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 37 #include <sys/cdefs.h> 38 #include "opt_capsicum.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/capsicum.h> 45 #include <sys/filedesc.h> 46 #include <sys/filio.h> 47 #include <sys/fcntl.h> 48 #include <sys/file.h> 49 #include <sys/lock.h> 50 #include <sys/proc.h> 51 #include <sys/signalvar.h> 52 #include <sys/protosw.h> 53 #include <sys/socketvar.h> 54 #include <sys/uio.h> 55 #include <sys/eventfd.h> 56 #include <sys/kernel.h> 57 #include <sys/ktr.h> 58 #include <sys/limits.h> 59 #include <sys/malloc.h> 60 #include <sys/poll.h> 61 #include <sys/resourcevar.h> 62 #include <sys/selinfo.h> 63 #include <sys/sleepqueue.h> 64 #include <sys/specialfd.h> 65 #include <sys/syscallsubr.h> 66 #include <sys/sysctl.h> 67 #include <sys/sysent.h> 68 #include <sys/vnode.h> 69 #include <sys/unistd.h> 70 #include <sys/bio.h> 71 #include <sys/buf.h> 72 #include <sys/condvar.h> 73 #ifdef KTRACE 74 #include <sys/ktrace.h> 75 #endif 76 77 #include <security/audit/audit.h> 78 79 /* 80 * The following macro defines how many bytes will be allocated from 81 * the stack instead of memory allocated when passing the IOCTL data 82 * structures from userspace and to the kernel. Some IOCTLs having 83 * small data structures are used very frequently and this small 84 * buffer on the stack gives a significant speedup improvement for 85 * those requests. The value of this define should be greater or equal 86 * to 64 bytes and should also be power of two. The data structure is 87 * currently hard-aligned to a 8-byte boundary on the stack. This 88 * should currently be sufficient for all supported platforms. 89 */ 90 #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */ 91 #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */ 92 93 #ifdef __LP64__ 94 static int iosize_max_clamp = 0; 95 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW, 96 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX"); 97 static int devfs_iosize_max_clamp = 1; 98 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW, 99 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices"); 100 #endif 101 102 /* 103 * Assert that the return value of read(2) and write(2) syscalls fits 104 * into a register. If not, an architecture will need to provide the 105 * usermode wrappers to reconstruct the result. 106 */ 107 CTASSERT(sizeof(register_t) >= sizeof(size_t)); 108 109 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); 110 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); 111 MALLOC_DEFINE(M_IOV, "iov", "large iov's"); 112 113 static int pollout(struct thread *, struct pollfd *, struct pollfd *, 114 u_int); 115 static int pollscan(struct thread *, struct pollfd *, u_int); 116 static int pollrescan(struct thread *); 117 static int selscan(struct thread *, fd_mask **, fd_mask **, int); 118 static int selrescan(struct thread *, fd_mask **, fd_mask **); 119 static void selfdalloc(struct thread *, void *); 120 static void selfdfree(struct seltd *, struct selfd *); 121 static int dofileread(struct thread *, int, struct file *, struct uio *, 122 off_t, int); 123 static int dofilewrite(struct thread *, int, struct file *, struct uio *, 124 off_t, int); 125 static void doselwakeup(struct selinfo *, int); 126 static void seltdinit(struct thread *); 127 static int seltdwait(struct thread *, sbintime_t, sbintime_t); 128 static void seltdclear(struct thread *); 129 130 /* 131 * One seltd per-thread allocated on demand as needed. 132 * 133 * t - protected by st_mtx 134 * k - Only accessed by curthread or read-only 135 */ 136 struct seltd { 137 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */ 138 struct selfd *st_free1; /* (k) free fd for read set. */ 139 struct selfd *st_free2; /* (k) free fd for write set. */ 140 struct mtx st_mtx; /* Protects struct seltd */ 141 struct cv st_wait; /* (t) Wait channel. */ 142 int st_flags; /* (t) SELTD_ flags. */ 143 }; 144 145 #define SELTD_PENDING 0x0001 /* We have pending events. */ 146 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */ 147 148 /* 149 * One selfd allocated per-thread per-file-descriptor. 150 * f - protected by sf_mtx 151 */ 152 struct selfd { 153 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */ 154 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */ 155 struct selinfo *sf_si; /* (f) selinfo when linked. */ 156 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */ 157 struct seltd *sf_td; /* (k) owning seltd. */ 158 void *sf_cookie; /* (k) fd or pollfd. */ 159 }; 160 161 MALLOC_DEFINE(M_SELFD, "selfd", "selfd"); 162 static struct mtx_pool *mtxpool_select; 163 164 #ifdef __LP64__ 165 size_t 166 devfs_iosize_max(void) 167 { 168 169 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 170 INT_MAX : SSIZE_MAX); 171 } 172 173 size_t 174 iosize_max(void) 175 { 176 177 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 178 INT_MAX : SSIZE_MAX); 179 } 180 #endif 181 182 #ifndef _SYS_SYSPROTO_H_ 183 struct read_args { 184 int fd; 185 void *buf; 186 size_t nbyte; 187 }; 188 #endif 189 int 190 sys_read(struct thread *td, struct read_args *uap) 191 { 192 struct uio auio; 193 struct iovec aiov; 194 int error; 195 196 if (uap->nbyte > IOSIZE_MAX) 197 return (EINVAL); 198 aiov.iov_base = uap->buf; 199 aiov.iov_len = uap->nbyte; 200 auio.uio_iov = &aiov; 201 auio.uio_iovcnt = 1; 202 auio.uio_resid = uap->nbyte; 203 auio.uio_segflg = UIO_USERSPACE; 204 error = kern_readv(td, uap->fd, &auio); 205 return (error); 206 } 207 208 /* 209 * Positioned read system call 210 */ 211 #ifndef _SYS_SYSPROTO_H_ 212 struct pread_args { 213 int fd; 214 void *buf; 215 size_t nbyte; 216 int pad; 217 off_t offset; 218 }; 219 #endif 220 int 221 sys_pread(struct thread *td, struct pread_args *uap) 222 { 223 224 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 225 } 226 227 int 228 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset) 229 { 230 struct uio auio; 231 struct iovec aiov; 232 int error; 233 234 if (nbyte > IOSIZE_MAX) 235 return (EINVAL); 236 aiov.iov_base = buf; 237 aiov.iov_len = nbyte; 238 auio.uio_iov = &aiov; 239 auio.uio_iovcnt = 1; 240 auio.uio_resid = nbyte; 241 auio.uio_segflg = UIO_USERSPACE; 242 error = kern_preadv(td, fd, &auio, offset); 243 return (error); 244 } 245 246 #if defined(COMPAT_FREEBSD6) 247 int 248 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap) 249 { 250 251 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 252 } 253 #endif 254 255 /* 256 * Scatter read system call. 257 */ 258 #ifndef _SYS_SYSPROTO_H_ 259 struct readv_args { 260 int fd; 261 struct iovec *iovp; 262 u_int iovcnt; 263 }; 264 #endif 265 int 266 sys_readv(struct thread *td, struct readv_args *uap) 267 { 268 struct uio *auio; 269 int error; 270 271 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 272 if (error) 273 return (error); 274 error = kern_readv(td, uap->fd, auio); 275 freeuio(auio); 276 return (error); 277 } 278 279 int 280 kern_readv(struct thread *td, int fd, struct uio *auio) 281 { 282 struct file *fp; 283 int error; 284 285 error = fget_read(td, fd, &cap_read_rights, &fp); 286 if (error) 287 return (error); 288 error = dofileread(td, fd, fp, auio, (off_t)-1, 0); 289 fdrop(fp, td); 290 return (error); 291 } 292 293 /* 294 * Scatter positioned read system call. 295 */ 296 #ifndef _SYS_SYSPROTO_H_ 297 struct preadv_args { 298 int fd; 299 struct iovec *iovp; 300 u_int iovcnt; 301 off_t offset; 302 }; 303 #endif 304 int 305 sys_preadv(struct thread *td, struct preadv_args *uap) 306 { 307 struct uio *auio; 308 int error; 309 310 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 311 if (error) 312 return (error); 313 error = kern_preadv(td, uap->fd, auio, uap->offset); 314 freeuio(auio); 315 return (error); 316 } 317 318 int 319 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset) 320 { 321 struct file *fp; 322 int error; 323 324 error = fget_read(td, fd, &cap_pread_rights, &fp); 325 if (error) 326 return (error); 327 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 328 error = ESPIPE; 329 else if (offset < 0 && 330 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 331 error = EINVAL; 332 else 333 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET); 334 fdrop(fp, td); 335 return (error); 336 } 337 338 /* 339 * Common code for readv and preadv that reads data in 340 * from a file using the passed in uio, offset, and flags. 341 */ 342 static int 343 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio, 344 off_t offset, int flags) 345 { 346 ssize_t cnt; 347 int error; 348 #ifdef KTRACE 349 struct uio *ktruio = NULL; 350 #endif 351 352 AUDIT_ARG_FD(fd); 353 354 /* Finish zero length reads right here */ 355 if (auio->uio_resid == 0) { 356 td->td_retval[0] = 0; 357 return (0); 358 } 359 auio->uio_rw = UIO_READ; 360 auio->uio_offset = offset; 361 auio->uio_td = td; 362 #ifdef KTRACE 363 if (KTRPOINT(td, KTR_GENIO)) 364 ktruio = cloneuio(auio); 365 #endif 366 cnt = auio->uio_resid; 367 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) { 368 if (auio->uio_resid != cnt && (error == ERESTART || 369 error == EINTR || error == EWOULDBLOCK)) 370 error = 0; 371 } 372 cnt -= auio->uio_resid; 373 #ifdef KTRACE 374 if (ktruio != NULL) { 375 ktruio->uio_resid = cnt; 376 ktrgenio(fd, UIO_READ, ktruio, error); 377 } 378 #endif 379 td->td_retval[0] = cnt; 380 return (error); 381 } 382 383 #ifndef _SYS_SYSPROTO_H_ 384 struct write_args { 385 int fd; 386 const void *buf; 387 size_t nbyte; 388 }; 389 #endif 390 int 391 sys_write(struct thread *td, struct write_args *uap) 392 { 393 struct uio auio; 394 struct iovec aiov; 395 int error; 396 397 if (uap->nbyte > IOSIZE_MAX) 398 return (EINVAL); 399 aiov.iov_base = (void *)(uintptr_t)uap->buf; 400 aiov.iov_len = uap->nbyte; 401 auio.uio_iov = &aiov; 402 auio.uio_iovcnt = 1; 403 auio.uio_resid = uap->nbyte; 404 auio.uio_segflg = UIO_USERSPACE; 405 error = kern_writev(td, uap->fd, &auio); 406 return (error); 407 } 408 409 /* 410 * Positioned write system call. 411 */ 412 #ifndef _SYS_SYSPROTO_H_ 413 struct pwrite_args { 414 int fd; 415 const void *buf; 416 size_t nbyte; 417 int pad; 418 off_t offset; 419 }; 420 #endif 421 int 422 sys_pwrite(struct thread *td, struct pwrite_args *uap) 423 { 424 425 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 426 } 427 428 int 429 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte, 430 off_t offset) 431 { 432 struct uio auio; 433 struct iovec aiov; 434 int error; 435 436 if (nbyte > IOSIZE_MAX) 437 return (EINVAL); 438 aiov.iov_base = (void *)(uintptr_t)buf; 439 aiov.iov_len = nbyte; 440 auio.uio_iov = &aiov; 441 auio.uio_iovcnt = 1; 442 auio.uio_resid = nbyte; 443 auio.uio_segflg = UIO_USERSPACE; 444 error = kern_pwritev(td, fd, &auio, offset); 445 return (error); 446 } 447 448 #if defined(COMPAT_FREEBSD6) 449 int 450 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap) 451 { 452 453 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 454 } 455 #endif 456 457 /* 458 * Gather write system call. 459 */ 460 #ifndef _SYS_SYSPROTO_H_ 461 struct writev_args { 462 int fd; 463 struct iovec *iovp; 464 u_int iovcnt; 465 }; 466 #endif 467 int 468 sys_writev(struct thread *td, struct writev_args *uap) 469 { 470 struct uio *auio; 471 int error; 472 473 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 474 if (error) 475 return (error); 476 error = kern_writev(td, uap->fd, auio); 477 freeuio(auio); 478 return (error); 479 } 480 481 int 482 kern_writev(struct thread *td, int fd, struct uio *auio) 483 { 484 struct file *fp; 485 int error; 486 487 error = fget_write(td, fd, &cap_write_rights, &fp); 488 if (error) 489 return (error); 490 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0); 491 fdrop(fp, td); 492 return (error); 493 } 494 495 /* 496 * Gather positioned write system call. 497 */ 498 #ifndef _SYS_SYSPROTO_H_ 499 struct pwritev_args { 500 int fd; 501 struct iovec *iovp; 502 u_int iovcnt; 503 off_t offset; 504 }; 505 #endif 506 int 507 sys_pwritev(struct thread *td, struct pwritev_args *uap) 508 { 509 struct uio *auio; 510 int error; 511 512 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 513 if (error) 514 return (error); 515 error = kern_pwritev(td, uap->fd, auio, uap->offset); 516 freeuio(auio); 517 return (error); 518 } 519 520 int 521 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset) 522 { 523 struct file *fp; 524 int error; 525 526 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 527 if (error) 528 return (error); 529 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 530 error = ESPIPE; 531 else if (offset < 0 && 532 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 533 error = EINVAL; 534 else 535 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET); 536 fdrop(fp, td); 537 return (error); 538 } 539 540 /* 541 * Common code for writev and pwritev that writes data to 542 * a file using the passed in uio, offset, and flags. 543 */ 544 static int 545 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio, 546 off_t offset, int flags) 547 { 548 ssize_t cnt; 549 int error; 550 #ifdef KTRACE 551 struct uio *ktruio = NULL; 552 #endif 553 554 AUDIT_ARG_FD(fd); 555 auio->uio_rw = UIO_WRITE; 556 auio->uio_td = td; 557 auio->uio_offset = offset; 558 #ifdef KTRACE 559 if (KTRPOINT(td, KTR_GENIO)) 560 ktruio = cloneuio(auio); 561 #endif 562 cnt = auio->uio_resid; 563 error = fo_write(fp, auio, td->td_ucred, flags, td); 564 /* 565 * Socket layer is responsible for special error handling, 566 * see sousrsend(). 567 */ 568 if (error != 0 && fp->f_type != DTYPE_SOCKET) { 569 if (auio->uio_resid != cnt && (error == ERESTART || 570 error == EINTR || error == EWOULDBLOCK)) 571 error = 0; 572 if (error == EPIPE) { 573 PROC_LOCK(td->td_proc); 574 tdsignal(td, SIGPIPE); 575 PROC_UNLOCK(td->td_proc); 576 } 577 } 578 cnt -= auio->uio_resid; 579 #ifdef KTRACE 580 if (ktruio != NULL) { 581 if (error == 0) 582 ktruio->uio_resid = cnt; 583 ktrgenio(fd, UIO_WRITE, ktruio, error); 584 } 585 #endif 586 td->td_retval[0] = cnt; 587 return (error); 588 } 589 590 /* 591 * Truncate a file given a file descriptor. 592 * 593 * Can't use fget_write() here, since must return EINVAL and not EBADF if the 594 * descriptor isn't writable. 595 */ 596 int 597 kern_ftruncate(struct thread *td, int fd, off_t length) 598 { 599 struct file *fp; 600 int error; 601 602 AUDIT_ARG_FD(fd); 603 if (length < 0) 604 return (EINVAL); 605 error = fget(td, fd, &cap_ftruncate_rights, &fp); 606 if (error) 607 return (error); 608 AUDIT_ARG_FILE(td->td_proc, fp); 609 if (!(fp->f_flag & FWRITE)) { 610 fdrop(fp, td); 611 return (EINVAL); 612 } 613 error = fo_truncate(fp, length, td->td_ucred, td); 614 fdrop(fp, td); 615 return (error); 616 } 617 618 #ifndef _SYS_SYSPROTO_H_ 619 struct ftruncate_args { 620 int fd; 621 int pad; 622 off_t length; 623 }; 624 #endif 625 int 626 sys_ftruncate(struct thread *td, struct ftruncate_args *uap) 627 { 628 629 return (kern_ftruncate(td, uap->fd, uap->length)); 630 } 631 632 #if defined(COMPAT_43) 633 #ifndef _SYS_SYSPROTO_H_ 634 struct oftruncate_args { 635 int fd; 636 long length; 637 }; 638 #endif 639 int 640 oftruncate(struct thread *td, struct oftruncate_args *uap) 641 { 642 643 return (kern_ftruncate(td, uap->fd, uap->length)); 644 } 645 #endif /* COMPAT_43 */ 646 647 #ifndef _SYS_SYSPROTO_H_ 648 struct ioctl_args { 649 int fd; 650 u_long com; 651 caddr_t data; 652 }; 653 #endif 654 /* ARGSUSED */ 655 int 656 sys_ioctl(struct thread *td, struct ioctl_args *uap) 657 { 658 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN); 659 uint32_t com; 660 int arg, error; 661 u_int size; 662 caddr_t data; 663 664 #ifdef INVARIANTS 665 if (uap->com > 0xffffffff) { 666 printf( 667 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n", 668 td->td_proc->p_pid, td->td_name, uap->com); 669 } 670 #endif 671 com = (uint32_t)uap->com; 672 673 /* 674 * Interpret high order word to find amount of data to be 675 * copied to/from the user's address space. 676 */ 677 size = IOCPARM_LEN(com); 678 if ((size > IOCPARM_MAX) || 679 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) || 680 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43) 681 ((com & IOC_OUT) && size == 0) || 682 #else 683 ((com & (IOC_IN | IOC_OUT)) && size == 0) || 684 #endif 685 ((com & IOC_VOID) && size > 0 && size != sizeof(int))) 686 return (ENOTTY); 687 688 if (size > 0) { 689 if (com & IOC_VOID) { 690 /* Integer argument. */ 691 arg = (intptr_t)uap->data; 692 data = (void *)&arg; 693 size = 0; 694 } else { 695 if (size > SYS_IOCTL_SMALL_SIZE) 696 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK); 697 else 698 data = smalldata; 699 } 700 } else 701 data = (void *)&uap->data; 702 if (com & IOC_IN) { 703 error = copyin(uap->data, data, (u_int)size); 704 if (error != 0) 705 goto out; 706 } else if (com & IOC_OUT) { 707 /* 708 * Zero the buffer so the user always 709 * gets back something deterministic. 710 */ 711 bzero(data, size); 712 } 713 714 error = kern_ioctl(td, uap->fd, com, data); 715 716 if (error == 0 && (com & IOC_OUT)) 717 error = copyout(data, uap->data, (u_int)size); 718 719 out: 720 if (size > SYS_IOCTL_SMALL_SIZE) 721 free(data, M_IOCTLOPS); 722 return (error); 723 } 724 725 int 726 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data) 727 { 728 struct file *fp; 729 struct filedesc *fdp; 730 int error, tmp, locked; 731 732 AUDIT_ARG_FD(fd); 733 AUDIT_ARG_CMD(com); 734 735 fdp = td->td_proc->p_fd; 736 737 switch (com) { 738 case FIONCLEX: 739 case FIOCLEX: 740 FILEDESC_XLOCK(fdp); 741 locked = LA_XLOCKED; 742 break; 743 default: 744 #ifdef CAPABILITIES 745 FILEDESC_SLOCK(fdp); 746 locked = LA_SLOCKED; 747 #else 748 locked = LA_UNLOCKED; 749 #endif 750 break; 751 } 752 753 #ifdef CAPABILITIES 754 if ((fp = fget_noref(fdp, fd)) == NULL) { 755 error = EBADF; 756 goto out; 757 } 758 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) { 759 fp = NULL; /* fhold() was not called yet */ 760 goto out; 761 } 762 if (!fhold(fp)) { 763 error = EBADF; 764 fp = NULL; 765 goto out; 766 } 767 if (locked == LA_SLOCKED) { 768 FILEDESC_SUNLOCK(fdp); 769 locked = LA_UNLOCKED; 770 } 771 #else 772 error = fget(td, fd, &cap_ioctl_rights, &fp); 773 if (error != 0) { 774 fp = NULL; 775 goto out; 776 } 777 #endif 778 if ((fp->f_flag & (FREAD | FWRITE)) == 0) { 779 error = EBADF; 780 goto out; 781 } 782 783 switch (com) { 784 case FIONCLEX: 785 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE; 786 goto out; 787 case FIOCLEX: 788 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE; 789 goto out; 790 case FIONBIO: 791 if ((tmp = *(int *)data)) 792 atomic_set_int(&fp->f_flag, FNONBLOCK); 793 else 794 atomic_clear_int(&fp->f_flag, FNONBLOCK); 795 data = (void *)&tmp; 796 break; 797 case FIOASYNC: 798 if ((tmp = *(int *)data)) 799 atomic_set_int(&fp->f_flag, FASYNC); 800 else 801 atomic_clear_int(&fp->f_flag, FASYNC); 802 data = (void *)&tmp; 803 break; 804 } 805 806 error = fo_ioctl(fp, com, data, td->td_ucred, td); 807 out: 808 switch (locked) { 809 case LA_XLOCKED: 810 FILEDESC_XUNLOCK(fdp); 811 break; 812 #ifdef CAPABILITIES 813 case LA_SLOCKED: 814 FILEDESC_SUNLOCK(fdp); 815 break; 816 #endif 817 default: 818 FILEDESC_UNLOCK_ASSERT(fdp); 819 break; 820 } 821 if (fp != NULL) 822 fdrop(fp, td); 823 return (error); 824 } 825 826 int 827 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) 828 { 829 int error; 830 831 error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len); 832 return (kern_posix_error(td, error)); 833 } 834 835 int 836 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len) 837 { 838 struct file *fp; 839 int error; 840 841 AUDIT_ARG_FD(fd); 842 if (offset < 0 || len <= 0) 843 return (EINVAL); 844 /* Check for wrap. */ 845 if (offset > OFF_MAX - len) 846 return (EFBIG); 847 AUDIT_ARG_FD(fd); 848 error = fget(td, fd, &cap_pwrite_rights, &fp); 849 if (error != 0) 850 return (error); 851 AUDIT_ARG_FILE(td->td_proc, fp); 852 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 853 error = ESPIPE; 854 goto out; 855 } 856 if ((fp->f_flag & FWRITE) == 0) { 857 error = EBADF; 858 goto out; 859 } 860 861 error = fo_fallocate(fp, offset, len, td); 862 out: 863 fdrop(fp, td); 864 return (error); 865 } 866 867 int 868 sys_fspacectl(struct thread *td, struct fspacectl_args *uap) 869 { 870 struct spacectl_range rqsr, rmsr; 871 int error, cerror; 872 873 error = copyin(uap->rqsr, &rqsr, sizeof(rqsr)); 874 if (error != 0) 875 return (error); 876 877 error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags, 878 &rmsr); 879 if (uap->rmsr != NULL) { 880 cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr)); 881 if (error == 0) 882 error = cerror; 883 } 884 return (error); 885 } 886 887 int 888 kern_fspacectl(struct thread *td, int fd, int cmd, 889 const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp) 890 { 891 struct file *fp; 892 struct spacectl_range rmsr; 893 int error; 894 895 AUDIT_ARG_FD(fd); 896 AUDIT_ARG_CMD(cmd); 897 AUDIT_ARG_FFLAGS(flags); 898 899 if (rqsr == NULL) 900 return (EINVAL); 901 rmsr = *rqsr; 902 if (rmsrp != NULL) 903 *rmsrp = rmsr; 904 905 if (cmd != SPACECTL_DEALLOC || 906 rqsr->r_offset < 0 || rqsr->r_len <= 0 || 907 rqsr->r_offset > OFF_MAX - rqsr->r_len || 908 (flags & ~SPACECTL_F_SUPPORTED) != 0) 909 return (EINVAL); 910 911 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 912 if (error != 0) 913 return (error); 914 AUDIT_ARG_FILE(td->td_proc, fp); 915 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 916 error = ESPIPE; 917 goto out; 918 } 919 if ((fp->f_flag & FWRITE) == 0) { 920 error = EBADF; 921 goto out; 922 } 923 924 error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags, 925 td->td_ucred, td); 926 /* fspacectl is not restarted after signals if the file is modified. */ 927 if (rmsr.r_len != rqsr->r_len && (error == ERESTART || 928 error == EINTR || error == EWOULDBLOCK)) 929 error = 0; 930 if (rmsrp != NULL) 931 *rmsrp = rmsr; 932 out: 933 fdrop(fp, td); 934 return (error); 935 } 936 937 int 938 kern_specialfd(struct thread *td, int type, void *arg) 939 { 940 struct file *fp; 941 struct specialfd_eventfd *ae; 942 int error, fd, fflags; 943 944 fflags = 0; 945 error = falloc_noinstall(td, &fp); 946 if (error != 0) 947 return (error); 948 949 switch (type) { 950 case SPECIALFD_EVENTFD: 951 ae = arg; 952 if ((ae->flags & EFD_CLOEXEC) != 0) 953 fflags |= O_CLOEXEC; 954 error = eventfd_create_file(td, fp, ae->initval, ae->flags); 955 break; 956 default: 957 error = EINVAL; 958 break; 959 } 960 961 if (error == 0) 962 error = finstall(td, fp, &fd, fflags, NULL); 963 fdrop(fp, td); 964 if (error == 0) 965 td->td_retval[0] = fd; 966 return (error); 967 } 968 969 int 970 sys___specialfd(struct thread *td, struct __specialfd_args *args) 971 { 972 struct specialfd_eventfd ae; 973 int error; 974 975 switch (args->type) { 976 case SPECIALFD_EVENTFD: 977 if (args->len != sizeof(struct specialfd_eventfd)) { 978 error = EINVAL; 979 break; 980 } 981 error = copyin(args->req, &ae, sizeof(ae)); 982 if (error != 0) 983 break; 984 if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK | 985 EFD_SEMAPHORE)) != 0) { 986 error = EINVAL; 987 break; 988 } 989 error = kern_specialfd(td, args->type, &ae); 990 break; 991 default: 992 error = EINVAL; 993 break; 994 } 995 return (error); 996 } 997 998 int 999 poll_no_poll(int events) 1000 { 1001 /* 1002 * Return true for read/write. If the user asked for something 1003 * special, return POLLNVAL, so that clients have a way of 1004 * determining reliably whether or not the extended 1005 * functionality is present without hard-coding knowledge 1006 * of specific filesystem implementations. 1007 */ 1008 if (events & ~POLLSTANDARD) 1009 return (POLLNVAL); 1010 1011 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 1012 } 1013 1014 int 1015 sys_pselect(struct thread *td, struct pselect_args *uap) 1016 { 1017 struct timespec ts; 1018 struct timeval tv, *tvp; 1019 sigset_t set, *uset; 1020 int error; 1021 1022 if (uap->ts != NULL) { 1023 error = copyin(uap->ts, &ts, sizeof(ts)); 1024 if (error != 0) 1025 return (error); 1026 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1027 tvp = &tv; 1028 } else 1029 tvp = NULL; 1030 if (uap->sm != NULL) { 1031 error = copyin(uap->sm, &set, sizeof(set)); 1032 if (error != 0) 1033 return (error); 1034 uset = &set; 1035 } else 1036 uset = NULL; 1037 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1038 uset, NFDBITS)); 1039 } 1040 1041 int 1042 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex, 1043 struct timeval *tvp, sigset_t *uset, int abi_nfdbits) 1044 { 1045 int error; 1046 1047 if (uset != NULL) { 1048 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1049 &td->td_oldsigmask, 0); 1050 if (error != 0) 1051 return (error); 1052 td->td_pflags |= TDP_OLDMASK; 1053 } 1054 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits); 1055 if (uset != NULL) { 1056 /* 1057 * Make sure that ast() is called on return to 1058 * usermode and TDP_OLDMASK is cleared, restoring old 1059 * sigmask. If we didn't get interrupted, then the caller is 1060 * likely not expecting a signal to hit that should normally be 1061 * blocked by its signal mask, so we restore the mask before 1062 * any signals could be delivered. 1063 */ 1064 if (error == EINTR) { 1065 ast_sched(td, TDA_SIGSUSPEND); 1066 } else { 1067 /* *select(2) should never restart. */ 1068 MPASS(error != ERESTART); 1069 ast_sched(td, TDA_PSELECT); 1070 } 1071 } 1072 1073 return (error); 1074 } 1075 1076 #ifndef _SYS_SYSPROTO_H_ 1077 struct select_args { 1078 int nd; 1079 fd_set *in, *ou, *ex; 1080 struct timeval *tv; 1081 }; 1082 #endif 1083 int 1084 sys_select(struct thread *td, struct select_args *uap) 1085 { 1086 struct timeval tv, *tvp; 1087 int error; 1088 1089 if (uap->tv != NULL) { 1090 error = copyin(uap->tv, &tv, sizeof(tv)); 1091 if (error) 1092 return (error); 1093 tvp = &tv; 1094 } else 1095 tvp = NULL; 1096 1097 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1098 NFDBITS)); 1099 } 1100 1101 /* 1102 * In the unlikely case when user specified n greater then the last 1103 * open file descriptor, check that no bits are set after the last 1104 * valid fd. We must return EBADF if any is set. 1105 * 1106 * There are applications that rely on the behaviour. 1107 * 1108 * nd is fd_nfiles. 1109 */ 1110 static int 1111 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits) 1112 { 1113 char *addr, *oaddr; 1114 int b, i, res; 1115 uint8_t bits; 1116 1117 if (nd >= ndu || fd_in == NULL) 1118 return (0); 1119 1120 oaddr = NULL; 1121 bits = 0; /* silence gcc */ 1122 for (i = nd; i < ndu; i++) { 1123 b = i / NBBY; 1124 #if BYTE_ORDER == LITTLE_ENDIAN 1125 addr = (char *)fd_in + b; 1126 #else 1127 addr = (char *)fd_in; 1128 if (abi_nfdbits == NFDBITS) { 1129 addr += rounddown(b, sizeof(fd_mask)) + 1130 sizeof(fd_mask) - 1 - b % sizeof(fd_mask); 1131 } else { 1132 addr += rounddown(b, sizeof(uint32_t)) + 1133 sizeof(uint32_t) - 1 - b % sizeof(uint32_t); 1134 } 1135 #endif 1136 if (addr != oaddr) { 1137 res = fubyte(addr); 1138 if (res == -1) 1139 return (EFAULT); 1140 oaddr = addr; 1141 bits = res; 1142 } 1143 if ((bits & (1 << (i % NBBY))) != 0) 1144 return (EBADF); 1145 } 1146 return (0); 1147 } 1148 1149 int 1150 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou, 1151 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits) 1152 { 1153 struct filedesc *fdp; 1154 /* 1155 * The magic 2048 here is chosen to be just enough for FD_SETSIZE 1156 * infds with the new FD_SETSIZE of 1024, and more than enough for 1157 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE 1158 * of 256. 1159 */ 1160 fd_mask s_selbits[howmany(2048, NFDBITS)]; 1161 fd_mask *ibits[3], *obits[3], *selbits, *sbp; 1162 struct timeval rtv; 1163 sbintime_t asbt, precision, rsbt; 1164 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits; 1165 int error, lf, ndu; 1166 1167 if (nd < 0) 1168 return (EINVAL); 1169 fdp = td->td_proc->p_fd; 1170 ndu = nd; 1171 lf = fdp->fd_nfiles; 1172 if (nd > lf) 1173 nd = lf; 1174 1175 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits); 1176 if (error != 0) 1177 return (error); 1178 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits); 1179 if (error != 0) 1180 return (error); 1181 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits); 1182 if (error != 0) 1183 return (error); 1184 1185 /* 1186 * Allocate just enough bits for the non-null fd_sets. Use the 1187 * preallocated auto buffer if possible. 1188 */ 1189 nfdbits = roundup(nd, NFDBITS); 1190 ncpbytes = nfdbits / NBBY; 1191 ncpubytes = roundup(nd, abi_nfdbits) / NBBY; 1192 nbufbytes = 0; 1193 if (fd_in != NULL) 1194 nbufbytes += 2 * ncpbytes; 1195 if (fd_ou != NULL) 1196 nbufbytes += 2 * ncpbytes; 1197 if (fd_ex != NULL) 1198 nbufbytes += 2 * ncpbytes; 1199 if (nbufbytes <= sizeof s_selbits) 1200 selbits = &s_selbits[0]; 1201 else 1202 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK); 1203 1204 /* 1205 * Assign pointers into the bit buffers and fetch the input bits. 1206 * Put the output buffers together so that they can be bzeroed 1207 * together. 1208 */ 1209 sbp = selbits; 1210 #define getbits(name, x) \ 1211 do { \ 1212 if (name == NULL) { \ 1213 ibits[x] = NULL; \ 1214 obits[x] = NULL; \ 1215 } else { \ 1216 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ 1217 obits[x] = sbp; \ 1218 sbp += ncpbytes / sizeof *sbp; \ 1219 error = copyin(name, ibits[x], ncpubytes); \ 1220 if (error != 0) \ 1221 goto done; \ 1222 if (ncpbytes != ncpubytes) \ 1223 bzero((char *)ibits[x] + ncpubytes, \ 1224 ncpbytes - ncpubytes); \ 1225 } \ 1226 } while (0) 1227 getbits(fd_in, 0); 1228 getbits(fd_ou, 1); 1229 getbits(fd_ex, 2); 1230 #undef getbits 1231 1232 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__) 1233 /* 1234 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS, 1235 * we are running under 32-bit emulation. This should be more 1236 * generic. 1237 */ 1238 #define swizzle_fdset(bits) \ 1239 if (abi_nfdbits != NFDBITS && bits != NULL) { \ 1240 int i; \ 1241 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \ 1242 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \ 1243 } 1244 #else 1245 #define swizzle_fdset(bits) 1246 #endif 1247 1248 /* Make sure the bit order makes it through an ABI transition */ 1249 swizzle_fdset(ibits[0]); 1250 swizzle_fdset(ibits[1]); 1251 swizzle_fdset(ibits[2]); 1252 1253 if (nbufbytes != 0) 1254 bzero(selbits, nbufbytes / 2); 1255 1256 precision = 0; 1257 if (tvp != NULL) { 1258 rtv = *tvp; 1259 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1260 rtv.tv_usec >= 1000000) { 1261 error = EINVAL; 1262 goto done; 1263 } 1264 if (!timevalisset(&rtv)) 1265 asbt = 0; 1266 else if (rtv.tv_sec <= INT32_MAX) { 1267 rsbt = tvtosbt(rtv); 1268 precision = rsbt; 1269 precision >>= tc_precexp; 1270 if (TIMESEL(&asbt, rsbt)) 1271 asbt += tc_tick_sbt; 1272 if (asbt <= SBT_MAX - rsbt) 1273 asbt += rsbt; 1274 else 1275 asbt = -1; 1276 } else 1277 asbt = -1; 1278 } else 1279 asbt = -1; 1280 seltdinit(td); 1281 /* Iterate until the timeout expires or descriptors become ready. */ 1282 for (;;) { 1283 error = selscan(td, ibits, obits, nd); 1284 if (error || td->td_retval[0] != 0) 1285 break; 1286 error = seltdwait(td, asbt, precision); 1287 if (error) 1288 break; 1289 error = selrescan(td, ibits, obits); 1290 if (error || td->td_retval[0] != 0) 1291 break; 1292 } 1293 seltdclear(td); 1294 1295 done: 1296 /* select is not restarted after signals... */ 1297 if (error == ERESTART) 1298 error = EINTR; 1299 if (error == EWOULDBLOCK) 1300 error = 0; 1301 1302 /* swizzle bit order back, if necessary */ 1303 swizzle_fdset(obits[0]); 1304 swizzle_fdset(obits[1]); 1305 swizzle_fdset(obits[2]); 1306 #undef swizzle_fdset 1307 1308 #define putbits(name, x) \ 1309 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \ 1310 error = error2; 1311 if (error == 0) { 1312 int error2; 1313 1314 putbits(fd_in, 0); 1315 putbits(fd_ou, 1); 1316 putbits(fd_ex, 2); 1317 #undef putbits 1318 } 1319 if (selbits != &s_selbits[0]) 1320 free(selbits, M_SELECT); 1321 1322 return (error); 1323 } 1324 /* 1325 * Convert a select bit set to poll flags. 1326 * 1327 * The backend always returns POLLHUP/POLLERR if appropriate and we 1328 * return this as a set bit in any set. 1329 */ 1330 static const int select_flags[3] = { 1331 POLLRDNORM | POLLHUP | POLLERR, 1332 POLLWRNORM | POLLHUP | POLLERR, 1333 POLLRDBAND | POLLERR 1334 }; 1335 1336 /* 1337 * Compute the fo_poll flags required for a fd given by the index and 1338 * bit position in the fd_mask array. 1339 */ 1340 static __inline int 1341 selflags(fd_mask **ibits, int idx, fd_mask bit) 1342 { 1343 int flags; 1344 int msk; 1345 1346 flags = 0; 1347 for (msk = 0; msk < 3; msk++) { 1348 if (ibits[msk] == NULL) 1349 continue; 1350 if ((ibits[msk][idx] & bit) == 0) 1351 continue; 1352 flags |= select_flags[msk]; 1353 } 1354 return (flags); 1355 } 1356 1357 /* 1358 * Set the appropriate output bits given a mask of fired events and the 1359 * input bits originally requested. 1360 */ 1361 static __inline int 1362 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events) 1363 { 1364 int msk; 1365 int n; 1366 1367 n = 0; 1368 for (msk = 0; msk < 3; msk++) { 1369 if ((events & select_flags[msk]) == 0) 1370 continue; 1371 if (ibits[msk] == NULL) 1372 continue; 1373 if ((ibits[msk][idx] & bit) == 0) 1374 continue; 1375 /* 1376 * XXX Check for a duplicate set. This can occur because a 1377 * socket calls selrecord() twice for each poll() call 1378 * resulting in two selfds per real fd. selrescan() will 1379 * call selsetbits twice as a result. 1380 */ 1381 if ((obits[msk][idx] & bit) != 0) 1382 continue; 1383 obits[msk][idx] |= bit; 1384 n++; 1385 } 1386 1387 return (n); 1388 } 1389 1390 /* 1391 * Traverse the list of fds attached to this thread's seltd and check for 1392 * completion. 1393 */ 1394 static int 1395 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits) 1396 { 1397 struct filedesc *fdp; 1398 struct selinfo *si; 1399 struct seltd *stp; 1400 struct selfd *sfp; 1401 struct selfd *sfn; 1402 struct file *fp; 1403 fd_mask bit; 1404 int fd, ev, n, idx; 1405 int error; 1406 bool only_user; 1407 1408 fdp = td->td_proc->p_fd; 1409 stp = td->td_sel; 1410 n = 0; 1411 only_user = FILEDESC_IS_ONLY_USER(fdp); 1412 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1413 fd = (int)(uintptr_t)sfp->sf_cookie; 1414 si = sfp->sf_si; 1415 selfdfree(stp, sfp); 1416 /* If the selinfo wasn't cleared the event didn't fire. */ 1417 if (si != NULL) 1418 continue; 1419 if (only_user) 1420 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1421 else 1422 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1423 if (__predict_false(error != 0)) 1424 return (error); 1425 idx = fd / NFDBITS; 1426 bit = (fd_mask)1 << (fd % NFDBITS); 1427 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td); 1428 if (only_user) 1429 fput_only_user(fdp, fp); 1430 else 1431 fdrop(fp, td); 1432 if (ev != 0) 1433 n += selsetbits(ibits, obits, idx, bit, ev); 1434 } 1435 stp->st_flags = 0; 1436 td->td_retval[0] = n; 1437 return (0); 1438 } 1439 1440 /* 1441 * Perform the initial filedescriptor scan and register ourselves with 1442 * each selinfo. 1443 */ 1444 static int 1445 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd) 1446 { 1447 struct filedesc *fdp; 1448 struct file *fp; 1449 fd_mask bit; 1450 int ev, flags, end, fd; 1451 int n, idx; 1452 int error; 1453 bool only_user; 1454 1455 fdp = td->td_proc->p_fd; 1456 n = 0; 1457 only_user = FILEDESC_IS_ONLY_USER(fdp); 1458 for (idx = 0, fd = 0; fd < nfd; idx++) { 1459 end = imin(fd + NFDBITS, nfd); 1460 for (bit = 1; fd < end; bit <<= 1, fd++) { 1461 /* Compute the list of events we're interested in. */ 1462 flags = selflags(ibits, idx, bit); 1463 if (flags == 0) 1464 continue; 1465 if (only_user) 1466 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1467 else 1468 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1469 if (__predict_false(error != 0)) 1470 return (error); 1471 selfdalloc(td, (void *)(uintptr_t)fd); 1472 ev = fo_poll(fp, flags, td->td_ucred, td); 1473 if (only_user) 1474 fput_only_user(fdp, fp); 1475 else 1476 fdrop(fp, td); 1477 if (ev != 0) 1478 n += selsetbits(ibits, obits, idx, bit, ev); 1479 } 1480 } 1481 1482 td->td_retval[0] = n; 1483 return (0); 1484 } 1485 1486 int 1487 sys_poll(struct thread *td, struct poll_args *uap) 1488 { 1489 struct timespec ts, *tsp; 1490 1491 if (uap->timeout != INFTIM) { 1492 if (uap->timeout < 0) 1493 return (EINVAL); 1494 ts.tv_sec = uap->timeout / 1000; 1495 ts.tv_nsec = (uap->timeout % 1000) * 1000000; 1496 tsp = &ts; 1497 } else 1498 tsp = NULL; 1499 1500 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL)); 1501 } 1502 1503 /* 1504 * kfds points to an array in the kernel. 1505 */ 1506 int 1507 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds, 1508 struct timespec *tsp, sigset_t *uset) 1509 { 1510 sbintime_t sbt, precision, tmp; 1511 time_t over; 1512 struct timespec ts; 1513 int error; 1514 1515 precision = 0; 1516 if (tsp != NULL) { 1517 if (!timespecvalid_interval(tsp)) 1518 return (EINVAL); 1519 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1520 sbt = 0; 1521 else { 1522 ts = *tsp; 1523 if (ts.tv_sec > INT32_MAX / 2) { 1524 over = ts.tv_sec - INT32_MAX / 2; 1525 ts.tv_sec -= over; 1526 } else 1527 over = 0; 1528 tmp = tstosbt(ts); 1529 precision = tmp; 1530 precision >>= tc_precexp; 1531 if (TIMESEL(&sbt, tmp)) 1532 sbt += tc_tick_sbt; 1533 sbt += tmp; 1534 } 1535 } else 1536 sbt = -1; 1537 1538 if (uset != NULL) { 1539 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1540 &td->td_oldsigmask, 0); 1541 if (error) 1542 return (error); 1543 td->td_pflags |= TDP_OLDMASK; 1544 } 1545 1546 seltdinit(td); 1547 /* Iterate until the timeout expires or descriptors become ready. */ 1548 for (;;) { 1549 error = pollscan(td, kfds, nfds); 1550 if (error || td->td_retval[0] != 0) 1551 break; 1552 error = seltdwait(td, sbt, precision); 1553 if (error) 1554 break; 1555 error = pollrescan(td); 1556 if (error || td->td_retval[0] != 0) 1557 break; 1558 } 1559 seltdclear(td); 1560 1561 /* poll is not restarted after signals... */ 1562 if (error == ERESTART) 1563 error = EINTR; 1564 if (error == EWOULDBLOCK) 1565 error = 0; 1566 1567 if (uset != NULL) { 1568 /* 1569 * Make sure that ast() is called on return to 1570 * usermode and TDP_OLDMASK is cleared, restoring old 1571 * sigmask. If we didn't get interrupted, then the caller is 1572 * likely not expecting a signal to hit that should normally be 1573 * blocked by its signal mask, so we restore the mask before 1574 * any signals could be delivered. 1575 */ 1576 if (error == EINTR) 1577 ast_sched(td, TDA_SIGSUSPEND); 1578 else 1579 ast_sched(td, TDA_PSELECT); 1580 } 1581 1582 return (error); 1583 } 1584 1585 int 1586 sys_ppoll(struct thread *td, struct ppoll_args *uap) 1587 { 1588 struct timespec ts, *tsp; 1589 sigset_t set, *ssp; 1590 int error; 1591 1592 if (uap->ts != NULL) { 1593 error = copyin(uap->ts, &ts, sizeof(ts)); 1594 if (error) 1595 return (error); 1596 tsp = &ts; 1597 } else 1598 tsp = NULL; 1599 if (uap->set != NULL) { 1600 error = copyin(uap->set, &set, sizeof(set)); 1601 if (error) 1602 return (error); 1603 ssp = &set; 1604 } else 1605 ssp = NULL; 1606 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp)); 1607 } 1608 1609 /* 1610 * ufds points to an array in user space. 1611 */ 1612 int 1613 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds, 1614 struct timespec *tsp, sigset_t *set) 1615 { 1616 struct pollfd *kfds; 1617 struct pollfd stackfds[32]; 1618 int error; 1619 1620 if (kern_poll_maxfds(nfds)) 1621 return (EINVAL); 1622 if (nfds > nitems(stackfds)) 1623 kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK); 1624 else 1625 kfds = stackfds; 1626 error = copyin(ufds, kfds, nfds * sizeof(*kfds)); 1627 if (error != 0) 1628 goto out; 1629 1630 error = kern_poll_kfds(td, kfds, nfds, tsp, set); 1631 if (error == 0) 1632 error = pollout(td, kfds, ufds, nfds); 1633 #ifdef KTRACE 1634 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1635 ktrstructarray("pollfd", UIO_USERSPACE, ufds, nfds, 1636 sizeof(*ufds)); 1637 #endif 1638 1639 out: 1640 if (nfds > nitems(stackfds)) 1641 free(kfds, M_TEMP); 1642 return (error); 1643 } 1644 1645 bool 1646 kern_poll_maxfds(u_int nfds) 1647 { 1648 1649 /* 1650 * This is kinda bogus. We have fd limits, but that is not 1651 * really related to the size of the pollfd array. Make sure 1652 * we let the process use at least FD_SETSIZE entries and at 1653 * least enough for the system-wide limits. We want to be reasonably 1654 * safe, but not overly restrictive. 1655 */ 1656 return (nfds > maxfilesperproc && nfds > FD_SETSIZE); 1657 } 1658 1659 static int 1660 pollrescan(struct thread *td) 1661 { 1662 struct seltd *stp; 1663 struct selfd *sfp; 1664 struct selfd *sfn; 1665 struct selinfo *si; 1666 struct filedesc *fdp; 1667 struct file *fp; 1668 struct pollfd *fd; 1669 int n, error; 1670 bool only_user; 1671 1672 n = 0; 1673 fdp = td->td_proc->p_fd; 1674 stp = td->td_sel; 1675 only_user = FILEDESC_IS_ONLY_USER(fdp); 1676 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1677 fd = (struct pollfd *)sfp->sf_cookie; 1678 si = sfp->sf_si; 1679 selfdfree(stp, sfp); 1680 /* If the selinfo wasn't cleared the event didn't fire. */ 1681 if (si != NULL) 1682 continue; 1683 if (only_user) 1684 error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp); 1685 else 1686 error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp); 1687 if (__predict_false(error != 0)) { 1688 fd->revents = POLLNVAL; 1689 n++; 1690 continue; 1691 } 1692 /* 1693 * Note: backend also returns POLLHUP and 1694 * POLLERR if appropriate. 1695 */ 1696 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td); 1697 if (only_user) 1698 fput_only_user(fdp, fp); 1699 else 1700 fdrop(fp, td); 1701 if (fd->revents != 0) 1702 n++; 1703 } 1704 stp->st_flags = 0; 1705 td->td_retval[0] = n; 1706 return (0); 1707 } 1708 1709 static int 1710 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd) 1711 { 1712 int error = 0; 1713 u_int i = 0; 1714 u_int n = 0; 1715 1716 for (i = 0; i < nfd; i++) { 1717 error = copyout(&fds->revents, &ufds->revents, 1718 sizeof(ufds->revents)); 1719 if (error) 1720 return (error); 1721 if (fds->revents != 0) 1722 n++; 1723 fds++; 1724 ufds++; 1725 } 1726 td->td_retval[0] = n; 1727 return (0); 1728 } 1729 1730 static int 1731 pollscan(struct thread *td, struct pollfd *fds, u_int nfd) 1732 { 1733 struct filedesc *fdp; 1734 struct file *fp; 1735 int i, n, error; 1736 bool only_user; 1737 1738 n = 0; 1739 fdp = td->td_proc->p_fd; 1740 only_user = FILEDESC_IS_ONLY_USER(fdp); 1741 for (i = 0; i < nfd; i++, fds++) { 1742 if (fds->fd < 0) { 1743 fds->revents = 0; 1744 continue; 1745 } 1746 if (only_user) 1747 error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp); 1748 else 1749 error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp); 1750 if (__predict_false(error != 0)) { 1751 fds->revents = POLLNVAL; 1752 n++; 1753 continue; 1754 } 1755 /* 1756 * Note: backend also returns POLLHUP and 1757 * POLLERR if appropriate. 1758 */ 1759 selfdalloc(td, fds); 1760 fds->revents = fo_poll(fp, fds->events, 1761 td->td_ucred, td); 1762 if (only_user) 1763 fput_only_user(fdp, fp); 1764 else 1765 fdrop(fp, td); 1766 /* 1767 * POSIX requires POLLOUT to be never 1768 * set simultaneously with POLLHUP. 1769 */ 1770 if ((fds->revents & POLLHUP) != 0) 1771 fds->revents &= ~POLLOUT; 1772 1773 if (fds->revents != 0) 1774 n++; 1775 } 1776 td->td_retval[0] = n; 1777 return (0); 1778 } 1779 1780 /* 1781 * XXX This was created specifically to support netncp and netsmb. This 1782 * allows the caller to specify a socket to wait for events on. It returns 1783 * 0 if any events matched and an error otherwise. There is no way to 1784 * determine which events fired. 1785 */ 1786 int 1787 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td) 1788 { 1789 struct timeval rtv; 1790 sbintime_t asbt, precision, rsbt; 1791 int error; 1792 1793 precision = 0; /* stupid gcc! */ 1794 if (tvp != NULL) { 1795 rtv = *tvp; 1796 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1797 rtv.tv_usec >= 1000000) 1798 return (EINVAL); 1799 if (!timevalisset(&rtv)) 1800 asbt = 0; 1801 else if (rtv.tv_sec <= INT32_MAX) { 1802 rsbt = tvtosbt(rtv); 1803 precision = rsbt; 1804 precision >>= tc_precexp; 1805 if (TIMESEL(&asbt, rsbt)) 1806 asbt += tc_tick_sbt; 1807 if (asbt <= SBT_MAX - rsbt) 1808 asbt += rsbt; 1809 else 1810 asbt = -1; 1811 } else 1812 asbt = -1; 1813 } else 1814 asbt = -1; 1815 seltdinit(td); 1816 /* 1817 * Iterate until the timeout expires or the socket becomes ready. 1818 */ 1819 for (;;) { 1820 selfdalloc(td, NULL); 1821 if (so->so_proto->pr_sopoll(so, events, td) != 0) { 1822 error = 0; 1823 break; 1824 } 1825 error = seltdwait(td, asbt, precision); 1826 if (error) 1827 break; 1828 } 1829 seltdclear(td); 1830 /* XXX Duplicates ncp/smb behavior. */ 1831 if (error == ERESTART) 1832 error = 0; 1833 return (error); 1834 } 1835 1836 /* 1837 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines 1838 * have two select sets, one for read and another for write. 1839 */ 1840 static void 1841 selfdalloc(struct thread *td, void *cookie) 1842 { 1843 struct seltd *stp; 1844 1845 stp = td->td_sel; 1846 if (stp->st_free1 == NULL) 1847 stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO); 1848 stp->st_free1->sf_td = stp; 1849 stp->st_free1->sf_cookie = cookie; 1850 if (stp->st_free2 == NULL) 1851 stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO); 1852 stp->st_free2->sf_td = stp; 1853 stp->st_free2->sf_cookie = cookie; 1854 } 1855 1856 static void 1857 selfdfree(struct seltd *stp, struct selfd *sfp) 1858 { 1859 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link); 1860 /* 1861 * Paired with doselwakeup. 1862 */ 1863 if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) { 1864 mtx_lock(sfp->sf_mtx); 1865 if (sfp->sf_si != NULL) { 1866 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads); 1867 } 1868 mtx_unlock(sfp->sf_mtx); 1869 } 1870 free(sfp, M_SELFD); 1871 } 1872 1873 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */ 1874 void 1875 seldrain(struct selinfo *sip) 1876 { 1877 1878 /* 1879 * This feature is already provided by doselwakeup(), thus it is 1880 * enough to go for it. 1881 * Eventually, the context, should take care to avoid races 1882 * between thread calling select()/poll() and file descriptor 1883 * detaching, but, again, the races are just the same as 1884 * selwakeup(). 1885 */ 1886 doselwakeup(sip, -1); 1887 } 1888 1889 /* 1890 * Record a select request. 1891 */ 1892 void 1893 selrecord(struct thread *selector, struct selinfo *sip) 1894 { 1895 struct selfd *sfp; 1896 struct seltd *stp; 1897 struct mtx *mtxp; 1898 1899 stp = selector->td_sel; 1900 /* 1901 * Don't record when doing a rescan. 1902 */ 1903 if (stp->st_flags & SELTD_RESCAN) 1904 return; 1905 /* 1906 * Grab one of the preallocated descriptors. 1907 */ 1908 sfp = NULL; 1909 if ((sfp = stp->st_free1) != NULL) 1910 stp->st_free1 = NULL; 1911 else if ((sfp = stp->st_free2) != NULL) 1912 stp->st_free2 = NULL; 1913 else 1914 panic("selrecord: No free selfd on selq"); 1915 mtxp = sip->si_mtx; 1916 if (mtxp == NULL) 1917 mtxp = mtx_pool_find(mtxpool_select, sip); 1918 /* 1919 * Initialize the sfp and queue it in the thread. 1920 */ 1921 sfp->sf_si = sip; 1922 sfp->sf_mtx = mtxp; 1923 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link); 1924 /* 1925 * Now that we've locked the sip, check for initialization. 1926 */ 1927 mtx_lock(mtxp); 1928 if (sip->si_mtx == NULL) { 1929 sip->si_mtx = mtxp; 1930 TAILQ_INIT(&sip->si_tdlist); 1931 } 1932 /* 1933 * Add this thread to the list of selfds listening on this selinfo. 1934 */ 1935 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads); 1936 mtx_unlock(sip->si_mtx); 1937 } 1938 1939 /* Wake up a selecting thread. */ 1940 void 1941 selwakeup(struct selinfo *sip) 1942 { 1943 doselwakeup(sip, -1); 1944 } 1945 1946 /* Wake up a selecting thread, and set its priority. */ 1947 void 1948 selwakeuppri(struct selinfo *sip, int pri) 1949 { 1950 doselwakeup(sip, pri); 1951 } 1952 1953 /* 1954 * Do a wakeup when a selectable event occurs. 1955 */ 1956 static void 1957 doselwakeup(struct selinfo *sip, int pri) 1958 { 1959 struct selfd *sfp; 1960 struct selfd *sfn; 1961 struct seltd *stp; 1962 1963 /* If it's not initialized there can't be any waiters. */ 1964 if (sip->si_mtx == NULL) 1965 return; 1966 /* 1967 * Locking the selinfo locks all selfds associated with it. 1968 */ 1969 mtx_lock(sip->si_mtx); 1970 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) { 1971 /* 1972 * Once we remove this sfp from the list and clear the 1973 * sf_si seltdclear will know to ignore this si. 1974 */ 1975 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads); 1976 stp = sfp->sf_td; 1977 mtx_lock(&stp->st_mtx); 1978 stp->st_flags |= SELTD_PENDING; 1979 cv_broadcastpri(&stp->st_wait, pri); 1980 mtx_unlock(&stp->st_mtx); 1981 /* 1982 * Paired with selfdfree. 1983 * 1984 * Storing this only after the wakeup provides an invariant that 1985 * stp is not used after selfdfree returns. 1986 */ 1987 atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL); 1988 } 1989 mtx_unlock(sip->si_mtx); 1990 } 1991 1992 static void 1993 seltdinit(struct thread *td) 1994 { 1995 struct seltd *stp; 1996 1997 stp = td->td_sel; 1998 if (stp != NULL) { 1999 MPASS(stp->st_flags == 0); 2000 MPASS(STAILQ_EMPTY(&stp->st_selq)); 2001 return; 2002 } 2003 stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO); 2004 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF); 2005 cv_init(&stp->st_wait, "select"); 2006 stp->st_flags = 0; 2007 STAILQ_INIT(&stp->st_selq); 2008 td->td_sel = stp; 2009 } 2010 2011 static int 2012 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision) 2013 { 2014 struct seltd *stp; 2015 int error; 2016 2017 stp = td->td_sel; 2018 /* 2019 * An event of interest may occur while we do not hold the seltd 2020 * locked so check the pending flag before we sleep. 2021 */ 2022 mtx_lock(&stp->st_mtx); 2023 /* 2024 * Any further calls to selrecord will be a rescan. 2025 */ 2026 stp->st_flags |= SELTD_RESCAN; 2027 if (stp->st_flags & SELTD_PENDING) { 2028 mtx_unlock(&stp->st_mtx); 2029 return (0); 2030 } 2031 if (sbt == 0) 2032 error = EWOULDBLOCK; 2033 else if (sbt != -1) 2034 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx, 2035 sbt, precision, C_ABSOLUTE); 2036 else 2037 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx); 2038 mtx_unlock(&stp->st_mtx); 2039 2040 return (error); 2041 } 2042 2043 void 2044 seltdfini(struct thread *td) 2045 { 2046 struct seltd *stp; 2047 2048 stp = td->td_sel; 2049 if (stp == NULL) 2050 return; 2051 MPASS(stp->st_flags == 0); 2052 MPASS(STAILQ_EMPTY(&stp->st_selq)); 2053 if (stp->st_free1) 2054 free(stp->st_free1, M_SELFD); 2055 if (stp->st_free2) 2056 free(stp->st_free2, M_SELFD); 2057 td->td_sel = NULL; 2058 cv_destroy(&stp->st_wait); 2059 mtx_destroy(&stp->st_mtx); 2060 free(stp, M_SELECT); 2061 } 2062 2063 /* 2064 * Remove the references to the thread from all of the objects we were 2065 * polling. 2066 */ 2067 static void 2068 seltdclear(struct thread *td) 2069 { 2070 struct seltd *stp; 2071 struct selfd *sfp; 2072 struct selfd *sfn; 2073 2074 stp = td->td_sel; 2075 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) 2076 selfdfree(stp, sfp); 2077 stp->st_flags = 0; 2078 } 2079 2080 static void selectinit(void *); 2081 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL); 2082 static void 2083 selectinit(void *dummy __unused) 2084 { 2085 2086 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF); 2087 } 2088 2089 /* 2090 * Set up a syscall return value that follows the convention specified for 2091 * posix_* functions. 2092 */ 2093 int 2094 kern_posix_error(struct thread *td, int error) 2095 { 2096 2097 if (error <= 0) 2098 return (error); 2099 td->td_errno = error; 2100 td->td_pflags |= TDP_NERRNO; 2101 td->td_retval[0] = error; 2102 return (0); 2103 } 2104 2105 int 2106 kcmp_cmp(uintptr_t a, uintptr_t b) 2107 { 2108 if (a == b) 2109 return (0); 2110 else if (a < b) 2111 return (1); 2112 return (2); 2113 } 2114 2115 static int 2116 kcmp_pget(struct thread *td, pid_t pid, struct proc **pp) 2117 { 2118 int error; 2119 2120 if (pid == td->td_proc->p_pid) { 2121 *pp = td->td_proc; 2122 return (0); 2123 } 2124 error = pget(pid, PGET_NOTID | PGET_CANDEBUG | PGET_NOTWEXIT | 2125 PGET_HOLD, pp); 2126 MPASS(*pp != td->td_proc); 2127 return (error); 2128 } 2129 2130 int 2131 kern_kcmp(struct thread *td, pid_t pid1, pid_t pid2, int type, 2132 uintptr_t idx1, uintptr_t idx2) 2133 { 2134 struct proc *p1, *p2; 2135 struct file *fp1, *fp2; 2136 int error, res; 2137 2138 res = -1; 2139 p1 = p2 = NULL; 2140 error = kcmp_pget(td, pid1, &p1); 2141 if (error == 0) 2142 error = kcmp_pget(td, pid2, &p2); 2143 if (error != 0) 2144 goto out; 2145 2146 switch (type) { 2147 case KCMP_FILE: 2148 case KCMP_FILEOBJ: 2149 error = fget_remote(td, p1, idx1, &fp1); 2150 if (error == 0) { 2151 error = fget_remote(td, p2, idx2, &fp2); 2152 if (error == 0) { 2153 if (type == KCMP_FILEOBJ) 2154 res = fo_cmp(fp1, fp2, td); 2155 else 2156 res = kcmp_cmp((uintptr_t)fp1, 2157 (uintptr_t)fp2); 2158 fdrop(fp2, td); 2159 } 2160 fdrop(fp1, td); 2161 } 2162 break; 2163 case KCMP_FILES: 2164 res = kcmp_cmp((uintptr_t)p1->p_fd, (uintptr_t)p2->p_fd); 2165 break; 2166 case KCMP_SIGHAND: 2167 res = kcmp_cmp((uintptr_t)p1->p_sigacts, 2168 (uintptr_t)p2->p_sigacts); 2169 break; 2170 case KCMP_VM: 2171 res = kcmp_cmp((uintptr_t)p1->p_vmspace, 2172 (uintptr_t)p2->p_vmspace); 2173 break; 2174 default: 2175 error = EINVAL; 2176 break; 2177 } 2178 2179 out: 2180 if (p1 != NULL && p1 != td->td_proc) 2181 PRELE(p1); 2182 if (p2 != NULL && p2 != td->td_proc) 2183 PRELE(p2); 2184 2185 td->td_retval[0] = res; 2186 return (error); 2187 } 2188 2189 int 2190 sys_kcmp(struct thread *td, struct kcmp_args *uap) 2191 { 2192 return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type, 2193 uap->idx1, uap->idx2)); 2194 } 2195 2196 int 2197 file_kcmp_generic(struct file *fp1, struct file *fp2, struct thread *td) 2198 { 2199 if (fp1->f_type != fp2->f_type) 2200 return (3); 2201 return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data)); 2202 } 2203