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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94 39 * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $ 40 * $DragonFly: src/sys/kern/sys_generic.c,v 1.48 2008/04/14 12:01:50 dillon Exp $ 41 */ 42 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/filedesc.h> 49 #include <sys/filio.h> 50 #include <sys/fcntl.h> 51 #include <sys/file.h> 52 #include <sys/proc.h> 53 #include <sys/signalvar.h> 54 #include <sys/socketvar.h> 55 #include <sys/uio.h> 56 #include <sys/kernel.h> 57 #include <sys/kern_syscall.h> 58 #include <sys/malloc.h> 59 #include <sys/mapped_ioctl.h> 60 #include <sys/poll.h> 61 #include <sys/queue.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sysctl.h> 64 #include <sys/sysent.h> 65 #include <sys/buf.h> 66 #ifdef KTRACE 67 #include <sys/ktrace.h> 68 #endif 69 #include <vm/vm.h> 70 #include <vm/vm_page.h> 71 #include <sys/file2.h> 72 73 #include <machine/limits.h> 74 75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); 76 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer"); 77 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); 78 MALLOC_DEFINE(M_IOV, "iov", "large iov's"); 79 80 static int doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex, 81 struct timeval *tv, int *res); 82 static int pollscan (struct proc *, struct pollfd *, u_int, int *); 83 static int selscan (struct proc *, fd_mask **, fd_mask **, 84 int, int *); 85 static int dofileread(int, struct file *, struct uio *, int, int *); 86 static int dofilewrite(int, struct file *, struct uio *, int, int *); 87 88 /* 89 * Read system call. 90 * 91 * MPSAFE 92 */ 93 int 94 sys_read(struct read_args *uap) 95 { 96 struct thread *td = curthread; 97 struct uio auio; 98 struct iovec aiov; 99 int error; 100 101 aiov.iov_base = uap->buf; 102 aiov.iov_len = uap->nbyte; 103 auio.uio_iov = &aiov; 104 auio.uio_iovcnt = 1; 105 auio.uio_offset = -1; 106 auio.uio_resid = uap->nbyte; 107 auio.uio_rw = UIO_READ; 108 auio.uio_segflg = UIO_USERSPACE; 109 auio.uio_td = td; 110 111 if (auio.uio_resid < 0) 112 error = EINVAL; 113 else 114 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_result); 115 return(error); 116 } 117 118 /* 119 * Positioned (Pread) read system call 120 * 121 * MPSAFE 122 */ 123 int 124 sys_extpread(struct extpread_args *uap) 125 { 126 struct thread *td = curthread; 127 struct uio auio; 128 struct iovec aiov; 129 int error; 130 int flags; 131 132 aiov.iov_base = uap->buf; 133 aiov.iov_len = uap->nbyte; 134 auio.uio_iov = &aiov; 135 auio.uio_iovcnt = 1; 136 auio.uio_offset = uap->offset; 137 auio.uio_resid = uap->nbyte; 138 auio.uio_rw = UIO_READ; 139 auio.uio_segflg = UIO_USERSPACE; 140 auio.uio_td = td; 141 142 flags = uap->flags & O_FMASK; 143 if (uap->offset != (off_t)-1) 144 flags |= O_FOFFSET; 145 146 if (auio.uio_resid < 0) 147 error = EINVAL; 148 else 149 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_result); 150 return(error); 151 } 152 153 /* 154 * Scatter read system call. 155 * 156 * MPSAFE 157 */ 158 int 159 sys_readv(struct readv_args *uap) 160 { 161 struct thread *td = curthread; 162 struct uio auio; 163 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 164 int error; 165 166 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt, 167 &auio.uio_resid); 168 if (error) 169 return (error); 170 auio.uio_iov = iov; 171 auio.uio_iovcnt = uap->iovcnt; 172 auio.uio_offset = -1; 173 auio.uio_rw = UIO_READ; 174 auio.uio_segflg = UIO_USERSPACE; 175 auio.uio_td = td; 176 177 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_result); 178 179 iovec_free(&iov, aiov); 180 return (error); 181 } 182 183 184 /* 185 * Scatter positioned read system call. 186 * 187 * MPSAFE 188 */ 189 int 190 sys_extpreadv(struct extpreadv_args *uap) 191 { 192 struct thread *td = curthread; 193 struct uio auio; 194 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 195 int error; 196 int flags; 197 198 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt, 199 &auio.uio_resid); 200 if (error) 201 return (error); 202 auio.uio_iov = iov; 203 auio.uio_iovcnt = uap->iovcnt; 204 auio.uio_offset = uap->offset; 205 auio.uio_rw = UIO_READ; 206 auio.uio_segflg = UIO_USERSPACE; 207 auio.uio_td = td; 208 209 flags = uap->flags & O_FMASK; 210 if (uap->offset != (off_t)-1) 211 flags |= O_FOFFSET; 212 213 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_result); 214 215 iovec_free(&iov, aiov); 216 return(error); 217 } 218 219 /* 220 * MPSAFE 221 */ 222 int 223 kern_preadv(int fd, struct uio *auio, int flags, int *res) 224 { 225 struct thread *td = curthread; 226 struct proc *p = td->td_proc; 227 struct file *fp; 228 int error; 229 230 KKASSERT(p); 231 232 fp = holdfp(p->p_fd, fd, FREAD); 233 if (fp == NULL) 234 return (EBADF); 235 if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) { 236 error = ESPIPE; 237 } else if (auio->uio_resid < 0) { 238 error = EINVAL; 239 } else { 240 error = dofileread(fd, fp, auio, flags, res); 241 } 242 fdrop(fp); 243 return(error); 244 } 245 246 /* 247 * Common code for readv and preadv that reads data in 248 * from a file using the passed in uio, offset, and flags. 249 * 250 * MPALMOSTSAFE - ktrace needs help 251 */ 252 static int 253 dofileread(int fd, struct file *fp, struct uio *auio, int flags, int *res) 254 { 255 struct thread *td = curthread; 256 int error; 257 int len; 258 #ifdef KTRACE 259 struct iovec *ktriov = NULL; 260 struct uio ktruio; 261 #endif 262 263 #ifdef KTRACE 264 /* 265 * if tracing, save a copy of iovec 266 */ 267 if (KTRPOINT(td, KTR_GENIO)) { 268 int iovlen = auio->uio_iovcnt * sizeof(struct iovec); 269 270 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 271 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 272 ktruio = *auio; 273 } 274 #endif 275 len = auio->uio_resid; 276 error = fo_read(fp, auio, fp->f_cred, flags); 277 if (error) { 278 if (auio->uio_resid != len && (error == ERESTART || 279 error == EINTR || error == EWOULDBLOCK)) 280 error = 0; 281 } 282 #ifdef KTRACE 283 if (ktriov != NULL) { 284 if (error == 0) { 285 ktruio.uio_iov = ktriov; 286 ktruio.uio_resid = len - auio->uio_resid; 287 get_mplock(); 288 ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error); 289 rel_mplock(); 290 } 291 FREE(ktriov, M_TEMP); 292 } 293 #endif 294 if (error == 0) 295 *res = len - auio->uio_resid; 296 297 return(error); 298 } 299 300 /* 301 * Write system call 302 * 303 * MPSAFE 304 */ 305 int 306 sys_write(struct write_args *uap) 307 { 308 struct thread *td = curthread; 309 struct uio auio; 310 struct iovec aiov; 311 int error; 312 313 aiov.iov_base = (void *)(uintptr_t)uap->buf; 314 aiov.iov_len = uap->nbyte; 315 auio.uio_iov = &aiov; 316 auio.uio_iovcnt = 1; 317 auio.uio_offset = -1; 318 auio.uio_resid = uap->nbyte; 319 auio.uio_rw = UIO_WRITE; 320 auio.uio_segflg = UIO_USERSPACE; 321 auio.uio_td = td; 322 323 if (auio.uio_resid < 0) 324 error = EINVAL; 325 else 326 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_result); 327 328 return(error); 329 } 330 331 /* 332 * Pwrite system call 333 * 334 * MPSAFE 335 */ 336 int 337 sys_extpwrite(struct extpwrite_args *uap) 338 { 339 struct thread *td = curthread; 340 struct uio auio; 341 struct iovec aiov; 342 int error; 343 int flags; 344 345 aiov.iov_base = (void *)(uintptr_t)uap->buf; 346 aiov.iov_len = uap->nbyte; 347 auio.uio_iov = &aiov; 348 auio.uio_iovcnt = 1; 349 auio.uio_offset = uap->offset; 350 auio.uio_resid = uap->nbyte; 351 auio.uio_rw = UIO_WRITE; 352 auio.uio_segflg = UIO_USERSPACE; 353 auio.uio_td = td; 354 355 flags = uap->flags & O_FMASK; 356 if (uap->offset != (off_t)-1) 357 flags |= O_FOFFSET; 358 359 if (auio.uio_resid < 0) 360 error = EINVAL; 361 else 362 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_result); 363 364 return(error); 365 } 366 367 /* 368 * MPSAFE 369 */ 370 int 371 sys_writev(struct writev_args *uap) 372 { 373 struct thread *td = curthread; 374 struct uio auio; 375 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 376 int error; 377 378 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt, 379 &auio.uio_resid); 380 if (error) 381 return (error); 382 auio.uio_iov = iov; 383 auio.uio_iovcnt = uap->iovcnt; 384 auio.uio_offset = -1; 385 auio.uio_rw = UIO_WRITE; 386 auio.uio_segflg = UIO_USERSPACE; 387 auio.uio_td = td; 388 389 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_result); 390 391 iovec_free(&iov, aiov); 392 return (error); 393 } 394 395 396 /* 397 * Gather positioned write system call 398 * 399 * MPSAFE 400 */ 401 int 402 sys_extpwritev(struct extpwritev_args *uap) 403 { 404 struct thread *td = curthread; 405 struct uio auio; 406 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 407 int error; 408 int flags; 409 410 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt, 411 &auio.uio_resid); 412 if (error) 413 return (error); 414 auio.uio_iov = iov; 415 auio.uio_iovcnt = uap->iovcnt; 416 auio.uio_offset = uap->offset; 417 auio.uio_rw = UIO_WRITE; 418 auio.uio_segflg = UIO_USERSPACE; 419 auio.uio_td = td; 420 421 flags = uap->flags & O_FMASK; 422 if (uap->offset != (off_t)-1) 423 flags |= O_FOFFSET; 424 425 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_result); 426 427 iovec_free(&iov, aiov); 428 return(error); 429 } 430 431 /* 432 * MPSAFE 433 */ 434 int 435 kern_pwritev(int fd, struct uio *auio, int flags, int *res) 436 { 437 struct thread *td = curthread; 438 struct proc *p = td->td_proc; 439 struct file *fp; 440 int error; 441 442 KKASSERT(p); 443 444 fp = holdfp(p->p_fd, fd, FWRITE); 445 if (fp == NULL) 446 return (EBADF); 447 else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) { 448 error = ESPIPE; 449 } else { 450 error = dofilewrite(fd, fp, auio, flags, res); 451 } 452 453 fdrop(fp); 454 return (error); 455 } 456 457 /* 458 * Common code for writev and pwritev that writes data to 459 * a file using the passed in uio, offset, and flags. 460 * 461 * MPALMOSTSAFE - ktrace needs help 462 */ 463 static int 464 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, int *res) 465 { 466 struct thread *td = curthread; 467 struct lwp *lp = td->td_lwp; 468 int error; 469 int len; 470 #ifdef KTRACE 471 struct iovec *ktriov = NULL; 472 struct uio ktruio; 473 #endif 474 475 #ifdef KTRACE 476 /* 477 * if tracing, save a copy of iovec and uio 478 */ 479 if (KTRPOINT(td, KTR_GENIO)) { 480 int iovlen = auio->uio_iovcnt * sizeof(struct iovec); 481 482 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK); 483 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 484 ktruio = *auio; 485 } 486 #endif 487 len = auio->uio_resid; 488 if (fp->f_type == DTYPE_VNODE) 489 bwillwrite(); 490 error = fo_write(fp, auio, fp->f_cred, flags); 491 if (error) { 492 if (auio->uio_resid != len && (error == ERESTART || 493 error == EINTR || error == EWOULDBLOCK)) 494 error = 0; 495 /* Socket layer is responsible for issuing SIGPIPE. */ 496 if (error == EPIPE) { 497 get_mplock(); 498 lwpsignal(lp->lwp_proc, lp, SIGPIPE); 499 rel_mplock(); 500 } 501 } 502 #ifdef KTRACE 503 if (ktriov != NULL) { 504 if (error == 0) { 505 ktruio.uio_iov = ktriov; 506 ktruio.uio_resid = len - auio->uio_resid; 507 get_mplock(); 508 ktrgenio(lp, fd, UIO_WRITE, &ktruio, error); 509 rel_mplock(); 510 } 511 FREE(ktriov, M_TEMP); 512 } 513 #endif 514 if (error == 0) 515 *res = len - auio->uio_resid; 516 517 return(error); 518 } 519 520 /* 521 * Ioctl system call 522 */ 523 /* ARGSUSED */ 524 int 525 sys_ioctl(struct ioctl_args *uap) 526 { 527 return(mapped_ioctl(uap->fd, uap->com, uap->data, NULL)); 528 } 529 530 struct ioctl_map_entry { 531 const char *subsys; 532 struct ioctl_map_range *cmd_ranges; 533 LIST_ENTRY(ioctl_map_entry) entries; 534 }; 535 536 /* 537 * The true heart of all ioctl syscall handlers (native, emulation). 538 * If map != NULL, it will be searched for a matching entry for com, 539 * and appropriate conversions/conversion functions will be utilized. 540 */ 541 int 542 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map) 543 { 544 struct thread *td = curthread; 545 struct proc *p = td->td_proc; 546 struct ucred *cred; 547 struct file *fp; 548 struct ioctl_map_range *iomc = NULL; 549 int error; 550 u_int size; 551 u_long ocom = com; 552 caddr_t data, memp; 553 int tmp; 554 #define STK_PARAMS 128 555 union { 556 char stkbuf[STK_PARAMS]; 557 long align; 558 } ubuf; 559 560 KKASSERT(p); 561 cred = p->p_ucred; 562 563 fp = holdfp(p->p_fd, fd, FREAD|FWRITE); 564 if (fp == NULL) 565 return(EBADF); 566 567 if (map != NULL) { /* obey translation map */ 568 u_long maskcmd; 569 struct ioctl_map_entry *e; 570 571 maskcmd = com & map->mask; 572 573 LIST_FOREACH(e, &map->mapping, entries) { 574 for (iomc = e->cmd_ranges; iomc->start != 0 || 575 iomc->maptocmd != 0 || iomc->wrapfunc != NULL || 576 iomc->mapfunc != NULL; 577 iomc++) { 578 if (maskcmd >= iomc->start && 579 maskcmd <= iomc->end) 580 break; 581 } 582 583 /* Did we find a match? */ 584 if (iomc->start != 0 || iomc->maptocmd != 0 || 585 iomc->wrapfunc != NULL || iomc->mapfunc != NULL) 586 break; 587 } 588 589 if (iomc == NULL || 590 (iomc->start == 0 && iomc->maptocmd == 0 591 && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) { 592 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n", 593 map->sys, fd, maskcmd, 594 (int)((maskcmd >> 8) & 0xff), 595 (int)(maskcmd & 0xff)); 596 error = EINVAL; 597 goto done; 598 } 599 600 /* 601 * If it's a non-range one to one mapping, maptocmd should be 602 * correct. If it's a ranged one to one mapping, we pass the 603 * original value of com, and for a range mapped to a different 604 * range, we always need a mapping function to translate the 605 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff 606 */ 607 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) { 608 com = iomc->maptocmd; 609 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) { 610 if (iomc->mapfunc != NULL) 611 com = iomc->mapfunc(iomc->start, iomc->end, 612 iomc->start, iomc->end, 613 com, com); 614 } else { 615 if (iomc->mapfunc != NULL) { 616 com = iomc->mapfunc(iomc->start, iomc->end, 617 iomc->maptocmd, iomc->maptoend, 618 com, ocom); 619 } else { 620 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n", 621 map->sys, fd, maskcmd, 622 (int)((maskcmd >> 8) & 0xff), 623 (int)(maskcmd & 0xff)); 624 error = EINVAL; 625 goto done; 626 } 627 } 628 } 629 630 switch (com) { 631 case FIONCLEX: 632 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); 633 goto done; 634 case FIOCLEX: 635 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); 636 goto done; 637 } 638 639 /* 640 * Interpret high order word to find amount of data to be 641 * copied to/from the user's address space. 642 */ 643 size = IOCPARM_LEN(com); 644 if (size > IOCPARM_MAX) { 645 error = ENOTTY; 646 goto done; 647 } 648 649 memp = NULL; 650 if (size > sizeof (ubuf.stkbuf)) { 651 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK); 652 data = memp; 653 } else { 654 data = ubuf.stkbuf; 655 } 656 if ((com & IOC_IN) != 0) { 657 if (size != 0) { 658 error = copyin(uspc_data, data, (u_int)size); 659 if (error) { 660 if (memp != NULL) 661 kfree(memp, M_IOCTLOPS); 662 goto done; 663 } 664 } else { 665 *(caddr_t *)data = uspc_data; 666 } 667 } else if ((com & IOC_OUT) != 0 && size) { 668 /* 669 * Zero the buffer so the user always 670 * gets back something deterministic. 671 */ 672 bzero(data, size); 673 } else if ((com & IOC_VOID) != 0) { 674 *(caddr_t *)data = uspc_data; 675 } 676 677 switch (com) { 678 case FIONBIO: 679 if ((tmp = *(int *)data)) 680 fp->f_flag |= FNONBLOCK; 681 else 682 fp->f_flag &= ~FNONBLOCK; 683 error = 0; 684 break; 685 686 case FIOASYNC: 687 if ((tmp = *(int *)data)) 688 fp->f_flag |= FASYNC; 689 else 690 fp->f_flag &= ~FASYNC; 691 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred); 692 break; 693 694 default: 695 /* 696 * If there is a override function, 697 * call it instead of directly routing the call 698 */ 699 if (map != NULL && iomc->wrapfunc != NULL) 700 error = iomc->wrapfunc(fp, com, ocom, data, cred); 701 else 702 error = fo_ioctl(fp, com, data, cred); 703 /* 704 * Copy any data to user, size was 705 * already set and checked above. 706 */ 707 if (error == 0 && (com & IOC_OUT) != 0 && size != 0) 708 error = copyout(data, uspc_data, (u_int)size); 709 break; 710 } 711 if (memp != NULL) 712 kfree(memp, M_IOCTLOPS); 713 done: 714 fdrop(fp); 715 return(error); 716 } 717 718 int 719 mapped_ioctl_register_handler(struct ioctl_map_handler *he) 720 { 721 struct ioctl_map_entry *ne; 722 723 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL && 724 he->subsys != NULL && *he->subsys != '\0'); 725 726 ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP, M_WAITOK); 727 728 ne->subsys = he->subsys; 729 ne->cmd_ranges = he->cmd_ranges; 730 731 LIST_INSERT_HEAD(&he->map->mapping, ne, entries); 732 733 return(0); 734 } 735 736 int 737 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he) 738 { 739 struct ioctl_map_entry *ne; 740 741 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL); 742 743 LIST_FOREACH(ne, &he->map->mapping, entries) { 744 if (ne->cmd_ranges != he->cmd_ranges) 745 continue; 746 LIST_REMOVE(ne, entries); 747 kfree(ne, M_IOCTLMAP); 748 return(0); 749 } 750 return(EINVAL); 751 } 752 753 static int nselcoll; /* Select collisions since boot */ 754 int selwait; 755 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, ""); 756 757 /* 758 * Select system call. 759 */ 760 int 761 sys_select(struct select_args *uap) 762 { 763 struct timeval ktv; 764 struct timeval *ktvp; 765 int error; 766 767 /* 768 * Get timeout if any. 769 */ 770 if (uap->tv != NULL) { 771 error = copyin(uap->tv, &ktv, sizeof (ktv)); 772 if (error) 773 return (error); 774 error = itimerfix(&ktv); 775 if (error) 776 return (error); 777 ktvp = &ktv; 778 } else { 779 ktvp = NULL; 780 } 781 782 /* 783 * Do real work. 784 */ 785 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktvp, 786 &uap->sysmsg_result); 787 788 return (error); 789 } 790 791 792 /* 793 * Pselect system call. 794 */ 795 int 796 sys_pselect(struct pselect_args *uap) 797 { 798 struct thread *td = curthread; 799 struct lwp *lp = td->td_lwp; 800 struct timespec kts; 801 struct timeval ktv; 802 struct timeval *ktvp; 803 sigset_t sigmask; 804 int error; 805 806 /* 807 * Get timeout if any and convert it. 808 * Round up during conversion to avoid timeout going off early. 809 */ 810 if (uap->ts != NULL) { 811 error = copyin(uap->ts, &kts, sizeof (kts)); 812 if (error) 813 return (error); 814 ktv.tv_sec = kts.tv_sec; 815 ktv.tv_usec = (kts.tv_nsec + 999) / 1000; 816 error = itimerfix(&ktv); 817 if (error) 818 return (error); 819 ktvp = &ktv; 820 } else { 821 ktvp = NULL; 822 } 823 824 /* 825 * Install temporary signal mask if any provided. 826 */ 827 if (uap->sigmask != NULL) { 828 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask)); 829 if (error) 830 return (error); 831 lp->lwp_oldsigmask = lp->lwp_sigmask; 832 SIG_CANTMASK(sigmask); 833 lp->lwp_sigmask = sigmask; 834 } 835 836 /* 837 * Do real job. 838 */ 839 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktvp, 840 &uap->sysmsg_result); 841 842 if (uap->sigmask != NULL) { 843 /* doselect() responsible for turning ERESTART into EINTR */ 844 KKASSERT(error != ERESTART); 845 if (error == EINTR) { 846 /* 847 * We can't restore the previous signal mask now 848 * because it could block the signal that interrupted 849 * us. So make a note to restore it after executing 850 * the handler. 851 */ 852 lp->lwp_flag |= LWP_OLDMASK; 853 } else { 854 /* 855 * No handler to run. Restore previous mask immediately. 856 */ 857 lp->lwp_sigmask = lp->lwp_oldsigmask; 858 } 859 } 860 861 return (error); 862 } 863 864 /* 865 * Common code for sys_select() and sys_pselect(). 866 * 867 * in, out and ex are userland pointers. tv must point to validated 868 * kernel-side timeout value or NULL for infinite timeout. res must 869 * point to syscall return value. 870 */ 871 static int 872 doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex, struct timeval *tv, 873 int *res) 874 { 875 struct lwp *lp = curthread->td_lwp; 876 struct proc *p = curproc; 877 878 /* 879 * The magic 2048 here is chosen to be just enough for FD_SETSIZE 880 * infds with the new FD_SETSIZE of 1024, and more than enough for 881 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE 882 * of 256. 883 */ 884 fd_mask s_selbits[howmany(2048, NFDBITS)]; 885 fd_mask *ibits[3], *obits[3], *selbits, *sbp; 886 struct timeval atv, rtv, ttv; 887 int ncoll, error, timo; 888 u_int nbufbytes, ncpbytes, nfdbits; 889 890 if (nd < 0) 891 return (EINVAL); 892 if (nd > p->p_fd->fd_nfiles) 893 nd = p->p_fd->fd_nfiles; /* forgiving; slightly wrong */ 894 895 /* 896 * Allocate just enough bits for the non-null fd_sets. Use the 897 * preallocated auto buffer if possible. 898 */ 899 nfdbits = roundup(nd, NFDBITS); 900 ncpbytes = nfdbits / NBBY; 901 nbufbytes = 0; 902 if (in != NULL) 903 nbufbytes += 2 * ncpbytes; 904 if (ou != NULL) 905 nbufbytes += 2 * ncpbytes; 906 if (ex != NULL) 907 nbufbytes += 2 * ncpbytes; 908 if (nbufbytes <= sizeof s_selbits) 909 selbits = &s_selbits[0]; 910 else 911 selbits = kmalloc(nbufbytes, M_SELECT, M_WAITOK); 912 913 /* 914 * Assign pointers into the bit buffers and fetch the input bits. 915 * Put the output buffers together so that they can be bzeroed 916 * together. 917 */ 918 sbp = selbits; 919 #define getbits(name, x) \ 920 do { \ 921 if (name == NULL) \ 922 ibits[x] = NULL; \ 923 else { \ 924 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ 925 obits[x] = sbp; \ 926 sbp += ncpbytes / sizeof *sbp; \ 927 error = copyin(name, ibits[x], ncpbytes); \ 928 if (error != 0) \ 929 goto done; \ 930 } \ 931 } while (0) 932 getbits(in, 0); 933 getbits(ou, 1); 934 getbits(ex, 2); 935 #undef getbits 936 if (nbufbytes != 0) 937 bzero(selbits, nbufbytes / 2); 938 939 if (tv != NULL) { 940 atv = *tv; 941 getmicrouptime(&rtv); 942 timevaladd(&atv, &rtv); 943 } else { 944 atv.tv_sec = 0; 945 atv.tv_usec = 0; 946 } 947 timo = 0; 948 retry: 949 ncoll = nselcoll; 950 lp->lwp_flag |= LWP_SELECT; 951 error = selscan(p, ibits, obits, nd, res); 952 if (error || *res) 953 goto done; 954 if (atv.tv_sec || atv.tv_usec) { 955 getmicrouptime(&rtv); 956 if (timevalcmp(&rtv, &atv, >=)) 957 goto done; 958 ttv = atv; 959 timevalsub(&ttv, &rtv); 960 timo = ttv.tv_sec > 24 * 60 * 60 ? 961 24 * 60 * 60 * hz : tvtohz_high(&ttv); 962 } 963 crit_enter(); 964 if ((lp->lwp_flag & LWP_SELECT) == 0 || nselcoll != ncoll) { 965 crit_exit(); 966 goto retry; 967 } 968 lp->lwp_flag &= ~LWP_SELECT; 969 970 error = tsleep((caddr_t)&selwait, PCATCH, "select", timo); 971 972 crit_exit(); 973 if (error == 0) 974 goto retry; 975 done: 976 lp->lwp_flag &= ~LWP_SELECT; 977 /* select is not restarted after signals... */ 978 if (error == ERESTART) 979 error = EINTR; 980 if (error == EWOULDBLOCK) 981 error = 0; 982 #define putbits(name, x) \ 983 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \ 984 error = error2; 985 if (error == 0) { 986 int error2; 987 988 putbits(in, 0); 989 putbits(ou, 1); 990 putbits(ex, 2); 991 #undef putbits 992 } 993 if (selbits != &s_selbits[0]) 994 kfree(selbits, M_SELECT); 995 return (error); 996 } 997 998 static int 999 selscan(struct proc *p, fd_mask **ibits, fd_mask **obits, int nfd, int *res) 1000 { 1001 int msk, i, fd; 1002 fd_mask bits; 1003 struct file *fp; 1004 int n = 0; 1005 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */ 1006 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND }; 1007 1008 for (msk = 0; msk < 3; msk++) { 1009 if (ibits[msk] == NULL) 1010 continue; 1011 for (i = 0; i < nfd; i += NFDBITS) { 1012 bits = ibits[msk][i/NFDBITS]; 1013 /* ffs(int mask) not portable, fd_mask is long */ 1014 for (fd = i; bits && fd < nfd; fd++, bits >>= 1) { 1015 if (!(bits & 1)) 1016 continue; 1017 fp = holdfp(p->p_fd, fd, -1); 1018 if (fp == NULL) 1019 return (EBADF); 1020 if (fo_poll(fp, flag[msk], fp->f_cred)) { 1021 obits[msk][(fd)/NFDBITS] |= 1022 ((fd_mask)1 << ((fd) % NFDBITS)); 1023 n++; 1024 } 1025 fdrop(fp); 1026 } 1027 } 1028 } 1029 *res = n; 1030 return (0); 1031 } 1032 1033 /* 1034 * Poll system call. 1035 */ 1036 int 1037 sys_poll(struct poll_args *uap) 1038 { 1039 struct pollfd *bits; 1040 struct pollfd smallbits[32]; 1041 struct timeval atv, rtv, ttv; 1042 int ncoll, error = 0, timo; 1043 u_int nfds; 1044 size_t ni; 1045 struct lwp *lp = curthread->td_lwp; 1046 struct proc *p = curproc; 1047 1048 nfds = uap->nfds; 1049 /* 1050 * This is kinda bogus. We have fd limits, but that is not 1051 * really related to the size of the pollfd array. Make sure 1052 * we let the process use at least FD_SETSIZE entries and at 1053 * least enough for the current limits. We want to be reasonably 1054 * safe, but not overly restrictive. 1055 */ 1056 if (nfds > p->p_rlimit[RLIMIT_NOFILE].rlim_cur && nfds > FD_SETSIZE) 1057 return (EINVAL); 1058 ni = nfds * sizeof(struct pollfd); 1059 if (ni > sizeof(smallbits)) 1060 bits = kmalloc(ni, M_TEMP, M_WAITOK); 1061 else 1062 bits = smallbits; 1063 error = copyin(uap->fds, bits, ni); 1064 if (error) 1065 goto done; 1066 if (uap->timeout != INFTIM) { 1067 atv.tv_sec = uap->timeout / 1000; 1068 atv.tv_usec = (uap->timeout % 1000) * 1000; 1069 if (itimerfix(&atv)) { 1070 error = EINVAL; 1071 goto done; 1072 } 1073 getmicrouptime(&rtv); 1074 timevaladd(&atv, &rtv); 1075 } else { 1076 atv.tv_sec = 0; 1077 atv.tv_usec = 0; 1078 } 1079 timo = 0; 1080 retry: 1081 ncoll = nselcoll; 1082 lp->lwp_flag |= LWP_SELECT; 1083 error = pollscan(p, bits, nfds, &uap->sysmsg_result); 1084 if (error || uap->sysmsg_result) 1085 goto done; 1086 if (atv.tv_sec || atv.tv_usec) { 1087 getmicrouptime(&rtv); 1088 if (timevalcmp(&rtv, &atv, >=)) 1089 goto done; 1090 ttv = atv; 1091 timevalsub(&ttv, &rtv); 1092 timo = ttv.tv_sec > 24 * 60 * 60 ? 1093 24 * 60 * 60 * hz : tvtohz_high(&ttv); 1094 } 1095 crit_enter(); 1096 if ((lp->lwp_flag & LWP_SELECT) == 0 || nselcoll != ncoll) { 1097 crit_exit(); 1098 goto retry; 1099 } 1100 lp->lwp_flag &= ~LWP_SELECT; 1101 error = tsleep((caddr_t)&selwait, PCATCH, "poll", timo); 1102 crit_exit(); 1103 if (error == 0) 1104 goto retry; 1105 done: 1106 lp->lwp_flag &= ~LWP_SELECT; 1107 /* poll is not restarted after signals... */ 1108 if (error == ERESTART) 1109 error = EINTR; 1110 if (error == EWOULDBLOCK) 1111 error = 0; 1112 if (error == 0) { 1113 error = copyout(bits, uap->fds, ni); 1114 if (error) 1115 goto out; 1116 } 1117 out: 1118 if (ni > sizeof(smallbits)) 1119 kfree(bits, M_TEMP); 1120 return (error); 1121 } 1122 1123 static int 1124 pollscan(struct proc *p, struct pollfd *fds, u_int nfd, int *res) 1125 { 1126 int i; 1127 struct file *fp; 1128 int n = 0; 1129 1130 for (i = 0; i < nfd; i++, fds++) { 1131 if (fds->fd >= p->p_fd->fd_nfiles) { 1132 fds->revents = POLLNVAL; 1133 n++; 1134 } else if (fds->fd < 0) { 1135 fds->revents = 0; 1136 } else { 1137 fp = holdfp(p->p_fd, fds->fd, -1); 1138 if (fp == NULL) { 1139 fds->revents = POLLNVAL; 1140 n++; 1141 } else { 1142 /* 1143 * Note: backend also returns POLLHUP and 1144 * POLLERR if appropriate. 1145 */ 1146 fds->revents = fo_poll(fp, fds->events, 1147 fp->f_cred); 1148 if (fds->revents != 0) 1149 n++; 1150 fdrop(fp); 1151 } 1152 } 1153 } 1154 *res = n; 1155 return (0); 1156 } 1157 1158 /* 1159 * OpenBSD poll system call. 1160 * XXX this isn't quite a true representation.. OpenBSD uses select ops. 1161 */ 1162 int 1163 sys_openbsd_poll(struct openbsd_poll_args *uap) 1164 { 1165 return (sys_poll((struct poll_args *)uap)); 1166 } 1167 1168 /*ARGSUSED*/ 1169 int 1170 seltrue(cdev_t dev, int events) 1171 { 1172 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 1173 } 1174 1175 /* 1176 * Record a select request. A global wait must be used since a process/thread 1177 * might go away after recording its request. 1178 */ 1179 void 1180 selrecord(struct thread *selector, struct selinfo *sip) 1181 { 1182 struct proc *p; 1183 struct lwp *lp = NULL; 1184 1185 if (selector->td_lwp == NULL) 1186 panic("selrecord: thread needs a process"); 1187 1188 if (sip->si_pid == selector->td_proc->p_pid && 1189 sip->si_tid == selector->td_lwp->lwp_tid) 1190 return; 1191 if (sip->si_pid && (p = pfind(sip->si_pid))) 1192 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, sip->si_tid); 1193 if (lp != NULL && lp->lwp_wchan == (caddr_t)&selwait) { 1194 sip->si_flags |= SI_COLL; 1195 } else { 1196 sip->si_pid = selector->td_proc->p_pid; 1197 sip->si_tid = selector->td_lwp->lwp_tid; 1198 } 1199 } 1200 1201 /* 1202 * Do a wakeup when a selectable event occurs. 1203 */ 1204 void 1205 selwakeup(struct selinfo *sip) 1206 { 1207 struct proc *p; 1208 struct lwp *lp = NULL; 1209 1210 if (sip->si_pid == 0) 1211 return; 1212 if (sip->si_flags & SI_COLL) { 1213 nselcoll++; 1214 sip->si_flags &= ~SI_COLL; 1215 wakeup((caddr_t)&selwait); /* YYY fixable */ 1216 } 1217 p = pfind(sip->si_pid); 1218 sip->si_pid = 0; 1219 if (p == NULL) 1220 return; 1221 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, sip->si_tid); 1222 if (lp == NULL) 1223 return; 1224 1225 crit_enter(); 1226 if (lp->lwp_wchan == (caddr_t)&selwait) { 1227 /* 1228 * Flag the process to break the tsleep when 1229 * setrunnable is called, but only call setrunnable 1230 * here if the process is not in a stopped state. 1231 */ 1232 lp->lwp_flag |= LWP_BREAKTSLEEP; 1233 if (p->p_stat != SSTOP) 1234 setrunnable(lp); 1235 } else if (lp->lwp_flag & LWP_SELECT) { 1236 lp->lwp_flag &= ~LWP_SELECT; 1237 } 1238 crit_exit(); 1239 } 1240 1241