1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_capsicum.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/capsicum.h> 49 #include <sys/filedesc.h> 50 #include <sys/filio.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/lock.h> 54 #include <sys/proc.h> 55 #include <sys/signalvar.h> 56 #include <sys/socketvar.h> 57 #include <sys/uio.h> 58 #include <sys/eventfd.h> 59 #include <sys/kernel.h> 60 #include <sys/ktr.h> 61 #include <sys/limits.h> 62 #include <sys/malloc.h> 63 #include <sys/poll.h> 64 #include <sys/resourcevar.h> 65 #include <sys/selinfo.h> 66 #include <sys/sleepqueue.h> 67 #include <sys/specialfd.h> 68 #include <sys/syscallsubr.h> 69 #include <sys/sysctl.h> 70 #include <sys/sysent.h> 71 #include <sys/vnode.h> 72 #include <sys/bio.h> 73 #include <sys/buf.h> 74 #include <sys/condvar.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 79 #include <security/audit/audit.h> 80 81 /* 82 * The following macro defines how many bytes will be allocated from 83 * the stack instead of memory allocated when passing the IOCTL data 84 * structures from userspace and to the kernel. Some IOCTLs having 85 * small data structures are used very frequently and this small 86 * buffer on the stack gives a significant speedup improvement for 87 * those requests. The value of this define should be greater or equal 88 * to 64 bytes and should also be power of two. The data structure is 89 * currently hard-aligned to a 8-byte boundary on the stack. This 90 * should currently be sufficient for all supported platforms. 91 */ 92 #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */ 93 #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */ 94 95 #ifdef __LP64__ 96 static int iosize_max_clamp = 0; 97 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW, 98 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX"); 99 static int devfs_iosize_max_clamp = 1; 100 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW, 101 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices"); 102 #endif 103 104 /* 105 * Assert that the return value of read(2) and write(2) syscalls fits 106 * into a register. If not, an architecture will need to provide the 107 * usermode wrappers to reconstruct the result. 108 */ 109 CTASSERT(sizeof(register_t) >= sizeof(size_t)); 110 111 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); 112 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); 113 MALLOC_DEFINE(M_IOV, "iov", "large iov's"); 114 115 static int pollout(struct thread *, struct pollfd *, struct pollfd *, 116 u_int); 117 static int pollscan(struct thread *, struct pollfd *, u_int); 118 static int pollrescan(struct thread *); 119 static int selscan(struct thread *, fd_mask **, fd_mask **, int); 120 static int selrescan(struct thread *, fd_mask **, fd_mask **); 121 static void selfdalloc(struct thread *, void *); 122 static void selfdfree(struct seltd *, struct selfd *); 123 static int dofileread(struct thread *, int, struct file *, struct uio *, 124 off_t, int); 125 static int dofilewrite(struct thread *, int, struct file *, struct uio *, 126 off_t, int); 127 static void doselwakeup(struct selinfo *, int); 128 static void seltdinit(struct thread *); 129 static int seltdwait(struct thread *, sbintime_t, sbintime_t); 130 static void seltdclear(struct thread *); 131 132 /* 133 * One seltd per-thread allocated on demand as needed. 134 * 135 * t - protected by st_mtx 136 * k - Only accessed by curthread or read-only 137 */ 138 struct seltd { 139 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */ 140 struct selfd *st_free1; /* (k) free fd for read set. */ 141 struct selfd *st_free2; /* (k) free fd for write set. */ 142 struct mtx st_mtx; /* Protects struct seltd */ 143 struct cv st_wait; /* (t) Wait channel. */ 144 int st_flags; /* (t) SELTD_ flags. */ 145 }; 146 147 #define SELTD_PENDING 0x0001 /* We have pending events. */ 148 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */ 149 150 /* 151 * One selfd allocated per-thread per-file-descriptor. 152 * f - protected by sf_mtx 153 */ 154 struct selfd { 155 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */ 156 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */ 157 struct selinfo *sf_si; /* (f) selinfo when linked. */ 158 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */ 159 struct seltd *sf_td; /* (k) owning seltd. */ 160 void *sf_cookie; /* (k) fd or pollfd. */ 161 }; 162 163 MALLOC_DEFINE(M_SELFD, "selfd", "selfd"); 164 static struct mtx_pool *mtxpool_select; 165 166 #ifdef __LP64__ 167 size_t 168 devfs_iosize_max(void) 169 { 170 171 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 172 INT_MAX : SSIZE_MAX); 173 } 174 175 size_t 176 iosize_max(void) 177 { 178 179 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 180 INT_MAX : SSIZE_MAX); 181 } 182 #endif 183 184 #ifndef _SYS_SYSPROTO_H_ 185 struct read_args { 186 int fd; 187 void *buf; 188 size_t nbyte; 189 }; 190 #endif 191 int 192 sys_read(struct thread *td, struct read_args *uap) 193 { 194 struct uio auio; 195 struct iovec aiov; 196 int error; 197 198 if (uap->nbyte > IOSIZE_MAX) 199 return (EINVAL); 200 aiov.iov_base = uap->buf; 201 aiov.iov_len = uap->nbyte; 202 auio.uio_iov = &aiov; 203 auio.uio_iovcnt = 1; 204 auio.uio_resid = uap->nbyte; 205 auio.uio_segflg = UIO_USERSPACE; 206 error = kern_readv(td, uap->fd, &auio); 207 return (error); 208 } 209 210 /* 211 * Positioned read system call 212 */ 213 #ifndef _SYS_SYSPROTO_H_ 214 struct pread_args { 215 int fd; 216 void *buf; 217 size_t nbyte; 218 int pad; 219 off_t offset; 220 }; 221 #endif 222 int 223 sys_pread(struct thread *td, struct pread_args *uap) 224 { 225 226 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 227 } 228 229 int 230 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset) 231 { 232 struct uio auio; 233 struct iovec aiov; 234 int error; 235 236 if (nbyte > IOSIZE_MAX) 237 return (EINVAL); 238 aiov.iov_base = buf; 239 aiov.iov_len = nbyte; 240 auio.uio_iov = &aiov; 241 auio.uio_iovcnt = 1; 242 auio.uio_resid = nbyte; 243 auio.uio_segflg = UIO_USERSPACE; 244 error = kern_preadv(td, fd, &auio, offset); 245 return (error); 246 } 247 248 #if defined(COMPAT_FREEBSD6) 249 int 250 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap) 251 { 252 253 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 254 } 255 #endif 256 257 /* 258 * Scatter read system call. 259 */ 260 #ifndef _SYS_SYSPROTO_H_ 261 struct readv_args { 262 int fd; 263 struct iovec *iovp; 264 u_int iovcnt; 265 }; 266 #endif 267 int 268 sys_readv(struct thread *td, struct readv_args *uap) 269 { 270 struct uio *auio; 271 int error; 272 273 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 274 if (error) 275 return (error); 276 error = kern_readv(td, uap->fd, auio); 277 free(auio, M_IOV); 278 return (error); 279 } 280 281 int 282 kern_readv(struct thread *td, int fd, struct uio *auio) 283 { 284 struct file *fp; 285 int error; 286 287 error = fget_read(td, fd, &cap_read_rights, &fp); 288 if (error) 289 return (error); 290 error = dofileread(td, fd, fp, auio, (off_t)-1, 0); 291 fdrop(fp, td); 292 return (error); 293 } 294 295 /* 296 * Scatter positioned read system call. 297 */ 298 #ifndef _SYS_SYSPROTO_H_ 299 struct preadv_args { 300 int fd; 301 struct iovec *iovp; 302 u_int iovcnt; 303 off_t offset; 304 }; 305 #endif 306 int 307 sys_preadv(struct thread *td, struct preadv_args *uap) 308 { 309 struct uio *auio; 310 int error; 311 312 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 313 if (error) 314 return (error); 315 error = kern_preadv(td, uap->fd, auio, uap->offset); 316 free(auio, M_IOV); 317 return (error); 318 } 319 320 int 321 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset) 322 { 323 struct file *fp; 324 int error; 325 326 error = fget_read(td, fd, &cap_pread_rights, &fp); 327 if (error) 328 return (error); 329 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 330 error = ESPIPE; 331 else if (offset < 0 && 332 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 333 error = EINVAL; 334 else 335 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET); 336 fdrop(fp, td); 337 return (error); 338 } 339 340 /* 341 * Common code for readv and preadv that reads data in 342 * from a file using the passed in uio, offset, and flags. 343 */ 344 static int 345 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio, 346 off_t offset, int flags) 347 { 348 ssize_t cnt; 349 int error; 350 #ifdef KTRACE 351 struct uio *ktruio = NULL; 352 #endif 353 354 AUDIT_ARG_FD(fd); 355 356 /* Finish zero length reads right here */ 357 if (auio->uio_resid == 0) { 358 td->td_retval[0] = 0; 359 return (0); 360 } 361 auio->uio_rw = UIO_READ; 362 auio->uio_offset = offset; 363 auio->uio_td = td; 364 #ifdef KTRACE 365 if (KTRPOINT(td, KTR_GENIO)) 366 ktruio = cloneuio(auio); 367 #endif 368 cnt = auio->uio_resid; 369 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) { 370 if (auio->uio_resid != cnt && (error == ERESTART || 371 error == EINTR || error == EWOULDBLOCK)) 372 error = 0; 373 } 374 cnt -= auio->uio_resid; 375 #ifdef KTRACE 376 if (ktruio != NULL) { 377 ktruio->uio_resid = cnt; 378 ktrgenio(fd, UIO_READ, ktruio, error); 379 } 380 #endif 381 td->td_retval[0] = cnt; 382 return (error); 383 } 384 385 #ifndef _SYS_SYSPROTO_H_ 386 struct write_args { 387 int fd; 388 const void *buf; 389 size_t nbyte; 390 }; 391 #endif 392 int 393 sys_write(struct thread *td, struct write_args *uap) 394 { 395 struct uio auio; 396 struct iovec aiov; 397 int error; 398 399 if (uap->nbyte > IOSIZE_MAX) 400 return (EINVAL); 401 aiov.iov_base = (void *)(uintptr_t)uap->buf; 402 aiov.iov_len = uap->nbyte; 403 auio.uio_iov = &aiov; 404 auio.uio_iovcnt = 1; 405 auio.uio_resid = uap->nbyte; 406 auio.uio_segflg = UIO_USERSPACE; 407 error = kern_writev(td, uap->fd, &auio); 408 return (error); 409 } 410 411 /* 412 * Positioned write system call. 413 */ 414 #ifndef _SYS_SYSPROTO_H_ 415 struct pwrite_args { 416 int fd; 417 const void *buf; 418 size_t nbyte; 419 int pad; 420 off_t offset; 421 }; 422 #endif 423 int 424 sys_pwrite(struct thread *td, struct pwrite_args *uap) 425 { 426 427 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 428 } 429 430 int 431 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte, 432 off_t offset) 433 { 434 struct uio auio; 435 struct iovec aiov; 436 int error; 437 438 if (nbyte > IOSIZE_MAX) 439 return (EINVAL); 440 aiov.iov_base = (void *)(uintptr_t)buf; 441 aiov.iov_len = nbyte; 442 auio.uio_iov = &aiov; 443 auio.uio_iovcnt = 1; 444 auio.uio_resid = nbyte; 445 auio.uio_segflg = UIO_USERSPACE; 446 error = kern_pwritev(td, fd, &auio, offset); 447 return (error); 448 } 449 450 #if defined(COMPAT_FREEBSD6) 451 int 452 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap) 453 { 454 455 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 456 } 457 #endif 458 459 /* 460 * Gather write system call. 461 */ 462 #ifndef _SYS_SYSPROTO_H_ 463 struct writev_args { 464 int fd; 465 struct iovec *iovp; 466 u_int iovcnt; 467 }; 468 #endif 469 int 470 sys_writev(struct thread *td, struct writev_args *uap) 471 { 472 struct uio *auio; 473 int error; 474 475 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 476 if (error) 477 return (error); 478 error = kern_writev(td, uap->fd, auio); 479 free(auio, M_IOV); 480 return (error); 481 } 482 483 int 484 kern_writev(struct thread *td, int fd, struct uio *auio) 485 { 486 struct file *fp; 487 int error; 488 489 error = fget_write(td, fd, &cap_write_rights, &fp); 490 if (error) 491 return (error); 492 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0); 493 fdrop(fp, td); 494 return (error); 495 } 496 497 /* 498 * Gather positioned write system call. 499 */ 500 #ifndef _SYS_SYSPROTO_H_ 501 struct pwritev_args { 502 int fd; 503 struct iovec *iovp; 504 u_int iovcnt; 505 off_t offset; 506 }; 507 #endif 508 int 509 sys_pwritev(struct thread *td, struct pwritev_args *uap) 510 { 511 struct uio *auio; 512 int error; 513 514 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 515 if (error) 516 return (error); 517 error = kern_pwritev(td, uap->fd, auio, uap->offset); 518 free(auio, M_IOV); 519 return (error); 520 } 521 522 int 523 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset) 524 { 525 struct file *fp; 526 int error; 527 528 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 529 if (error) 530 return (error); 531 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 532 error = ESPIPE; 533 else if (offset < 0 && 534 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 535 error = EINVAL; 536 else 537 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET); 538 fdrop(fp, td); 539 return (error); 540 } 541 542 /* 543 * Common code for writev and pwritev that writes data to 544 * a file using the passed in uio, offset, and flags. 545 */ 546 static int 547 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio, 548 off_t offset, int flags) 549 { 550 ssize_t cnt; 551 int error; 552 #ifdef KTRACE 553 struct uio *ktruio = NULL; 554 #endif 555 556 AUDIT_ARG_FD(fd); 557 auio->uio_rw = UIO_WRITE; 558 auio->uio_td = td; 559 auio->uio_offset = offset; 560 #ifdef KTRACE 561 if (KTRPOINT(td, KTR_GENIO)) 562 ktruio = cloneuio(auio); 563 #endif 564 cnt = auio->uio_resid; 565 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) { 566 if (auio->uio_resid != cnt && (error == ERESTART || 567 error == EINTR || error == EWOULDBLOCK)) 568 error = 0; 569 /* Socket layer is responsible for issuing SIGPIPE. */ 570 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) { 571 PROC_LOCK(td->td_proc); 572 tdsignal(td, SIGPIPE); 573 PROC_UNLOCK(td->td_proc); 574 } 575 } 576 cnt -= auio->uio_resid; 577 #ifdef KTRACE 578 if (ktruio != NULL) { 579 ktruio->uio_resid = cnt; 580 ktrgenio(fd, UIO_WRITE, ktruio, error); 581 } 582 #endif 583 td->td_retval[0] = cnt; 584 return (error); 585 } 586 587 /* 588 * Truncate a file given a file descriptor. 589 * 590 * Can't use fget_write() here, since must return EINVAL and not EBADF if the 591 * descriptor isn't writable. 592 */ 593 int 594 kern_ftruncate(struct thread *td, int fd, off_t length) 595 { 596 struct file *fp; 597 int error; 598 599 AUDIT_ARG_FD(fd); 600 if (length < 0) 601 return (EINVAL); 602 error = fget(td, fd, &cap_ftruncate_rights, &fp); 603 if (error) 604 return (error); 605 AUDIT_ARG_FILE(td->td_proc, fp); 606 if (!(fp->f_flag & FWRITE)) { 607 fdrop(fp, td); 608 return (EINVAL); 609 } 610 error = fo_truncate(fp, length, td->td_ucred, td); 611 fdrop(fp, td); 612 return (error); 613 } 614 615 #ifndef _SYS_SYSPROTO_H_ 616 struct ftruncate_args { 617 int fd; 618 int pad; 619 off_t length; 620 }; 621 #endif 622 int 623 sys_ftruncate(struct thread *td, struct ftruncate_args *uap) 624 { 625 626 return (kern_ftruncate(td, uap->fd, uap->length)); 627 } 628 629 #if defined(COMPAT_43) 630 #ifndef _SYS_SYSPROTO_H_ 631 struct oftruncate_args { 632 int fd; 633 long length; 634 }; 635 #endif 636 int 637 oftruncate(struct thread *td, struct oftruncate_args *uap) 638 { 639 640 return (kern_ftruncate(td, uap->fd, uap->length)); 641 } 642 #endif /* COMPAT_43 */ 643 644 #ifndef _SYS_SYSPROTO_H_ 645 struct ioctl_args { 646 int fd; 647 u_long com; 648 caddr_t data; 649 }; 650 #endif 651 /* ARGSUSED */ 652 int 653 sys_ioctl(struct thread *td, struct ioctl_args *uap) 654 { 655 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN); 656 uint32_t com; 657 int arg, error; 658 u_int size; 659 caddr_t data; 660 661 #ifdef INVARIANTS 662 if (uap->com > 0xffffffff) { 663 printf( 664 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n", 665 td->td_proc->p_pid, td->td_name, uap->com); 666 } 667 #endif 668 com = (uint32_t)uap->com; 669 670 /* 671 * Interpret high order word to find amount of data to be 672 * copied to/from the user's address space. 673 */ 674 size = IOCPARM_LEN(com); 675 if ((size > IOCPARM_MAX) || 676 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) || 677 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43) 678 ((com & IOC_OUT) && size == 0) || 679 #else 680 ((com & (IOC_IN | IOC_OUT)) && size == 0) || 681 #endif 682 ((com & IOC_VOID) && size > 0 && size != sizeof(int))) 683 return (ENOTTY); 684 685 if (size > 0) { 686 if (com & IOC_VOID) { 687 /* Integer argument. */ 688 arg = (intptr_t)uap->data; 689 data = (void *)&arg; 690 size = 0; 691 } else { 692 if (size > SYS_IOCTL_SMALL_SIZE) 693 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK); 694 else 695 data = smalldata; 696 } 697 } else 698 data = (void *)&uap->data; 699 if (com & IOC_IN) { 700 error = copyin(uap->data, data, (u_int)size); 701 if (error != 0) 702 goto out; 703 } else if (com & IOC_OUT) { 704 /* 705 * Zero the buffer so the user always 706 * gets back something deterministic. 707 */ 708 bzero(data, size); 709 } 710 711 error = kern_ioctl(td, uap->fd, com, data); 712 713 if (error == 0 && (com & IOC_OUT)) 714 error = copyout(data, uap->data, (u_int)size); 715 716 out: 717 if (size > SYS_IOCTL_SMALL_SIZE) 718 free(data, M_IOCTLOPS); 719 return (error); 720 } 721 722 int 723 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data) 724 { 725 struct file *fp; 726 struct filedesc *fdp; 727 int error, tmp, locked; 728 729 AUDIT_ARG_FD(fd); 730 AUDIT_ARG_CMD(com); 731 732 fdp = td->td_proc->p_fd; 733 734 switch (com) { 735 case FIONCLEX: 736 case FIOCLEX: 737 FILEDESC_XLOCK(fdp); 738 locked = LA_XLOCKED; 739 break; 740 default: 741 #ifdef CAPABILITIES 742 FILEDESC_SLOCK(fdp); 743 locked = LA_SLOCKED; 744 #else 745 locked = LA_UNLOCKED; 746 #endif 747 break; 748 } 749 750 #ifdef CAPABILITIES 751 if ((fp = fget_noref(fdp, fd)) == NULL) { 752 error = EBADF; 753 goto out; 754 } 755 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) { 756 fp = NULL; /* fhold() was not called yet */ 757 goto out; 758 } 759 if (!fhold(fp)) { 760 error = EBADF; 761 fp = NULL; 762 goto out; 763 } 764 if (locked == LA_SLOCKED) { 765 FILEDESC_SUNLOCK(fdp); 766 locked = LA_UNLOCKED; 767 } 768 #else 769 error = fget(td, fd, &cap_ioctl_rights, &fp); 770 if (error != 0) { 771 fp = NULL; 772 goto out; 773 } 774 #endif 775 if ((fp->f_flag & (FREAD | FWRITE)) == 0) { 776 error = EBADF; 777 goto out; 778 } 779 780 switch (com) { 781 case FIONCLEX: 782 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE; 783 goto out; 784 case FIOCLEX: 785 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE; 786 goto out; 787 case FIONBIO: 788 if ((tmp = *(int *)data)) 789 atomic_set_int(&fp->f_flag, FNONBLOCK); 790 else 791 atomic_clear_int(&fp->f_flag, FNONBLOCK); 792 data = (void *)&tmp; 793 break; 794 case FIOASYNC: 795 if ((tmp = *(int *)data)) 796 atomic_set_int(&fp->f_flag, FASYNC); 797 else 798 atomic_clear_int(&fp->f_flag, FASYNC); 799 data = (void *)&tmp; 800 break; 801 } 802 803 error = fo_ioctl(fp, com, data, td->td_ucred, td); 804 out: 805 switch (locked) { 806 case LA_XLOCKED: 807 FILEDESC_XUNLOCK(fdp); 808 break; 809 #ifdef CAPABILITIES 810 case LA_SLOCKED: 811 FILEDESC_SUNLOCK(fdp); 812 break; 813 #endif 814 default: 815 FILEDESC_UNLOCK_ASSERT(fdp); 816 break; 817 } 818 if (fp != NULL) 819 fdrop(fp, td); 820 return (error); 821 } 822 823 int 824 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) 825 { 826 int error; 827 828 error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len); 829 return (kern_posix_error(td, error)); 830 } 831 832 int 833 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len) 834 { 835 struct file *fp; 836 int error; 837 838 AUDIT_ARG_FD(fd); 839 if (offset < 0 || len <= 0) 840 return (EINVAL); 841 /* Check for wrap. */ 842 if (offset > OFF_MAX - len) 843 return (EFBIG); 844 AUDIT_ARG_FD(fd); 845 error = fget(td, fd, &cap_pwrite_rights, &fp); 846 if (error != 0) 847 return (error); 848 AUDIT_ARG_FILE(td->td_proc, fp); 849 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 850 error = ESPIPE; 851 goto out; 852 } 853 if ((fp->f_flag & FWRITE) == 0) { 854 error = EBADF; 855 goto out; 856 } 857 858 error = fo_fallocate(fp, offset, len, td); 859 out: 860 fdrop(fp, td); 861 return (error); 862 } 863 864 int 865 sys_fspacectl(struct thread *td, struct fspacectl_args *uap) 866 { 867 struct spacectl_range rqsr, rmsr; 868 int error, cerror; 869 870 error = copyin(uap->rqsr, &rqsr, sizeof(rqsr)); 871 if (error != 0) 872 return (error); 873 874 error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags, 875 &rmsr); 876 if (uap->rmsr != NULL) { 877 cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr)); 878 if (error == 0) 879 error = cerror; 880 } 881 return (error); 882 } 883 884 int 885 kern_fspacectl(struct thread *td, int fd, int cmd, 886 const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp) 887 { 888 struct file *fp; 889 struct spacectl_range rmsr; 890 int error; 891 892 AUDIT_ARG_FD(fd); 893 AUDIT_ARG_CMD(cmd); 894 AUDIT_ARG_FFLAGS(flags); 895 896 if (rqsr == NULL) 897 return (EINVAL); 898 rmsr = *rqsr; 899 if (rmsrp != NULL) 900 *rmsrp = rmsr; 901 902 if (cmd != SPACECTL_DEALLOC || 903 rqsr->r_offset < 0 || rqsr->r_len <= 0 || 904 rqsr->r_offset > OFF_MAX - rqsr->r_len || 905 (flags & ~SPACECTL_F_SUPPORTED) != 0) 906 return (EINVAL); 907 908 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 909 if (error != 0) 910 return (error); 911 AUDIT_ARG_FILE(td->td_proc, fp); 912 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 913 error = ESPIPE; 914 goto out; 915 } 916 if ((fp->f_flag & FWRITE) == 0) { 917 error = EBADF; 918 goto out; 919 } 920 921 error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags, 922 td->td_ucred, td); 923 /* fspacectl is not restarted after signals if the file is modified. */ 924 if (rmsr.r_len != rqsr->r_len && (error == ERESTART || 925 error == EINTR || error == EWOULDBLOCK)) 926 error = 0; 927 if (rmsrp != NULL) 928 *rmsrp = rmsr; 929 out: 930 fdrop(fp, td); 931 return (error); 932 } 933 934 int 935 kern_specialfd(struct thread *td, int type, void *arg) 936 { 937 struct file *fp; 938 struct specialfd_eventfd *ae; 939 int error, fd, fflags; 940 941 fflags = 0; 942 error = falloc_noinstall(td, &fp); 943 if (error != 0) 944 return (error); 945 946 switch (type) { 947 case SPECIALFD_EVENTFD: 948 ae = arg; 949 if ((ae->flags & EFD_CLOEXEC) != 0) 950 fflags |= O_CLOEXEC; 951 error = eventfd_create_file(td, fp, ae->initval, ae->flags); 952 break; 953 default: 954 error = EINVAL; 955 break; 956 } 957 958 if (error == 0) 959 error = finstall(td, fp, &fd, fflags, NULL); 960 fdrop(fp, td); 961 if (error == 0) 962 td->td_retval[0] = fd; 963 return (error); 964 } 965 966 int 967 sys___specialfd(struct thread *td, struct __specialfd_args *args) 968 { 969 struct specialfd_eventfd ae; 970 int error; 971 972 switch (args->type) { 973 case SPECIALFD_EVENTFD: 974 if (args->len != sizeof(struct specialfd_eventfd)) { 975 error = EINVAL; 976 break; 977 } 978 error = copyin(args->req, &ae, sizeof(ae)); 979 if (error != 0) 980 break; 981 if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK | 982 EFD_SEMAPHORE)) != 0) { 983 error = EINVAL; 984 break; 985 } 986 error = kern_specialfd(td, args->type, &ae); 987 break; 988 default: 989 error = EINVAL; 990 break; 991 } 992 return (error); 993 } 994 995 int 996 poll_no_poll(int events) 997 { 998 /* 999 * Return true for read/write. If the user asked for something 1000 * special, return POLLNVAL, so that clients have a way of 1001 * determining reliably whether or not the extended 1002 * functionality is present without hard-coding knowledge 1003 * of specific filesystem implementations. 1004 */ 1005 if (events & ~POLLSTANDARD) 1006 return (POLLNVAL); 1007 1008 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 1009 } 1010 1011 int 1012 sys_pselect(struct thread *td, struct pselect_args *uap) 1013 { 1014 struct timespec ts; 1015 struct timeval tv, *tvp; 1016 sigset_t set, *uset; 1017 int error; 1018 1019 if (uap->ts != NULL) { 1020 error = copyin(uap->ts, &ts, sizeof(ts)); 1021 if (error != 0) 1022 return (error); 1023 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1024 tvp = &tv; 1025 } else 1026 tvp = NULL; 1027 if (uap->sm != NULL) { 1028 error = copyin(uap->sm, &set, sizeof(set)); 1029 if (error != 0) 1030 return (error); 1031 uset = &set; 1032 } else 1033 uset = NULL; 1034 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1035 uset, NFDBITS)); 1036 } 1037 1038 int 1039 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex, 1040 struct timeval *tvp, sigset_t *uset, int abi_nfdbits) 1041 { 1042 int error; 1043 1044 if (uset != NULL) { 1045 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1046 &td->td_oldsigmask, 0); 1047 if (error != 0) 1048 return (error); 1049 td->td_pflags |= TDP_OLDMASK; 1050 /* 1051 * Make sure that ast() is called on return to 1052 * usermode and TDP_OLDMASK is cleared, restoring old 1053 * sigmask. 1054 */ 1055 thread_lock(td); 1056 td->td_flags |= TDF_ASTPENDING; 1057 thread_unlock(td); 1058 } 1059 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits); 1060 return (error); 1061 } 1062 1063 #ifndef _SYS_SYSPROTO_H_ 1064 struct select_args { 1065 int nd; 1066 fd_set *in, *ou, *ex; 1067 struct timeval *tv; 1068 }; 1069 #endif 1070 int 1071 sys_select(struct thread *td, struct select_args *uap) 1072 { 1073 struct timeval tv, *tvp; 1074 int error; 1075 1076 if (uap->tv != NULL) { 1077 error = copyin(uap->tv, &tv, sizeof(tv)); 1078 if (error) 1079 return (error); 1080 tvp = &tv; 1081 } else 1082 tvp = NULL; 1083 1084 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1085 NFDBITS)); 1086 } 1087 1088 /* 1089 * In the unlikely case when user specified n greater then the last 1090 * open file descriptor, check that no bits are set after the last 1091 * valid fd. We must return EBADF if any is set. 1092 * 1093 * There are applications that rely on the behaviour. 1094 * 1095 * nd is fd_nfiles. 1096 */ 1097 static int 1098 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits) 1099 { 1100 char *addr, *oaddr; 1101 int b, i, res; 1102 uint8_t bits; 1103 1104 if (nd >= ndu || fd_in == NULL) 1105 return (0); 1106 1107 oaddr = NULL; 1108 bits = 0; /* silence gcc */ 1109 for (i = nd; i < ndu; i++) { 1110 b = i / NBBY; 1111 #if BYTE_ORDER == LITTLE_ENDIAN 1112 addr = (char *)fd_in + b; 1113 #else 1114 addr = (char *)fd_in; 1115 if (abi_nfdbits == NFDBITS) { 1116 addr += rounddown(b, sizeof(fd_mask)) + 1117 sizeof(fd_mask) - 1 - b % sizeof(fd_mask); 1118 } else { 1119 addr += rounddown(b, sizeof(uint32_t)) + 1120 sizeof(uint32_t) - 1 - b % sizeof(uint32_t); 1121 } 1122 #endif 1123 if (addr != oaddr) { 1124 res = fubyte(addr); 1125 if (res == -1) 1126 return (EFAULT); 1127 oaddr = addr; 1128 bits = res; 1129 } 1130 if ((bits & (1 << (i % NBBY))) != 0) 1131 return (EBADF); 1132 } 1133 return (0); 1134 } 1135 1136 int 1137 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou, 1138 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits) 1139 { 1140 struct filedesc *fdp; 1141 /* 1142 * The magic 2048 here is chosen to be just enough for FD_SETSIZE 1143 * infds with the new FD_SETSIZE of 1024, and more than enough for 1144 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE 1145 * of 256. 1146 */ 1147 fd_mask s_selbits[howmany(2048, NFDBITS)]; 1148 fd_mask *ibits[3], *obits[3], *selbits, *sbp; 1149 struct timeval rtv; 1150 sbintime_t asbt, precision, rsbt; 1151 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits; 1152 int error, lf, ndu; 1153 1154 if (nd < 0) 1155 return (EINVAL); 1156 fdp = td->td_proc->p_fd; 1157 ndu = nd; 1158 lf = fdp->fd_nfiles; 1159 if (nd > lf) 1160 nd = lf; 1161 1162 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits); 1163 if (error != 0) 1164 return (error); 1165 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits); 1166 if (error != 0) 1167 return (error); 1168 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits); 1169 if (error != 0) 1170 return (error); 1171 1172 /* 1173 * Allocate just enough bits for the non-null fd_sets. Use the 1174 * preallocated auto buffer if possible. 1175 */ 1176 nfdbits = roundup(nd, NFDBITS); 1177 ncpbytes = nfdbits / NBBY; 1178 ncpubytes = roundup(nd, abi_nfdbits) / NBBY; 1179 nbufbytes = 0; 1180 if (fd_in != NULL) 1181 nbufbytes += 2 * ncpbytes; 1182 if (fd_ou != NULL) 1183 nbufbytes += 2 * ncpbytes; 1184 if (fd_ex != NULL) 1185 nbufbytes += 2 * ncpbytes; 1186 if (nbufbytes <= sizeof s_selbits) 1187 selbits = &s_selbits[0]; 1188 else 1189 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK); 1190 1191 /* 1192 * Assign pointers into the bit buffers and fetch the input bits. 1193 * Put the output buffers together so that they can be bzeroed 1194 * together. 1195 */ 1196 sbp = selbits; 1197 #define getbits(name, x) \ 1198 do { \ 1199 if (name == NULL) { \ 1200 ibits[x] = NULL; \ 1201 obits[x] = NULL; \ 1202 } else { \ 1203 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ 1204 obits[x] = sbp; \ 1205 sbp += ncpbytes / sizeof *sbp; \ 1206 error = copyin(name, ibits[x], ncpubytes); \ 1207 if (error != 0) \ 1208 goto done; \ 1209 if (ncpbytes != ncpubytes) \ 1210 bzero((char *)ibits[x] + ncpubytes, \ 1211 ncpbytes - ncpubytes); \ 1212 } \ 1213 } while (0) 1214 getbits(fd_in, 0); 1215 getbits(fd_ou, 1); 1216 getbits(fd_ex, 2); 1217 #undef getbits 1218 1219 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__) 1220 /* 1221 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS, 1222 * we are running under 32-bit emulation. This should be more 1223 * generic. 1224 */ 1225 #define swizzle_fdset(bits) \ 1226 if (abi_nfdbits != NFDBITS && bits != NULL) { \ 1227 int i; \ 1228 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \ 1229 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \ 1230 } 1231 #else 1232 #define swizzle_fdset(bits) 1233 #endif 1234 1235 /* Make sure the bit order makes it through an ABI transition */ 1236 swizzle_fdset(ibits[0]); 1237 swizzle_fdset(ibits[1]); 1238 swizzle_fdset(ibits[2]); 1239 1240 if (nbufbytes != 0) 1241 bzero(selbits, nbufbytes / 2); 1242 1243 precision = 0; 1244 if (tvp != NULL) { 1245 rtv = *tvp; 1246 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1247 rtv.tv_usec >= 1000000) { 1248 error = EINVAL; 1249 goto done; 1250 } 1251 if (!timevalisset(&rtv)) 1252 asbt = 0; 1253 else if (rtv.tv_sec <= INT32_MAX) { 1254 rsbt = tvtosbt(rtv); 1255 precision = rsbt; 1256 precision >>= tc_precexp; 1257 if (TIMESEL(&asbt, rsbt)) 1258 asbt += tc_tick_sbt; 1259 if (asbt <= SBT_MAX - rsbt) 1260 asbt += rsbt; 1261 else 1262 asbt = -1; 1263 } else 1264 asbt = -1; 1265 } else 1266 asbt = -1; 1267 seltdinit(td); 1268 /* Iterate until the timeout expires or descriptors become ready. */ 1269 for (;;) { 1270 error = selscan(td, ibits, obits, nd); 1271 if (error || td->td_retval[0] != 0) 1272 break; 1273 error = seltdwait(td, asbt, precision); 1274 if (error) 1275 break; 1276 error = selrescan(td, ibits, obits); 1277 if (error || td->td_retval[0] != 0) 1278 break; 1279 } 1280 seltdclear(td); 1281 1282 done: 1283 /* select is not restarted after signals... */ 1284 if (error == ERESTART) 1285 error = EINTR; 1286 if (error == EWOULDBLOCK) 1287 error = 0; 1288 1289 /* swizzle bit order back, if necessary */ 1290 swizzle_fdset(obits[0]); 1291 swizzle_fdset(obits[1]); 1292 swizzle_fdset(obits[2]); 1293 #undef swizzle_fdset 1294 1295 #define putbits(name, x) \ 1296 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \ 1297 error = error2; 1298 if (error == 0) { 1299 int error2; 1300 1301 putbits(fd_in, 0); 1302 putbits(fd_ou, 1); 1303 putbits(fd_ex, 2); 1304 #undef putbits 1305 } 1306 if (selbits != &s_selbits[0]) 1307 free(selbits, M_SELECT); 1308 1309 return (error); 1310 } 1311 /* 1312 * Convert a select bit set to poll flags. 1313 * 1314 * The backend always returns POLLHUP/POLLERR if appropriate and we 1315 * return this as a set bit in any set. 1316 */ 1317 static const int select_flags[3] = { 1318 POLLRDNORM | POLLHUP | POLLERR, 1319 POLLWRNORM | POLLHUP | POLLERR, 1320 POLLRDBAND | POLLERR 1321 }; 1322 1323 /* 1324 * Compute the fo_poll flags required for a fd given by the index and 1325 * bit position in the fd_mask array. 1326 */ 1327 static __inline int 1328 selflags(fd_mask **ibits, int idx, fd_mask bit) 1329 { 1330 int flags; 1331 int msk; 1332 1333 flags = 0; 1334 for (msk = 0; msk < 3; msk++) { 1335 if (ibits[msk] == NULL) 1336 continue; 1337 if ((ibits[msk][idx] & bit) == 0) 1338 continue; 1339 flags |= select_flags[msk]; 1340 } 1341 return (flags); 1342 } 1343 1344 /* 1345 * Set the appropriate output bits given a mask of fired events and the 1346 * input bits originally requested. 1347 */ 1348 static __inline int 1349 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events) 1350 { 1351 int msk; 1352 int n; 1353 1354 n = 0; 1355 for (msk = 0; msk < 3; msk++) { 1356 if ((events & select_flags[msk]) == 0) 1357 continue; 1358 if (ibits[msk] == NULL) 1359 continue; 1360 if ((ibits[msk][idx] & bit) == 0) 1361 continue; 1362 /* 1363 * XXX Check for a duplicate set. This can occur because a 1364 * socket calls selrecord() twice for each poll() call 1365 * resulting in two selfds per real fd. selrescan() will 1366 * call selsetbits twice as a result. 1367 */ 1368 if ((obits[msk][idx] & bit) != 0) 1369 continue; 1370 obits[msk][idx] |= bit; 1371 n++; 1372 } 1373 1374 return (n); 1375 } 1376 1377 /* 1378 * Traverse the list of fds attached to this thread's seltd and check for 1379 * completion. 1380 */ 1381 static int 1382 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits) 1383 { 1384 struct filedesc *fdp; 1385 struct selinfo *si; 1386 struct seltd *stp; 1387 struct selfd *sfp; 1388 struct selfd *sfn; 1389 struct file *fp; 1390 fd_mask bit; 1391 int fd, ev, n, idx; 1392 int error; 1393 bool only_user; 1394 1395 fdp = td->td_proc->p_fd; 1396 stp = td->td_sel; 1397 n = 0; 1398 only_user = FILEDESC_IS_ONLY_USER(fdp); 1399 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1400 fd = (int)(uintptr_t)sfp->sf_cookie; 1401 si = sfp->sf_si; 1402 selfdfree(stp, sfp); 1403 /* If the selinfo wasn't cleared the event didn't fire. */ 1404 if (si != NULL) 1405 continue; 1406 if (only_user) 1407 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1408 else 1409 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1410 if (__predict_false(error != 0)) 1411 return (error); 1412 idx = fd / NFDBITS; 1413 bit = (fd_mask)1 << (fd % NFDBITS); 1414 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td); 1415 if (only_user) 1416 fput_only_user(fdp, fp); 1417 else 1418 fdrop(fp, td); 1419 if (ev != 0) 1420 n += selsetbits(ibits, obits, idx, bit, ev); 1421 } 1422 stp->st_flags = 0; 1423 td->td_retval[0] = n; 1424 return (0); 1425 } 1426 1427 /* 1428 * Perform the initial filedescriptor scan and register ourselves with 1429 * each selinfo. 1430 */ 1431 static int 1432 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd) 1433 { 1434 struct filedesc *fdp; 1435 struct file *fp; 1436 fd_mask bit; 1437 int ev, flags, end, fd; 1438 int n, idx; 1439 int error; 1440 bool only_user; 1441 1442 fdp = td->td_proc->p_fd; 1443 n = 0; 1444 only_user = FILEDESC_IS_ONLY_USER(fdp); 1445 for (idx = 0, fd = 0; fd < nfd; idx++) { 1446 end = imin(fd + NFDBITS, nfd); 1447 for (bit = 1; fd < end; bit <<= 1, fd++) { 1448 /* Compute the list of events we're interested in. */ 1449 flags = selflags(ibits, idx, bit); 1450 if (flags == 0) 1451 continue; 1452 if (only_user) 1453 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1454 else 1455 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1456 if (__predict_false(error != 0)) 1457 return (error); 1458 selfdalloc(td, (void *)(uintptr_t)fd); 1459 ev = fo_poll(fp, flags, td->td_ucred, td); 1460 if (only_user) 1461 fput_only_user(fdp, fp); 1462 else 1463 fdrop(fp, td); 1464 if (ev != 0) 1465 n += selsetbits(ibits, obits, idx, bit, ev); 1466 } 1467 } 1468 1469 td->td_retval[0] = n; 1470 return (0); 1471 } 1472 1473 int 1474 sys_poll(struct thread *td, struct poll_args *uap) 1475 { 1476 struct timespec ts, *tsp; 1477 1478 if (uap->timeout != INFTIM) { 1479 if (uap->timeout < 0) 1480 return (EINVAL); 1481 ts.tv_sec = uap->timeout / 1000; 1482 ts.tv_nsec = (uap->timeout % 1000) * 1000000; 1483 tsp = &ts; 1484 } else 1485 tsp = NULL; 1486 1487 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL)); 1488 } 1489 1490 /* 1491 * kfds points to an array in the kernel. 1492 */ 1493 int 1494 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds, 1495 struct timespec *tsp, sigset_t *uset) 1496 { 1497 sbintime_t sbt, precision, tmp; 1498 time_t over; 1499 struct timespec ts; 1500 int error; 1501 1502 precision = 0; 1503 if (tsp != NULL) { 1504 if (tsp->tv_sec < 0) 1505 return (EINVAL); 1506 if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000) 1507 return (EINVAL); 1508 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1509 sbt = 0; 1510 else { 1511 ts = *tsp; 1512 if (ts.tv_sec > INT32_MAX / 2) { 1513 over = ts.tv_sec - INT32_MAX / 2; 1514 ts.tv_sec -= over; 1515 } else 1516 over = 0; 1517 tmp = tstosbt(ts); 1518 precision = tmp; 1519 precision >>= tc_precexp; 1520 if (TIMESEL(&sbt, tmp)) 1521 sbt += tc_tick_sbt; 1522 sbt += tmp; 1523 } 1524 } else 1525 sbt = -1; 1526 1527 if (uset != NULL) { 1528 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1529 &td->td_oldsigmask, 0); 1530 if (error) 1531 return (error); 1532 td->td_pflags |= TDP_OLDMASK; 1533 /* 1534 * Make sure that ast() is called on return to 1535 * usermode and TDP_OLDMASK is cleared, restoring old 1536 * sigmask. 1537 */ 1538 thread_lock(td); 1539 td->td_flags |= TDF_ASTPENDING; 1540 thread_unlock(td); 1541 } 1542 1543 seltdinit(td); 1544 /* Iterate until the timeout expires or descriptors become ready. */ 1545 for (;;) { 1546 error = pollscan(td, kfds, nfds); 1547 if (error || td->td_retval[0] != 0) 1548 break; 1549 error = seltdwait(td, sbt, precision); 1550 if (error) 1551 break; 1552 error = pollrescan(td); 1553 if (error || td->td_retval[0] != 0) 1554 break; 1555 } 1556 seltdclear(td); 1557 1558 /* poll is not restarted after signals... */ 1559 if (error == ERESTART) 1560 error = EINTR; 1561 if (error == EWOULDBLOCK) 1562 error = 0; 1563 return (error); 1564 } 1565 1566 int 1567 sys_ppoll(struct thread *td, struct ppoll_args *uap) 1568 { 1569 struct timespec ts, *tsp; 1570 sigset_t set, *ssp; 1571 int error; 1572 1573 if (uap->ts != NULL) { 1574 error = copyin(uap->ts, &ts, sizeof(ts)); 1575 if (error) 1576 return (error); 1577 tsp = &ts; 1578 } else 1579 tsp = NULL; 1580 if (uap->set != NULL) { 1581 error = copyin(uap->set, &set, sizeof(set)); 1582 if (error) 1583 return (error); 1584 ssp = &set; 1585 } else 1586 ssp = NULL; 1587 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp)); 1588 } 1589 1590 /* 1591 * ufds points to an array in user space. 1592 */ 1593 int 1594 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds, 1595 struct timespec *tsp, sigset_t *set) 1596 { 1597 struct pollfd *kfds; 1598 struct pollfd stackfds[32]; 1599 int error; 1600 1601 if (kern_poll_maxfds(nfds)) 1602 return (EINVAL); 1603 if (nfds > nitems(stackfds)) 1604 kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK); 1605 else 1606 kfds = stackfds; 1607 error = copyin(ufds, kfds, nfds * sizeof(*kfds)); 1608 if (error != 0) 1609 goto out; 1610 1611 error = kern_poll_kfds(td, kfds, nfds, tsp, set); 1612 if (error == 0) 1613 error = pollout(td, kfds, ufds, nfds); 1614 1615 out: 1616 if (nfds > nitems(stackfds)) 1617 free(kfds, M_TEMP); 1618 return (error); 1619 } 1620 1621 bool 1622 kern_poll_maxfds(u_int nfds) 1623 { 1624 1625 /* 1626 * This is kinda bogus. We have fd limits, but that is not 1627 * really related to the size of the pollfd array. Make sure 1628 * we let the process use at least FD_SETSIZE entries and at 1629 * least enough for the system-wide limits. We want to be reasonably 1630 * safe, but not overly restrictive. 1631 */ 1632 return (nfds > maxfilesperproc && nfds > FD_SETSIZE); 1633 } 1634 1635 static int 1636 pollrescan(struct thread *td) 1637 { 1638 struct seltd *stp; 1639 struct selfd *sfp; 1640 struct selfd *sfn; 1641 struct selinfo *si; 1642 struct filedesc *fdp; 1643 struct file *fp; 1644 struct pollfd *fd; 1645 int n, error; 1646 bool only_user; 1647 1648 n = 0; 1649 fdp = td->td_proc->p_fd; 1650 stp = td->td_sel; 1651 only_user = FILEDESC_IS_ONLY_USER(fdp); 1652 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1653 fd = (struct pollfd *)sfp->sf_cookie; 1654 si = sfp->sf_si; 1655 selfdfree(stp, sfp); 1656 /* If the selinfo wasn't cleared the event didn't fire. */ 1657 if (si != NULL) 1658 continue; 1659 if (only_user) 1660 error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp); 1661 else 1662 error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp); 1663 if (__predict_false(error != 0)) { 1664 fd->revents = POLLNVAL; 1665 n++; 1666 continue; 1667 } 1668 /* 1669 * Note: backend also returns POLLHUP and 1670 * POLLERR if appropriate. 1671 */ 1672 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td); 1673 if (only_user) 1674 fput_only_user(fdp, fp); 1675 else 1676 fdrop(fp, td); 1677 if (fd->revents != 0) 1678 n++; 1679 } 1680 stp->st_flags = 0; 1681 td->td_retval[0] = n; 1682 return (0); 1683 } 1684 1685 static int 1686 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd) 1687 { 1688 int error = 0; 1689 u_int i = 0; 1690 u_int n = 0; 1691 1692 for (i = 0; i < nfd; i++) { 1693 error = copyout(&fds->revents, &ufds->revents, 1694 sizeof(ufds->revents)); 1695 if (error) 1696 return (error); 1697 if (fds->revents != 0) 1698 n++; 1699 fds++; 1700 ufds++; 1701 } 1702 td->td_retval[0] = n; 1703 return (0); 1704 } 1705 1706 static int 1707 pollscan(struct thread *td, struct pollfd *fds, u_int nfd) 1708 { 1709 struct filedesc *fdp; 1710 struct file *fp; 1711 int i, n, error; 1712 bool only_user; 1713 1714 n = 0; 1715 fdp = td->td_proc->p_fd; 1716 only_user = FILEDESC_IS_ONLY_USER(fdp); 1717 for (i = 0; i < nfd; i++, fds++) { 1718 if (fds->fd < 0) { 1719 fds->revents = 0; 1720 continue; 1721 } 1722 if (only_user) 1723 error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp); 1724 else 1725 error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp); 1726 if (__predict_false(error != 0)) { 1727 fds->revents = POLLNVAL; 1728 n++; 1729 continue; 1730 } 1731 /* 1732 * Note: backend also returns POLLHUP and 1733 * POLLERR if appropriate. 1734 */ 1735 selfdalloc(td, fds); 1736 fds->revents = fo_poll(fp, fds->events, 1737 td->td_ucred, td); 1738 if (only_user) 1739 fput_only_user(fdp, fp); 1740 else 1741 fdrop(fp, td); 1742 /* 1743 * POSIX requires POLLOUT to be never 1744 * set simultaneously with POLLHUP. 1745 */ 1746 if ((fds->revents & POLLHUP) != 0) 1747 fds->revents &= ~POLLOUT; 1748 1749 if (fds->revents != 0) 1750 n++; 1751 } 1752 td->td_retval[0] = n; 1753 return (0); 1754 } 1755 1756 /* 1757 * XXX This was created specifically to support netncp and netsmb. This 1758 * allows the caller to specify a socket to wait for events on. It returns 1759 * 0 if any events matched and an error otherwise. There is no way to 1760 * determine which events fired. 1761 */ 1762 int 1763 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td) 1764 { 1765 struct timeval rtv; 1766 sbintime_t asbt, precision, rsbt; 1767 int error; 1768 1769 precision = 0; /* stupid gcc! */ 1770 if (tvp != NULL) { 1771 rtv = *tvp; 1772 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1773 rtv.tv_usec >= 1000000) 1774 return (EINVAL); 1775 if (!timevalisset(&rtv)) 1776 asbt = 0; 1777 else if (rtv.tv_sec <= INT32_MAX) { 1778 rsbt = tvtosbt(rtv); 1779 precision = rsbt; 1780 precision >>= tc_precexp; 1781 if (TIMESEL(&asbt, rsbt)) 1782 asbt += tc_tick_sbt; 1783 if (asbt <= SBT_MAX - rsbt) 1784 asbt += rsbt; 1785 else 1786 asbt = -1; 1787 } else 1788 asbt = -1; 1789 } else 1790 asbt = -1; 1791 seltdinit(td); 1792 /* 1793 * Iterate until the timeout expires or the socket becomes ready. 1794 */ 1795 for (;;) { 1796 selfdalloc(td, NULL); 1797 if (sopoll(so, events, NULL, td) != 0) { 1798 error = 0; 1799 break; 1800 } 1801 error = seltdwait(td, asbt, precision); 1802 if (error) 1803 break; 1804 } 1805 seltdclear(td); 1806 /* XXX Duplicates ncp/smb behavior. */ 1807 if (error == ERESTART) 1808 error = 0; 1809 return (error); 1810 } 1811 1812 /* 1813 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines 1814 * have two select sets, one for read and another for write. 1815 */ 1816 static void 1817 selfdalloc(struct thread *td, void *cookie) 1818 { 1819 struct seltd *stp; 1820 1821 stp = td->td_sel; 1822 if (stp->st_free1 == NULL) 1823 stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO); 1824 stp->st_free1->sf_td = stp; 1825 stp->st_free1->sf_cookie = cookie; 1826 if (stp->st_free2 == NULL) 1827 stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO); 1828 stp->st_free2->sf_td = stp; 1829 stp->st_free2->sf_cookie = cookie; 1830 } 1831 1832 static void 1833 selfdfree(struct seltd *stp, struct selfd *sfp) 1834 { 1835 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link); 1836 /* 1837 * Paired with doselwakeup. 1838 */ 1839 if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) { 1840 mtx_lock(sfp->sf_mtx); 1841 if (sfp->sf_si != NULL) { 1842 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads); 1843 } 1844 mtx_unlock(sfp->sf_mtx); 1845 } 1846 free(sfp, M_SELFD); 1847 } 1848 1849 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */ 1850 void 1851 seldrain(struct selinfo *sip) 1852 { 1853 1854 /* 1855 * This feature is already provided by doselwakeup(), thus it is 1856 * enough to go for it. 1857 * Eventually, the context, should take care to avoid races 1858 * between thread calling select()/poll() and file descriptor 1859 * detaching, but, again, the races are just the same as 1860 * selwakeup(). 1861 */ 1862 doselwakeup(sip, -1); 1863 } 1864 1865 /* 1866 * Record a select request. 1867 */ 1868 void 1869 selrecord(struct thread *selector, struct selinfo *sip) 1870 { 1871 struct selfd *sfp; 1872 struct seltd *stp; 1873 struct mtx *mtxp; 1874 1875 stp = selector->td_sel; 1876 /* 1877 * Don't record when doing a rescan. 1878 */ 1879 if (stp->st_flags & SELTD_RESCAN) 1880 return; 1881 /* 1882 * Grab one of the preallocated descriptors. 1883 */ 1884 sfp = NULL; 1885 if ((sfp = stp->st_free1) != NULL) 1886 stp->st_free1 = NULL; 1887 else if ((sfp = stp->st_free2) != NULL) 1888 stp->st_free2 = NULL; 1889 else 1890 panic("selrecord: No free selfd on selq"); 1891 mtxp = sip->si_mtx; 1892 if (mtxp == NULL) 1893 mtxp = mtx_pool_find(mtxpool_select, sip); 1894 /* 1895 * Initialize the sfp and queue it in the thread. 1896 */ 1897 sfp->sf_si = sip; 1898 sfp->sf_mtx = mtxp; 1899 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link); 1900 /* 1901 * Now that we've locked the sip, check for initialization. 1902 */ 1903 mtx_lock(mtxp); 1904 if (sip->si_mtx == NULL) { 1905 sip->si_mtx = mtxp; 1906 TAILQ_INIT(&sip->si_tdlist); 1907 } 1908 /* 1909 * Add this thread to the list of selfds listening on this selinfo. 1910 */ 1911 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads); 1912 mtx_unlock(sip->si_mtx); 1913 } 1914 1915 /* Wake up a selecting thread. */ 1916 void 1917 selwakeup(struct selinfo *sip) 1918 { 1919 doselwakeup(sip, -1); 1920 } 1921 1922 /* Wake up a selecting thread, and set its priority. */ 1923 void 1924 selwakeuppri(struct selinfo *sip, int pri) 1925 { 1926 doselwakeup(sip, pri); 1927 } 1928 1929 /* 1930 * Do a wakeup when a selectable event occurs. 1931 */ 1932 static void 1933 doselwakeup(struct selinfo *sip, int pri) 1934 { 1935 struct selfd *sfp; 1936 struct selfd *sfn; 1937 struct seltd *stp; 1938 1939 /* If it's not initialized there can't be any waiters. */ 1940 if (sip->si_mtx == NULL) 1941 return; 1942 /* 1943 * Locking the selinfo locks all selfds associated with it. 1944 */ 1945 mtx_lock(sip->si_mtx); 1946 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) { 1947 /* 1948 * Once we remove this sfp from the list and clear the 1949 * sf_si seltdclear will know to ignore this si. 1950 */ 1951 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads); 1952 stp = sfp->sf_td; 1953 mtx_lock(&stp->st_mtx); 1954 stp->st_flags |= SELTD_PENDING; 1955 cv_broadcastpri(&stp->st_wait, pri); 1956 mtx_unlock(&stp->st_mtx); 1957 /* 1958 * Paired with selfdfree. 1959 * 1960 * Storing this only after the wakeup provides an invariant that 1961 * stp is not used after selfdfree returns. 1962 */ 1963 atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL); 1964 } 1965 mtx_unlock(sip->si_mtx); 1966 } 1967 1968 static void 1969 seltdinit(struct thread *td) 1970 { 1971 struct seltd *stp; 1972 1973 stp = td->td_sel; 1974 if (stp != NULL) { 1975 MPASS(stp->st_flags == 0); 1976 MPASS(STAILQ_EMPTY(&stp->st_selq)); 1977 return; 1978 } 1979 stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO); 1980 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF); 1981 cv_init(&stp->st_wait, "select"); 1982 stp->st_flags = 0; 1983 STAILQ_INIT(&stp->st_selq); 1984 td->td_sel = stp; 1985 } 1986 1987 static int 1988 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision) 1989 { 1990 struct seltd *stp; 1991 int error; 1992 1993 stp = td->td_sel; 1994 /* 1995 * An event of interest may occur while we do not hold the seltd 1996 * locked so check the pending flag before we sleep. 1997 */ 1998 mtx_lock(&stp->st_mtx); 1999 /* 2000 * Any further calls to selrecord will be a rescan. 2001 */ 2002 stp->st_flags |= SELTD_RESCAN; 2003 if (stp->st_flags & SELTD_PENDING) { 2004 mtx_unlock(&stp->st_mtx); 2005 return (0); 2006 } 2007 if (sbt == 0) 2008 error = EWOULDBLOCK; 2009 else if (sbt != -1) 2010 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx, 2011 sbt, precision, C_ABSOLUTE); 2012 else 2013 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx); 2014 mtx_unlock(&stp->st_mtx); 2015 2016 return (error); 2017 } 2018 2019 void 2020 seltdfini(struct thread *td) 2021 { 2022 struct seltd *stp; 2023 2024 stp = td->td_sel; 2025 if (stp == NULL) 2026 return; 2027 MPASS(stp->st_flags == 0); 2028 MPASS(STAILQ_EMPTY(&stp->st_selq)); 2029 if (stp->st_free1) 2030 free(stp->st_free1, M_SELFD); 2031 if (stp->st_free2) 2032 free(stp->st_free2, M_SELFD); 2033 td->td_sel = NULL; 2034 cv_destroy(&stp->st_wait); 2035 mtx_destroy(&stp->st_mtx); 2036 free(stp, M_SELECT); 2037 } 2038 2039 /* 2040 * Remove the references to the thread from all of the objects we were 2041 * polling. 2042 */ 2043 static void 2044 seltdclear(struct thread *td) 2045 { 2046 struct seltd *stp; 2047 struct selfd *sfp; 2048 struct selfd *sfn; 2049 2050 stp = td->td_sel; 2051 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) 2052 selfdfree(stp, sfp); 2053 stp->st_flags = 0; 2054 } 2055 2056 static void selectinit(void *); 2057 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL); 2058 static void 2059 selectinit(void *dummy __unused) 2060 { 2061 2062 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF); 2063 } 2064 2065 /* 2066 * Set up a syscall return value that follows the convention specified for 2067 * posix_* functions. 2068 */ 2069 int 2070 kern_posix_error(struct thread *td, int error) 2071 { 2072 2073 if (error <= 0) 2074 return (error); 2075 td->td_errno = error; 2076 td->td_pflags |= TDP_NERRNO; 2077 td->td_retval[0] = error; 2078 return (0); 2079 } 2080