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 "opt_capsicum.h" 38 #include "opt_ktrace.h" 39 40 #define EXTERR_CATEGORY EXTERR_CAT_FILEDESC 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/exterrvar.h> 50 #include <sys/lock.h> 51 #include <sys/proc.h> 52 #include <sys/signalvar.h> 53 #include <sys/protosw.h> 54 #include <sys/socketvar.h> 55 #include <sys/uio.h> 56 #include <sys/eventfd.h> 57 #include <sys/kernel.h> 58 #include <sys/ktr.h> 59 #include <sys/limits.h> 60 #include <sys/malloc.h> 61 #include <sys/poll.h> 62 #include <sys/resourcevar.h> 63 #include <sys/selinfo.h> 64 #include <sys/sleepqueue.h> 65 #include <sys/specialfd.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/sysctl.h> 68 #include <sys/sysent.h> 69 #include <sys/vnode.h> 70 #include <sys/unistd.h> 71 #include <sys/bio.h> 72 #include <sys/buf.h> 73 #include <sys/condvar.h> 74 #ifdef KTRACE 75 #include <sys/ktrace.h> 76 #endif 77 78 #include <security/audit/audit.h> 79 80 /* 81 * The following macro defines how many bytes will be allocated from 82 * the stack instead of memory allocated when passing the IOCTL data 83 * structures from userspace and to the kernel. Some IOCTLs having 84 * small data structures are used very frequently and this small 85 * buffer on the stack gives a significant speedup improvement for 86 * those requests. The value of this define should be greater or equal 87 * to 64 bytes and should also be power of two. The data structure is 88 * currently hard-aligned to a 8-byte boundary on the stack. This 89 * should currently be sufficient for all supported platforms. 90 */ 91 #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */ 92 #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */ 93 94 #ifdef __LP64__ 95 static int iosize_max_clamp = 0; 96 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW, 97 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX"); 98 static int devfs_iosize_max_clamp = 1; 99 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW, 100 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices"); 101 #endif 102 103 /* 104 * Assert that the return value of read(2) and write(2) syscalls fits 105 * into a register. If not, an architecture will need to provide the 106 * usermode wrappers to reconstruct the result. 107 */ 108 CTASSERT(sizeof(register_t) >= sizeof(size_t)); 109 110 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); 111 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); 112 MALLOC_DEFINE(M_IOV, "iov", "large iov's"); 113 114 static int pollout(struct thread *, struct pollfd *, struct pollfd *, 115 u_int); 116 static int pollscan(struct thread *, struct pollfd *, u_int); 117 static int pollrescan(struct thread *); 118 static int selscan(struct thread *, fd_mask **, fd_mask **, int); 119 static int selrescan(struct thread *, fd_mask **, fd_mask **); 120 static void selfdalloc(struct thread *, void *); 121 static void selfdfree(struct seltd *, struct selfd *); 122 static int dofileread(struct thread *, int, struct file *, struct uio *, 123 off_t, int); 124 static int dofilewrite(struct thread *, int, struct file *, struct uio *, 125 off_t, int); 126 static void doselwakeup(struct selinfo *, int); 127 static void seltdinit(struct thread *); 128 static int seltdwait(struct thread *, sbintime_t, sbintime_t); 129 static void seltdclear(struct thread *); 130 131 /* 132 * One seltd per-thread allocated on demand as needed. 133 * 134 * t - protected by st_mtx 135 * k - Only accessed by curthread or read-only 136 */ 137 struct seltd { 138 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */ 139 struct selfd *st_free1; /* (k) free fd for read set. */ 140 struct selfd *st_free2; /* (k) free fd for write set. */ 141 struct mtx st_mtx; /* Protects struct seltd */ 142 struct cv st_wait; /* (t) Wait channel. */ 143 int st_flags; /* (t) SELTD_ flags. */ 144 }; 145 146 #define SELTD_PENDING 0x0001 /* We have pending events. */ 147 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */ 148 149 /* 150 * One selfd allocated per-thread per-file-descriptor. 151 * f - protected by sf_mtx 152 */ 153 struct selfd { 154 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */ 155 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */ 156 struct selinfo *sf_si; /* (f) selinfo when linked. */ 157 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */ 158 struct seltd *sf_td; /* (k) owning seltd. */ 159 void *sf_cookie; /* (k) fd or pollfd. */ 160 }; 161 162 MALLOC_DEFINE(M_SELFD, "selfd", "selfd"); 163 static struct mtx_pool *mtxpool_select; 164 165 #ifdef __LP64__ 166 size_t 167 devfs_iosize_max(void) 168 { 169 170 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 171 INT_MAX : SSIZE_MAX); 172 } 173 174 size_t 175 iosize_max(void) 176 { 177 178 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? 179 INT_MAX : SSIZE_MAX); 180 } 181 #endif 182 183 #ifndef _SYS_SYSPROTO_H_ 184 struct read_args { 185 int fd; 186 void *buf; 187 size_t nbyte; 188 }; 189 #endif 190 int 191 sys_read(struct thread *td, struct read_args *uap) 192 { 193 struct uio auio; 194 struct iovec aiov; 195 int error; 196 197 if (uap->nbyte > IOSIZE_MAX) 198 return (EINVAL); 199 aiov.iov_base = uap->buf; 200 aiov.iov_len = uap->nbyte; 201 auio.uio_iov = &aiov; 202 auio.uio_iovcnt = 1; 203 auio.uio_resid = uap->nbyte; 204 auio.uio_segflg = UIO_USERSPACE; 205 error = kern_readv(td, uap->fd, &auio); 206 return (error); 207 } 208 209 /* 210 * Positioned read system call 211 */ 212 #ifndef _SYS_SYSPROTO_H_ 213 struct pread_args { 214 int fd; 215 void *buf; 216 size_t nbyte; 217 int pad; 218 off_t offset; 219 }; 220 #endif 221 int 222 sys_pread(struct thread *td, struct pread_args *uap) 223 { 224 225 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 226 } 227 228 int 229 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset) 230 { 231 struct uio auio; 232 struct iovec aiov; 233 int error; 234 235 if (nbyte > IOSIZE_MAX) 236 return (EINVAL); 237 aiov.iov_base = buf; 238 aiov.iov_len = nbyte; 239 auio.uio_iov = &aiov; 240 auio.uio_iovcnt = 1; 241 auio.uio_resid = nbyte; 242 auio.uio_segflg = UIO_USERSPACE; 243 error = kern_preadv(td, fd, &auio, offset); 244 return (error); 245 } 246 247 #if defined(COMPAT_FREEBSD6) 248 int 249 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap) 250 { 251 252 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 253 } 254 #endif 255 256 /* 257 * Scatter read system call. 258 */ 259 #ifndef _SYS_SYSPROTO_H_ 260 struct readv_args { 261 int fd; 262 struct iovec *iovp; 263 u_int iovcnt; 264 }; 265 #endif 266 int 267 sys_readv(struct thread *td, struct readv_args *uap) 268 { 269 struct uio *auio; 270 int error; 271 272 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 273 if (error) 274 return (error); 275 error = kern_readv(td, uap->fd, auio); 276 freeuio(auio); 277 return (error); 278 } 279 280 int 281 kern_readv(struct thread *td, int fd, struct uio *auio) 282 { 283 struct file *fp; 284 int error; 285 286 error = fget_read(td, fd, &cap_read_rights, &fp); 287 if (error) 288 return (error); 289 error = dofileread(td, fd, fp, auio, (off_t)-1, 0); 290 fdrop(fp, td); 291 return (error); 292 } 293 294 /* 295 * Scatter positioned read system call. 296 */ 297 #ifndef _SYS_SYSPROTO_H_ 298 struct preadv_args { 299 int fd; 300 struct iovec *iovp; 301 u_int iovcnt; 302 off_t offset; 303 }; 304 #endif 305 int 306 sys_preadv(struct thread *td, struct preadv_args *uap) 307 { 308 struct uio *auio; 309 int error; 310 311 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 312 if (error) 313 return (error); 314 error = kern_preadv(td, uap->fd, auio, uap->offset); 315 freeuio(auio); 316 return (error); 317 } 318 319 int 320 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset) 321 { 322 struct file *fp; 323 int error; 324 325 error = fget_read(td, fd, &cap_pread_rights, &fp); 326 if (error) 327 return (error); 328 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 329 error = ESPIPE; 330 else if (offset < 0 && 331 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 332 error = EINVAL; 333 else 334 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET); 335 fdrop(fp, td); 336 return (error); 337 } 338 339 /* 340 * Common code for readv and preadv that reads data in 341 * from a file using the passed in uio, offset, and flags. 342 */ 343 static int 344 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio, 345 off_t offset, int flags) 346 { 347 ssize_t cnt; 348 int error; 349 #ifdef KTRACE 350 struct uio *ktruio = NULL; 351 #endif 352 353 AUDIT_ARG_FD(fd); 354 355 /* Finish zero length reads right here */ 356 if (auio->uio_resid == 0) { 357 td->td_retval[0] = 0; 358 return (0); 359 } 360 auio->uio_rw = UIO_READ; 361 auio->uio_offset = offset; 362 auio->uio_td = td; 363 #ifdef KTRACE 364 if (KTRPOINT(td, KTR_GENIO)) 365 ktruio = cloneuio(auio); 366 #endif 367 cnt = auio->uio_resid; 368 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) { 369 if (auio->uio_resid != cnt && (error == ERESTART || 370 error == EINTR || error == EWOULDBLOCK)) 371 error = 0; 372 } 373 cnt -= auio->uio_resid; 374 #ifdef KTRACE 375 if (ktruio != NULL) { 376 ktruio->uio_resid = cnt; 377 ktrgenio(fd, UIO_READ, ktruio, error); 378 } 379 #endif 380 td->td_retval[0] = cnt; 381 return (error); 382 } 383 384 #ifndef _SYS_SYSPROTO_H_ 385 struct write_args { 386 int fd; 387 const void *buf; 388 size_t nbyte; 389 }; 390 #endif 391 int 392 sys_write(struct thread *td, struct write_args *uap) 393 { 394 struct uio auio; 395 struct iovec aiov; 396 int error; 397 398 if (uap->nbyte > IOSIZE_MAX) 399 return (EINVAL); 400 aiov.iov_base = (void *)(uintptr_t)uap->buf; 401 aiov.iov_len = uap->nbyte; 402 auio.uio_iov = &aiov; 403 auio.uio_iovcnt = 1; 404 auio.uio_resid = uap->nbyte; 405 auio.uio_segflg = UIO_USERSPACE; 406 error = kern_writev(td, uap->fd, &auio); 407 return (error); 408 } 409 410 /* 411 * Positioned write system call. 412 */ 413 #ifndef _SYS_SYSPROTO_H_ 414 struct pwrite_args { 415 int fd; 416 const void *buf; 417 size_t nbyte; 418 int pad; 419 off_t offset; 420 }; 421 #endif 422 int 423 sys_pwrite(struct thread *td, struct pwrite_args *uap) 424 { 425 426 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 427 } 428 429 int 430 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte, 431 off_t offset) 432 { 433 struct uio auio; 434 struct iovec aiov; 435 int error; 436 437 if (nbyte > IOSIZE_MAX) 438 return (EINVAL); 439 aiov.iov_base = (void *)(uintptr_t)buf; 440 aiov.iov_len = nbyte; 441 auio.uio_iov = &aiov; 442 auio.uio_iovcnt = 1; 443 auio.uio_resid = nbyte; 444 auio.uio_segflg = UIO_USERSPACE; 445 error = kern_pwritev(td, fd, &auio, offset); 446 return (error); 447 } 448 449 #if defined(COMPAT_FREEBSD6) 450 int 451 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap) 452 { 453 454 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset)); 455 } 456 #endif 457 458 /* 459 * Gather write system call. 460 */ 461 #ifndef _SYS_SYSPROTO_H_ 462 struct writev_args { 463 int fd; 464 struct iovec *iovp; 465 u_int iovcnt; 466 }; 467 #endif 468 int 469 sys_writev(struct thread *td, struct writev_args *uap) 470 { 471 struct uio *auio; 472 int error; 473 474 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 475 if (error) 476 return (error); 477 error = kern_writev(td, uap->fd, auio); 478 freeuio(auio); 479 return (error); 480 } 481 482 int 483 kern_writev(struct thread *td, int fd, struct uio *auio) 484 { 485 struct file *fp; 486 int error; 487 488 error = fget_write(td, fd, &cap_write_rights, &fp); 489 if (error) 490 return (error); 491 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0); 492 fdrop(fp, td); 493 return (error); 494 } 495 496 /* 497 * Gather positioned write system call. 498 */ 499 #ifndef _SYS_SYSPROTO_H_ 500 struct pwritev_args { 501 int fd; 502 struct iovec *iovp; 503 u_int iovcnt; 504 off_t offset; 505 }; 506 #endif 507 int 508 sys_pwritev(struct thread *td, struct pwritev_args *uap) 509 { 510 struct uio *auio; 511 int error; 512 513 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 514 if (error) 515 return (error); 516 error = kern_pwritev(td, uap->fd, auio, uap->offset); 517 freeuio(auio); 518 return (error); 519 } 520 521 int 522 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset) 523 { 524 struct file *fp; 525 int error; 526 527 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 528 if (error) 529 return (error); 530 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) 531 error = ESPIPE; 532 else if (offset < 0 && 533 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) 534 error = EINVAL; 535 else 536 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET); 537 fdrop(fp, td); 538 return (error); 539 } 540 541 /* 542 * Common code for writev and pwritev that writes data to 543 * a file using the passed in uio, offset, and flags. 544 */ 545 static int 546 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio, 547 off_t offset, int flags) 548 { 549 ssize_t cnt; 550 int error; 551 #ifdef KTRACE 552 struct uio *ktruio = NULL; 553 #endif 554 555 AUDIT_ARG_FD(fd); 556 auio->uio_rw = UIO_WRITE; 557 auio->uio_td = td; 558 auio->uio_offset = offset; 559 #ifdef KTRACE 560 if (KTRPOINT(td, KTR_GENIO)) 561 ktruio = cloneuio(auio); 562 #endif 563 cnt = auio->uio_resid; 564 error = fo_write(fp, auio, td->td_ucred, flags, td); 565 /* 566 * Socket layer is responsible for special error handling, 567 * see sousrsend(). 568 */ 569 if (error != 0 && fp->f_type != DTYPE_SOCKET) { 570 if (auio->uio_resid != cnt && (error == ERESTART || 571 error == EINTR || error == EWOULDBLOCK)) 572 error = 0; 573 if (error == EPIPE) { 574 PROC_LOCK(td->td_proc); 575 tdsignal(td, SIGPIPE); 576 PROC_UNLOCK(td->td_proc); 577 } 578 } 579 cnt -= auio->uio_resid; 580 #ifdef KTRACE 581 if (ktruio != NULL) { 582 if (error == 0) 583 ktruio->uio_resid = cnt; 584 ktrgenio(fd, UIO_WRITE, ktruio, error); 585 } 586 #endif 587 td->td_retval[0] = cnt; 588 return (error); 589 } 590 591 /* 592 * Truncate a file given a file descriptor. 593 * 594 * Can't use fget_write() here, since must return EINVAL and not EBADF if the 595 * descriptor isn't writable. 596 */ 597 int 598 kern_ftruncate(struct thread *td, int fd, off_t length) 599 { 600 struct file *fp; 601 int error; 602 603 AUDIT_ARG_FD(fd); 604 if (length < 0) 605 return (EINVAL); 606 error = fget(td, fd, &cap_ftruncate_rights, &fp); 607 if (error) 608 return (error); 609 AUDIT_ARG_FILE(td->td_proc, fp); 610 if (!(fp->f_flag & FWRITE)) { 611 fdrop(fp, td); 612 return (EINVAL); 613 } 614 error = fo_truncate(fp, length, td->td_ucred, td); 615 fdrop(fp, td); 616 return (error); 617 } 618 619 #ifndef _SYS_SYSPROTO_H_ 620 struct ftruncate_args { 621 int fd; 622 int pad; 623 off_t length; 624 }; 625 #endif 626 int 627 sys_ftruncate(struct thread *td, struct ftruncate_args *uap) 628 { 629 630 return (kern_ftruncate(td, uap->fd, uap->length)); 631 } 632 633 #if defined(COMPAT_43) 634 #ifndef _SYS_SYSPROTO_H_ 635 struct oftruncate_args { 636 int fd; 637 long length; 638 }; 639 #endif 640 int 641 oftruncate(struct thread *td, struct oftruncate_args *uap) 642 { 643 644 return (kern_ftruncate(td, uap->fd, uap->length)); 645 } 646 #endif /* COMPAT_43 */ 647 648 #ifndef _SYS_SYSPROTO_H_ 649 struct ioctl_args { 650 int fd; 651 u_long com; 652 caddr_t data; 653 }; 654 #endif 655 /* ARGSUSED */ 656 int 657 sys_ioctl(struct thread *td, struct ioctl_args *uap) 658 { 659 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN); 660 uint32_t com; 661 int arg, error; 662 u_int size; 663 caddr_t data; 664 665 #ifdef INVARIANTS 666 if (uap->com > 0xffffffff) { 667 printf( 668 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n", 669 td->td_proc->p_pid, td->td_name, uap->com); 670 } 671 #endif 672 com = (uint32_t)uap->com; 673 674 /* 675 * Interpret high order word to find amount of data to be 676 * copied to/from the user's address space. 677 */ 678 size = IOCPARM_LEN(com); 679 if ((size > IOCPARM_MAX) || 680 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) || 681 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43) 682 ((com & IOC_OUT) && size == 0) || 683 #else 684 ((com & (IOC_IN | IOC_OUT)) && size == 0) || 685 #endif 686 ((com & IOC_VOID) && size > 0 && size != sizeof(int))) 687 return (ENOTTY); 688 689 if (size > 0) { 690 if (com & IOC_VOID) { 691 /* Integer argument. */ 692 arg = (intptr_t)uap->data; 693 data = (void *)&arg; 694 size = 0; 695 } else { 696 if (size > SYS_IOCTL_SMALL_SIZE) 697 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK); 698 else 699 data = smalldata; 700 } 701 } else 702 data = (void *)&uap->data; 703 if (com & IOC_IN) { 704 error = copyin(uap->data, data, (u_int)size); 705 if (error != 0) 706 goto out; 707 } else if (com & IOC_OUT) { 708 /* 709 * Zero the buffer so the user always 710 * gets back something deterministic. 711 */ 712 bzero(data, size); 713 } 714 715 error = kern_ioctl(td, uap->fd, com, data); 716 717 if (error == 0 && (com & IOC_OUT)) 718 error = copyout(data, uap->data, (u_int)size); 719 720 out: 721 if (size > SYS_IOCTL_SMALL_SIZE) 722 free(data, M_IOCTLOPS); 723 return (error); 724 } 725 726 int 727 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data) 728 { 729 struct file *fp; 730 struct filedesc *fdp; 731 int error, tmp, locked; 732 733 AUDIT_ARG_FD(fd); 734 AUDIT_ARG_CMD(com); 735 736 fdp = td->td_proc->p_fd; 737 738 switch (com) { 739 case FIONCLEX: 740 case FIOCLEX: 741 FILEDESC_XLOCK(fdp); 742 locked = LA_XLOCKED; 743 break; 744 default: 745 #ifdef CAPABILITIES 746 FILEDESC_SLOCK(fdp); 747 locked = LA_SLOCKED; 748 #else 749 locked = LA_UNLOCKED; 750 #endif 751 break; 752 } 753 754 #ifdef CAPABILITIES 755 if ((fp = fget_noref(fdp, fd)) == NULL) { 756 error = EBADF; 757 goto out; 758 } 759 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) { 760 fp = NULL; /* fhold() was not called yet */ 761 goto out; 762 } 763 if (!fhold(fp)) { 764 error = EBADF; 765 fp = NULL; 766 goto out; 767 } 768 if (locked == LA_SLOCKED) { 769 FILEDESC_SUNLOCK(fdp); 770 locked = LA_UNLOCKED; 771 } 772 #else 773 error = fget(td, fd, &cap_ioctl_rights, &fp); 774 if (error != 0) { 775 fp = NULL; 776 goto out; 777 } 778 #endif 779 if ((fp->f_flag & (FREAD | FWRITE)) == 0) { 780 error = EBADF; 781 goto out; 782 } 783 784 switch (com) { 785 case FIONCLEX: 786 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE; 787 goto out; 788 case FIOCLEX: 789 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE; 790 goto out; 791 case FIONBIO: 792 if ((tmp = *(int *)data)) 793 atomic_set_int(&fp->f_flag, FNONBLOCK); 794 else 795 atomic_clear_int(&fp->f_flag, FNONBLOCK); 796 data = (void *)&tmp; 797 break; 798 case FIOASYNC: 799 if ((tmp = *(int *)data)) 800 atomic_set_int(&fp->f_flag, FASYNC); 801 else 802 atomic_clear_int(&fp->f_flag, FASYNC); 803 data = (void *)&tmp; 804 break; 805 } 806 807 error = fo_ioctl(fp, com, data, td->td_ucred, td); 808 out: 809 switch (locked) { 810 case LA_XLOCKED: 811 FILEDESC_XUNLOCK(fdp); 812 break; 813 #ifdef CAPABILITIES 814 case LA_SLOCKED: 815 FILEDESC_SUNLOCK(fdp); 816 break; 817 #endif 818 default: 819 FILEDESC_UNLOCK_ASSERT(fdp); 820 break; 821 } 822 if (fp != NULL) 823 fdrop(fp, td); 824 return (error); 825 } 826 827 int 828 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) 829 { 830 int error; 831 832 error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len); 833 return (kern_posix_error(td, error)); 834 } 835 836 int 837 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len) 838 { 839 struct file *fp; 840 int error; 841 842 AUDIT_ARG_FD(fd); 843 if (offset < 0 || len <= 0) 844 return (EINVAL); 845 /* Check for wrap. */ 846 if (offset > OFF_MAX - len) 847 return (EFBIG); 848 AUDIT_ARG_FD(fd); 849 error = fget(td, fd, &cap_pwrite_rights, &fp); 850 if (error != 0) 851 return (error); 852 AUDIT_ARG_FILE(td->td_proc, fp); 853 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 854 error = ESPIPE; 855 goto out; 856 } 857 if ((fp->f_flag & FWRITE) == 0) { 858 error = EBADF; 859 goto out; 860 } 861 862 error = fo_fallocate(fp, offset, len, td); 863 out: 864 fdrop(fp, td); 865 return (error); 866 } 867 868 int 869 sys_fspacectl(struct thread *td, struct fspacectl_args *uap) 870 { 871 struct spacectl_range rqsr, rmsr; 872 int error, cerror; 873 874 error = copyin(uap->rqsr, &rqsr, sizeof(rqsr)); 875 if (error != 0) 876 return (error); 877 878 error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags, 879 &rmsr); 880 if (uap->rmsr != NULL) { 881 cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr)); 882 if (error == 0) 883 error = cerror; 884 } 885 return (error); 886 } 887 888 int 889 kern_fspacectl(struct thread *td, int fd, int cmd, 890 const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp) 891 { 892 struct file *fp; 893 struct spacectl_range rmsr; 894 int error; 895 896 AUDIT_ARG_FD(fd); 897 AUDIT_ARG_CMD(cmd); 898 AUDIT_ARG_FFLAGS(flags); 899 900 if (rqsr == NULL) 901 return (EINVAL); 902 rmsr = *rqsr; 903 if (rmsrp != NULL) 904 *rmsrp = rmsr; 905 906 if (cmd != SPACECTL_DEALLOC || 907 rqsr->r_offset < 0 || rqsr->r_len <= 0 || 908 rqsr->r_offset > OFF_MAX - rqsr->r_len || 909 (flags & ~SPACECTL_F_SUPPORTED) != 0) 910 return (EINVAL); 911 912 error = fget_write(td, fd, &cap_pwrite_rights, &fp); 913 if (error != 0) 914 return (error); 915 AUDIT_ARG_FILE(td->td_proc, fp); 916 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { 917 error = ESPIPE; 918 goto out; 919 } 920 if ((fp->f_flag & FWRITE) == 0) { 921 error = EBADF; 922 goto out; 923 } 924 925 error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags, 926 td->td_ucred, td); 927 /* fspacectl is not restarted after signals if the file is modified. */ 928 if (rmsr.r_len != rqsr->r_len && (error == ERESTART || 929 error == EINTR || error == EWOULDBLOCK)) 930 error = 0; 931 if (rmsrp != NULL) 932 *rmsrp = rmsr; 933 out: 934 fdrop(fp, td); 935 return (error); 936 } 937 938 int 939 kern_specialfd(struct thread *td, int type, void *arg) 940 { 941 struct file *fp; 942 struct specialfd_eventfd *ae; 943 int error, fd, fflags; 944 945 fflags = 0; 946 error = falloc_noinstall(td, &fp); 947 if (error != 0) 948 return (error); 949 950 switch (type) { 951 case SPECIALFD_EVENTFD: 952 ae = arg; 953 if ((ae->flags & EFD_CLOEXEC) != 0) 954 fflags |= O_CLOEXEC; 955 error = eventfd_create_file(td, fp, ae->initval, ae->flags); 956 break; 957 default: 958 error = EINVAL; 959 break; 960 } 961 962 if (error == 0) 963 error = finstall(td, fp, &fd, fflags, NULL); 964 fdrop(fp, td); 965 if (error == 0) 966 td->td_retval[0] = fd; 967 return (error); 968 } 969 970 int 971 sys___specialfd(struct thread *td, struct __specialfd_args *args) 972 { 973 struct specialfd_eventfd ae; 974 int error; 975 976 switch (args->type) { 977 case SPECIALFD_EVENTFD: 978 if (args->len != sizeof(struct specialfd_eventfd)) { 979 error = EINVAL; 980 break; 981 } 982 error = copyin(args->req, &ae, sizeof(ae)); 983 if (error != 0) 984 break; 985 if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK | 986 EFD_SEMAPHORE)) != 0) { 987 error = EINVAL; 988 break; 989 } 990 error = kern_specialfd(td, args->type, &ae); 991 break; 992 default: 993 error = EINVAL; 994 break; 995 } 996 return (error); 997 } 998 999 int 1000 poll_no_poll(int events) 1001 { 1002 /* 1003 * Return true for read/write. If the user asked for something 1004 * special, return POLLNVAL, so that clients have a way of 1005 * determining reliably whether or not the extended 1006 * functionality is present without hard-coding knowledge 1007 * of specific filesystem implementations. 1008 */ 1009 if (events & ~POLLSTANDARD) 1010 return (POLLNVAL); 1011 1012 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 1013 } 1014 1015 int 1016 sys_pselect(struct thread *td, struct pselect_args *uap) 1017 { 1018 struct timespec ts; 1019 struct timeval tv, *tvp; 1020 sigset_t set, *uset; 1021 int error; 1022 1023 if (uap->ts != NULL) { 1024 error = copyin(uap->ts, &ts, sizeof(ts)); 1025 if (error != 0) 1026 return (error); 1027 TIMESPEC_TO_TIMEVAL(&tv, &ts); 1028 tvp = &tv; 1029 } else 1030 tvp = NULL; 1031 if (uap->sm != NULL) { 1032 error = copyin(uap->sm, &set, sizeof(set)); 1033 if (error != 0) 1034 return (error); 1035 uset = &set; 1036 } else 1037 uset = NULL; 1038 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1039 uset, NFDBITS)); 1040 } 1041 1042 int 1043 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex, 1044 struct timeval *tvp, sigset_t *uset, int abi_nfdbits) 1045 { 1046 int error; 1047 1048 if (uset != NULL) { 1049 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1050 &td->td_oldsigmask, 0); 1051 if (error != 0) 1052 return (error); 1053 td->td_pflags |= TDP_OLDMASK; 1054 } 1055 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits); 1056 if (uset != NULL) { 1057 /* 1058 * Make sure that ast() is called on return to 1059 * usermode and TDP_OLDMASK is cleared, restoring old 1060 * sigmask. If we didn't get interrupted, then the caller is 1061 * likely not expecting a signal to hit that should normally be 1062 * blocked by its signal mask, so we restore the mask before 1063 * any signals could be delivered. 1064 */ 1065 if (error == EINTR) { 1066 ast_sched(td, TDA_SIGSUSPEND); 1067 } else { 1068 /* *select(2) should never restart. */ 1069 MPASS(error != ERESTART); 1070 ast_sched(td, TDA_PSELECT); 1071 } 1072 } 1073 1074 return (error); 1075 } 1076 1077 #ifndef _SYS_SYSPROTO_H_ 1078 struct select_args { 1079 int nd; 1080 fd_set *in, *ou, *ex; 1081 struct timeval *tv; 1082 }; 1083 #endif 1084 int 1085 sys_select(struct thread *td, struct select_args *uap) 1086 { 1087 struct timeval tv, *tvp; 1088 int error; 1089 1090 if (uap->tv != NULL) { 1091 error = copyin(uap->tv, &tv, sizeof(tv)); 1092 if (error) 1093 return (error); 1094 tvp = &tv; 1095 } else 1096 tvp = NULL; 1097 1098 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, 1099 NFDBITS)); 1100 } 1101 1102 /* 1103 * In the unlikely case when user specified n greater then the last 1104 * open file descriptor, check that no bits are set after the last 1105 * valid fd. We must return EBADF if any is set. 1106 * 1107 * There are applications that rely on the behaviour. 1108 * 1109 * nd is fd_nfiles. 1110 */ 1111 static int 1112 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits) 1113 { 1114 char *addr, *oaddr; 1115 int b, i, res; 1116 uint8_t bits; 1117 1118 if (nd >= ndu || fd_in == NULL) 1119 return (0); 1120 1121 oaddr = NULL; 1122 bits = 0; /* silence gcc */ 1123 for (i = nd; i < ndu; i++) { 1124 b = i / NBBY; 1125 #if BYTE_ORDER == LITTLE_ENDIAN 1126 addr = (char *)fd_in + b; 1127 #else 1128 addr = (char *)fd_in; 1129 if (abi_nfdbits == NFDBITS) { 1130 addr += rounddown(b, sizeof(fd_mask)) + 1131 sizeof(fd_mask) - 1 - b % sizeof(fd_mask); 1132 } else { 1133 addr += rounddown(b, sizeof(uint32_t)) + 1134 sizeof(uint32_t) - 1 - b % sizeof(uint32_t); 1135 } 1136 #endif 1137 if (addr != oaddr) { 1138 res = fubyte(addr); 1139 if (res == -1) 1140 return (EFAULT); 1141 oaddr = addr; 1142 bits = res; 1143 } 1144 if ((bits & (1 << (i % NBBY))) != 0) 1145 return (EBADF); 1146 } 1147 return (0); 1148 } 1149 1150 int 1151 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou, 1152 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits) 1153 { 1154 struct filedesc *fdp; 1155 /* 1156 * The magic 2048 here is chosen to be just enough for FD_SETSIZE 1157 * infds with the new FD_SETSIZE of 1024, and more than enough for 1158 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE 1159 * of 256. 1160 */ 1161 fd_mask s_selbits[howmany(2048, NFDBITS)]; 1162 fd_mask *ibits[3], *obits[3], *selbits, *sbp; 1163 struct timeval rtv; 1164 sbintime_t asbt, precision, rsbt; 1165 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits; 1166 int error, lf, ndu; 1167 1168 if (nd < 0) 1169 return (EINVAL); 1170 fdp = td->td_proc->p_fd; 1171 ndu = nd; 1172 lf = fdp->fd_nfiles; 1173 if (nd > lf) 1174 nd = lf; 1175 1176 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits); 1177 if (error != 0) 1178 return (error); 1179 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits); 1180 if (error != 0) 1181 return (error); 1182 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits); 1183 if (error != 0) 1184 return (error); 1185 1186 /* 1187 * Allocate just enough bits for the non-null fd_sets. Use the 1188 * preallocated auto buffer if possible. 1189 */ 1190 nfdbits = roundup(nd, NFDBITS); 1191 ncpbytes = nfdbits / NBBY; 1192 ncpubytes = roundup(nd, abi_nfdbits) / NBBY; 1193 nbufbytes = 0; 1194 if (fd_in != NULL) 1195 nbufbytes += 2 * ncpbytes; 1196 if (fd_ou != NULL) 1197 nbufbytes += 2 * ncpbytes; 1198 if (fd_ex != NULL) 1199 nbufbytes += 2 * ncpbytes; 1200 if (nbufbytes <= sizeof s_selbits) 1201 selbits = &s_selbits[0]; 1202 else 1203 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK); 1204 1205 /* 1206 * Assign pointers into the bit buffers and fetch the input bits. 1207 * Put the output buffers together so that they can be bzeroed 1208 * together. 1209 */ 1210 sbp = selbits; 1211 #define getbits(name, x) \ 1212 do { \ 1213 if (name == NULL) { \ 1214 ibits[x] = NULL; \ 1215 obits[x] = NULL; \ 1216 } else { \ 1217 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ 1218 obits[x] = sbp; \ 1219 sbp += ncpbytes / sizeof *sbp; \ 1220 error = copyin(name, ibits[x], ncpubytes); \ 1221 if (error != 0) \ 1222 goto done; \ 1223 if (ncpbytes != ncpubytes) \ 1224 bzero((char *)ibits[x] + ncpubytes, \ 1225 ncpbytes - ncpubytes); \ 1226 } \ 1227 } while (0) 1228 getbits(fd_in, 0); 1229 getbits(fd_ou, 1); 1230 getbits(fd_ex, 2); 1231 #undef getbits 1232 1233 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__) 1234 /* 1235 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS, 1236 * we are running under 32-bit emulation. This should be more 1237 * generic. 1238 */ 1239 #define swizzle_fdset(bits) \ 1240 if (abi_nfdbits != NFDBITS && bits != NULL) { \ 1241 int i; \ 1242 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \ 1243 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \ 1244 } 1245 #else 1246 #define swizzle_fdset(bits) 1247 #endif 1248 1249 /* Make sure the bit order makes it through an ABI transition */ 1250 swizzle_fdset(ibits[0]); 1251 swizzle_fdset(ibits[1]); 1252 swizzle_fdset(ibits[2]); 1253 1254 if (nbufbytes != 0) 1255 bzero(selbits, nbufbytes / 2); 1256 1257 precision = 0; 1258 if (tvp != NULL) { 1259 rtv = *tvp; 1260 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1261 rtv.tv_usec >= 1000000) { 1262 error = EINVAL; 1263 goto done; 1264 } 1265 if (!timevalisset(&rtv)) 1266 asbt = 0; 1267 else if (rtv.tv_sec <= INT32_MAX) { 1268 rsbt = tvtosbt(rtv); 1269 precision = rsbt; 1270 precision >>= tc_precexp; 1271 if (TIMESEL(&asbt, rsbt)) 1272 asbt += tc_tick_sbt; 1273 if (asbt <= SBT_MAX - rsbt) 1274 asbt += rsbt; 1275 else 1276 asbt = -1; 1277 } else 1278 asbt = -1; 1279 } else 1280 asbt = -1; 1281 seltdinit(td); 1282 /* Iterate until the timeout expires or descriptors become ready. */ 1283 for (;;) { 1284 error = selscan(td, ibits, obits, nd); 1285 if (error || td->td_retval[0] != 0) 1286 break; 1287 error = seltdwait(td, asbt, precision); 1288 if (error) 1289 break; 1290 error = selrescan(td, ibits, obits); 1291 if (error || td->td_retval[0] != 0) 1292 break; 1293 } 1294 seltdclear(td); 1295 1296 done: 1297 /* select is not restarted after signals... */ 1298 if (error == ERESTART) 1299 error = EINTR; 1300 if (error == EWOULDBLOCK) 1301 error = 0; 1302 1303 /* swizzle bit order back, if necessary */ 1304 swizzle_fdset(obits[0]); 1305 swizzle_fdset(obits[1]); 1306 swizzle_fdset(obits[2]); 1307 #undef swizzle_fdset 1308 1309 #define putbits(name, x) \ 1310 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \ 1311 error = error2; 1312 if (error == 0) { 1313 int error2; 1314 1315 putbits(fd_in, 0); 1316 putbits(fd_ou, 1); 1317 putbits(fd_ex, 2); 1318 #undef putbits 1319 } 1320 if (selbits != &s_selbits[0]) 1321 free(selbits, M_SELECT); 1322 1323 return (error); 1324 } 1325 /* 1326 * Convert a select bit set to poll flags. 1327 * 1328 * The backend always returns POLLHUP/POLLERR if appropriate and we 1329 * return this as a set bit in any set. 1330 */ 1331 static const int select_flags[3] = { 1332 POLLRDNORM | POLLHUP | POLLERR, 1333 POLLWRNORM | POLLHUP | POLLERR, 1334 POLLRDBAND | POLLERR 1335 }; 1336 1337 /* 1338 * Compute the fo_poll flags required for a fd given by the index and 1339 * bit position in the fd_mask array. 1340 */ 1341 static __inline int 1342 selflags(fd_mask **ibits, int idx, fd_mask bit) 1343 { 1344 int flags; 1345 int msk; 1346 1347 flags = 0; 1348 for (msk = 0; msk < 3; msk++) { 1349 if (ibits[msk] == NULL) 1350 continue; 1351 if ((ibits[msk][idx] & bit) == 0) 1352 continue; 1353 flags |= select_flags[msk]; 1354 } 1355 return (flags); 1356 } 1357 1358 /* 1359 * Set the appropriate output bits given a mask of fired events and the 1360 * input bits originally requested. 1361 */ 1362 static __inline int 1363 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events) 1364 { 1365 int msk; 1366 int n; 1367 1368 n = 0; 1369 for (msk = 0; msk < 3; msk++) { 1370 if ((events & select_flags[msk]) == 0) 1371 continue; 1372 if (ibits[msk] == NULL) 1373 continue; 1374 if ((ibits[msk][idx] & bit) == 0) 1375 continue; 1376 /* 1377 * XXX Check for a duplicate set. This can occur because a 1378 * socket calls selrecord() twice for each poll() call 1379 * resulting in two selfds per real fd. selrescan() will 1380 * call selsetbits twice as a result. 1381 */ 1382 if ((obits[msk][idx] & bit) != 0) 1383 continue; 1384 obits[msk][idx] |= bit; 1385 n++; 1386 } 1387 1388 return (n); 1389 } 1390 1391 /* 1392 * Traverse the list of fds attached to this thread's seltd and check for 1393 * completion. 1394 */ 1395 static int 1396 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits) 1397 { 1398 struct filedesc *fdp; 1399 struct selinfo *si; 1400 struct seltd *stp; 1401 struct selfd *sfp; 1402 struct selfd *sfn; 1403 struct file *fp; 1404 fd_mask bit; 1405 int fd, ev, n, idx; 1406 int error; 1407 bool only_user; 1408 1409 fdp = td->td_proc->p_fd; 1410 stp = td->td_sel; 1411 n = 0; 1412 only_user = FILEDESC_IS_ONLY_USER(fdp); 1413 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1414 fd = (int)(uintptr_t)sfp->sf_cookie; 1415 si = sfp->sf_si; 1416 selfdfree(stp, sfp); 1417 /* If the selinfo wasn't cleared the event didn't fire. */ 1418 if (si != NULL) 1419 continue; 1420 if (only_user) 1421 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1422 else 1423 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1424 if (__predict_false(error != 0)) 1425 return (error); 1426 idx = fd / NFDBITS; 1427 bit = (fd_mask)1 << (fd % NFDBITS); 1428 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td); 1429 if (only_user) 1430 fput_only_user(fdp, fp); 1431 else 1432 fdrop(fp, td); 1433 if (ev != 0) 1434 n += selsetbits(ibits, obits, idx, bit, ev); 1435 } 1436 stp->st_flags = 0; 1437 td->td_retval[0] = n; 1438 return (0); 1439 } 1440 1441 /* 1442 * Perform the initial filedescriptor scan and register ourselves with 1443 * each selinfo. 1444 */ 1445 static int 1446 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd) 1447 { 1448 struct filedesc *fdp; 1449 struct file *fp; 1450 fd_mask bit; 1451 int ev, flags, end, fd; 1452 int n, idx; 1453 int error; 1454 bool only_user; 1455 1456 fdp = td->td_proc->p_fd; 1457 n = 0; 1458 only_user = FILEDESC_IS_ONLY_USER(fdp); 1459 for (idx = 0, fd = 0; fd < nfd; idx++) { 1460 end = imin(fd + NFDBITS, nfd); 1461 for (bit = 1; fd < end; bit <<= 1, fd++) { 1462 /* Compute the list of events we're interested in. */ 1463 flags = selflags(ibits, idx, bit); 1464 if (flags == 0) 1465 continue; 1466 if (only_user) 1467 error = fget_only_user(fdp, fd, &cap_event_rights, &fp); 1468 else 1469 error = fget_unlocked(td, fd, &cap_event_rights, &fp); 1470 if (__predict_false(error != 0)) 1471 return (error); 1472 selfdalloc(td, (void *)(uintptr_t)fd); 1473 ev = fo_poll(fp, flags, td->td_ucred, td); 1474 if (only_user) 1475 fput_only_user(fdp, fp); 1476 else 1477 fdrop(fp, td); 1478 if (ev != 0) 1479 n += selsetbits(ibits, obits, idx, bit, ev); 1480 } 1481 } 1482 1483 td->td_retval[0] = n; 1484 return (0); 1485 } 1486 1487 int 1488 sys_poll(struct thread *td, struct poll_args *uap) 1489 { 1490 struct timespec ts, *tsp; 1491 1492 if (uap->timeout != INFTIM) { 1493 if (uap->timeout < 0) 1494 return (EINVAL); 1495 ts.tv_sec = uap->timeout / 1000; 1496 ts.tv_nsec = (uap->timeout % 1000) * 1000000; 1497 tsp = &ts; 1498 } else 1499 tsp = NULL; 1500 1501 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL)); 1502 } 1503 1504 /* 1505 * kfds points to an array in the kernel. 1506 */ 1507 int 1508 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds, 1509 struct timespec *tsp, sigset_t *uset) 1510 { 1511 sbintime_t sbt, precision, tmp; 1512 time_t over; 1513 struct timespec ts; 1514 int error; 1515 1516 precision = 0; 1517 if (tsp != NULL) { 1518 if (!timespecvalid_interval(tsp)) 1519 return (EINVAL); 1520 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1521 sbt = 0; 1522 else { 1523 ts = *tsp; 1524 if (ts.tv_sec > INT32_MAX / 2) { 1525 over = ts.tv_sec - INT32_MAX / 2; 1526 ts.tv_sec -= over; 1527 } else 1528 over = 0; 1529 tmp = tstosbt(ts); 1530 precision = tmp; 1531 precision >>= tc_precexp; 1532 if (TIMESEL(&sbt, tmp)) 1533 sbt += tc_tick_sbt; 1534 sbt += tmp; 1535 } 1536 } else 1537 sbt = -1; 1538 1539 if (uset != NULL) { 1540 error = kern_sigprocmask(td, SIG_SETMASK, uset, 1541 &td->td_oldsigmask, 0); 1542 if (error) 1543 return (error); 1544 td->td_pflags |= TDP_OLDMASK; 1545 } 1546 1547 seltdinit(td); 1548 /* Iterate until the timeout expires or descriptors become ready. */ 1549 for (;;) { 1550 error = pollscan(td, kfds, nfds); 1551 if (error || td->td_retval[0] != 0) 1552 break; 1553 error = seltdwait(td, sbt, precision); 1554 if (error) 1555 break; 1556 error = pollrescan(td); 1557 if (error || td->td_retval[0] != 0) 1558 break; 1559 } 1560 seltdclear(td); 1561 1562 /* poll is not restarted after signals... */ 1563 if (error == ERESTART) 1564 error = EINTR; 1565 if (error == EWOULDBLOCK) 1566 error = 0; 1567 1568 if (uset != NULL) { 1569 /* 1570 * Make sure that ast() is called on return to 1571 * usermode and TDP_OLDMASK is cleared, restoring old 1572 * sigmask. If we didn't get interrupted, then the caller is 1573 * likely not expecting a signal to hit that should normally be 1574 * blocked by its signal mask, so we restore the mask before 1575 * any signals could be delivered. 1576 */ 1577 if (error == EINTR) 1578 ast_sched(td, TDA_SIGSUSPEND); 1579 else 1580 ast_sched(td, TDA_PSELECT); 1581 } 1582 1583 return (error); 1584 } 1585 1586 int 1587 sys_ppoll(struct thread *td, struct ppoll_args *uap) 1588 { 1589 struct timespec ts, *tsp; 1590 sigset_t set, *ssp; 1591 int error; 1592 1593 if (uap->ts != NULL) { 1594 error = copyin(uap->ts, &ts, sizeof(ts)); 1595 if (error) 1596 return (error); 1597 tsp = &ts; 1598 } else 1599 tsp = NULL; 1600 if (uap->set != NULL) { 1601 error = copyin(uap->set, &set, sizeof(set)); 1602 if (error) 1603 return (error); 1604 ssp = &set; 1605 } else 1606 ssp = NULL; 1607 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp)); 1608 } 1609 1610 /* 1611 * ufds points to an array in user space. 1612 */ 1613 int 1614 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds, 1615 struct timespec *tsp, sigset_t *set) 1616 { 1617 struct pollfd *kfds; 1618 struct pollfd stackfds[32]; 1619 int error; 1620 1621 if (kern_poll_maxfds(nfds)) 1622 return (EINVAL); 1623 if (nfds > nitems(stackfds)) 1624 kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK); 1625 else 1626 kfds = stackfds; 1627 error = copyin(ufds, kfds, nfds * sizeof(*kfds)); 1628 if (error != 0) 1629 goto out; 1630 1631 error = kern_poll_kfds(td, kfds, nfds, tsp, set); 1632 if (error == 0) 1633 error = pollout(td, kfds, ufds, nfds); 1634 #ifdef KTRACE 1635 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1636 ktrstructarray("pollfd", UIO_USERSPACE, ufds, nfds, 1637 sizeof(*ufds)); 1638 #endif 1639 1640 out: 1641 if (nfds > nitems(stackfds)) 1642 free(kfds, M_TEMP); 1643 return (error); 1644 } 1645 1646 bool 1647 kern_poll_maxfds(u_int nfds) 1648 { 1649 1650 /* 1651 * This is kinda bogus. We have fd limits, but that is not 1652 * really related to the size of the pollfd array. Make sure 1653 * we let the process use at least FD_SETSIZE entries and at 1654 * least enough for the system-wide limits. We want to be reasonably 1655 * safe, but not overly restrictive. 1656 */ 1657 return (nfds > maxfilesperproc && nfds > FD_SETSIZE); 1658 } 1659 1660 static int 1661 pollrescan(struct thread *td) 1662 { 1663 struct seltd *stp; 1664 struct selfd *sfp; 1665 struct selfd *sfn; 1666 struct selinfo *si; 1667 struct filedesc *fdp; 1668 struct file *fp; 1669 struct pollfd *fd; 1670 int n, error; 1671 bool only_user; 1672 1673 n = 0; 1674 fdp = td->td_proc->p_fd; 1675 stp = td->td_sel; 1676 only_user = FILEDESC_IS_ONLY_USER(fdp); 1677 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { 1678 fd = (struct pollfd *)sfp->sf_cookie; 1679 si = sfp->sf_si; 1680 selfdfree(stp, sfp); 1681 /* If the selinfo wasn't cleared the event didn't fire. */ 1682 if (si != NULL) 1683 continue; 1684 if (only_user) 1685 error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp); 1686 else 1687 error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp); 1688 if (__predict_false(error != 0)) { 1689 fd->revents = POLLNVAL; 1690 n++; 1691 continue; 1692 } 1693 /* 1694 * Note: backend also returns POLLHUP and 1695 * POLLERR if appropriate. 1696 */ 1697 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td); 1698 if (only_user) 1699 fput_only_user(fdp, fp); 1700 else 1701 fdrop(fp, td); 1702 if (fd->revents != 0) 1703 n++; 1704 } 1705 stp->st_flags = 0; 1706 td->td_retval[0] = n; 1707 return (0); 1708 } 1709 1710 static int 1711 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd) 1712 { 1713 int error = 0; 1714 u_int i = 0; 1715 u_int n = 0; 1716 1717 for (i = 0; i < nfd; i++) { 1718 error = copyout(&fds->revents, &ufds->revents, 1719 sizeof(ufds->revents)); 1720 if (error) 1721 return (error); 1722 if (fds->revents != 0) 1723 n++; 1724 fds++; 1725 ufds++; 1726 } 1727 td->td_retval[0] = n; 1728 return (0); 1729 } 1730 1731 static int 1732 pollscan(struct thread *td, struct pollfd *fds, u_int nfd) 1733 { 1734 struct filedesc *fdp; 1735 struct file *fp; 1736 int i, n, error; 1737 bool only_user; 1738 1739 n = 0; 1740 fdp = td->td_proc->p_fd; 1741 only_user = FILEDESC_IS_ONLY_USER(fdp); 1742 for (i = 0; i < nfd; i++, fds++) { 1743 if (fds->fd < 0) { 1744 fds->revents = 0; 1745 continue; 1746 } 1747 if (only_user) 1748 error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp); 1749 else 1750 error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp); 1751 if (__predict_false(error != 0)) { 1752 fds->revents = POLLNVAL; 1753 n++; 1754 continue; 1755 } 1756 /* 1757 * Note: backend also returns POLLHUP and 1758 * POLLERR if appropriate. 1759 */ 1760 selfdalloc(td, fds); 1761 fds->revents = fo_poll(fp, fds->events, 1762 td->td_ucred, td); 1763 if (only_user) 1764 fput_only_user(fdp, fp); 1765 else 1766 fdrop(fp, td); 1767 /* 1768 * POSIX requires POLLOUT to be never 1769 * set simultaneously with POLLHUP. 1770 */ 1771 if ((fds->revents & POLLHUP) != 0) 1772 fds->revents &= ~POLLOUT; 1773 1774 if (fds->revents != 0) 1775 n++; 1776 } 1777 td->td_retval[0] = n; 1778 return (0); 1779 } 1780 1781 /* 1782 * XXX This was created specifically to support netncp and netsmb. This 1783 * allows the caller to specify a socket to wait for events on. It returns 1784 * 0 if any events matched and an error otherwise. There is no way to 1785 * determine which events fired. 1786 */ 1787 int 1788 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td) 1789 { 1790 struct timeval rtv; 1791 sbintime_t asbt, precision, rsbt; 1792 int error; 1793 1794 precision = 0; /* stupid gcc! */ 1795 if (tvp != NULL) { 1796 rtv = *tvp; 1797 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || 1798 rtv.tv_usec >= 1000000) 1799 return (EINVAL); 1800 if (!timevalisset(&rtv)) 1801 asbt = 0; 1802 else if (rtv.tv_sec <= INT32_MAX) { 1803 rsbt = tvtosbt(rtv); 1804 precision = rsbt; 1805 precision >>= tc_precexp; 1806 if (TIMESEL(&asbt, rsbt)) 1807 asbt += tc_tick_sbt; 1808 if (asbt <= SBT_MAX - rsbt) 1809 asbt += rsbt; 1810 else 1811 asbt = -1; 1812 } else 1813 asbt = -1; 1814 } else 1815 asbt = -1; 1816 seltdinit(td); 1817 /* 1818 * Iterate until the timeout expires or the socket becomes ready. 1819 */ 1820 for (;;) { 1821 selfdalloc(td, NULL); 1822 if (so->so_proto->pr_sopoll(so, events, td) != 0) { 1823 error = 0; 1824 break; 1825 } 1826 error = seltdwait(td, asbt, precision); 1827 if (error) 1828 break; 1829 } 1830 seltdclear(td); 1831 /* XXX Duplicates ncp/smb behavior. */ 1832 if (error == ERESTART) 1833 error = 0; 1834 return (error); 1835 } 1836 1837 /* 1838 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines 1839 * have two select sets, one for read and another for write. 1840 */ 1841 static void 1842 selfdalloc(struct thread *td, void *cookie) 1843 { 1844 struct seltd *stp; 1845 1846 stp = td->td_sel; 1847 if (stp->st_free1 == NULL) 1848 stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO); 1849 stp->st_free1->sf_td = stp; 1850 stp->st_free1->sf_cookie = cookie; 1851 if (stp->st_free2 == NULL) 1852 stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO); 1853 stp->st_free2->sf_td = stp; 1854 stp->st_free2->sf_cookie = cookie; 1855 } 1856 1857 static void 1858 selfdfree(struct seltd *stp, struct selfd *sfp) 1859 { 1860 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link); 1861 /* 1862 * Paired with doselwakeup. 1863 */ 1864 if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) { 1865 mtx_lock(sfp->sf_mtx); 1866 if (sfp->sf_si != NULL) { 1867 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads); 1868 } 1869 mtx_unlock(sfp->sf_mtx); 1870 } 1871 free(sfp, M_SELFD); 1872 } 1873 1874 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */ 1875 void 1876 seldrain(struct selinfo *sip) 1877 { 1878 1879 /* 1880 * This feature is already provided by doselwakeup(), thus it is 1881 * enough to go for it. 1882 * Eventually, the context, should take care to avoid races 1883 * between thread calling select()/poll() and file descriptor 1884 * detaching, but, again, the races are just the same as 1885 * selwakeup(). 1886 */ 1887 doselwakeup(sip, -1); 1888 } 1889 1890 /* 1891 * Record a select request. 1892 */ 1893 void 1894 selrecord(struct thread *selector, struct selinfo *sip) 1895 { 1896 struct selfd *sfp; 1897 struct seltd *stp; 1898 struct mtx *mtxp; 1899 1900 stp = selector->td_sel; 1901 /* 1902 * Don't record when doing a rescan. 1903 */ 1904 if (stp->st_flags & SELTD_RESCAN) 1905 return; 1906 /* 1907 * Grab one of the preallocated descriptors. 1908 */ 1909 sfp = NULL; 1910 if ((sfp = stp->st_free1) != NULL) 1911 stp->st_free1 = NULL; 1912 else if ((sfp = stp->st_free2) != NULL) 1913 stp->st_free2 = NULL; 1914 else 1915 panic("selrecord: No free selfd on selq"); 1916 mtxp = sip->si_mtx; 1917 if (mtxp == NULL) 1918 mtxp = mtx_pool_find(mtxpool_select, sip); 1919 /* 1920 * Initialize the sfp and queue it in the thread. 1921 */ 1922 sfp->sf_si = sip; 1923 sfp->sf_mtx = mtxp; 1924 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link); 1925 /* 1926 * Now that we've locked the sip, check for initialization. 1927 */ 1928 mtx_lock(mtxp); 1929 if (sip->si_mtx == NULL) { 1930 sip->si_mtx = mtxp; 1931 TAILQ_INIT(&sip->si_tdlist); 1932 } 1933 /* 1934 * Add this thread to the list of selfds listening on this selinfo. 1935 */ 1936 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads); 1937 mtx_unlock(sip->si_mtx); 1938 } 1939 1940 /* Wake up a selecting thread. */ 1941 void 1942 selwakeup(struct selinfo *sip) 1943 { 1944 doselwakeup(sip, -1); 1945 } 1946 1947 /* Wake up a selecting thread, and set its priority. */ 1948 void 1949 selwakeuppri(struct selinfo *sip, int pri) 1950 { 1951 doselwakeup(sip, pri); 1952 } 1953 1954 /* 1955 * Do a wakeup when a selectable event occurs. 1956 */ 1957 static void 1958 doselwakeup(struct selinfo *sip, int pri) 1959 { 1960 struct selfd *sfp; 1961 struct selfd *sfn; 1962 struct seltd *stp; 1963 1964 /* If it's not initialized there can't be any waiters. */ 1965 if (sip->si_mtx == NULL) 1966 return; 1967 /* 1968 * Locking the selinfo locks all selfds associated with it. 1969 */ 1970 mtx_lock(sip->si_mtx); 1971 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) { 1972 /* 1973 * Once we remove this sfp from the list and clear the 1974 * sf_si seltdclear will know to ignore this si. 1975 */ 1976 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads); 1977 stp = sfp->sf_td; 1978 mtx_lock(&stp->st_mtx); 1979 stp->st_flags |= SELTD_PENDING; 1980 cv_broadcastpri(&stp->st_wait, pri); 1981 mtx_unlock(&stp->st_mtx); 1982 /* 1983 * Paired with selfdfree. 1984 * 1985 * Storing this only after the wakeup provides an invariant that 1986 * stp is not used after selfdfree returns. 1987 */ 1988 atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL); 1989 } 1990 mtx_unlock(sip->si_mtx); 1991 } 1992 1993 static void 1994 seltdinit(struct thread *td) 1995 { 1996 struct seltd *stp; 1997 1998 stp = td->td_sel; 1999 if (stp != NULL) { 2000 MPASS(stp->st_flags == 0); 2001 MPASS(STAILQ_EMPTY(&stp->st_selq)); 2002 return; 2003 } 2004 stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO); 2005 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF); 2006 cv_init(&stp->st_wait, "select"); 2007 stp->st_flags = 0; 2008 STAILQ_INIT(&stp->st_selq); 2009 td->td_sel = stp; 2010 } 2011 2012 static int 2013 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision) 2014 { 2015 struct seltd *stp; 2016 int error; 2017 2018 stp = td->td_sel; 2019 /* 2020 * An event of interest may occur while we do not hold the seltd 2021 * locked so check the pending flag before we sleep. 2022 */ 2023 mtx_lock(&stp->st_mtx); 2024 /* 2025 * Any further calls to selrecord will be a rescan. 2026 */ 2027 stp->st_flags |= SELTD_RESCAN; 2028 if (stp->st_flags & SELTD_PENDING) { 2029 mtx_unlock(&stp->st_mtx); 2030 return (0); 2031 } 2032 if (sbt == 0) 2033 error = EWOULDBLOCK; 2034 else if (sbt != -1) 2035 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx, 2036 sbt, precision, C_ABSOLUTE); 2037 else 2038 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx); 2039 mtx_unlock(&stp->st_mtx); 2040 2041 return (error); 2042 } 2043 2044 void 2045 seltdfini(struct thread *td) 2046 { 2047 struct seltd *stp; 2048 2049 stp = td->td_sel; 2050 if (stp == NULL) 2051 return; 2052 MPASS(stp->st_flags == 0); 2053 MPASS(STAILQ_EMPTY(&stp->st_selq)); 2054 if (stp->st_free1) 2055 free(stp->st_free1, M_SELFD); 2056 if (stp->st_free2) 2057 free(stp->st_free2, M_SELFD); 2058 td->td_sel = NULL; 2059 cv_destroy(&stp->st_wait); 2060 mtx_destroy(&stp->st_mtx); 2061 free(stp, M_SELECT); 2062 } 2063 2064 /* 2065 * Remove the references to the thread from all of the objects we were 2066 * polling. 2067 */ 2068 static void 2069 seltdclear(struct thread *td) 2070 { 2071 struct seltd *stp; 2072 struct selfd *sfp; 2073 struct selfd *sfn; 2074 2075 stp = td->td_sel; 2076 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) 2077 selfdfree(stp, sfp); 2078 stp->st_flags = 0; 2079 } 2080 2081 static void selectinit(void *); 2082 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL); 2083 static void 2084 selectinit(void *dummy __unused) 2085 { 2086 2087 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF); 2088 } 2089 2090 /* 2091 * Set up a syscall return value that follows the convention specified for 2092 * posix_* functions. 2093 */ 2094 int 2095 kern_posix_error(struct thread *td, int error) 2096 { 2097 2098 if (error <= 0) 2099 return (error); 2100 td->td_errno = error; 2101 td->td_pflags |= TDP_NERRNO; 2102 td->td_retval[0] = error; 2103 return (0); 2104 } 2105 2106 int 2107 kcmp_cmp(uintptr_t a, uintptr_t b) 2108 { 2109 if (a == b) 2110 return (0); 2111 else if (a < b) 2112 return (1); 2113 return (2); 2114 } 2115 2116 static int 2117 kcmp_pget(struct thread *td, pid_t pid, struct proc **pp) 2118 { 2119 int error; 2120 2121 if (pid == td->td_proc->p_pid) { 2122 *pp = td->td_proc; 2123 return (0); 2124 } 2125 error = pget(pid, PGET_NOTID | PGET_CANDEBUG | PGET_NOTWEXIT | 2126 PGET_HOLD, pp); 2127 MPASS(*pp != td->td_proc); 2128 return (error); 2129 } 2130 2131 int 2132 kern_kcmp(struct thread *td, pid_t pid1, pid_t pid2, int type, 2133 uintptr_t idx1, uintptr_t idx2) 2134 { 2135 struct proc *p1, *p2; 2136 struct file *fp1, *fp2; 2137 int error, res; 2138 2139 res = -1; 2140 p1 = p2 = NULL; 2141 error = kcmp_pget(td, pid1, &p1); 2142 if (error == 0) 2143 error = kcmp_pget(td, pid2, &p2); 2144 if (error != 0) 2145 goto out; 2146 2147 switch (type) { 2148 case KCMP_FILE: 2149 case KCMP_FILEOBJ: 2150 error = fget_remote(td, p1, idx1, &fp1); 2151 if (error == 0) { 2152 error = fget_remote(td, p2, idx2, &fp2); 2153 if (error == 0) { 2154 if (type == KCMP_FILEOBJ) 2155 res = fo_cmp(fp1, fp2, td); 2156 else 2157 res = kcmp_cmp((uintptr_t)fp1, 2158 (uintptr_t)fp2); 2159 fdrop(fp2, td); 2160 } 2161 fdrop(fp1, td); 2162 } 2163 break; 2164 case KCMP_FILES: 2165 res = kcmp_cmp((uintptr_t)p1->p_fd, (uintptr_t)p2->p_fd); 2166 break; 2167 case KCMP_SIGHAND: 2168 res = kcmp_cmp((uintptr_t)p1->p_sigacts, 2169 (uintptr_t)p2->p_sigacts); 2170 break; 2171 case KCMP_VM: 2172 res = kcmp_cmp((uintptr_t)p1->p_vmspace, 2173 (uintptr_t)p2->p_vmspace); 2174 break; 2175 default: 2176 error = EINVAL; 2177 break; 2178 } 2179 2180 out: 2181 if (p1 != NULL && p1 != td->td_proc) 2182 PRELE(p1); 2183 if (p2 != NULL && p2 != td->td_proc) 2184 PRELE(p2); 2185 2186 td->td_retval[0] = res; 2187 return (error); 2188 } 2189 2190 int 2191 sys_kcmp(struct thread *td, struct kcmp_args *uap) 2192 { 2193 return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type, 2194 uap->idx1, uap->idx2)); 2195 } 2196 2197 int 2198 file_kcmp_generic(struct file *fp1, struct file *fp2, struct thread *td) 2199 { 2200 if (fp1->f_type != fp2->f_type) 2201 return (3); 2202 return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data)); 2203 } 2204 2205 int 2206 exterr_to_ue(struct thread *td, struct uexterror *ue) 2207 { 2208 if ((td->td_pflags2 & TDP2_EXTERR) == 0) 2209 return (ENOENT); 2210 2211 memset(ue, 0, sizeof(*ue)); 2212 ue->error = td->td_kexterr.error; 2213 ue->cat = td->td_kexterr.cat; 2214 ue->src_line = td->td_kexterr.src_line; 2215 ue->p1 = td->td_kexterr.p1; 2216 ue->p2 = td->td_kexterr.p2; 2217 if (td->td_kexterr.msg != NULL) 2218 strlcpy(ue->msg, td->td_kexterr.msg, sizeof(ue->msg)); 2219 return (0); 2220 } 2221 2222 void 2223 exterr_copyout(struct thread *td) 2224 { 2225 struct uexterror ue; 2226 ksiginfo_t ksi; 2227 void *uloc; 2228 size_t sz; 2229 int error; 2230 2231 MPASS((td->td_pflags2 & TDP2_UEXTERR) != 0); 2232 2233 uloc = (char *)td->td_exterr_ptr + __offsetof(struct uexterror, 2234 error); 2235 error = exterr_to_ue(td, &ue); 2236 if (error != 0) { 2237 ue.error = 0; 2238 sz = sizeof(ue.error); 2239 } else { 2240 sz = sizeof(ue) - __offsetof(struct uexterror, error); 2241 } 2242 error = copyout(&ue.error, uloc, sz); 2243 if (error != 0) { 2244 td->td_pflags2 &= ~TDP2_UEXTERR; 2245 ksiginfo_init_trap(&ksi); 2246 ksi.ksi_signo = SIGSEGV; 2247 ksi.ksi_code = SEGV_ACCERR; 2248 ksi.ksi_addr = uloc; 2249 trapsignal(td, &ksi); 2250 } 2251 } 2252 2253 int 2254 sys_exterrctl(struct thread *td, struct exterrctl_args *uap) 2255 { 2256 uint32_t ver; 2257 int error; 2258 2259 if ((uap->flags & ~(EXTERRCTLF_FORCE)) != 0) 2260 return (EINVAL); 2261 switch (uap->op) { 2262 case EXTERRCTL_ENABLE: 2263 if ((td->td_pflags2 & TDP2_UEXTERR) != 0 && 2264 (uap->flags & EXTERRCTLF_FORCE) == 0) 2265 return (EBUSY); 2266 td->td_pflags2 &= ~TDP2_UEXTERR; 2267 error = copyin(uap->ptr, &ver, sizeof(ver)); 2268 if (error != 0) 2269 return (error); 2270 if (ver != UEXTERROR_VER) 2271 return (EINVAL); 2272 td->td_pflags2 |= TDP2_UEXTERR; 2273 td->td_exterr_ptr = uap->ptr; 2274 return (0); 2275 case EXTERRCTL_DISABLE: 2276 if ((td->td_pflags2 & TDP2_UEXTERR) == 0) 2277 return (EINVAL); 2278 td->td_pflags2 &= ~TDP2_UEXTERR; 2279 return (0); 2280 default: 2281 return (EINVAL); 2282 } 2283 } 2284 2285 int 2286 exterr_set(int eerror, int category, const char *mmsg, uintptr_t pp1, 2287 uintptr_t pp2, int line) 2288 { 2289 struct thread *td; 2290 2291 td = curthread; 2292 if ((td->td_pflags2 & TDP2_UEXTERR) != 0) { 2293 td->td_pflags2 |= TDP2_EXTERR; 2294 td->td_kexterr.error = eerror; 2295 td->td_kexterr.cat = category; 2296 td->td_kexterr.msg = mmsg; 2297 td->td_kexterr.p1 = pp1; 2298 td->td_kexterr.p2 = pp2; 2299 td->td_kexterr.src_line = line; 2300 ktrexterr(td); 2301 } 2302 return (eerror); 2303 } 2304