, td); 194 return (error); 195 } 196 197 /* 198 * Scatter positioned read system call. 199 */ 200 #ifndef _SYS_SYSPROTO_H_ 201 struct preadv_args { 202 int fd; 203 struct iovec *iovp; 204 u_int iovcnt; 205 off_t offset; 206 }; 207 #endif 208 /* 209 * MPSAFE 210 */ 211 int 212 preadv(struct thread *td, struct preadv_args *uap) 213 { 214 struct uio *auio; 215 int error; 216 217 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 218 if (error) 219 return (error); 220 error = kern_preadv(td, uap->fd, auio, uap->offset); 221 free(auio, M_IOV); 222 return (error); 223 } 224 225 int 226 kern_preadv(td, fd, auio, offset) 227 struct thread *td; 228 int fd; 229 struct uio *auio; 230 off_t offset; 231 { 232 struct file *fp; 233 int error; 234 235 error = fget_read(td, fd, &fp); 236 if (error) 237 return (error); 238 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 239 error = ESPIPE; 240 else if (offset < 0 && fp->f_vnode->v_type != VCHR) 241 error = EINVAL; 242 else 243 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET); 244 fdrop(fp, td); 245 return (error); 246 } 247 248 /* 249 * Common code for readv and preadv that reads data in 250 * from a file using the passed in uio, offset, and flags. 251 */ 252 static int 253 dofileread(td, fd, fp, auio, offset, flags) 254 struct thread *td; 255 int fd; 256 struct file *fp; 257 struct uio *auio; 258 off_t offset; 259 int flags; 260 { 261 ssize_t cnt; 262 int error; 263 #ifdef KTRACE 264 struct uio *ktruio = NULL; 265 #endif 266 267 /* Finish zero length reads right here */ 268 if (auio->uio_resid == 0) { 269 td->td_retval[0] = 0; 270 return(0); 271 } 272 auio->uio_rw = UIO_READ; 273 auio->uio_offset = offset; 274 auio->uio_td = td; 275 #ifdef KTRACE 276 if (KTRPOINT(td, KTR_GENIO)) 277 ktruio = cloneuio(auio); 278 #endif 279 cnt = auio->uio_resid; 280 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) { 281 if (auio->uio_resid != cnt && (error == ERESTART || 282 error == EINTR || error == EWOULDBLOCK)) 283 error = 0; 284 } 285 cnt -= auio->uio_resid; 286 #ifdef KTRACE 287 if (ktruio != NULL) { 288 ktruio->uio_resid = cnt; 289 ktrgenio(fd, UIO_READ, ktruio, error); 290 } 291 #endif 292 td->td_retval[0] = cnt; 293 return (error); 294 } 295 296 /* 297 * Write system call 298 */ 299 #ifndef _SYS_SYSPROTO_H_ 300 struct write_args { 301 int fd; 302 const void *buf; 303 size_t nbyte; 304 }; 305 #endif 306 /* 307 * MPSAFE 308 */ 309 int 310 write(td, uap) 311 struct thread *td; 312 struct write_args *uap; 313 { 314 struct uio auio; 315 struct iovec aiov; 316 int error; 317 318 if (uap->nbyte > INT_MAX) 319 return (EINVAL); 320 aiov.iov_base = (void *)(uintptr_t)uap->buf; 321 aiov.iov_len = uap->nbyte; 322 auio.uio_iov = &aiov; 323 auio.uio_iovcnt = 1; 324 auio.uio_resid = uap->nbyte; 325 auio.uio_segflg = UIO_USERSPACE; 326 error = kern_writev(td, uap->fd, &auio); 327 return(error); 328 } 329 330 /* 331 * Positioned write system call 332 */ 333 #ifndef _SYS_SYSPROTO_H_ 334 struct pwrite_args { 335 int fd; 336 const void *buf; 337 size_t nbyte; 338 int pad; 339 off_t offset; 340 }; 341 #endif 342 /* 343 * MPSAFE 344 */ 345 int 346 pwrite(td, uap) 347 struct thread *td; 348 struct pwrite_args *uap; 349 { 350 struct uio auio; 351 struct iovec aiov; 352 int error; 353 354 if (uap->nbyte > INT_MAX) 355 return (EINVAL); 356 aiov.iov_base = (void *)(uintptr_t)uap->buf; 357 aiov.iov_len = uap->nbyte; 358 auio.uio_iov = &aiov; 359 auio.uio_iovcnt = 1; 360 auio.uio_resid = uap->nbyte; 361 auio.uio_segflg = UIO_USERSPACE; 362 error = kern_pwritev(td, uap->fd, &auio, uap->offset); 363 return(error); 364 } 365 366 /* 367 * Gather write system call 368 */ 369 #ifndef _SYS_SYSPROTO_H_ 370 struct writev_args { 371 int fd; 372 struct iovec *iovp; 373 u_int iovcnt; 374 }; 375 #endif 376 /* 377 * MPSAFE 378 */ 379 int 380 writev(struct thread *td, struct writev_args *uap) 381 { 382 struct uio *auio; 383 int error; 384 385 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 386 if (error) 387 return (error); 388 error = kern_writev(td, uap->fd, auio); 389 free(auio, M_IOV); 390 return (error); 391 } 392 393 int 394 kern_writev(struct thread *td, int fd, struct uio *auio) 395 { 396 struct file *fp; 397 int error; 398 399 error = fget_write(td, fd, &fp); 400 if (error) 401 return (EBADF); /* XXX this can't be right */ 402 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0); 403 fdrop(fp, td); 404 return (error); 405 } 406 407 /* 408 * Gather positioned write system call 409 */ 410 #ifndef _SYS_SYSPROTO_H_ 411 struct pwritev_args { 412 int fd; 413 struct iovec *iovp; 414 u_int iovcnt; 415 off_t offset; 416 }; 417 #endif 418 /* 419 * MPSAFE 420 */ 421 int 422 pwritev(struct thread *td, struct pwritev_args *uap) 423 { 424 struct uio *auio; 425 int error; 426 427 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 428 if (error) 429 return (error); 430 error = kern_pwritev(td, uap->fd, auio, uap->offset); 431 free(auio, M_IOV); 432 return (error); 433 } 434 435 int 436 kern_pwritev(td, fd, auio, offset) 437 struct thread *td; 438 struct uio *auio; 439 int fd; 440 off_t offset; 441 { 442 struct file *fp; 443 int error; 444 445 error = fget_write(td, fd, &fp); 446 if (error) 447 return (EBADF); /* XXX this can't be right */ 448 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 449 error = ESPIPE; 450 else if (offset < 0 && fp->f_vnode->v_type != VCHR) 451 error = EINVAL; 452 else 453 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET); 454 fdrop(fp, td); 455 return (error); 456 } 457 458 /* 459 * Common code for writev and pwritev that writes data to 460 * a file using the passed in uio, offset, and flags. 461 */ 462 static int 463 dofilewrite(td, fd, fp, auio, offset, flags) 464 struct thread *td; 465 int fd; 466 struct file *fp; 467 struct uio *auio; 468 off_t offset; 469 int flags; 470 { 471 ssize_t cnt; 472 int error; 473 #ifdef KTRACE 474 struct uio *ktruio = NULL; 475 #endif 476 477 auio->uio_rw = UIO_WRITE; 478 auio->uio_td = td; 479 auio->uio_offset = offset; 480 #ifdef KTRACE 481 if (KTRPOINT(td, KTR_GENIO)) 482 ktruio = cloneuio(auio); 483 #endif 484 cnt = auio->uio_resid; 485 if (fp->f_type == DTYPE_VNODE) 486 bwillwrite(); 487 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) { 488 if (auio->uio_resid != cnt && (error == ERESTART || 489 error == EINTR || error == EWOULDBLOCK)) 490 error = 0; 491 /* Socket layer is responsible for issuing SIGPIPE. */ 492 if (error == EPIPE) { 493 PROC_LOCK(td->td_proc); 494 psignal(td->td_proc, SIGPIPE); 495 PROC_UNLOCK(td->td_proc); 496 } 497 } 498 cnt -= auio->uio_resid; 499 #ifdef KTRACE 500 if (ktruio != NULL) { 501 ktruio->uio_resid = cnt; 502 ktrgenio(fd, UIO_WRITE, ktruio, error); 503 } 504 #endif 505 td->td_retval[0] = cnt; 506 return (error); 507 } 508 509 /* 510 * Ioctl system call 511 */ 512 #ifndef _SYS_SYSPROTO_H_ 513 struct ioctl_args { 514 int fd; 515 u_long com; 516 caddr_t data; 517 }; 518 #endif 519 /* 520 * MPSAFE 521 */ 522 /* ARGSUSED */ 523 int 524 ioctl(struct thread *td, struct ioctl_args *uap) 525 { 526 struct file *fp; 527 struct filedesc *fdp; 528 u_long com; 529 int error = 0; 530 u_int size; 531 caddr_t data, memp; 532 int tmp; 533 534 if (uap->com > 0xffffffff) { 535 printf( 536 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n", 537 td->td_proc->p_pid, td->td_proc->p_comm, uap->com); 538 uap->com &= 0xffffffff; 539 } 540 if ((error = fget(td, uap->fd, &fp)) != 0) 541 return (error); 542 if ((fp->f_flag & (FREAD | FWRITE)) == 0) { 543 fdrop(fp, td); 544 return (EBADF); 545 } 546 fdp = td->td_proc->p_fd; 547 switch (com = uap->com) { 548 case FIONCLEX: 549 FILEDESC_LOCK_FAST(fdp); 550 fdp->fd_ofileflags[uap->fd] &= ~UF_EXCLOSE; 551 FILEDESC_UNLOCK_FAST(fdp); 552 fdrop(fp, td); 553 return (0); 554 case FIOCLEX: 555 FILEDESC_LOCK_FAST(fdp); 556 fdp->fd_ofileflags[uap->fd] |= UF_EXCLOSE; 557 FILEDESC_UNLOCK_FAST(fdp); 558 fdrop(fp, td); 559 return (0); 560 } 561 562 /* 563 * Interpret high order word to find amount of data to be 564 * copied to/from the user's address space. 565 */ 566 size = IOCPARM_LEN(com); 567 if ((size > IOCPARM_MAX) || 568 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) || 569 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43) 570 ((com & IOC_OUT) && size == 0) || 571 #else 572 ((com & (IOC_IN | IOC_OUT)) && size == 0) || 573 #endif 574 ((com & IOC_VOID) && size > 0)) { 575 fdrop(fp, td); 576 return (ENOTTY); 577 } 578 579 if (size > 0) { 580 memp = malloc((u_long)size, M_IOCTLOPS, M_WAITOK); 581 data = memp; 582 } else { 583 memp = NULL; 584 data = (void *)&uap->data; 585 } 586 if (com & IOC_IN) { 587 error = copyin(uap->data, data, (u_int)size); 588 if (error) { 589 free(memp, M_IOCTLOPS); 590 fdrop(fp, td); 591 return (error); 592 } 593 } else if (com & IOC_OUT) { 594 /* 595 * Zero the buffer so the user always 596 * gets back something deterministic. 597 */ 598 bzero(data, size); 599 } 600 601 if (com == FIONBIO) { 602 FILE_LOCK(fp); 603 if ((tmp = *(int *)data)) 604 fp->f_flag |= FNONBLOCK; 605 else 606 fp->f_flag &= ~FNONBLOCK; 607 FILE_UNLOCK(fp); 608 data = (void *)&tmp; 609 } else if (com == FIOASYNC) { 610 FILE_LOCK(fp); 611 if ((tmp = *(int *)data)) 612 fp->f_flag |= FASYNC; 613 else 614 fp->f_flag &= ~FASYNC; 615 FILE_UNLOCK(fp); 616 data = (void *)&tmp; 617 } 618 619 error = fo_ioctl(fp, com, data, td->td_ucred, td); 620 621 if (error == 0 && (com & IOC_OUT)) 622 error = copyout(data, uap->data, (u_int)size); 623 624 if (memp != NULL) 625 free(memp, M_IOCTLOPS); 626 fdrop(fp, td); 627 return (error); 628 } 629 630 /* 631 * sellock and selwait are initialized in selectinit() via SYSINIT. 632 */ 633 struct mtx sellock; 634 struct cv selwait; 635 u_int nselcoll; /* Select collisions since boot */ 636 SYSCTL_UINT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, ""); 637 638 /* 639 * Select system call. 640 */ 641 #ifndef _SYS_SYSPROTO_H_ 642 struct select_args { 643 int nd; 644 fd_set *in, *ou, *ex; 645 struct timeval *tv; 646 }; 647 #endif 648 /* 649 * MPSAFE 650 */ 651 int 652 select(td, uap) 653 register struct thread *td; 654 register struct select_args *uap; 655 { 656 struct timeval tv, *tvp; 657 int error; 658 659 if (uap->tv != NULL) { 660 error = copyin(uap->tv, &tv, sizeof(tv)); 661 if (error) 662 return (error); 663 tvp = &tv; 664 } else 665 tvp = NULL; 666 667 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp)); 668 } 669 670 int 671 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou, 672 fd_set *fd_ex, struct timeval *tvp) 673 { 674 struct filedesc *fdp; 675 /* 676 * The magic 2048 here is chosen to be just enough for FD_SETSIZE 677 * infds with the new FD_SETSIZE of 1024, and more than enough for 678 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE 679 * of 256. 680 */ 681 fd_mask s_selbits[howmany(2048, NFDBITS)]; 682 fd_mask *ibits[3], *obits[3], *selbits, *sbp; 683 struct timeval atv, rtv, ttv; 684 int error, timo; 685 u_int ncoll, nbufbytes, ncpbytes, nfdbits; 686 687 if (nd < 0) 688 return (EINVAL); 689 fdp = td->td_proc->p_fd; 690 691 FILEDESC_LOCK_FAST(fdp); 692 693 if (nd > td->td_proc->p_fd->fd_nfiles) 694 nd = td->td_proc->p_fd->fd_nfiles; /* forgiving; slightly wrong */ 695 FILEDESC_UNLOCK_FAST(fdp); 696 697 /* 698 * Allocate just enough bits for the non-null fd_sets. Use the 699 * preallocated auto buffer if possible. 700 */ 701 nfdbits = roundup(nd, NFDBITS); 702 ncpbytes = nfdbits / NBBY; 703 nbufbytes = 0; 704 if (fd_in != NULL) 705 nbufbytes += 2 * ncpbytes; 706 if (fd_ou != NULL) 707 nbufbytes += 2 * ncpbytes; 708 if (fd_ex != NULL) 709 nbufbytes += 2 * ncpbytes; 710 if (nbufbytes <= sizeof s_selbits) 711 selbits = &s_selbits[0]; 712 else 713 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK); 714 715 /* 716 * Assign pointers into the bit buffers and fetch the input bits. 717 * Put the output buffers together so that they can be bzeroed 718 * together. 719 */ 720 sbp = selbits; 721 #define getbits(name, x) \ 722 do { \ 723 if (name == NULL) \ 724 ibits[x] = NULL; \ 725 else { \ 726 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ 727 obits[x] = sbp; \ 728 sbp += ncpbytes / sizeof *sbp; \ 729 error = copyin(name, ibits[x], ncpbytes); \ 730 if (error != 0) \ 731 goto done_nosellock; \ 732 } \ 733 } while (0) 734 getbits(fd_in, 0); 735 getbits(fd_ou, 1); 736 getbits(fd_ex, 2); 737 #undef getbits 738 if (nbufbytes != 0) 739 bzero(selbits, nbufbytes / 2); 740 741 if (tvp != NULL) { 742 atv = *tvp; 743 if (itimerfix(&atv)) { 744 error = EINVAL; 745 goto done_nosellock; 746 } 747 getmicrouptime(&rtv); 748 timevaladd(&atv, &rtv); 749 } else { 750 atv.tv_sec = 0; 751 atv.tv_usec = 0; 752 } 753 timo = 0; 754 TAILQ_INIT(&td->td_selq); 755 mtx_lock(&sellock); 756 retry: 757 ncoll = nselcoll; 758 mtx_lock_spin(&sched_lock); 759 td->td_flags |= TDF_SELECT; 760 mtx_unlock_spin(&sched_lock); 761 mtx_unlock(&sellock); 762 763 error = selscan(td, ibits, obits, nd); 764 mtx_lock(&sellock); 765 if (error || td->td_retval[0]) 766 goto done; 767 if (atv.tv_sec || atv.tv_usec) { 768 getmicrouptime(&rtv); 769 if (timevalcmp(&rtv, &atv, >=)) 770 goto done; 771 ttv = atv; 772 timevalsub(&ttv, &rtv); 773 timo = ttv.tv_sec > 24 * 60 * 60 ? 774 24 * 60 * 60 * hz : tvtohz(&ttv); 775 } 776 777 /* 778 * An event of interest may occur while we do not hold 779 * sellock, so check TDF_SELECT and the number of 780 * collisions and rescan the file descriptors if 781 * necessary. 782 */ 783 mtx_lock_spin(&sched_lock); 784 if ((td->td_flags & TDF_SELECT) == 0 || nselcoll != ncoll) { 785 mtx_unlock_spin(&sched_lock); 786 goto retry; 787 } 788 mtx_unlock_spin(&sched_lock); 789 790 if (timo > 0) 791 error = cv_timedwait_sig(&selwait, &sellock, timo); 792 else 793 error = cv_wait_sig(&selwait, &sellock); 794 795 if (error == 0) 796 goto retry; 797 798 done: 799 clear_selinfo_list(td); 800 mtx_lock_spin(&sched_lock); 801 td->td_flags &= ~TDF_SELECT; 802 mtx_unlock_spin(&sched_lock); 803 mtx_unlock(&sellock); 804 805 done_nosellock: 806 /* select is not restarted after signals... */ 807 if (error == ERESTART) 808 error = EINTR; 809 if (error == EWOULDBLOCK) 810 error = 0; 811 #define putbits(name, x) \ 812 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \ 813 error = error2; 814 if (error == 0) { 815 int error2; 816 817 putbits(fd_in, 0); 818 putbits(fd_ou, 1); 819 putbits(fd_ex, 2); 820 #undef putbits 821 } 822 if (selbits != &s_selbits[0]) 823 free(selbits, M_SELECT); 824 825 return (error); 826 } 827 828 static int 829 selscan(td, ibits, obits, nfd) 830 struct thread *td; 831 fd_mask **ibits, **obits; 832 int nfd; 833 { 834 int msk, i, fd; 835 fd_mask bits; 836 struct file *fp; 837 int n = 0; 838 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */ 839 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND }; 840 struct filedesc *fdp = td->td_proc->p_fd; 841 842 FILEDESC_LOCK(fdp); 843 for (msk = 0; msk < 3; msk++) { 844 if (ibits[msk] == NULL) 845 continue; 846 for (i = 0; i < nfd; i += NFDBITS) { 847 bits = ibits[msk][i/NFDBITS]; 848 /* ffs(int mask) not portable, fd_mask is long */ 849 for (fd = i; bits && fd < nfd; fd++, bits >>= 1) { 850 if (!(bits & 1)) 851 continue; 852 if ((fp = fget_locked(fdp, fd)) == NULL) { 853 FILEDESC_UNLOCK(fdp); 854 return (EBADF); 855 } 856 if (fo_poll(fp, flag[msk], td->td_ucred, 857 td)) { 858 obits[msk][(fd)/NFDBITS] |= 859 ((fd_mask)1 << ((fd) % NFDBITS)); 860 n++; 861 } 862 } 863 } 864 } 865 FILEDESC_UNLOCK(fdp); 866 td->td_retval[0] = n; 867 return (0); 868 } 869 870 /* 871 * Poll system call. 872 */ 873 #ifndef _SYS_SYSPROTO_H_ 874 struct poll_args { 875 struct pollfd *fds; 876 u_int nfds; 877 int timeout; 878 }; 879 #endif 880 /* 881 * MPSAFE 882 */ 883 int 884 poll(td, uap) 885 struct thread *td; 886 struct poll_args *uap; 887 { 888 struct pollfd *bits; 889 struct pollfd smallbits[32]; 890 struct timeval atv, rtv, ttv; 891 int error = 0, timo; 892 u_int ncoll, nfds; 893 size_t ni; 894 895 nfds = uap->nfds; 896 897 /* 898 * This is kinda bogus. We have fd limits, but that is not 899 * really related to the size of the pollfd array. Make sure 900 * we let the process use at least FD_SETSIZE entries and at 901 * least enough for the current limits. We want to be reasonably 902 * safe, but not overly restrictive. 903 */ 904 PROC_LOCK(td->td_proc); 905 if ((nfds > lim_cur(td->td_proc, RLIMIT_NOFILE)) && 906 (nfds > FD_SETSIZE)) { 907 PROC_UNLOCK(td->td_proc); 908 error = EINVAL; 909 goto done2; 910 } 911 PROC_UNLOCK(td->td_proc); 912 ni = nfds * sizeof(struct pollfd); 913 if (ni > sizeof(smallbits)) 914 bits = malloc(ni, M_TEMP, M_WAITOK); 915 else 916 bits = smallbits; 917 error = copyin(uap->fds, bits, ni); 918 if (error) 919 goto done_nosellock; 920 if (uap->timeout != INFTIM) { 921 atv.tv_sec = uap->timeout / 1000; 922 atv.tv_usec = (uap->timeout % 1000) * 1000; 923 if (itimerfix(&atv)) { 924 error = EINVAL; 925 goto done_nosellock; 926 } 927 getmicrouptime(&rtv); 928 timevaladd(&atv, &rtv); 929 } else { 930 atv.tv_sec = 0; 931 atv.tv_usec = 0; 932 } 933 timo = 0; 934 TAILQ_INIT(&td->td_selq); 935 mtx_lock(&sellock); 936 retry: 937 ncoll = nselcoll; 938 mtx_lock_spin(&sched_lock); 939 td->td_flags |= TDF_SELECT; 940 mtx_unlock_spin(&sched_lock); 941 mtx_unlock(&sellock); 942 943 error = pollscan(td, bits, nfds); 944 mtx_lock(&sellock); 945 if (error || td->td_retval[0]) 946 goto done; 947 if (atv.tv_sec || atv.tv_usec) { 948 getmicrouptime(&rtv); 949 if (timevalcmp(&rtv, &atv, >=)) 950 goto done; 951 ttv = atv; 952 timevalsub(&ttv, &rtv); 953 timo = ttv.tv_sec > 24 * 60 * 60 ? 954 24 * 60 * 60 * hz : tvtohz(&ttv); 955 } 956 /* 957 * An event of interest may occur while we do not hold 958 * sellock, so check TDF_SELECT and the number of collisions 959 * and rescan the file descriptors if necessary. 960 */ 961 mtx_lock_spin(&sched_lock); 962 if ((td->td_flags & TDF_SELECT) == 0 || nselcoll != ncoll) { 963 mtx_unlock_spin(&sched_lock); 964 goto retry; 965 } 966 mtx_unlock_spin(&sched_lock); 967 968 if (timo > 0) 969 error = cv_timedwait_sig(&selwait, &sellock, timo); 970 else 971 error = cv_wait_sig(&selwait, &sellock); 972 973 if (error == 0) 974 goto retry; 975 976 done: 977 clear_selinfo_list(td); 978 mtx_lock_spin(&sched_lock); 979 td->td_flags &= ~TDF_SELECT; 980 mtx_unlock_spin(&sched_lock); 981 mtx_unlock(&sellock); 982 983 done_nosellock: 984 /* poll is not restarted after signals... */ 985 if (error == ERESTART) 986 error = EINTR; 987 if (error == EWOULDBLOCK) 988 error = 0; 989 if (error == 0) { 990 error = copyout(bits, uap->fds, ni); 991 if (error) 992 goto out; 993 } 994 out: 995 if (ni > sizeof(smallbits)) 996 free(bits, M_TEMP); 997 done2: 998 return (error); 999 } 1000 1001 static int 1002 pollscan(td, fds, nfd) 1003 struct thread *td; 1004 struct pollfd *fds; 1005 u_int nfd; 1006 { 1007 register struct filedesc *fdp = td->td_proc->p_fd; 1008 int i; 1009 struct file *fp; 1010 int n = 0; 1011 1012 FILEDESC_LOCK(fdp); 1013 for (i = 0; i < nfd; i++, fds++) { 1014 if (fds->fd >= fdp->fd_nfiles) { 1015 fds->revents = POLLNVAL; 1016 n++; 1017 } else if (fds->fd < 0) { 1018 fds->revents = 0; 1019 } else { 1020 fp = fdp->fd_ofiles[fds->fd]; 1021 if (fp == NULL) { 1022 fds->revents = POLLNVAL; 1023 n++; 1024 } else { 1025 /* 1026 * Note: backend also returns POLLHUP and 1027 * POLLERR if appropriate. 1028 */ 1029 fds->revents = fo_poll(fp, fds->events, 1030 td->td_ucred, td); 1031 if (fds->revents != 0) 1032 n++; 1033 } 1034 } 1035 } 1036 FILEDESC_UNLOCK(fdp); 1037 td->td_retval[0] = n; 1038 return (0); 1039 } 1040 1041 /* 1042 * OpenBSD poll system call. 1043 * XXX this isn't quite a true representation.. OpenBSD uses select ops. 1044 */ 1045 #ifndef _SYS_SYSPROTO_H_ 1046 struct openbsd_poll_args { 1047 struct pollfd *fds; 1048 u_int nfds; 1049 int timeout; 1050 }; 1051 #endif 1052 /* 1053 * MPSAFE 1054 */ 1055 int 1056 openbsd_poll(td, uap) 1057 register struct thread *td; 1058 register struct openbsd_poll_args *uap; 1059 { 1060 return (poll(td, (struct poll_args *)uap)); 1061 } 1062 1063 /* 1064 * Remove the references to the thread from all of the objects 1065 * we were polling. 1066 * 1067 * This code assumes that the underlying owner of the selinfo 1068 * structure will hold sellock before it changes it, and that 1069 * it will unlink itself from our list if it goes away. 1070 */ 1071 void 1072 clear_selinfo_list(td) 1073 struct thread *td; 1074 { 1075 struct selinfo *si; 1076 1077 mtx_assert(&sellock, MA_OWNED); 1078 TAILQ_FOREACH(si, &td->td_selq, si_thrlist) 1079 si->si_thread = NULL; 1080 TAILQ_INIT(&td->td_selq); 1081 } 1082 1083 /* 1084 * Record a select request. 1085 */ 1086 void 1087 selrecord(selector, sip) 1088 struct thread *selector; 1089 struct selinfo *sip; 1090 { 1091 1092 mtx_lock(&sellock); 1093 /* 1094 * If the selinfo's thread pointer is NULL then take ownership of it. 1095 * 1096 * If the thread pointer is not NULL and it points to another 1097 * thread, then we have a collision. 1098 * 1099 * If the thread pointer is not NULL and points back to us then leave 1100 * it alone as we've already added pointed it at us and added it to 1101 * our list. 1102 */ 1103 if (sip->si_thread == NULL) { 1104 sip->si_thread = selector; 1105 TAILQ_INSERT_TAIL(&selector->td_selq, sip, si_thrlist); 1106 } else if (sip->si_thread != selector) { 1107 sip->si_flags |= SI_COLL; 1108 } 1109 1110 mtx_unlock(&sellock); 1111 } 1112 1113 /* Wake up a selecting thread. */ 1114 void 1115 selwakeup(sip) 1116 struct selinfo *sip; 1117 { 1118 doselwakeup(sip, -1); 1119 } 1120 1121 /* Wake up a selecting thread, and set its priority. */ 1122 void 1123 selwakeuppri(sip, pri) 1124 struct selinfo *sip; 1125 int pri; 1126 { 1127 doselwakeup(sip, pri); 1128 } 1129 1130 /* 1131 * Do a wakeup when a selectable event occurs. 1132 */ 1133 static void 1134 doselwakeup(sip, pri) 1135 struct selinfo *sip; 1136 int pri; 1137 { 1138 struct thread *td; 1139 1140 mtx_lock(&sellock); 1141 td = sip->si_thread; 1142 if ((sip->si_flags & SI_COLL) != 0) { 1143 nselcoll++; 1144 sip->si_flags &= ~SI_COLL; 1145 cv_broadcastpri(&selwait, pri); 1146 } 1147 if (td == NULL) { 1148 mtx_unlock(&sellock); 1149 return; 1150 } 1151 TAILQ_REMOVE(&td->td_selq, sip, si_thrlist); 1152 sip->si_thread = NULL; 1153 mtx_lock_spin(&sched_lock); 1154 td->td_flags &= ~TDF_SELECT; 1155 mtx_unlock_spin(&sched_lock); 1156 sleepq_remove(td, &selwait); 1157 mtx_unlock(&sellock); 1158 } 1159 1160 static void selectinit(void *); 1161 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, selectinit, NULL) 1162 1163 /* ARGSUSED*/ 1164 static void 1165 selectinit(dummy) 1166 void *dummy; 1167 { 1168 cv_init(&selwait, "select"); 1169 mtx_init(&sellock, "sellck", NULL, MTX_DEF); 1170 } 1171