1 /* $OpenBSD: sys_generic.c,v 1.160 2024/12/30 02:46:00 guenther 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 ptsignal(p, SIGPIPE, STHREAD); 387 } 388 cnt -= uio->uio_resid; 389 390 mtx_enter(&fp->f_mtx); 391 fp->f_wxfer++; 392 fp->f_wbytes += cnt; 393 mtx_leave(&fp->f_mtx); 394 #ifdef KTRACE 395 if (ktriov != NULL) { 396 if (error == 0) 397 ktrgenio(p, fd, UIO_WRITE, ktriov, cnt); 398 free(ktriov, M_TEMP, iovlen); 399 } 400 #endif 401 *retval = cnt; 402 done: 403 FRELE(fp, p); 404 return (error); 405 } 406 407 /* 408 * Ioctl system call 409 */ 410 int 411 sys_ioctl(struct proc *p, void *v, register_t *retval) 412 { 413 struct sys_ioctl_args /* { 414 syscallarg(int) fd; 415 syscallarg(u_long) com; 416 syscallarg(void *) data; 417 } */ *uap = v; 418 struct file *fp; 419 struct filedesc *fdp = p->p_fd; 420 u_long com = SCARG(uap, com); 421 int error = 0; 422 u_int size = 0; 423 caddr_t data, memp = NULL; 424 int tmp; 425 #define STK_PARAMS 128 426 long long stkbuf[STK_PARAMS / sizeof(long long)]; 427 428 if ((fp = fd_getfile_mode(fdp, SCARG(uap, fd), FREAD|FWRITE)) == NULL) 429 return (EBADF); 430 431 if (fp->f_type == DTYPE_SOCKET) { 432 struct socket *so = fp->f_data; 433 434 if (so->so_state & SS_DNS) { 435 error = EINVAL; 436 goto out; 437 } 438 } 439 440 error = pledge_ioctl(p, com, fp); 441 if (error) 442 goto out; 443 444 switch (com) { 445 case FIONCLEX: 446 case FIOCLEX: 447 fdplock(fdp); 448 if (com == FIONCLEX) 449 fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE; 450 else 451 fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE; 452 fdpunlock(fdp); 453 goto out; 454 } 455 456 /* 457 * Interpret high order word to find amount of data to be 458 * copied to/from the user's address space. 459 */ 460 size = IOCPARM_LEN(com); 461 if (size > IOCPARM_MAX) { 462 error = ENOTTY; 463 goto out; 464 } 465 if (size > sizeof (stkbuf)) { 466 memp = malloc(size, M_IOCTLOPS, M_WAITOK); 467 data = memp; 468 } else 469 data = (caddr_t)stkbuf; 470 if (com&IOC_IN) { 471 if (size) { 472 error = copyin(SCARG(uap, data), data, size); 473 if (error) { 474 goto out; 475 } 476 } else 477 *(caddr_t *)data = SCARG(uap, data); 478 } else if ((com&IOC_OUT) && size) 479 /* 480 * Zero the buffer so the user always 481 * gets back something deterministic. 482 */ 483 memset(data, 0, size); 484 else if (com&IOC_VOID) 485 *(caddr_t *)data = SCARG(uap, data); 486 487 switch (com) { 488 489 case FIONBIO: 490 if ((tmp = *(int *)data) != 0) 491 atomic_setbits_int(&fp->f_flag, FNONBLOCK); 492 else 493 atomic_clearbits_int(&fp->f_flag, FNONBLOCK); 494 error = 0; 495 break; 496 497 case FIOASYNC: 498 if ((tmp = *(int *)data) != 0) 499 atomic_setbits_int(&fp->f_flag, FASYNC); 500 else 501 atomic_clearbits_int(&fp->f_flag, FASYNC); 502 error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p); 503 break; 504 505 default: 506 error = (*fp->f_ops->fo_ioctl)(fp, com, data, p); 507 break; 508 } 509 /* 510 * Copy any data to user, size was 511 * already set and checked above. 512 */ 513 if (error == 0 && (com&IOC_OUT) && size) 514 error = copyout(data, SCARG(uap, data), size); 515 out: 516 FRELE(fp, p); 517 free(memp, M_IOCTLOPS, size); 518 return (error); 519 } 520 521 /* 522 * Select system call. 523 */ 524 int 525 sys_select(struct proc *p, void *v, register_t *retval) 526 { 527 struct sys_select_args /* { 528 syscallarg(int) nd; 529 syscallarg(fd_set *) in; 530 syscallarg(fd_set *) ou; 531 syscallarg(fd_set *) ex; 532 syscallarg(struct timeval *) tv; 533 } */ *uap = v; 534 535 struct timespec ts, *tsp = NULL; 536 int error; 537 538 if (SCARG(uap, tv) != NULL) { 539 struct timeval tv; 540 if ((error = copyin(SCARG(uap, tv), &tv, sizeof tv)) != 0) 541 return (error); 542 #ifdef KTRACE 543 if (KTRPOINT(p, KTR_STRUCT)) 544 ktrreltimeval(p, &tv); 545 #endif 546 if (tv.tv_sec < 0 || !timerisvalid(&tv)) 547 return (EINVAL); 548 TIMEVAL_TO_TIMESPEC(&tv, &ts); 549 tsp = &ts; 550 } 551 552 return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou), 553 SCARG(uap, ex), tsp, NULL, retval)); 554 } 555 556 int 557 sys_pselect(struct proc *p, void *v, register_t *retval) 558 { 559 struct sys_pselect_args /* { 560 syscallarg(int) nd; 561 syscallarg(fd_set *) in; 562 syscallarg(fd_set *) ou; 563 syscallarg(fd_set *) ex; 564 syscallarg(const struct timespec *) ts; 565 syscallarg(const sigset_t *) mask; 566 } */ *uap = v; 567 568 struct timespec ts, *tsp = NULL; 569 sigset_t ss, *ssp = NULL; 570 int error; 571 572 if (SCARG(uap, ts) != NULL) { 573 if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0) 574 return (error); 575 #ifdef KTRACE 576 if (KTRPOINT(p, KTR_STRUCT)) 577 ktrreltimespec(p, &ts); 578 #endif 579 if (ts.tv_sec < 0 || !timespecisvalid(&ts)) 580 return (EINVAL); 581 tsp = &ts; 582 } 583 if (SCARG(uap, mask) != NULL) { 584 if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0) 585 return (error); 586 ssp = &ss; 587 } 588 589 return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou), 590 SCARG(uap, ex), tsp, ssp, retval)); 591 } 592 593 int 594 dopselect(struct proc *p, int nd, fd_set *in, fd_set *ou, fd_set *ex, 595 struct timespec *timeout, const sigset_t *sigmask, register_t *retval) 596 { 597 struct kqueue_scan_state scan; 598 struct timespec zerots = {}; 599 fd_mask bits[6]; 600 fd_set *pibits[3], *pobits[3]; 601 int error, nfiles, ncollected = 0, nevents = 0; 602 u_int ni; 603 604 if (nd < 0) 605 return (EINVAL); 606 607 nfiles = READ_ONCE(p->p_fd->fd_nfiles); 608 if (nd > nfiles) 609 nd = nfiles; 610 611 ni = howmany(nd, NFDBITS) * sizeof(fd_mask); 612 if (ni > sizeof(bits[0])) { 613 caddr_t mbits; 614 615 mbits = mallocarray(6, ni, M_TEMP, M_WAITOK|M_ZERO); 616 pibits[0] = (fd_set *)&mbits[ni * 0]; 617 pibits[1] = (fd_set *)&mbits[ni * 1]; 618 pibits[2] = (fd_set *)&mbits[ni * 2]; 619 pobits[0] = (fd_set *)&mbits[ni * 3]; 620 pobits[1] = (fd_set *)&mbits[ni * 4]; 621 pobits[2] = (fd_set *)&mbits[ni * 5]; 622 } else { 623 memset(bits, 0, sizeof(bits)); 624 pibits[0] = (fd_set *)&bits[0]; 625 pibits[1] = (fd_set *)&bits[1]; 626 pibits[2] = (fd_set *)&bits[2]; 627 pobits[0] = (fd_set *)&bits[3]; 628 pobits[1] = (fd_set *)&bits[4]; 629 pobits[2] = (fd_set *)&bits[5]; 630 } 631 632 kqpoll_init(nd); 633 634 #define getbits(name, x) \ 635 if (name && (error = copyin(name, pibits[x], ni))) \ 636 goto done; 637 getbits(in, 0); 638 getbits(ou, 1); 639 getbits(ex, 2); 640 #undef getbits 641 #ifdef KTRACE 642 if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) { 643 if (in) ktrfdset(p, pibits[0], ni); 644 if (ou) ktrfdset(p, pibits[1], ni); 645 if (ex) ktrfdset(p, pibits[2], ni); 646 } 647 #endif 648 649 if (sigmask) 650 dosigsuspend(p, *sigmask &~ sigcantmask); 651 652 /* Register kqueue events */ 653 error = pselregister(p, pibits, pobits, nd, &nevents, &ncollected); 654 if (error != 0) 655 goto done; 656 657 /* 658 * The poll/select family of syscalls has been designed to 659 * block when file descriptors are not available, even if 660 * there's nothing to wait for. 661 */ 662 if (nevents == 0 && ncollected == 0) { 663 uint64_t nsecs = INFSLP; 664 665 if (timeout != NULL) { 666 if (!timespecisset(timeout)) 667 goto done; 668 nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP)); 669 } 670 error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqsel", nsecs); 671 /* select is not restarted after signals... */ 672 if (error == ERESTART) 673 error = EINTR; 674 if (error == EWOULDBLOCK) 675 error = 0; 676 goto done; 677 } 678 679 /* Do not block if registering found pending events. */ 680 if (ncollected > 0) 681 timeout = &zerots; 682 683 /* Collect at most `nevents' possibly waiting in kqueue_scan() */ 684 kqueue_scan_setup(&scan, p->p_kq); 685 while (nevents > 0) { 686 struct kevent kev[KQ_NEVENTS]; 687 int i, ready, count; 688 689 /* Maximum number of events per iteration */ 690 count = MIN(nitems(kev), nevents); 691 ready = kqueue_scan(&scan, count, kev, timeout, p, &error); 692 693 /* Convert back events that are ready. */ 694 for (i = 0; i < ready && error == 0; i++) 695 error = pselcollect(p, &kev[i], pobits, &ncollected); 696 /* 697 * Stop if there was an error or if we had enough 698 * space to collect all events that were ready. 699 */ 700 if (error || ready < count) 701 break; 702 703 nevents -= ready; 704 } 705 kqueue_scan_finish(&scan); 706 *retval = ncollected; 707 done: 708 #define putbits(name, x) \ 709 if (name && (error2 = copyout(pobits[x], name, ni))) \ 710 error = error2; 711 if (error == 0) { 712 int error2; 713 714 putbits(in, 0); 715 putbits(ou, 1); 716 putbits(ex, 2); 717 #undef putbits 718 #ifdef KTRACE 719 if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) { 720 if (in) ktrfdset(p, pobits[0], ni); 721 if (ou) ktrfdset(p, pobits[1], ni); 722 if (ex) ktrfdset(p, pobits[2], ni); 723 } 724 #endif 725 } 726 727 if (pibits[0] != (fd_set *)&bits[0]) 728 free(pibits[0], M_TEMP, 6 * ni); 729 730 kqpoll_done(nd); 731 732 return (error); 733 } 734 735 /* 736 * Convert fd_set into kqueue events and register them on the 737 * per-thread queue. 738 */ 739 int 740 pselregister(struct proc *p, fd_set *pibits[3], fd_set *pobits[3], int nfd, 741 int *nregistered, int *ncollected) 742 { 743 static const int evf[] = { EVFILT_READ, EVFILT_WRITE, EVFILT_EXCEPT }; 744 static const int evff[] = { 0, 0, NOTE_OOB }; 745 int msk, i, j, fd, nevents = 0, error = 0; 746 struct kevent kev; 747 fd_mask bits; 748 749 for (msk = 0; msk < 3; msk++) { 750 for (i = 0; i < nfd; i += NFDBITS) { 751 bits = pibits[msk]->fds_bits[i / NFDBITS]; 752 while ((j = ffs(bits)) && (fd = i + --j) < nfd) { 753 bits &= ~(1 << j); 754 755 DPRINTFN(2, "select fd %d mask %d serial %lu\n", 756 fd, msk, p->p_kq_serial); 757 EV_SET(&kev, fd, evf[msk], 758 EV_ADD|EV_ENABLE|__EV_SELECT, 759 evff[msk], 0, (void *)(p->p_kq_serial)); 760 error = kqueue_register(p->p_kq, &kev, 0, p); 761 switch (error) { 762 case 0: 763 nevents++; 764 /* FALLTHROUGH */ 765 case EOPNOTSUPP:/* No underlying kqfilter */ 766 case EINVAL: /* Unimplemented filter */ 767 case EPERM: /* Specific to FIFO and 768 * __EV_SELECT */ 769 error = 0; 770 break; 771 case ENXIO: /* Device has been detached */ 772 default: 773 goto bad; 774 } 775 } 776 } 777 } 778 779 *nregistered = nevents; 780 return (0); 781 bad: 782 DPRINTFN(0, "select fd %u filt %d error %d\n", (int)kev.ident, 783 kev.filter, error); 784 return (error); 785 } 786 787 /* 788 * Convert given kqueue event into corresponding select(2) bit. 789 */ 790 int 791 pselcollect(struct proc *p, struct kevent *kevp, fd_set *pobits[3], 792 int *ncollected) 793 { 794 if ((unsigned long)kevp->udata != p->p_kq_serial) { 795 panic("%s: spurious kevp %p fd %d udata 0x%lx serial 0x%lx", 796 __func__, kevp, (int)kevp->ident, 797 (unsigned long)kevp->udata, p->p_kq_serial); 798 } 799 800 if (kevp->flags & EV_ERROR) { 801 DPRINTFN(2, "select fd %d filt %d error %d\n", 802 (int)kevp->ident, kevp->filter, (int)kevp->data); 803 return (kevp->data); 804 } 805 806 switch (kevp->filter) { 807 case EVFILT_READ: 808 FD_SET(kevp->ident, pobits[0]); 809 break; 810 case EVFILT_WRITE: 811 FD_SET(kevp->ident, pobits[1]); 812 break; 813 case EVFILT_EXCEPT: 814 FD_SET(kevp->ident, pobits[2]); 815 break; 816 default: 817 KASSERT(0); 818 } 819 (*ncollected)++; 820 821 DPRINTFN(2, "select fd %d filt %d\n", (int)kevp->ident, kevp->filter); 822 return (0); 823 } 824 825 /* 826 * Do a wakeup when a selectable event occurs. 827 */ 828 void 829 selwakeup(struct selinfo *sip) 830 { 831 KERNEL_LOCK(); 832 knote_locked(&sip->si_note, NOTE_SUBMIT); 833 KERNEL_UNLOCK(); 834 } 835 836 /* 837 * Only copyout the revents field. 838 */ 839 int 840 pollout(struct pollfd *pl, struct pollfd *upl, u_int nfds) 841 { 842 int error = 0; 843 u_int i = 0; 844 845 while (!error && i++ < nfds) { 846 error = copyout(&pl->revents, &upl->revents, 847 sizeof(upl->revents)); 848 pl++; 849 upl++; 850 } 851 852 return (error); 853 } 854 855 /* 856 * We are using the same mechanism as select only we encode/decode args 857 * differently. 858 */ 859 int 860 sys_poll(struct proc *p, void *v, register_t *retval) 861 { 862 struct sys_poll_args /* { 863 syscallarg(struct pollfd *) fds; 864 syscallarg(u_int) nfds; 865 syscallarg(int) timeout; 866 } */ *uap = v; 867 868 struct timespec ts, *tsp = NULL; 869 int msec = SCARG(uap, timeout); 870 871 if (msec != INFTIM) { 872 if (msec < 0) 873 return (EINVAL); 874 ts.tv_sec = msec / 1000; 875 ts.tv_nsec = (msec - (ts.tv_sec * 1000)) * 1000000; 876 tsp = &ts; 877 } 878 879 return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, NULL, 880 retval)); 881 } 882 883 int 884 sys_ppoll(struct proc *p, void *v, register_t *retval) 885 { 886 struct sys_ppoll_args /* { 887 syscallarg(struct pollfd *) fds; 888 syscallarg(u_int) nfds; 889 syscallarg(const struct timespec *) ts; 890 syscallarg(const sigset_t *) mask; 891 } */ *uap = v; 892 893 int error; 894 struct timespec ts, *tsp = NULL; 895 sigset_t ss, *ssp = NULL; 896 897 if (SCARG(uap, ts) != NULL) { 898 if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0) 899 return (error); 900 #ifdef KTRACE 901 if (KTRPOINT(p, KTR_STRUCT)) 902 ktrreltimespec(p, &ts); 903 #endif 904 if (ts.tv_sec < 0 || !timespecisvalid(&ts)) 905 return (EINVAL); 906 tsp = &ts; 907 } 908 909 if (SCARG(uap, mask) != NULL) { 910 if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0) 911 return (error); 912 ssp = &ss; 913 } 914 915 return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, ssp, 916 retval)); 917 } 918 919 int 920 doppoll(struct proc *p, struct pollfd *fds, u_int nfds, 921 struct timespec *timeout, const sigset_t *sigmask, register_t *retval) 922 { 923 struct kqueue_scan_state scan; 924 struct timespec zerots = {}; 925 struct pollfd pfds[4], *pl = pfds; 926 int error, ncollected = 0, nevents = 0; 927 size_t sz; 928 929 /* Standards say no more than MAX_OPEN; this is possibly better. */ 930 if (nfds > min((int)lim_cur(RLIMIT_NOFILE), maxfiles)) 931 return (EINVAL); 932 933 /* optimize for the default case, of a small nfds value */ 934 if (nfds > nitems(pfds)) { 935 pl = mallocarray(nfds, sizeof(*pl), M_TEMP, 936 M_WAITOK | M_CANFAIL); 937 if (pl == NULL) 938 return (EINVAL); 939 } 940 941 kqpoll_init(nfds); 942 943 sz = nfds * sizeof(*pl); 944 945 if ((error = copyin(fds, pl, sz)) != 0) 946 goto bad; 947 948 if (sigmask) 949 dosigsuspend(p, *sigmask &~ sigcantmask); 950 951 /* Register kqueue events */ 952 ppollregister(p, pl, nfds, &nevents, &ncollected); 953 954 /* 955 * The poll/select family of syscalls has been designed to 956 * block when file descriptors are not available, even if 957 * there's nothing to wait for. 958 */ 959 if (nevents == 0 && ncollected == 0) { 960 uint64_t nsecs = INFSLP; 961 962 if (timeout != NULL) { 963 if (!timespecisset(timeout)) 964 goto done; 965 nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP)); 966 } 967 968 error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqpoll", nsecs); 969 if (error == ERESTART) 970 error = EINTR; 971 if (error == EWOULDBLOCK) 972 error = 0; 973 goto done; 974 } 975 976 /* Do not block if registering found pending events. */ 977 if (ncollected > 0) 978 timeout = &zerots; 979 980 /* Collect at most `nevents' possibly waiting in kqueue_scan() */ 981 kqueue_scan_setup(&scan, p->p_kq); 982 while (nevents > 0) { 983 struct kevent kev[KQ_NEVENTS]; 984 int i, ready, count; 985 986 /* Maximum number of events per iteration */ 987 count = MIN(nitems(kev), nevents); 988 ready = kqueue_scan(&scan, count, kev, timeout, p, &error); 989 990 /* Convert back events that are ready. */ 991 for (i = 0; i < ready; i++) 992 ncollected += ppollcollect(p, &kev[i], pl, nfds); 993 994 /* 995 * Stop if there was an error or if we had enough 996 * place to collect all events that were ready. 997 */ 998 if (error || ready < count) 999 break; 1000 1001 nevents -= ready; 1002 } 1003 kqueue_scan_finish(&scan); 1004 *retval = ncollected; 1005 done: 1006 /* 1007 * NOTE: poll(2) is not restarted after a signal and EWOULDBLOCK is 1008 * ignored (since the whole point is to see what would block). 1009 */ 1010 switch (error) { 1011 case EINTR: 1012 error = pollout(pl, fds, nfds); 1013 if (error == 0) 1014 error = EINTR; 1015 break; 1016 case EWOULDBLOCK: 1017 case 0: 1018 error = pollout(pl, fds, nfds); 1019 break; 1020 } 1021 #ifdef KTRACE 1022 if (KTRPOINT(p, KTR_STRUCT)) 1023 ktrpollfd(p, pl, nfds); 1024 #endif /* KTRACE */ 1025 bad: 1026 if (pl != pfds) 1027 free(pl, M_TEMP, sz); 1028 1029 kqpoll_done(nfds); 1030 1031 return (error); 1032 } 1033 1034 int 1035 ppollregister_evts(struct proc *p, struct kevent *kevp, int nkev, 1036 struct pollfd *pl, unsigned int pollid) 1037 { 1038 int i, error, nevents = 0; 1039 1040 KASSERT(pl->revents == 0); 1041 1042 for (i = 0; i < nkev; i++, kevp++) { 1043 again: 1044 error = kqueue_register(p->p_kq, kevp, pollid, p); 1045 switch (error) { 1046 case 0: 1047 nevents++; 1048 break; 1049 case EOPNOTSUPP:/* No underlying kqfilter */ 1050 case EINVAL: /* Unimplemented filter */ 1051 break; 1052 case EBADF: /* Bad file descriptor */ 1053 pl->revents |= POLLNVAL; 1054 break; 1055 case EPERM: /* Specific to FIFO */ 1056 KASSERT(kevp->filter == EVFILT_WRITE); 1057 if (nkev == 1) { 1058 /* 1059 * If this is the only filter make sure 1060 * POLLHUP is passed to userland. 1061 */ 1062 kevp->filter = EVFILT_EXCEPT; 1063 goto again; 1064 } 1065 break; 1066 default: 1067 DPRINTFN(0, "poll err %lu fd %d revents %02x serial" 1068 " %lu filt %d ERROR=%d\n", 1069 ((unsigned long)kevp->udata - p->p_kq_serial), 1070 pl->fd, pl->revents, p->p_kq_serial, kevp->filter, 1071 error); 1072 /* FALLTHROUGH */ 1073 case ENXIO: /* Device has been detached */ 1074 pl->revents |= POLLERR; 1075 break; 1076 } 1077 } 1078 1079 return (nevents); 1080 } 1081 1082 /* 1083 * Convert pollfd into kqueue events and register them on the 1084 * per-thread queue. 1085 * 1086 * At most 3 events can correspond to a single pollfd. 1087 */ 1088 void 1089 ppollregister(struct proc *p, struct pollfd *pl, int nfds, int *nregistered, 1090 int *ncollected) 1091 { 1092 int i, nkev, nevt, forcehup; 1093 struct kevent kev[3], *kevp; 1094 1095 for (i = 0; i < nfds; i++) { 1096 pl[i].events &= ~POLL_NOHUP; 1097 pl[i].revents = 0; 1098 1099 if (pl[i].fd < 0) 1100 continue; 1101 1102 /* 1103 * POLLHUP checking is implicit in the event filters. 1104 * However, the checking must be even if no events are 1105 * requested. 1106 */ 1107 forcehup = ((pl[i].events & ~POLLHUP) == 0); 1108 1109 DPRINTFN(1, "poll set %d/%d fd %d events %02x serial %lu\n", 1110 i+1, nfds, pl[i].fd, pl[i].events, p->p_kq_serial); 1111 1112 nevt = 0; 1113 nkev = 0; 1114 kevp = kev; 1115 if (pl[i].events & (POLLIN | POLLRDNORM)) { 1116 EV_SET(kevp, pl[i].fd, EVFILT_READ, 1117 EV_ADD|EV_ENABLE|__EV_POLL, 0, 0, 1118 (void *)(p->p_kq_serial + i)); 1119 nkev++; 1120 kevp++; 1121 } 1122 if (pl[i].events & (POLLOUT | POLLWRNORM)) { 1123 EV_SET(kevp, pl[i].fd, EVFILT_WRITE, 1124 EV_ADD|EV_ENABLE|__EV_POLL, 0, 0, 1125 (void *)(p->p_kq_serial + i)); 1126 nkev++; 1127 kevp++; 1128 } 1129 if ((pl[i].events & (POLLPRI | POLLRDBAND)) || forcehup) { 1130 int evff = forcehup ? 0 : NOTE_OOB; 1131 1132 EV_SET(kevp, pl[i].fd, EVFILT_EXCEPT, 1133 EV_ADD|EV_ENABLE|__EV_POLL, evff, 0, 1134 (void *)(p->p_kq_serial + i)); 1135 nkev++; 1136 kevp++; 1137 } 1138 1139 if (nkev == 0) 1140 continue; 1141 1142 *nregistered += ppollregister_evts(p, kev, nkev, &pl[i], i); 1143 1144 if (pl[i].revents != 0) 1145 (*ncollected)++; 1146 } 1147 1148 DPRINTFN(1, "poll registered = %d, collected = %d\n", *nregistered, 1149 *ncollected); 1150 } 1151 1152 /* 1153 * Convert given kqueue event into corresponding poll(2) revents bit. 1154 */ 1155 int 1156 ppollcollect(struct proc *p, struct kevent *kevp, struct pollfd *pl, u_int nfds) 1157 { 1158 static struct timeval poll_errintvl = { 5, 0 }; 1159 static struct timeval poll_lasterr; 1160 int already_seen; 1161 unsigned long i; 1162 1163 /* Extract poll array index */ 1164 i = (unsigned long)kevp->udata - p->p_kq_serial; 1165 1166 if (i >= nfds) { 1167 panic("%s: spurious kevp %p nfds %u udata 0x%lx serial 0x%lx", 1168 __func__, kevp, nfds, 1169 (unsigned long)kevp->udata, p->p_kq_serial); 1170 } 1171 if ((int)kevp->ident != pl[i].fd) { 1172 panic("%s: kevp %p %lu/%d mismatch fd %d!=%d serial 0x%lx", 1173 __func__, kevp, i + 1, nfds, (int)kevp->ident, pl[i].fd, 1174 p->p_kq_serial); 1175 } 1176 1177 /* 1178 * A given descriptor may already have generated an error 1179 * against another filter during kqueue_register(). 1180 * 1181 * Make sure to set the appropriate flags but do not 1182 * increment `*retval' more than once. 1183 */ 1184 already_seen = (pl[i].revents != 0); 1185 1186 /* POLLNVAL preempts other events. */ 1187 if ((kevp->flags & EV_ERROR) && kevp->data == EBADF) { 1188 pl[i].revents = POLLNVAL; 1189 goto done; 1190 } else if (pl[i].revents & POLLNVAL) { 1191 goto done; 1192 } 1193 1194 switch (kevp->filter) { 1195 case EVFILT_READ: 1196 if (kevp->flags & __EV_HUP) 1197 pl[i].revents |= POLLHUP; 1198 if (pl[i].events & (POLLIN | POLLRDNORM)) 1199 pl[i].revents |= pl[i].events & (POLLIN | POLLRDNORM); 1200 break; 1201 case EVFILT_WRITE: 1202 /* POLLHUP and POLLOUT/POLLWRNORM are mutually exclusive */ 1203 if (kevp->flags & __EV_HUP) { 1204 pl[i].revents |= POLLHUP; 1205 } else if (pl[i].events & (POLLOUT | POLLWRNORM)) { 1206 pl[i].revents |= pl[i].events & (POLLOUT | POLLWRNORM); 1207 } 1208 break; 1209 case EVFILT_EXCEPT: 1210 if (kevp->flags & __EV_HUP) { 1211 if (pl[i].events != 0 && pl[i].events != POLLOUT) 1212 DPRINTFN(0, "weird events %x\n", pl[i].events); 1213 pl[i].revents |= POLLHUP; 1214 break; 1215 } 1216 if (pl[i].events & (POLLPRI | POLLRDBAND)) 1217 pl[i].revents |= pl[i].events & (POLLPRI | POLLRDBAND); 1218 break; 1219 default: 1220 KASSERT(0); 1221 } 1222 1223 done: 1224 DPRINTFN(1, "poll get %lu/%d fd %d revents %02x serial %lu filt %d\n", 1225 i+1, nfds, pl[i].fd, pl[i].revents, (unsigned long)kevp->udata, 1226 kevp->filter); 1227 1228 /* 1229 * Make noise about unclaimed events as they might indicate a bug 1230 * and can result in spurious-looking wakeups of poll(2). 1231 * 1232 * Live-locking within the system call should not happen because 1233 * the scan loop in doppoll() has an upper limit for the number 1234 * of events to process. 1235 */ 1236 if (pl[i].revents == 0 && ratecheck(&poll_lasterr, &poll_errintvl)) { 1237 printf("%s[%d]: poll index %lu fd %d events 0x%x " 1238 "filter %d/0x%x unclaimed\n", 1239 p->p_p->ps_comm, p->p_tid, i, pl[i].fd, 1240 pl[i].events, kevp->filter, kevp->flags); 1241 } 1242 1243 if (!already_seen && (pl[i].revents != 0)) 1244 return (1); 1245 1246 return (0); 1247 } 1248 1249 /* 1250 * utrace system call 1251 */ 1252 int 1253 sys_utrace(struct proc *curp, void *v, register_t *retval) 1254 { 1255 #ifdef KTRACE 1256 struct sys_utrace_args /* { 1257 syscallarg(const char *) label; 1258 syscallarg(const void *) addr; 1259 syscallarg(size_t) len; 1260 } */ *uap = v; 1261 1262 return (ktruser(curp, SCARG(uap, label), SCARG(uap, addr), 1263 SCARG(uap, len))); 1264 #else 1265 return (0); 1266 #endif 1267 } 1268