1 /* $OpenBSD: sys_generic.c,v 1.158 2024/08/12 19:32:05 anton Exp $ */ 2 /* $NetBSD: sys_generic.c,v 1.24 1996/03/29 00:25:32 cgd Exp $ */ 3 4 /* 5 * Copyright (c) 1996 Theo de Raadt 6 * Copyright (c) 1982, 1986, 1989, 1993 7 * The Regents of the University of California. All rights reserved. 8 * (c) UNIX System Laboratories, Inc. 9 * All or some portions of this file are derived from material licensed 10 * to the University of California by American Telephone and Telegraph 11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 12 * the permission of UNIX System Laboratories, Inc. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. 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 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/filedesc.h> 44 #include <sys/ioctl.h> 45 #include <sys/fcntl.h> 46 #include <sys/vnode.h> 47 #include <sys/file.h> 48 #include <sys/proc.h> 49 #include <sys/resourcevar.h> 50 #include <sys/socketvar.h> 51 #include <sys/signalvar.h> 52 #include <sys/uio.h> 53 #include <sys/time.h> 54 #include <sys/malloc.h> 55 #include <sys/poll.h> 56 #include <sys/eventvar.h> 57 #ifdef KTRACE 58 #include <sys/ktrace.h> 59 #endif 60 #include <sys/pledge.h> 61 62 #include <sys/mount.h> 63 #include <sys/syscallargs.h> 64 65 /* 66 * Debug values: 67 * 1 - print implementation errors, things that should not happen. 68 * 2 - print ppoll(2) information, somewhat verbose 69 * 3 - print pselect(2) and ppoll(2) information, very verbose 70 */ 71 /* #define KQPOLL_DEBUG */ 72 #ifdef KQPOLL_DEBUG 73 int kqpoll_debug = 1; 74 #define DPRINTFN(v, x...) if (kqpoll_debug > v) { \ 75 printf("%s(%d): ", curproc->p_p->ps_comm, curproc->p_tid); \ 76 printf(x); \ 77 } 78 #else 79 #define DPRINTFN(v, x...) do {} while (0); 80 #endif 81 82 int pselregister(struct proc *, fd_set **, fd_set **, int, int *, int *); 83 int pselcollect(struct proc *, struct kevent *, fd_set **, int *); 84 void ppollregister(struct proc *, struct pollfd *, int, int *, int *); 85 int ppollcollect(struct proc *, struct kevent *, struct pollfd *, u_int); 86 87 int pollout(struct pollfd *, struct pollfd *, u_int); 88 int dopselect(struct proc *, int, fd_set *, fd_set *, fd_set *, 89 struct timespec *, const sigset_t *, register_t *); 90 int doppoll(struct proc *, struct pollfd *, u_int, struct timespec *, 91 const sigset_t *, register_t *); 92 93 int 94 iovec_copyin(const struct iovec *uiov, struct iovec **iovp, struct iovec *aiov, 95 unsigned int iovcnt, size_t *residp) 96 { 97 #ifdef KTRACE 98 struct proc *p = curproc; 99 #endif 100 struct iovec *iov; 101 int error, i; 102 size_t resid = 0; 103 104 if (iovcnt > UIO_SMALLIOV) { 105 if (iovcnt > IOV_MAX) 106 return (EINVAL); 107 iov = mallocarray(iovcnt, sizeof(*iov), M_IOV, M_WAITOK); 108 } else if (iovcnt > 0) { 109 iov = aiov; 110 } else { 111 return (EINVAL); 112 } 113 *iovp = iov; 114 115 if ((error = copyin(uiov, iov, iovcnt * sizeof(*iov)))) 116 return (error); 117 118 #ifdef KTRACE 119 if (KTRPOINT(p, KTR_STRUCT)) 120 ktriovec(p, iov, iovcnt); 121 #endif 122 123 for (i = 0; i < iovcnt; i++) { 124 resid += iov->iov_len; 125 /* 126 * Writes return ssize_t because -1 is returned on error. 127 * Therefore we must restrict the length to SSIZE_MAX to 128 * avoid garbage return values. Note that the addition is 129 * guaranteed to not wrap because SSIZE_MAX * 2 < SIZE_MAX. 130 */ 131 if (iov->iov_len > SSIZE_MAX || resid > SSIZE_MAX) 132 return (EINVAL); 133 iov++; 134 } 135 136 if (residp != NULL) 137 *residp = resid; 138 139 return (0); 140 } 141 142 void 143 iovec_free(struct iovec *iov, unsigned int iovcnt) 144 { 145 if (iovcnt > UIO_SMALLIOV) 146 free(iov, M_IOV, iovcnt * sizeof(*iov)); 147 } 148 149 /* 150 * Read system call. 151 */ 152 int 153 sys_read(struct proc *p, void *v, register_t *retval) 154 { 155 struct sys_read_args /* { 156 syscallarg(int) fd; 157 syscallarg(void *) buf; 158 syscallarg(size_t) nbyte; 159 } */ *uap = v; 160 struct iovec iov; 161 struct uio auio; 162 163 iov.iov_base = SCARG(uap, buf); 164 iov.iov_len = SCARG(uap, nbyte); 165 if (iov.iov_len > SSIZE_MAX) 166 return (EINVAL); 167 168 auio.uio_iov = &iov; 169 auio.uio_iovcnt = 1; 170 auio.uio_resid = iov.iov_len; 171 172 return (dofilereadv(p, SCARG(uap, fd), &auio, 0, retval)); 173 } 174 175 /* 176 * Scatter read system call. 177 */ 178 int 179 sys_readv(struct proc *p, void *v, register_t *retval) 180 { 181 struct sys_readv_args /* { 182 syscallarg(int) fd; 183 syscallarg(const struct iovec *) iovp; 184 syscallarg(int) iovcnt; 185 } */ *uap = v; 186 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 187 int error, iovcnt = SCARG(uap, iovcnt); 188 struct uio auio; 189 size_t resid; 190 191 error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid); 192 if (error) 193 goto done; 194 195 auio.uio_iov = iov; 196 auio.uio_iovcnt = iovcnt; 197 auio.uio_resid = resid; 198 199 error = dofilereadv(p, SCARG(uap, fd), &auio, 0, retval); 200 done: 201 iovec_free(iov, iovcnt); 202 return (error); 203 } 204 205 int 206 dofilereadv(struct proc *p, int fd, struct uio *uio, int flags, 207 register_t *retval) 208 { 209 struct filedesc *fdp = p->p_fd; 210 struct file *fp; 211 long cnt, error = 0; 212 u_int iovlen; 213 #ifdef KTRACE 214 struct iovec *ktriov = NULL; 215 #endif 216 217 KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0); 218 iovlen = uio->uio_iovcnt * sizeof(struct iovec); 219 220 if ((fp = fd_getfile_mode(fdp, fd, FREAD)) == NULL) 221 return (EBADF); 222 223 /* Checks for positioned read. */ 224 if (flags & FO_POSITION) { 225 struct vnode *vp = fp->f_data; 226 227 if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO || 228 (vp->v_flag & VISTTY)) { 229 error = ESPIPE; 230 goto done; 231 } 232 233 if (uio->uio_offset < 0 && vp->v_type != VCHR) { 234 error = EINVAL; 235 goto done; 236 } 237 } 238 239 uio->uio_rw = UIO_READ; 240 uio->uio_segflg = UIO_USERSPACE; 241 uio->uio_procp = p; 242 #ifdef KTRACE 243 /* 244 * if tracing, save a copy of iovec 245 */ 246 if (KTRPOINT(p, KTR_GENIO)) { 247 ktriov = malloc(iovlen, M_TEMP, M_WAITOK); 248 memcpy(ktriov, uio->uio_iov, iovlen); 249 } 250 #endif 251 cnt = uio->uio_resid; 252 error = (*fp->f_ops->fo_read)(fp, uio, flags); 253 if (error) { 254 if (uio->uio_resid != cnt && (error == ERESTART || 255 error == EINTR || error == EWOULDBLOCK)) 256 error = 0; 257 } 258 cnt -= uio->uio_resid; 259 260 mtx_enter(&fp->f_mtx); 261 fp->f_rxfer++; 262 fp->f_rbytes += cnt; 263 mtx_leave(&fp->f_mtx); 264 #ifdef KTRACE 265 if (ktriov != NULL) { 266 if (error == 0) 267 ktrgenio(p, fd, UIO_READ, ktriov, cnt); 268 free(ktriov, M_TEMP, iovlen); 269 } 270 #endif 271 *retval = cnt; 272 done: 273 FRELE(fp, p); 274 return (error); 275 } 276 277 /* 278 * Write system call 279 */ 280 int 281 sys_write(struct proc *p, void *v, register_t *retval) 282 { 283 struct sys_write_args /* { 284 syscallarg(int) fd; 285 syscallarg(const void *) buf; 286 syscallarg(size_t) nbyte; 287 } */ *uap = v; 288 struct iovec iov; 289 struct uio auio; 290 291 iov.iov_base = (void *)SCARG(uap, buf); 292 iov.iov_len = SCARG(uap, nbyte); 293 if (iov.iov_len > SSIZE_MAX) 294 return (EINVAL); 295 296 auio.uio_iov = &iov; 297 auio.uio_iovcnt = 1; 298 auio.uio_resid = iov.iov_len; 299 300 return (dofilewritev(p, SCARG(uap, fd), &auio, 0, retval)); 301 } 302 303 /* 304 * Gather write system call 305 */ 306 int 307 sys_writev(struct proc *p, void *v, register_t *retval) 308 { 309 struct sys_writev_args /* { 310 syscallarg(int) fd; 311 syscallarg(const struct iovec *) iovp; 312 syscallarg(int) iovcnt; 313 } */ *uap = v; 314 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 315 int error, iovcnt = SCARG(uap, iovcnt); 316 struct uio auio; 317 size_t resid; 318 319 error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid); 320 if (error) 321 goto done; 322 323 auio.uio_iov = iov; 324 auio.uio_iovcnt = iovcnt; 325 auio.uio_resid = resid; 326 327 error = dofilewritev(p, SCARG(uap, fd), &auio, 0, retval); 328 done: 329 iovec_free(iov, iovcnt); 330 return (error); 331 } 332 333 int 334 dofilewritev(struct proc *p, int fd, struct uio *uio, int flags, 335 register_t *retval) 336 { 337 struct filedesc *fdp = p->p_fd; 338 struct file *fp; 339 long cnt, error = 0; 340 u_int iovlen; 341 #ifdef KTRACE 342 struct iovec *ktriov = NULL; 343 #endif 344 345 KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0); 346 iovlen = uio->uio_iovcnt * sizeof(struct iovec); 347 348 if ((fp = fd_getfile_mode(fdp, fd, FWRITE)) == NULL) 349 return (EBADF); 350 351 /* Checks for positioned write. */ 352 if (flags & FO_POSITION) { 353 struct vnode *vp = fp->f_data; 354 355 if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO || 356 (vp->v_flag & VISTTY)) { 357 error = ESPIPE; 358 goto done; 359 } 360 361 if (uio->uio_offset < 0 && vp->v_type != VCHR) { 362 error = EINVAL; 363 goto done; 364 } 365 } 366 367 uio->uio_rw = UIO_WRITE; 368 uio->uio_segflg = UIO_USERSPACE; 369 uio->uio_procp = p; 370 #ifdef KTRACE 371 /* 372 * if tracing, save a copy of iovec 373 */ 374 if (KTRPOINT(p, KTR_GENIO)) { 375 ktriov = malloc(iovlen, M_TEMP, M_WAITOK); 376 memcpy(ktriov, uio->uio_iov, iovlen); 377 } 378 #endif 379 cnt = uio->uio_resid; 380 error = (*fp->f_ops->fo_write)(fp, uio, flags); 381 if (error) { 382 if (uio->uio_resid != cnt && (error == ERESTART || 383 error == EINTR || error == EWOULDBLOCK)) 384 error = 0; 385 if (error == EPIPE) { 386 KERNEL_LOCK(); 387 ptsignal(p, SIGPIPE, STHREAD); 388 KERNEL_UNLOCK(); 389 } 390 } 391 cnt -= uio->uio_resid; 392 393 mtx_enter(&fp->f_mtx); 394 fp->f_wxfer++; 395 fp->f_wbytes += cnt; 396 mtx_leave(&fp->f_mtx); 397 #ifdef KTRACE 398 if (ktriov != NULL) { 399 if (error == 0) 400 ktrgenio(p, fd, UIO_WRITE, ktriov, cnt); 401 free(ktriov, M_TEMP, iovlen); 402 } 403 #endif 404 *retval = cnt; 405 done: 406 FRELE(fp, p); 407 return (error); 408 } 409 410 /* 411 * Ioctl system call 412 */ 413 int 414 sys_ioctl(struct proc *p, void *v, register_t *retval) 415 { 416 struct sys_ioctl_args /* { 417 syscallarg(int) fd; 418 syscallarg(u_long) com; 419 syscallarg(void *) data; 420 } */ *uap = v; 421 struct file *fp; 422 struct filedesc *fdp = p->p_fd; 423 u_long com = SCARG(uap, com); 424 int error = 0; 425 u_int size = 0; 426 caddr_t data, memp = NULL; 427 int tmp; 428 #define STK_PARAMS 128 429 long long stkbuf[STK_PARAMS / sizeof(long long)]; 430 431 if ((fp = fd_getfile_mode(fdp, SCARG(uap, fd), FREAD|FWRITE)) == NULL) 432 return (EBADF); 433 434 if (fp->f_type == DTYPE_SOCKET) { 435 struct socket *so = fp->f_data; 436 437 if (so->so_state & SS_DNS) { 438 error = EINVAL; 439 goto out; 440 } 441 } 442 443 error = pledge_ioctl(p, com, fp); 444 if (error) 445 goto out; 446 447 switch (com) { 448 case FIONCLEX: 449 case FIOCLEX: 450 fdplock(fdp); 451 if (com == FIONCLEX) 452 fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE; 453 else 454 fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE; 455 fdpunlock(fdp); 456 goto out; 457 } 458 459 /* 460 * Interpret high order word to find amount of data to be 461 * copied to/from the user's address space. 462 */ 463 size = IOCPARM_LEN(com); 464 if (size > IOCPARM_MAX) { 465 error = ENOTTY; 466 goto out; 467 } 468 if (size > sizeof (stkbuf)) { 469 memp = malloc(size, M_IOCTLOPS, M_WAITOK); 470 data = memp; 471 } else 472 data = (caddr_t)stkbuf; 473 if (com&IOC_IN) { 474 if (size) { 475 error = copyin(SCARG(uap, data), data, size); 476 if (error) { 477 goto out; 478 } 479 } else 480 *(caddr_t *)data = SCARG(uap, data); 481 } else if ((com&IOC_OUT) && size) 482 /* 483 * Zero the buffer so the user always 484 * gets back something deterministic. 485 */ 486 memset(data, 0, size); 487 else if (com&IOC_VOID) 488 *(caddr_t *)data = SCARG(uap, data); 489 490 switch (com) { 491 492 case FIONBIO: 493 if ((tmp = *(int *)data) != 0) 494 atomic_setbits_int(&fp->f_flag, FNONBLOCK); 495 else 496 atomic_clearbits_int(&fp->f_flag, FNONBLOCK); 497 error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p); 498 break; 499 500 case FIOASYNC: 501 if ((tmp = *(int *)data) != 0) 502 atomic_setbits_int(&fp->f_flag, FASYNC); 503 else 504 atomic_clearbits_int(&fp->f_flag, FASYNC); 505 error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p); 506 break; 507 508 default: 509 error = (*fp->f_ops->fo_ioctl)(fp, com, data, p); 510 break; 511 } 512 /* 513 * Copy any data to user, size was 514 * already set and checked above. 515 */ 516 if (error == 0 && (com&IOC_OUT) && size) 517 error = copyout(data, SCARG(uap, data), size); 518 out: 519 FRELE(fp, p); 520 free(memp, M_IOCTLOPS, size); 521 return (error); 522 } 523 524 /* 525 * Select system call. 526 */ 527 int 528 sys_select(struct proc *p, void *v, register_t *retval) 529 { 530 struct sys_select_args /* { 531 syscallarg(int) nd; 532 syscallarg(fd_set *) in; 533 syscallarg(fd_set *) ou; 534 syscallarg(fd_set *) ex; 535 syscallarg(struct timeval *) tv; 536 } */ *uap = v; 537 538 struct timespec ts, *tsp = NULL; 539 int error; 540 541 if (SCARG(uap, tv) != NULL) { 542 struct timeval tv; 543 if ((error = copyin(SCARG(uap, tv), &tv, sizeof tv)) != 0) 544 return (error); 545 #ifdef KTRACE 546 if (KTRPOINT(p, KTR_STRUCT)) 547 ktrreltimeval(p, &tv); 548 #endif 549 if (tv.tv_sec < 0 || !timerisvalid(&tv)) 550 return (EINVAL); 551 TIMEVAL_TO_TIMESPEC(&tv, &ts); 552 tsp = &ts; 553 } 554 555 return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou), 556 SCARG(uap, ex), tsp, NULL, retval)); 557 } 558 559 int 560 sys_pselect(struct proc *p, void *v, register_t *retval) 561 { 562 struct sys_pselect_args /* { 563 syscallarg(int) nd; 564 syscallarg(fd_set *) in; 565 syscallarg(fd_set *) ou; 566 syscallarg(fd_set *) ex; 567 syscallarg(const struct timespec *) ts; 568 syscallarg(const sigset_t *) mask; 569 } */ *uap = v; 570 571 struct timespec ts, *tsp = NULL; 572 sigset_t ss, *ssp = NULL; 573 int error; 574 575 if (SCARG(uap, ts) != NULL) { 576 if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0) 577 return (error); 578 #ifdef KTRACE 579 if (KTRPOINT(p, KTR_STRUCT)) 580 ktrreltimespec(p, &ts); 581 #endif 582 if (ts.tv_sec < 0 || !timespecisvalid(&ts)) 583 return (EINVAL); 584 tsp = &ts; 585 } 586 if (SCARG(uap, mask) != NULL) { 587 if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0) 588 return (error); 589 ssp = &ss; 590 } 591 592 return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou), 593 SCARG(uap, ex), tsp, ssp, retval)); 594 } 595 596 int 597 dopselect(struct proc *p, int nd, fd_set *in, fd_set *ou, fd_set *ex, 598 struct timespec *timeout, const sigset_t *sigmask, register_t *retval) 599 { 600 struct kqueue_scan_state scan; 601 struct timespec zerots = {}; 602 fd_mask bits[6]; 603 fd_set *pibits[3], *pobits[3]; 604 int error, nfiles, ncollected = 0, nevents = 0; 605 u_int ni; 606 607 if (nd < 0) 608 return (EINVAL); 609 610 nfiles = READ_ONCE(p->p_fd->fd_nfiles); 611 if (nd > nfiles) 612 nd = nfiles; 613 614 ni = howmany(nd, NFDBITS) * sizeof(fd_mask); 615 if (ni > sizeof(bits[0])) { 616 caddr_t mbits; 617 618 mbits = mallocarray(6, ni, M_TEMP, M_WAITOK|M_ZERO); 619 pibits[0] = (fd_set *)&mbits[ni * 0]; 620 pibits[1] = (fd_set *)&mbits[ni * 1]; 621 pibits[2] = (fd_set *)&mbits[ni * 2]; 622 pobits[0] = (fd_set *)&mbits[ni * 3]; 623 pobits[1] = (fd_set *)&mbits[ni * 4]; 624 pobits[2] = (fd_set *)&mbits[ni * 5]; 625 } else { 626 memset(bits, 0, sizeof(bits)); 627 pibits[0] = (fd_set *)&bits[0]; 628 pibits[1] = (fd_set *)&bits[1]; 629 pibits[2] = (fd_set *)&bits[2]; 630 pobits[0] = (fd_set *)&bits[3]; 631 pobits[1] = (fd_set *)&bits[4]; 632 pobits[2] = (fd_set *)&bits[5]; 633 } 634 635 kqpoll_init(nd); 636 637 #define getbits(name, x) \ 638 if (name && (error = copyin(name, pibits[x], ni))) \ 639 goto done; 640 getbits(in, 0); 641 getbits(ou, 1); 642 getbits(ex, 2); 643 #undef getbits 644 #ifdef KTRACE 645 if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) { 646 if (in) ktrfdset(p, pibits[0], ni); 647 if (ou) ktrfdset(p, pibits[1], ni); 648 if (ex) ktrfdset(p, pibits[2], ni); 649 } 650 #endif 651 652 if (sigmask) 653 dosigsuspend(p, *sigmask &~ sigcantmask); 654 655 /* Register kqueue events */ 656 error = pselregister(p, pibits, pobits, nd, &nevents, &ncollected); 657 if (error != 0) 658 goto done; 659 660 /* 661 * The poll/select family of syscalls has been designed to 662 * block when file descriptors are not available, even if 663 * there's nothing to wait for. 664 */ 665 if (nevents == 0 && ncollected == 0) { 666 uint64_t nsecs = INFSLP; 667 668 if (timeout != NULL) { 669 if (!timespecisset(timeout)) 670 goto done; 671 nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP)); 672 } 673 error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqsel", nsecs); 674 /* select is not restarted after signals... */ 675 if (error == ERESTART) 676 error = EINTR; 677 if (error == EWOULDBLOCK) 678 error = 0; 679 goto done; 680 } 681 682 /* Do not block if registering found pending events. */ 683 if (ncollected > 0) 684 timeout = &zerots; 685 686 /* Collect at most `nevents' possibly waiting in kqueue_scan() */ 687 kqueue_scan_setup(&scan, p->p_kq); 688 while (nevents > 0) { 689 struct kevent kev[KQ_NEVENTS]; 690 int i, ready, count; 691 692 /* Maximum number of events per iteration */ 693 count = MIN(nitems(kev), nevents); 694 ready = kqueue_scan(&scan, count, kev, timeout, p, &error); 695 696 /* Convert back events that are ready. */ 697 for (i = 0; i < ready && error == 0; i++) 698 error = pselcollect(p, &kev[i], pobits, &ncollected); 699 /* 700 * Stop if there was an error or if we had enough 701 * space to collect all events that were ready. 702 */ 703 if (error || ready < count) 704 break; 705 706 nevents -= ready; 707 } 708 kqueue_scan_finish(&scan); 709 *retval = ncollected; 710 done: 711 #define putbits(name, x) \ 712 if (name && (error2 = copyout(pobits[x], name, ni))) \ 713 error = error2; 714 if (error == 0) { 715 int error2; 716 717 putbits(in, 0); 718 putbits(ou, 1); 719 putbits(ex, 2); 720 #undef putbits 721 #ifdef KTRACE 722 if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) { 723 if (in) ktrfdset(p, pobits[0], ni); 724 if (ou) ktrfdset(p, pobits[1], ni); 725 if (ex) ktrfdset(p, pobits[2], ni); 726 } 727 #endif 728 } 729 730 if (pibits[0] != (fd_set *)&bits[0]) 731 free(pibits[0], M_TEMP, 6 * ni); 732 733 kqpoll_done(nd); 734 735 return (error); 736 } 737 738 /* 739 * Convert fd_set into kqueue events and register them on the 740 * per-thread queue. 741 */ 742 int 743 pselregister(struct proc *p, fd_set *pibits[3], fd_set *pobits[3], int nfd, 744 int *nregistered, int *ncollected) 745 { 746 static const int evf[] = { EVFILT_READ, EVFILT_WRITE, EVFILT_EXCEPT }; 747 static const int evff[] = { 0, 0, NOTE_OOB }; 748 int msk, i, j, fd, nevents = 0, error = 0; 749 struct kevent kev; 750 fd_mask bits; 751 752 for (msk = 0; msk < 3; msk++) { 753 for (i = 0; i < nfd; i += NFDBITS) { 754 bits = pibits[msk]->fds_bits[i / NFDBITS]; 755 while ((j = ffs(bits)) && (fd = i + --j) < nfd) { 756 bits &= ~(1 << j); 757 758 DPRINTFN(2, "select fd %d mask %d serial %lu\n", 759 fd, msk, p->p_kq_serial); 760 EV_SET(&kev, fd, evf[msk], 761 EV_ADD|EV_ENABLE|__EV_SELECT, 762 evff[msk], 0, (void *)(p->p_kq_serial)); 763 error = kqueue_register(p->p_kq, &kev, 0, p); 764 switch (error) { 765 case 0: 766 nevents++; 767 /* FALLTHROUGH */ 768 case EOPNOTSUPP:/* No underlying kqfilter */ 769 case EINVAL: /* Unimplemented filter */ 770 case EPERM: /* Specific to FIFO and 771 * __EV_SELECT */ 772 error = 0; 773 break; 774 case ENXIO: /* Device has been detached */ 775 default: 776 goto bad; 777 } 778 } 779 } 780 } 781 782 *nregistered = nevents; 783 return (0); 784 bad: 785 DPRINTFN(0, "select fd %u filt %d error %d\n", (int)kev.ident, 786 kev.filter, error); 787 return (error); 788 } 789 790 /* 791 * Convert given kqueue event into corresponding select(2) bit. 792 */ 793 int 794 pselcollect(struct proc *p, struct kevent *kevp, fd_set *pobits[3], 795 int *ncollected) 796 { 797 if ((unsigned long)kevp->udata != p->p_kq_serial) { 798 panic("%s: spurious kevp %p fd %d udata 0x%lx serial 0x%lx", 799 __func__, kevp, (int)kevp->ident, 800 (unsigned long)kevp->udata, p->p_kq_serial); 801 } 802 803 if (kevp->flags & EV_ERROR) { 804 DPRINTFN(2, "select fd %d filt %d error %d\n", 805 (int)kevp->ident, kevp->filter, (int)kevp->data); 806 return (kevp->data); 807 } 808 809 switch (kevp->filter) { 810 case EVFILT_READ: 811 FD_SET(kevp->ident, pobits[0]); 812 break; 813 case EVFILT_WRITE: 814 FD_SET(kevp->ident, pobits[1]); 815 break; 816 case EVFILT_EXCEPT: 817 FD_SET(kevp->ident, pobits[2]); 818 break; 819 default: 820 KASSERT(0); 821 } 822 (*ncollected)++; 823 824 DPRINTFN(2, "select fd %d filt %d\n", (int)kevp->ident, kevp->filter); 825 return (0); 826 } 827 828 /* 829 * Do a wakeup when a selectable event occurs. 830 */ 831 void 832 selwakeup(struct selinfo *sip) 833 { 834 KERNEL_LOCK(); 835 knote_locked(&sip->si_note, NOTE_SUBMIT); 836 KERNEL_UNLOCK(); 837 } 838 839 /* 840 * Only copyout the revents field. 841 */ 842 int 843 pollout(struct pollfd *pl, struct pollfd *upl, u_int nfds) 844 { 845 int error = 0; 846 u_int i = 0; 847 848 while (!error && i++ < nfds) { 849 error = copyout(&pl->revents, &upl->revents, 850 sizeof(upl->revents)); 851 pl++; 852 upl++; 853 } 854 855 return (error); 856 } 857 858 /* 859 * We are using the same mechanism as select only we encode/decode args 860 * differently. 861 */ 862 int 863 sys_poll(struct proc *p, void *v, register_t *retval) 864 { 865 struct sys_poll_args /* { 866 syscallarg(struct pollfd *) fds; 867 syscallarg(u_int) nfds; 868 syscallarg(int) timeout; 869 } */ *uap = v; 870 871 struct timespec ts, *tsp = NULL; 872 int msec = SCARG(uap, timeout); 873 874 if (msec != INFTIM) { 875 if (msec < 0) 876 return (EINVAL); 877 ts.tv_sec = msec / 1000; 878 ts.tv_nsec = (msec - (ts.tv_sec * 1000)) * 1000000; 879 tsp = &ts; 880 } 881 882 return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, NULL, 883 retval)); 884 } 885 886 int 887 sys_ppoll(struct proc *p, void *v, register_t *retval) 888 { 889 struct sys_ppoll_args /* { 890 syscallarg(struct pollfd *) fds; 891 syscallarg(u_int) nfds; 892 syscallarg(const struct timespec *) ts; 893 syscallarg(const sigset_t *) mask; 894 } */ *uap = v; 895 896 int error; 897 struct timespec ts, *tsp = NULL; 898 sigset_t ss, *ssp = NULL; 899 900 if (SCARG(uap, ts) != NULL) { 901 if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0) 902 return (error); 903 #ifdef KTRACE 904 if (KTRPOINT(p, KTR_STRUCT)) 905 ktrreltimespec(p, &ts); 906 #endif 907 if (ts.tv_sec < 0 || !timespecisvalid(&ts)) 908 return (EINVAL); 909 tsp = &ts; 910 } 911 912 if (SCARG(uap, mask) != NULL) { 913 if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0) 914 return (error); 915 ssp = &ss; 916 } 917 918 return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, ssp, 919 retval)); 920 } 921 922 int 923 doppoll(struct proc *p, struct pollfd *fds, u_int nfds, 924 struct timespec *timeout, const sigset_t *sigmask, register_t *retval) 925 { 926 struct kqueue_scan_state scan; 927 struct timespec zerots = {}; 928 struct pollfd pfds[4], *pl = pfds; 929 int error, ncollected = 0, nevents = 0; 930 size_t sz; 931 932 /* Standards say no more than MAX_OPEN; this is possibly better. */ 933 if (nfds > min((int)lim_cur(RLIMIT_NOFILE), maxfiles)) 934 return (EINVAL); 935 936 /* optimize for the default case, of a small nfds value */ 937 if (nfds > nitems(pfds)) { 938 pl = mallocarray(nfds, sizeof(*pl), M_TEMP, 939 M_WAITOK | M_CANFAIL); 940 if (pl == NULL) 941 return (EINVAL); 942 } 943 944 kqpoll_init(nfds); 945 946 sz = nfds * sizeof(*pl); 947 948 if ((error = copyin(fds, pl, sz)) != 0) 949 goto bad; 950 951 if (sigmask) 952 dosigsuspend(p, *sigmask &~ sigcantmask); 953 954 /* Register kqueue events */ 955 ppollregister(p, pl, nfds, &nevents, &ncollected); 956 957 /* 958 * The poll/select family of syscalls has been designed to 959 * block when file descriptors are not available, even if 960 * there's nothing to wait for. 961 */ 962 if (nevents == 0 && ncollected == 0) { 963 uint64_t nsecs = INFSLP; 964 965 if (timeout != NULL) { 966 if (!timespecisset(timeout)) 967 goto done; 968 nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP)); 969 } 970 971 error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqpoll", nsecs); 972 if (error == ERESTART) 973 error = EINTR; 974 if (error == EWOULDBLOCK) 975 error = 0; 976 goto done; 977 } 978 979 /* Do not block if registering found pending events. */ 980 if (ncollected > 0) 981 timeout = &zerots; 982 983 /* Collect at most `nevents' possibly waiting in kqueue_scan() */ 984 kqueue_scan_setup(&scan, p->p_kq); 985 while (nevents > 0) { 986 struct kevent kev[KQ_NEVENTS]; 987 int i, ready, count; 988 989 /* Maximum number of events per iteration */ 990 count = MIN(nitems(kev), nevents); 991 ready = kqueue_scan(&scan, count, kev, timeout, p, &error); 992 993 /* Convert back events that are ready. */ 994 for (i = 0; i < ready; i++) 995 ncollected += ppollcollect(p, &kev[i], pl, nfds); 996 997 /* 998 * Stop if there was an error or if we had enough 999 * place to collect all events that were ready. 1000 */ 1001 if (error || ready < count) 1002 break; 1003 1004 nevents -= ready; 1005 } 1006 kqueue_scan_finish(&scan); 1007 *retval = ncollected; 1008 done: 1009 /* 1010 * NOTE: poll(2) is not restarted after a signal and EWOULDBLOCK is 1011 * ignored (since the whole point is to see what would block). 1012 */ 1013 switch (error) { 1014 case EINTR: 1015 error = pollout(pl, fds, nfds); 1016 if (error == 0) 1017 error = EINTR; 1018 break; 1019 case EWOULDBLOCK: 1020 case 0: 1021 error = pollout(pl, fds, nfds); 1022 break; 1023 } 1024 #ifdef KTRACE 1025 if (KTRPOINT(p, KTR_STRUCT)) 1026 ktrpollfd(p, pl, nfds); 1027 #endif /* KTRACE */ 1028 bad: 1029 if (pl != pfds) 1030 free(pl, M_TEMP, sz); 1031 1032 kqpoll_done(nfds); 1033 1034 return (error); 1035 } 1036 1037 int 1038 ppollregister_evts(struct proc *p, struct kevent *kevp, int nkev, 1039 struct pollfd *pl, unsigned int pollid) 1040 { 1041 int i, error, nevents = 0; 1042 1043 KASSERT(pl->revents == 0); 1044 1045 for (i = 0; i < nkev; i++, kevp++) { 1046 again: 1047 error = kqueue_register(p->p_kq, kevp, pollid, p); 1048 switch (error) { 1049 case 0: 1050 nevents++; 1051 break; 1052 case EOPNOTSUPP:/* No underlying kqfilter */ 1053 case EINVAL: /* Unimplemented filter */ 1054 break; 1055 case EBADF: /* Bad file descriptor */ 1056 pl->revents |= POLLNVAL; 1057 break; 1058 case EPERM: /* Specific to FIFO */ 1059 KASSERT(kevp->filter == EVFILT_WRITE); 1060 if (nkev == 1) { 1061 /* 1062 * If this is the only filter make sure 1063 * POLLHUP is passed to userland. 1064 */ 1065 kevp->filter = EVFILT_EXCEPT; 1066 goto again; 1067 } 1068 break; 1069 default: 1070 DPRINTFN(0, "poll err %lu fd %d revents %02x serial" 1071 " %lu filt %d ERROR=%d\n", 1072 ((unsigned long)kevp->udata - p->p_kq_serial), 1073 pl->fd, pl->revents, p->p_kq_serial, kevp->filter, 1074 error); 1075 /* FALLTHROUGH */ 1076 case ENXIO: /* Device has been detached */ 1077 pl->revents |= POLLERR; 1078 break; 1079 } 1080 } 1081 1082 return (nevents); 1083 } 1084 1085 /* 1086 * Convert pollfd into kqueue events and register them on the 1087 * per-thread queue. 1088 * 1089 * At most 3 events can correspond to a single pollfd. 1090 */ 1091 void 1092 ppollregister(struct proc *p, struct pollfd *pl, int nfds, int *nregistered, 1093 int *ncollected) 1094 { 1095 int i, nkev, nevt, forcehup; 1096 struct kevent kev[3], *kevp; 1097 1098 for (i = 0; i < nfds; i++) { 1099 pl[i].events &= ~POLL_NOHUP; 1100 pl[i].revents = 0; 1101 1102 if (pl[i].fd < 0) 1103 continue; 1104 1105 /* 1106 * POLLHUP checking is implicit in the event filters. 1107 * However, the checking must be even if no events are 1108 * requested. 1109 */ 1110 forcehup = ((pl[i].events & ~POLLHUP) == 0); 1111 1112 DPRINTFN(1, "poll set %d/%d fd %d events %02x serial %lu\n", 1113 i+1, nfds, pl[i].fd, pl[i].events, p->p_kq_serial); 1114 1115 nevt = 0; 1116 nkev = 0; 1117 kevp = kev; 1118 if (pl[i].events & (POLLIN | POLLRDNORM)) { 1119 EV_SET(kevp, pl[i].fd, EVFILT_READ, 1120 EV_ADD|EV_ENABLE|__EV_POLL, 0, 0, 1121 (void *)(p->p_kq_serial + i)); 1122 nkev++; 1123 kevp++; 1124 } 1125 if (pl[i].events & (POLLOUT | POLLWRNORM)) { 1126 EV_SET(kevp, pl[i].fd, EVFILT_WRITE, 1127 EV_ADD|EV_ENABLE|__EV_POLL, 0, 0, 1128 (void *)(p->p_kq_serial + i)); 1129 nkev++; 1130 kevp++; 1131 } 1132 if ((pl[i].events & (POLLPRI | POLLRDBAND)) || forcehup) { 1133 int evff = forcehup ? 0 : NOTE_OOB; 1134 1135 EV_SET(kevp, pl[i].fd, EVFILT_EXCEPT, 1136 EV_ADD|EV_ENABLE|__EV_POLL, evff, 0, 1137 (void *)(p->p_kq_serial + i)); 1138 nkev++; 1139 kevp++; 1140 } 1141 1142 if (nkev == 0) 1143 continue; 1144 1145 *nregistered += ppollregister_evts(p, kev, nkev, &pl[i], i); 1146 1147 if (pl[i].revents != 0) 1148 (*ncollected)++; 1149 } 1150 1151 DPRINTFN(1, "poll registered = %d, collected = %d\n", *nregistered, 1152 *ncollected); 1153 } 1154 1155 /* 1156 * Convert given kqueue event into corresponding poll(2) revents bit. 1157 */ 1158 int 1159 ppollcollect(struct proc *p, struct kevent *kevp, struct pollfd *pl, u_int nfds) 1160 { 1161 static struct timeval poll_errintvl = { 5, 0 }; 1162 static struct timeval poll_lasterr; 1163 int already_seen; 1164 unsigned long i; 1165 1166 /* Extract poll array index */ 1167 i = (unsigned long)kevp->udata - p->p_kq_serial; 1168 1169 if (i >= nfds) { 1170 panic("%s: spurious kevp %p nfds %u udata 0x%lx serial 0x%lx", 1171 __func__, kevp, nfds, 1172 (unsigned long)kevp->udata, p->p_kq_serial); 1173 } 1174 if ((int)kevp->ident != pl[i].fd) { 1175 panic("%s: kevp %p %lu/%d mismatch fd %d!=%d serial 0x%lx", 1176 __func__, kevp, i + 1, nfds, (int)kevp->ident, pl[i].fd, 1177 p->p_kq_serial); 1178 } 1179 1180 /* 1181 * A given descriptor may already have generated an error 1182 * against another filter during kqueue_register(). 1183 * 1184 * Make sure to set the appropriate flags but do not 1185 * increment `*retval' more than once. 1186 */ 1187 already_seen = (pl[i].revents != 0); 1188 1189 /* POLLNVAL preempts other events. */ 1190 if ((kevp->flags & EV_ERROR) && kevp->data == EBADF) { 1191 pl[i].revents = POLLNVAL; 1192 goto done; 1193 } else if (pl[i].revents & POLLNVAL) { 1194 goto done; 1195 } 1196 1197 switch (kevp->filter) { 1198 case EVFILT_READ: 1199 if (kevp->flags & __EV_HUP) 1200 pl[i].revents |= POLLHUP; 1201 if (pl[i].events & (POLLIN | POLLRDNORM)) 1202 pl[i].revents |= pl[i].events & (POLLIN | POLLRDNORM); 1203 break; 1204 case EVFILT_WRITE: 1205 /* POLLHUP and POLLOUT/POLLWRNORM are mutually exclusive */ 1206 if (kevp->flags & __EV_HUP) { 1207 pl[i].revents |= POLLHUP; 1208 } else if (pl[i].events & (POLLOUT | POLLWRNORM)) { 1209 pl[i].revents |= pl[i].events & (POLLOUT | POLLWRNORM); 1210 } 1211 break; 1212 case EVFILT_EXCEPT: 1213 if (kevp->flags & __EV_HUP) { 1214 if (pl[i].events != 0 && pl[i].events != POLLOUT) 1215 DPRINTFN(0, "weird events %x\n", pl[i].events); 1216 pl[i].revents |= POLLHUP; 1217 break; 1218 } 1219 if (pl[i].events & (POLLPRI | POLLRDBAND)) 1220 pl[i].revents |= pl[i].events & (POLLPRI | POLLRDBAND); 1221 break; 1222 default: 1223 KASSERT(0); 1224 } 1225 1226 done: 1227 DPRINTFN(1, "poll get %lu/%d fd %d revents %02x serial %lu filt %d\n", 1228 i+1, nfds, pl[i].fd, pl[i].revents, (unsigned long)kevp->udata, 1229 kevp->filter); 1230 1231 /* 1232 * Make noise about unclaimed events as they might indicate a bug 1233 * and can result in spurious-looking wakeups of poll(2). 1234 * 1235 * Live-locking within the system call should not happen because 1236 * the scan loop in doppoll() has an upper limit for the number 1237 * of events to process. 1238 */ 1239 if (pl[i].revents == 0 && ratecheck(&poll_lasterr, &poll_errintvl)) { 1240 printf("%s[%d]: poll index %lu fd %d events 0x%x " 1241 "filter %d/0x%x unclaimed\n", 1242 p->p_p->ps_comm, p->p_tid, i, pl[i].fd, 1243 pl[i].events, kevp->filter, kevp->flags); 1244 } 1245 1246 if (!already_seen && (pl[i].revents != 0)) 1247 return (1); 1248 1249 return (0); 1250 } 1251 1252 /* 1253 * utrace system call 1254 */ 1255 int 1256 sys_utrace(struct proc *curp, void *v, register_t *retval) 1257 { 1258 #ifdef KTRACE 1259 struct sys_utrace_args /* { 1260 syscallarg(const char *) label; 1261 syscallarg(const void *) addr; 1262 syscallarg(size_t) len; 1263 } */ *uap = v; 1264 1265 return (ktruser(curp, SCARG(uap, label), SCARG(uap, addr), 1266 SCARG(uap, len))); 1267 #else 1268 return (0); 1269 #endif 1270 } 1271