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