kern_descrip.c revision 124602
1/* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 39 */ 40 41#include <sys/cdefs.h> 42__FBSDID("$FreeBSD: head/sys/kern/kern_descrip.c 124602 2004-01-17 00:59:04Z des $"); 43 44#include "opt_compat.h" 45 46#include <sys/param.h> 47#include <sys/limits.h> 48#include <sys/systm.h> 49#include <sys/syscallsubr.h> 50#include <sys/sysproto.h> 51#include <sys/conf.h> 52#include <sys/filedesc.h> 53#include <sys/lock.h> 54#include <sys/kernel.h> 55#include <sys/limits.h> 56#include <sys/malloc.h> 57#include <sys/mutex.h> 58#include <sys/sysctl.h> 59#include <sys/vnode.h> 60#include <sys/mount.h> 61#include <sys/proc.h> 62#include <sys/namei.h> 63#include <sys/file.h> 64#include <sys/stat.h> 65#include <sys/filio.h> 66#include <sys/fcntl.h> 67#include <sys/unistd.h> 68#include <sys/resourcevar.h> 69#include <sys/event.h> 70#include <sys/sx.h> 71#include <sys/socketvar.h> 72#include <sys/signalvar.h> 73 74#include <vm/vm.h> 75#include <vm/vm_extern.h> 76#include <vm/uma.h> 77 78static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 79static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 80 "file desc to leader structures"); 81static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 82 83static uma_zone_t file_zone; 84 85static d_open_t fdopen; 86#define NUMFDESC 64 87 88#define CDEV_MAJOR 22 89static struct cdevsw fildesc_cdevsw = { 90 .d_open = fdopen, 91 .d_name = "FD", 92 .d_maj = CDEV_MAJOR, 93}; 94 95/* How to treat 'new' parameter when allocating a fd for do_dup(). */ 96enum dup_type { DUP_VARIABLE, DUP_FIXED }; 97 98static int do_dup(struct thread *td, enum dup_type type, int old, int new, 99 register_t *retval); 100static int fd_first_free(struct filedesc *, int, int); 101static int fd_last_used(struct filedesc *, int, int); 102static void fdgrowtable(struct filedesc *, int); 103 104/* 105 * Descriptor management. 106 */ 107struct filelist filehead; /* head of list of open files */ 108int nfiles; /* actual number of open files */ 109struct sx filelist_lock; /* sx to protect filelist */ 110struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 111 112/* 113 * Find the first zero bit in the given bitmap, starting at low and not 114 * exceeding size - 1. 115 */ 116static int 117fd_first_free(struct filedesc *fdp, int low, int size) 118{ 119 NDSLOTTYPE *map = fdp->fd_map; 120 NDSLOTTYPE mask; 121 int off, maxoff; 122 123 if (low >= size) 124 return (low); 125 126 off = NDSLOT(low); 127 if (low % NDENTRIES) { 128 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 129 if ((mask &= ~map[off]) != 0) 130 return (off * NDENTRIES + ffsl(mask) - 1); 131 ++off; 132 } 133 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 134 if ((mask = ~map[off]) != 0) 135 return (off * NDENTRIES + ffsl(mask) - 1); 136 return (size); 137} 138 139/* 140 * Find the highest non-zero bit in the given bitmap, starting at low and 141 * not exceeding size - 1. 142 */ 143static int 144fd_last_used(struct filedesc *fdp, int low, int size) 145{ 146 NDSLOTTYPE *map = fdp->fd_map; 147 NDSLOTTYPE mask; 148 int off, minoff; 149 150 if (low >= size) 151 return (-1); 152 153 off = NDSLOT(size); 154 if (size % NDENTRIES) { 155 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 156 if ((mask &= map[off]) != 0) 157 return (off * NDENTRIES + flsl(mask) - 1); 158 --off; 159 } 160 for (minoff = NDSLOT(low); off >= minoff; --off) 161 if ((mask = map[off]) != 0) 162 return (off * NDENTRIES + flsl(mask) - 1); 163 return (size - 1); 164} 165 166static int 167fdisused(struct filedesc *fdp, int fd) 168{ 169 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 170 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 171 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 172} 173 174/* 175 * Mark a file descriptor as used. 176 */ 177void 178fdused(struct filedesc *fdp, int fd) 179{ 180 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 181 KASSERT(!fdisused(fdp, fd), 182 ("fd already used")); 183 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 184 if (fd > fdp->fd_lastfile) 185 fdp->fd_lastfile = fd; 186 if (fd == fdp->fd_freefile) 187 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 188} 189 190/* 191 * Mark a file descriptor as unused. 192 */ 193void 194fdunused(struct filedesc *fdp, int fd) 195{ 196 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 197 KASSERT(fdisused(fdp, fd), 198 ("fd is already unused")); 199 KASSERT(fdp->fd_ofiles[fd] == NULL, 200 ("fd is still in use")); 201 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 202 if (fd < fdp->fd_freefile) 203 fdp->fd_freefile = fd; 204 if (fd == fdp->fd_lastfile) 205 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 206} 207 208/* 209 * System calls on descriptors. 210 */ 211#ifndef _SYS_SYSPROTO_H_ 212struct getdtablesize_args { 213 int dummy; 214}; 215#endif 216/* 217 * MPSAFE 218 */ 219/* ARGSUSED */ 220int 221getdtablesize(td, uap) 222 struct thread *td; 223 struct getdtablesize_args *uap; 224{ 225 struct proc *p = td->td_proc; 226 227 mtx_lock(&Giant); 228 td->td_retval[0] = 229 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 230 mtx_unlock(&Giant); 231 return (0); 232} 233 234/* 235 * Duplicate a file descriptor to a particular value. 236 * 237 * note: keep in mind that a potential race condition exists when closing 238 * descriptors from a shared descriptor table (via rfork). 239 */ 240#ifndef _SYS_SYSPROTO_H_ 241struct dup2_args { 242 u_int from; 243 u_int to; 244}; 245#endif 246/* 247 * MPSAFE 248 */ 249/* ARGSUSED */ 250int 251dup2(td, uap) 252 struct thread *td; 253 struct dup2_args *uap; 254{ 255 256 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 257 td->td_retval)); 258} 259 260/* 261 * Duplicate a file descriptor. 262 */ 263#ifndef _SYS_SYSPROTO_H_ 264struct dup_args { 265 u_int fd; 266}; 267#endif 268/* 269 * MPSAFE 270 */ 271/* ARGSUSED */ 272int 273dup(td, uap) 274 struct thread *td; 275 struct dup_args *uap; 276{ 277 278 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval)); 279} 280 281/* 282 * The file control system call. 283 */ 284#ifndef _SYS_SYSPROTO_H_ 285struct fcntl_args { 286 int fd; 287 int cmd; 288 long arg; 289}; 290#endif 291/* 292 * MPSAFE 293 */ 294/* ARGSUSED */ 295int 296fcntl(td, uap) 297 struct thread *td; 298 struct fcntl_args *uap; 299{ 300 struct flock fl; 301 intptr_t arg; 302 int error; 303 304 error = 0; 305 switch (uap->cmd) { 306 case F_GETLK: 307 case F_SETLK: 308 case F_SETLKW: 309 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 310 arg = (intptr_t)&fl; 311 break; 312 default: 313 arg = uap->arg; 314 break; 315 } 316 if (error) 317 return (error); 318 error = kern_fcntl(td, uap->fd, uap->cmd, arg); 319 if (error) 320 return (error); 321 if (uap->cmd == F_GETLK) 322 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 323 return (error); 324} 325 326int 327kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 328{ 329 struct filedesc *fdp; 330 struct flock *flp; 331 struct file *fp; 332 struct proc *p; 333 char *pop; 334 struct vnode *vp; 335 u_int newmin; 336 int error, flg, tmp; 337 338 error = 0; 339 flg = F_POSIX; 340 p = td->td_proc; 341 fdp = p->p_fd; 342 mtx_lock(&Giant); 343 FILEDESC_LOCK(fdp); 344 if ((unsigned)fd >= fdp->fd_nfiles || 345 (fp = fdp->fd_ofiles[fd]) == NULL) { 346 FILEDESC_UNLOCK(fdp); 347 error = EBADF; 348 goto done2; 349 } 350 pop = &fdp->fd_ofileflags[fd]; 351 352 switch (cmd) { 353 case F_DUPFD: 354 FILEDESC_UNLOCK(fdp); 355 newmin = arg; 356 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || 357 newmin >= maxfilesperproc) { 358 error = EINVAL; 359 break; 360 } 361 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval); 362 break; 363 364 case F_GETFD: 365 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 366 FILEDESC_UNLOCK(fdp); 367 break; 368 369 case F_SETFD: 370 *pop = (*pop &~ UF_EXCLOSE) | 371 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 372 FILEDESC_UNLOCK(fdp); 373 break; 374 375 case F_GETFL: 376 FILE_LOCK(fp); 377 FILEDESC_UNLOCK(fdp); 378 td->td_retval[0] = OFLAGS(fp->f_flag); 379 FILE_UNLOCK(fp); 380 break; 381 382 case F_SETFL: 383 FILE_LOCK(fp); 384 FILEDESC_UNLOCK(fdp); 385 fhold_locked(fp); 386 fp->f_flag &= ~FCNTLFLAGS; 387 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 388 FILE_UNLOCK(fp); 389 tmp = fp->f_flag & FNONBLOCK; 390 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 391 if (error) { 392 fdrop(fp, td); 393 break; 394 } 395 tmp = fp->f_flag & FASYNC; 396 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 397 if (error == 0) { 398 fdrop(fp, td); 399 break; 400 } 401 FILE_LOCK(fp); 402 fp->f_flag &= ~FNONBLOCK; 403 FILE_UNLOCK(fp); 404 tmp = 0; 405 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 406 fdrop(fp, td); 407 break; 408 409 case F_GETOWN: 410 fhold(fp); 411 FILEDESC_UNLOCK(fdp); 412 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 413 if (error == 0) 414 td->td_retval[0] = tmp; 415 fdrop(fp, td); 416 break; 417 418 case F_SETOWN: 419 fhold(fp); 420 FILEDESC_UNLOCK(fdp); 421 tmp = arg; 422 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 423 fdrop(fp, td); 424 break; 425 426 case F_SETLKW: 427 flg |= F_WAIT; 428 /* FALLTHROUGH F_SETLK */ 429 430 case F_SETLK: 431 if (fp->f_type != DTYPE_VNODE) { 432 FILEDESC_UNLOCK(fdp); 433 error = EBADF; 434 break; 435 } 436 437 flp = (struct flock *)arg; 438 if (flp->l_whence == SEEK_CUR) { 439 if (fp->f_offset < 0 || 440 (flp->l_start > 0 && 441 fp->f_offset > OFF_MAX - flp->l_start)) { 442 FILEDESC_UNLOCK(fdp); 443 error = EOVERFLOW; 444 break; 445 } 446 flp->l_start += fp->f_offset; 447 } 448 449 /* 450 * VOP_ADVLOCK() may block. 451 */ 452 fhold(fp); 453 FILEDESC_UNLOCK(fdp); 454 vp = fp->f_vnode; 455 456 switch (flp->l_type) { 457 case F_RDLCK: 458 if ((fp->f_flag & FREAD) == 0) { 459 error = EBADF; 460 break; 461 } 462 PROC_LOCK(p->p_leader); 463 p->p_leader->p_flag |= P_ADVLOCK; 464 PROC_UNLOCK(p->p_leader); 465 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 466 flp, flg); 467 break; 468 case F_WRLCK: 469 if ((fp->f_flag & FWRITE) == 0) { 470 error = EBADF; 471 break; 472 } 473 PROC_LOCK(p->p_leader); 474 p->p_leader->p_flag |= P_ADVLOCK; 475 PROC_UNLOCK(p->p_leader); 476 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 477 flp, flg); 478 break; 479 case F_UNLCK: 480 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 481 flp, F_POSIX); 482 break; 483 default: 484 error = EINVAL; 485 break; 486 } 487 /* Check for race with close */ 488 FILEDESC_LOCK(fdp); 489 if ((unsigned) fd >= fdp->fd_nfiles || 490 fp != fdp->fd_ofiles[fd]) { 491 FILEDESC_UNLOCK(fdp); 492 flp->l_whence = SEEK_SET; 493 flp->l_start = 0; 494 flp->l_len = 0; 495 flp->l_type = F_UNLCK; 496 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 497 F_UNLCK, flp, F_POSIX); 498 } else 499 FILEDESC_UNLOCK(fdp); 500 fdrop(fp, td); 501 break; 502 503 case F_GETLK: 504 if (fp->f_type != DTYPE_VNODE) { 505 FILEDESC_UNLOCK(fdp); 506 error = EBADF; 507 break; 508 } 509 flp = (struct flock *)arg; 510 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 511 flp->l_type != F_UNLCK) { 512 FILEDESC_UNLOCK(fdp); 513 error = EINVAL; 514 break; 515 } 516 if (flp->l_whence == SEEK_CUR) { 517 if ((flp->l_start > 0 && 518 fp->f_offset > OFF_MAX - flp->l_start) || 519 (flp->l_start < 0 && 520 fp->f_offset < OFF_MIN - flp->l_start)) { 521 FILEDESC_UNLOCK(fdp); 522 error = EOVERFLOW; 523 break; 524 } 525 flp->l_start += fp->f_offset; 526 } 527 /* 528 * VOP_ADVLOCK() may block. 529 */ 530 fhold(fp); 531 FILEDESC_UNLOCK(fdp); 532 vp = fp->f_vnode; 533 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 534 F_POSIX); 535 fdrop(fp, td); 536 break; 537 default: 538 FILEDESC_UNLOCK(fdp); 539 error = EINVAL; 540 break; 541 } 542done2: 543 mtx_unlock(&Giant); 544 return (error); 545} 546 547/* 548 * Common code for dup, dup2, and fcntl(F_DUPFD). 549 */ 550static int 551do_dup(td, type, old, new, retval) 552 enum dup_type type; 553 int old, new; 554 register_t *retval; 555 struct thread *td; 556{ 557 struct filedesc *fdp; 558 struct proc *p; 559 struct file *fp; 560 struct file *delfp; 561 int error, holdleaders, maxfd; 562 563 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED), 564 ("invalid dup type %d", type)); 565 566 p = td->td_proc; 567 fdp = p->p_fd; 568 569 /* 570 * Verify we have a valid descriptor to dup from and possibly to 571 * dup to. 572 */ 573 if (old < 0 || new < 0) 574 return (EBADF); 575 maxfd = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 576 if (new >= maxfd) 577 return (EMFILE); 578 579 FILEDESC_LOCK(fdp); 580 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 581 FILEDESC_UNLOCK(fdp); 582 return (EBADF); 583 } 584 if (type == DUP_FIXED && old == new) { 585 *retval = new; 586 FILEDESC_UNLOCK(fdp); 587 return (0); 588 } 589 fp = fdp->fd_ofiles[old]; 590 fhold(fp); 591 592 /* 593 * If the caller specified a file descriptor, make sure the file 594 * table is large enough to hold it, and grab it. Otherwise, just 595 * allocate a new descriptor the usual way. Since the filedesc 596 * lock may be temporarily dropped in the process, we have to look 597 * out for a race. 598 */ 599 if (type == DUP_FIXED) { 600 if (new >= fdp->fd_nfiles) 601 fdgrowtable(fdp, new + 1); 602 if (fdp->fd_ofiles[new] == NULL) 603 fdused(fdp, new); 604 } else { 605 if ((error = fdalloc(td, new, &new)) != 0) { 606 FILEDESC_UNLOCK(fdp); 607 fdrop(fp, td); 608 return (error); 609 } 610 } 611 612 /* 613 * If the old file changed out from under us then treat it as a 614 * bad file descriptor. Userland should do its own locking to 615 * avoid this case. 616 */ 617 if (fdp->fd_ofiles[old] != fp) { 618 /* we've allocated a descriptor which we won't use */ 619 if (fdp->fd_ofiles[new] == NULL) 620 fdunused(fdp, new); 621 FILEDESC_UNLOCK(fdp); 622 fdrop(fp, td); 623 return (EBADF); 624 } 625 KASSERT(old != new, 626 ("new fd is same as old")); 627 628 /* 629 * Save info on the descriptor being overwritten. We cannot close 630 * it without introducing an ownership race for the slot, since we 631 * need to drop the filedesc lock to call closef(). 632 * 633 * XXX this duplicates parts of close(). 634 */ 635 delfp = fdp->fd_ofiles[new]; 636 holdleaders = 0; 637 if (delfp != NULL) { 638 if (td->td_proc->p_fdtol != NULL) { 639 /* 640 * Ask fdfree() to sleep to ensure that all relevant 641 * process leaders can be traversed in closef(). 642 */ 643 fdp->fd_holdleaderscount++; 644 holdleaders = 1; 645 } 646 } 647 648 /* 649 * Duplicate the source descriptor 650 */ 651 fdp->fd_ofiles[new] = fp; 652 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 653 if (new > fdp->fd_lastfile) 654 fdp->fd_lastfile = new; 655 FILEDESC_UNLOCK(fdp); 656 *retval = new; 657 658 /* 659 * If we dup'd over a valid file, we now own the reference to it 660 * and must dispose of it using closef() semantics (as if a 661 * close() were performed on it). 662 * 663 * XXX this duplicates parts of close(). 664 */ 665 if (delfp != NULL) { 666 /* XXX need to call knote_fdclose() */ 667 mtx_lock(&Giant); 668 (void) closef(delfp, td); 669 mtx_unlock(&Giant); 670 if (holdleaders) { 671 FILEDESC_LOCK(fdp); 672 fdp->fd_holdleaderscount--; 673 if (fdp->fd_holdleaderscount == 0 && 674 fdp->fd_holdleaderswakeup != 0) { 675 fdp->fd_holdleaderswakeup = 0; 676 wakeup(&fdp->fd_holdleaderscount); 677 } 678 FILEDESC_UNLOCK(fdp); 679 } 680 } 681 return (0); 682} 683 684/* 685 * If sigio is on the list associated with a process or process group, 686 * disable signalling from the device, remove sigio from the list and 687 * free sigio. 688 */ 689void 690funsetown(sigiop) 691 struct sigio **sigiop; 692{ 693 struct sigio *sigio; 694 695 SIGIO_LOCK(); 696 sigio = *sigiop; 697 if (sigio == NULL) { 698 SIGIO_UNLOCK(); 699 return; 700 } 701 *(sigio->sio_myref) = NULL; 702 if ((sigio)->sio_pgid < 0) { 703 struct pgrp *pg = (sigio)->sio_pgrp; 704 PGRP_LOCK(pg); 705 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 706 sigio, sio_pgsigio); 707 PGRP_UNLOCK(pg); 708 } else { 709 struct proc *p = (sigio)->sio_proc; 710 PROC_LOCK(p); 711 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 712 sigio, sio_pgsigio); 713 PROC_UNLOCK(p); 714 } 715 SIGIO_UNLOCK(); 716 crfree(sigio->sio_ucred); 717 FREE(sigio, M_SIGIO); 718} 719 720/* 721 * Free a list of sigio structures. 722 * We only need to lock the SIGIO_LOCK because we have made ourselves 723 * inaccessable to callers of fsetown and therefore do not need to lock 724 * the proc or pgrp struct for the list manipulation. 725 */ 726void 727funsetownlst(sigiolst) 728 struct sigiolst *sigiolst; 729{ 730 struct proc *p; 731 struct pgrp *pg; 732 struct sigio *sigio; 733 734 sigio = SLIST_FIRST(sigiolst); 735 if (sigio == NULL) 736 return; 737 p = NULL; 738 pg = NULL; 739 740 /* 741 * Every entry of the list should belong 742 * to a single proc or pgrp. 743 */ 744 if (sigio->sio_pgid < 0) { 745 pg = sigio->sio_pgrp; 746 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 747 } else /* if (sigio->sio_pgid > 0) */ { 748 p = sigio->sio_proc; 749 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 750 } 751 752 SIGIO_LOCK(); 753 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 754 *(sigio->sio_myref) = NULL; 755 if (pg != NULL) { 756 KASSERT(sigio->sio_pgid < 0, 757 ("Proc sigio in pgrp sigio list")); 758 KASSERT(sigio->sio_pgrp == pg, 759 ("Bogus pgrp in sigio list")); 760 PGRP_LOCK(pg); 761 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 762 sio_pgsigio); 763 PGRP_UNLOCK(pg); 764 } else /* if (p != NULL) */ { 765 KASSERT(sigio->sio_pgid > 0, 766 ("Pgrp sigio in proc sigio list")); 767 KASSERT(sigio->sio_proc == p, 768 ("Bogus proc in sigio list")); 769 PROC_LOCK(p); 770 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 771 sio_pgsigio); 772 PROC_UNLOCK(p); 773 } 774 SIGIO_UNLOCK(); 775 crfree(sigio->sio_ucred); 776 FREE(sigio, M_SIGIO); 777 SIGIO_LOCK(); 778 } 779 SIGIO_UNLOCK(); 780} 781 782/* 783 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 784 * 785 * After permission checking, add a sigio structure to the sigio list for 786 * the process or process group. 787 */ 788int 789fsetown(pgid, sigiop) 790 pid_t pgid; 791 struct sigio **sigiop; 792{ 793 struct proc *proc; 794 struct pgrp *pgrp; 795 struct sigio *sigio; 796 int ret; 797 798 if (pgid == 0) { 799 funsetown(sigiop); 800 return (0); 801 } 802 803 ret = 0; 804 805 /* Allocate and fill in the new sigio out of locks. */ 806 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK); 807 sigio->sio_pgid = pgid; 808 sigio->sio_ucred = crhold(curthread->td_ucred); 809 sigio->sio_myref = sigiop; 810 811 sx_slock(&proctree_lock); 812 if (pgid > 0) { 813 proc = pfind(pgid); 814 if (proc == NULL) { 815 ret = ESRCH; 816 goto fail; 817 } 818 819 /* 820 * Policy - Don't allow a process to FSETOWN a process 821 * in another session. 822 * 823 * Remove this test to allow maximum flexibility or 824 * restrict FSETOWN to the current process or process 825 * group for maximum safety. 826 */ 827 PROC_UNLOCK(proc); 828 if (proc->p_session != curthread->td_proc->p_session) { 829 ret = EPERM; 830 goto fail; 831 } 832 833 pgrp = NULL; 834 } else /* if (pgid < 0) */ { 835 pgrp = pgfind(-pgid); 836 if (pgrp == NULL) { 837 ret = ESRCH; 838 goto fail; 839 } 840 PGRP_UNLOCK(pgrp); 841 842 /* 843 * Policy - Don't allow a process to FSETOWN a process 844 * in another session. 845 * 846 * Remove this test to allow maximum flexibility or 847 * restrict FSETOWN to the current process or process 848 * group for maximum safety. 849 */ 850 if (pgrp->pg_session != curthread->td_proc->p_session) { 851 ret = EPERM; 852 goto fail; 853 } 854 855 proc = NULL; 856 } 857 funsetown(sigiop); 858 if (pgid > 0) { 859 PROC_LOCK(proc); 860 /* 861 * Since funsetownlst() is called without the proctree 862 * locked, we need to check for P_WEXIT. 863 * XXX: is ESRCH correct? 864 */ 865 if ((proc->p_flag & P_WEXIT) != 0) { 866 PROC_UNLOCK(proc); 867 ret = ESRCH; 868 goto fail; 869 } 870 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 871 sigio->sio_proc = proc; 872 PROC_UNLOCK(proc); 873 } else { 874 PGRP_LOCK(pgrp); 875 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 876 sigio->sio_pgrp = pgrp; 877 PGRP_UNLOCK(pgrp); 878 } 879 sx_sunlock(&proctree_lock); 880 SIGIO_LOCK(); 881 *sigiop = sigio; 882 SIGIO_UNLOCK(); 883 return (0); 884 885fail: 886 sx_sunlock(&proctree_lock); 887 crfree(sigio->sio_ucred); 888 FREE(sigio, M_SIGIO); 889 return (ret); 890} 891 892/* 893 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 894 */ 895pid_t 896fgetown(sigiop) 897 struct sigio **sigiop; 898{ 899 pid_t pgid; 900 901 SIGIO_LOCK(); 902 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 903 SIGIO_UNLOCK(); 904 return (pgid); 905} 906 907/* 908 * Close a file descriptor. 909 */ 910#ifndef _SYS_SYSPROTO_H_ 911struct close_args { 912 int fd; 913}; 914#endif 915/* 916 * MPSAFE 917 */ 918/* ARGSUSED */ 919int 920close(td, uap) 921 struct thread *td; 922 struct close_args *uap; 923{ 924 struct filedesc *fdp; 925 struct file *fp; 926 int fd, error; 927 int holdleaders; 928 929 fd = uap->fd; 930 error = 0; 931 holdleaders = 0; 932 fdp = td->td_proc->p_fd; 933 mtx_lock(&Giant); 934 FILEDESC_LOCK(fdp); 935 if ((unsigned)fd >= fdp->fd_nfiles || 936 (fp = fdp->fd_ofiles[fd]) == NULL) { 937 FILEDESC_UNLOCK(fdp); 938 mtx_unlock(&Giant); 939 return (EBADF); 940 } 941 fdp->fd_ofiles[fd] = NULL; 942 fdp->fd_ofileflags[fd] = 0; 943 fdunused(fdp, fd); 944 if (td->td_proc->p_fdtol != NULL) { 945 /* 946 * Ask fdfree() to sleep to ensure that all relevant 947 * process leaders can be traversed in closef(). 948 */ 949 fdp->fd_holdleaderscount++; 950 holdleaders = 1; 951 } 952 953 /* 954 * we now hold the fp reference that used to be owned by the descriptor 955 * array. 956 */ 957 if (fd < fdp->fd_knlistsize) { 958 FILEDESC_UNLOCK(fdp); 959 knote_fdclose(td, fd); 960 } else 961 FILEDESC_UNLOCK(fdp); 962 963 error = closef(fp, td); 964 mtx_unlock(&Giant); 965 if (holdleaders) { 966 FILEDESC_LOCK(fdp); 967 fdp->fd_holdleaderscount--; 968 if (fdp->fd_holdleaderscount == 0 && 969 fdp->fd_holdleaderswakeup != 0) { 970 fdp->fd_holdleaderswakeup = 0; 971 wakeup(&fdp->fd_holdleaderscount); 972 } 973 FILEDESC_UNLOCK(fdp); 974 } 975 return (error); 976} 977 978#if defined(COMPAT_43) || defined(COMPAT_SUNOS) 979/* 980 * Return status information about a file descriptor. 981 */ 982#ifndef _SYS_SYSPROTO_H_ 983struct ofstat_args { 984 int fd; 985 struct ostat *sb; 986}; 987#endif 988/* 989 * MPSAFE 990 */ 991/* ARGSUSED */ 992int 993ofstat(td, uap) 994 struct thread *td; 995 struct ofstat_args *uap; 996{ 997 struct file *fp; 998 struct stat ub; 999 struct ostat oub; 1000 int error; 1001 1002 if ((error = fget(td, uap->fd, &fp)) != 0) 1003 goto done2; 1004 mtx_lock(&Giant); 1005 error = fo_stat(fp, &ub, td->td_ucred, td); 1006 mtx_unlock(&Giant); 1007 if (error == 0) { 1008 cvtstat(&ub, &oub); 1009 error = copyout(&oub, uap->sb, sizeof(oub)); 1010 } 1011 fdrop(fp, td); 1012done2: 1013 return (error); 1014} 1015#endif /* COMPAT_43 || COMPAT_SUNOS */ 1016 1017/* 1018 * Return status information about a file descriptor. 1019 */ 1020#ifndef _SYS_SYSPROTO_H_ 1021struct fstat_args { 1022 int fd; 1023 struct stat *sb; 1024}; 1025#endif 1026/* 1027 * MPSAFE 1028 */ 1029/* ARGSUSED */ 1030int 1031fstat(td, uap) 1032 struct thread *td; 1033 struct fstat_args *uap; 1034{ 1035 struct file *fp; 1036 struct stat ub; 1037 int error; 1038 1039 if ((error = fget(td, uap->fd, &fp)) != 0) 1040 goto done2; 1041 mtx_lock(&Giant); 1042 error = fo_stat(fp, &ub, td->td_ucred, td); 1043 mtx_unlock(&Giant); 1044 if (error == 0) 1045 error = copyout(&ub, uap->sb, sizeof(ub)); 1046 fdrop(fp, td); 1047done2: 1048 return (error); 1049} 1050 1051/* 1052 * Return status information about a file descriptor. 1053 */ 1054#ifndef _SYS_SYSPROTO_H_ 1055struct nfstat_args { 1056 int fd; 1057 struct nstat *sb; 1058}; 1059#endif 1060/* 1061 * MPSAFE 1062 */ 1063/* ARGSUSED */ 1064int 1065nfstat(td, uap) 1066 struct thread *td; 1067 struct nfstat_args *uap; 1068{ 1069 struct file *fp; 1070 struct stat ub; 1071 struct nstat nub; 1072 int error; 1073 1074 if ((error = fget(td, uap->fd, &fp)) != 0) 1075 goto done2; 1076 mtx_lock(&Giant); 1077 error = fo_stat(fp, &ub, td->td_ucred, td); 1078 mtx_unlock(&Giant); 1079 if (error == 0) { 1080 cvtnstat(&ub, &nub); 1081 error = copyout(&nub, uap->sb, sizeof(nub)); 1082 } 1083 fdrop(fp, td); 1084done2: 1085 return (error); 1086} 1087 1088/* 1089 * Return pathconf information about a file descriptor. 1090 */ 1091#ifndef _SYS_SYSPROTO_H_ 1092struct fpathconf_args { 1093 int fd; 1094 int name; 1095}; 1096#endif 1097/* 1098 * MPSAFE 1099 */ 1100/* ARGSUSED */ 1101int 1102fpathconf(td, uap) 1103 struct thread *td; 1104 struct fpathconf_args *uap; 1105{ 1106 struct file *fp; 1107 struct vnode *vp; 1108 int error; 1109 1110 if ((error = fget(td, uap->fd, &fp)) != 0) 1111 return (error); 1112 1113 /* If asynchronous I/O is available, it works for all descriptors. */ 1114 if (uap->name == _PC_ASYNC_IO) { 1115 td->td_retval[0] = async_io_version; 1116 goto out; 1117 } 1118 vp = fp->f_vnode; 1119 if (vp != NULL) { 1120 mtx_lock(&Giant); 1121 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1122 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1123 VOP_UNLOCK(vp, 0, td); 1124 mtx_unlock(&Giant); 1125 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1126 if (uap->name != _PC_PIPE_BUF) { 1127 error = EINVAL; 1128 } else { 1129 td->td_retval[0] = PIPE_BUF; 1130 error = 0; 1131 } 1132 } else { 1133 error = EOPNOTSUPP; 1134 } 1135out: 1136 fdrop(fp, td); 1137 return (error); 1138} 1139 1140/* 1141 * Grow the file table to accomodate (at least) nfd descriptors. This may 1142 * block and drop the filedesc lock, but it will reacquire it before 1143 * returing. 1144 */ 1145static void 1146fdgrowtable(struct filedesc *fdp, int nfd) 1147{ 1148 struct file **ntable; 1149 char *nfileflags; 1150 int nnfiles, onfiles; 1151 NDSLOTTYPE *nmap; 1152 1153 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1154 1155 KASSERT(fdp->fd_nfiles > 0, 1156 ("zero-length file table")); 1157 1158 /* compute the size of the new table */ 1159 onfiles = fdp->fd_nfiles; 1160 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1161 if (nnfiles <= onfiles) 1162 /* the table is already large enough */ 1163 return; 1164 1165 /* allocate a new table and (if required) new bitmaps */ 1166 FILEDESC_UNLOCK(fdp); 1167 MALLOC(ntable, struct file **, nnfiles * OFILESIZE, 1168 M_FILEDESC, M_ZERO | M_WAITOK); 1169 nfileflags = (char *)&ntable[nnfiles]; 1170 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1171 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE, 1172 M_FILEDESC, M_ZERO | M_WAITOK); 1173 else 1174 nmap = NULL; 1175 FILEDESC_LOCK(fdp); 1176 1177 /* 1178 * We now have new tables ready to go. Since we dropped the 1179 * filedesc lock to call malloc(), watch out for a race. 1180 */ 1181 onfiles = fdp->fd_nfiles; 1182 if (onfiles >= nnfiles) { 1183 /* we lost the race, but that's OK */ 1184 free(ntable, M_FILEDESC); 1185 if (nmap != NULL) 1186 free(nmap, M_FILEDESC); 1187 return; 1188 } 1189 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1190 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1191 if (onfiles > NDFILE) 1192 free(fdp->fd_ofiles, M_FILEDESC); 1193 fdp->fd_ofiles = ntable; 1194 fdp->fd_ofileflags = nfileflags; 1195 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1196 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1197 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1198 free(fdp->fd_map, M_FILEDESC); 1199 fdp->fd_map = nmap; 1200 } 1201 fdp->fd_nfiles = nnfiles; 1202} 1203 1204/* 1205 * Allocate a file descriptor for the process. 1206 */ 1207int 1208fdalloc(struct thread *td, int minfd, int *result) 1209{ 1210 struct proc *p = td->td_proc; 1211 struct filedesc *fdp = p->p_fd; 1212 int fd = -1, maxfd; 1213 1214 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1215 1216 maxfd = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1217 1218 /* 1219 * Search the bitmap for a free descriptor. If none is found, try 1220 * to grow the file table. Keep at it until we either get a file 1221 * descriptor or run into process or system limits; fdgrowtable() 1222 * may drop the filedesc lock, so we're in a race. 1223 */ 1224 for (;;) { 1225 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1226 if (fd >= maxfd) 1227 return (EMFILE); 1228 if (fd < fdp->fd_nfiles) 1229 break; 1230 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1231 } 1232 1233 /* 1234 * Perform some sanity checks, then mark the file descriptor as 1235 * used and return it to the caller. 1236 */ 1237 KASSERT(!fdisused(fdp, fd), 1238 ("fd_first_free() returned non-free descriptor")); 1239 KASSERT(fdp->fd_ofiles[fd] == NULL, 1240 ("free descriptor isn't")); 1241 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */ 1242 fdused(fdp, fd); 1243 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 1244 *result = fd; 1245 return (0); 1246} 1247 1248/* 1249 * Check to see whether n user file descriptors 1250 * are available to the process p. 1251 */ 1252int 1253fdavail(td, n) 1254 struct thread *td; 1255 int n; 1256{ 1257 struct proc *p = td->td_proc; 1258 struct filedesc *fdp = td->td_proc->p_fd; 1259 struct file **fpp; 1260 int i, lim, last; 1261 1262 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1263 1264 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc); 1265 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1266 return (1); 1267 last = min(fdp->fd_nfiles, lim); 1268 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 1269 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 1270 if (*fpp == NULL && --n <= 0) 1271 return (1); 1272 } 1273 return (0); 1274} 1275 1276/* 1277 * Create a new open file structure and allocate 1278 * a file decriptor for the process that refers to it. 1279 * We add one reference to the file for the descriptor table 1280 * and one reference for resultfp. This is to prevent us being 1281 * prempted and the entry in the descriptor table closed after 1282 * we release the FILEDESC lock. 1283 */ 1284int 1285falloc(td, resultfp, resultfd) 1286 struct thread *td; 1287 struct file **resultfp; 1288 int *resultfd; 1289{ 1290 struct proc *p = td->td_proc; 1291 struct file *fp, *fq; 1292 int error, i; 1293 int maxuserfiles = maxfiles - (maxfiles / 20); 1294 static struct timeval lastfail; 1295 static int curfail; 1296 1297 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1298 sx_xlock(&filelist_lock); 1299 if ((nfiles >= maxuserfiles && td->td_ucred->cr_ruid != 0) 1300 || nfiles >= maxfiles) { 1301 if (ppsratecheck(&lastfail, &curfail, 1)) { 1302 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n", 1303 td->td_ucred->cr_ruid); 1304 } 1305 sx_xunlock(&filelist_lock); 1306 uma_zfree(file_zone, fp); 1307 return (ENFILE); 1308 } 1309 nfiles++; 1310 1311 /* 1312 * If the process has file descriptor zero open, add the new file 1313 * descriptor to the list of open files at that point, otherwise 1314 * put it at the front of the list of open files. 1315 */ 1316 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep); 1317 fp->f_count = 1; 1318 if (resultfp) 1319 fp->f_count++; 1320 fp->f_cred = crhold(td->td_ucred); 1321 fp->f_ops = &badfileops; 1322 FILEDESC_LOCK(p->p_fd); 1323 if ((fq = p->p_fd->fd_ofiles[0])) { 1324 LIST_INSERT_AFTER(fq, fp, f_list); 1325 } else { 1326 LIST_INSERT_HEAD(&filehead, fp, f_list); 1327 } 1328 sx_xunlock(&filelist_lock); 1329 if ((error = fdalloc(td, 0, &i))) { 1330 FILEDESC_UNLOCK(p->p_fd); 1331 fdrop(fp, td); 1332 if (resultfp) 1333 fdrop(fp, td); 1334 return (error); 1335 } 1336 p->p_fd->fd_ofiles[i] = fp; 1337 FILEDESC_UNLOCK(p->p_fd); 1338 if (resultfp) 1339 *resultfp = fp; 1340 if (resultfd) 1341 *resultfd = i; 1342 return (0); 1343} 1344 1345/* 1346 * Free a file descriptor. 1347 */ 1348void 1349ffree(fp) 1350 struct file *fp; 1351{ 1352 1353 KASSERT(fp->f_count == 0, ("ffree: fp_fcount not 0!")); 1354 sx_xlock(&filelist_lock); 1355 LIST_REMOVE(fp, f_list); 1356 nfiles--; 1357 sx_xunlock(&filelist_lock); 1358 crfree(fp->f_cred); 1359 uma_zfree(file_zone, fp); 1360} 1361 1362/* 1363 * Build a new filedesc structure from another. 1364 * Copy the current, root, and jail root vnode references. 1365 */ 1366struct filedesc * 1367fdinit(fdp) 1368 struct filedesc *fdp; 1369{ 1370 struct filedesc0 *newfdp; 1371 1372 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1373 1374 FILEDESC_UNLOCK(fdp); 1375 MALLOC(newfdp, struct filedesc0 *, sizeof(struct filedesc0), 1376 M_FILEDESC, M_WAITOK | M_ZERO); 1377 FILEDESC_LOCK(fdp); 1378 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF); 1379 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1380 if (newfdp->fd_fd.fd_cdir) 1381 VREF(newfdp->fd_fd.fd_cdir); 1382 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1383 if (newfdp->fd_fd.fd_rdir) 1384 VREF(newfdp->fd_fd.fd_rdir); 1385 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1386 if (newfdp->fd_fd.fd_jdir) 1387 VREF(newfdp->fd_fd.fd_jdir); 1388 1389 /* Create the file descriptor table. */ 1390 newfdp->fd_fd.fd_refcnt = 1; 1391 newfdp->fd_fd.fd_cmask = CMASK; 1392 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1393 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1394 newfdp->fd_fd.fd_nfiles = NDFILE; 1395 newfdp->fd_fd.fd_knlistsize = -1; 1396 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1397 return (&newfdp->fd_fd); 1398} 1399 1400/* 1401 * Share a filedesc structure. 1402 */ 1403struct filedesc * 1404fdshare(fdp) 1405 struct filedesc *fdp; 1406{ 1407 FILEDESC_LOCK(fdp); 1408 fdp->fd_refcnt++; 1409 FILEDESC_UNLOCK(fdp); 1410 return (fdp); 1411} 1412 1413/* 1414 * Copy a filedesc structure. 1415 * A NULL pointer in returns a NULL reference, this is to ease callers, 1416 * not catch errors. 1417 */ 1418struct filedesc * 1419fdcopy(fdp) 1420 struct filedesc *fdp; 1421{ 1422 struct filedesc *newfdp; 1423 int i; 1424 1425 /* Certain daemons might not have file descriptors. */ 1426 if (fdp == NULL) 1427 return (NULL); 1428 1429 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1430 newfdp = fdinit(fdp); 1431 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1432 FILEDESC_UNLOCK(fdp); 1433 FILEDESC_LOCK(newfdp); 1434 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1435 FILEDESC_UNLOCK(newfdp); 1436 FILEDESC_LOCK(fdp); 1437 } 1438 /* copy everything except kqueue descriptors */ 1439 newfdp->fd_freefile = -1; 1440 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1441 if (fdisused(fdp, i) && 1442 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) { 1443 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1444 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1445 fhold(newfdp->fd_ofiles[i]); 1446 newfdp->fd_lastfile = i; 1447 } else { 1448 if (newfdp->fd_freefile == -1) 1449 newfdp->fd_freefile = i; 1450 } 1451 } 1452 FILEDESC_UNLOCK(fdp); 1453 FILEDESC_LOCK(newfdp); 1454 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1455 if (newfdp->fd_ofiles[i] != NULL) 1456 fdused(newfdp, i); 1457 FILEDESC_UNLOCK(newfdp); 1458 FILEDESC_LOCK(fdp); 1459 if (newfdp->fd_freefile == -1) 1460 newfdp->fd_freefile = i; 1461 newfdp->fd_cmask = fdp->fd_cmask; 1462 return (newfdp); 1463} 1464 1465/* A mutex to protect the association between a proc and filedesc. */ 1466struct mtx fdesc_mtx; 1467MTX_SYSINIT(fdesc, &fdesc_mtx, "fdesc", MTX_DEF); 1468 1469/* 1470 * Release a filedesc structure. 1471 */ 1472void 1473fdfree(td) 1474 struct thread *td; 1475{ 1476 struct filedesc *fdp; 1477 struct file **fpp; 1478 int i; 1479 struct filedesc_to_leader *fdtol; 1480 struct file *fp; 1481 struct vnode *vp; 1482 struct flock lf; 1483 1484 /* Certain daemons might not have file descriptors. */ 1485 fdp = td->td_proc->p_fd; 1486 if (fdp == NULL) 1487 return; 1488 1489 /* Check for special need to clear POSIX style locks */ 1490 fdtol = td->td_proc->p_fdtol; 1491 if (fdtol != NULL) { 1492 FILEDESC_LOCK(fdp); 1493 KASSERT(fdtol->fdl_refcount > 0, 1494 ("filedesc_to_refcount botch: fdl_refcount=%d", 1495 fdtol->fdl_refcount)); 1496 if (fdtol->fdl_refcount == 1 && 1497 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1498 i = 0; 1499 fpp = fdp->fd_ofiles; 1500 for (i = 0, fpp = fdp->fd_ofiles; 1501 i <= fdp->fd_lastfile; 1502 i++, fpp++) { 1503 if (*fpp == NULL || 1504 (*fpp)->f_type != DTYPE_VNODE) 1505 continue; 1506 fp = *fpp; 1507 fhold(fp); 1508 FILEDESC_UNLOCK(fdp); 1509 lf.l_whence = SEEK_SET; 1510 lf.l_start = 0; 1511 lf.l_len = 0; 1512 lf.l_type = F_UNLCK; 1513 vp = fp->f_vnode; 1514 (void) VOP_ADVLOCK(vp, 1515 (caddr_t)td->td_proc-> 1516 p_leader, 1517 F_UNLCK, 1518 &lf, 1519 F_POSIX); 1520 FILEDESC_LOCK(fdp); 1521 fdrop(fp, td); 1522 fpp = fdp->fd_ofiles + i; 1523 } 1524 } 1525 retry: 1526 if (fdtol->fdl_refcount == 1) { 1527 if (fdp->fd_holdleaderscount > 0 && 1528 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1529 /* 1530 * close() or do_dup() has cleared a reference 1531 * in a shared file descriptor table. 1532 */ 1533 fdp->fd_holdleaderswakeup = 1; 1534 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx, 1535 PLOCK, "fdlhold", 0); 1536 goto retry; 1537 } 1538 if (fdtol->fdl_holdcount > 0) { 1539 /* 1540 * Ensure that fdtol->fdl_leader 1541 * remains valid in closef(). 1542 */ 1543 fdtol->fdl_wakeup = 1; 1544 msleep(fdtol, &fdp->fd_mtx, 1545 PLOCK, "fdlhold", 0); 1546 goto retry; 1547 } 1548 } 1549 fdtol->fdl_refcount--; 1550 if (fdtol->fdl_refcount == 0 && 1551 fdtol->fdl_holdcount == 0) { 1552 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1553 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1554 } else 1555 fdtol = NULL; 1556 td->td_proc->p_fdtol = NULL; 1557 FILEDESC_UNLOCK(fdp); 1558 if (fdtol != NULL) 1559 FREE(fdtol, M_FILEDESC_TO_LEADER); 1560 } 1561 FILEDESC_LOCK(fdp); 1562 if (--fdp->fd_refcnt > 0) { 1563 FILEDESC_UNLOCK(fdp); 1564 return; 1565 } 1566 1567 /* 1568 * We are the last reference to the structure, so we can 1569 * safely assume it will not change out from under us. 1570 */ 1571 FILEDESC_UNLOCK(fdp); 1572 fpp = fdp->fd_ofiles; 1573 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1574 if (*fpp) 1575 (void) closef(*fpp, td); 1576 } 1577 1578 /* XXX This should happen earlier. */ 1579 mtx_lock(&fdesc_mtx); 1580 td->td_proc->p_fd = NULL; 1581 mtx_unlock(&fdesc_mtx); 1582 1583 if (fdp->fd_nfiles > NDFILE) 1584 FREE(fdp->fd_ofiles, M_FILEDESC); 1585 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1586 FREE(fdp->fd_map, M_FILEDESC); 1587 if (fdp->fd_cdir) 1588 vrele(fdp->fd_cdir); 1589 if (fdp->fd_rdir) 1590 vrele(fdp->fd_rdir); 1591 if (fdp->fd_jdir) 1592 vrele(fdp->fd_jdir); 1593 if (fdp->fd_knlist) 1594 FREE(fdp->fd_knlist, M_KQUEUE); 1595 if (fdp->fd_knhash) 1596 FREE(fdp->fd_knhash, M_KQUEUE); 1597 mtx_destroy(&fdp->fd_mtx); 1598 FREE(fdp, M_FILEDESC); 1599} 1600 1601/* 1602 * For setugid programs, we don't want to people to use that setugidness 1603 * to generate error messages which write to a file which otherwise would 1604 * otherwise be off-limits to the process. We check for filesystems where 1605 * the vnode can change out from under us after execve (like [lin]procfs). 1606 * 1607 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1608 * sufficient. We also don't for check setugidness since we know we are. 1609 */ 1610static int 1611is_unsafe(struct file *fp) 1612{ 1613 if (fp->f_type == DTYPE_VNODE) { 1614 struct vnode *vp = fp->f_vnode; 1615 1616 if ((vp->v_vflag & VV_PROCDEP) != 0) 1617 return (1); 1618 } 1619 return (0); 1620} 1621 1622/* 1623 * Make this setguid thing safe, if at all possible. 1624 */ 1625void 1626setugidsafety(td) 1627 struct thread *td; 1628{ 1629 struct filedesc *fdp; 1630 int i; 1631 1632 /* Certain daemons might not have file descriptors. */ 1633 fdp = td->td_proc->p_fd; 1634 if (fdp == NULL) 1635 return; 1636 1637 /* 1638 * Note: fdp->fd_ofiles may be reallocated out from under us while 1639 * we are blocked in a close. Be careful! 1640 */ 1641 FILEDESC_LOCK(fdp); 1642 for (i = 0; i <= fdp->fd_lastfile; i++) { 1643 if (i > 2) 1644 break; 1645 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1646 struct file *fp; 1647 1648 if (i < fdp->fd_knlistsize) { 1649 FILEDESC_UNLOCK(fdp); 1650 knote_fdclose(td, i); 1651 FILEDESC_LOCK(fdp); 1652 } 1653 /* 1654 * NULL-out descriptor prior to close to avoid 1655 * a race while close blocks. 1656 */ 1657 fp = fdp->fd_ofiles[i]; 1658 fdp->fd_ofiles[i] = NULL; 1659 fdp->fd_ofileflags[i] = 0; 1660 fdunused(fdp, i); 1661 FILEDESC_UNLOCK(fdp); 1662 (void) closef(fp, td); 1663 FILEDESC_LOCK(fdp); 1664 } 1665 } 1666 FILEDESC_UNLOCK(fdp); 1667} 1668 1669/* 1670 * Close any files on exec? 1671 */ 1672void 1673fdcloseexec(td) 1674 struct thread *td; 1675{ 1676 struct filedesc *fdp; 1677 int i; 1678 1679 /* Certain daemons might not have file descriptors. */ 1680 fdp = td->td_proc->p_fd; 1681 if (fdp == NULL) 1682 return; 1683 1684 FILEDESC_LOCK(fdp); 1685 1686 /* 1687 * We cannot cache fd_ofiles or fd_ofileflags since operations 1688 * may block and rip them out from under us. 1689 */ 1690 for (i = 0; i <= fdp->fd_lastfile; i++) { 1691 if (fdp->fd_ofiles[i] != NULL && 1692 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) { 1693 struct file *fp; 1694 1695 if (i < fdp->fd_knlistsize) { 1696 FILEDESC_UNLOCK(fdp); 1697 knote_fdclose(td, i); 1698 FILEDESC_LOCK(fdp); 1699 } 1700 /* 1701 * NULL-out descriptor prior to close to avoid 1702 * a race while close blocks. 1703 */ 1704 fp = fdp->fd_ofiles[i]; 1705 fdp->fd_ofiles[i] = NULL; 1706 fdp->fd_ofileflags[i] = 0; 1707 fdunused(fdp, i); 1708 FILEDESC_UNLOCK(fdp); 1709 (void) closef(fp, td); 1710 FILEDESC_LOCK(fdp); 1711 } 1712 } 1713 FILEDESC_UNLOCK(fdp); 1714} 1715 1716/* 1717 * It is unsafe for set[ug]id processes to be started with file 1718 * descriptors 0..2 closed, as these descriptors are given implicit 1719 * significance in the Standard C library. fdcheckstd() will create a 1720 * descriptor referencing /dev/null for each of stdin, stdout, and 1721 * stderr that is not already open. 1722 */ 1723int 1724fdcheckstd(td) 1725 struct thread *td; 1726{ 1727 struct nameidata nd; 1728 struct filedesc *fdp; 1729 struct file *fp; 1730 register_t retval; 1731 int fd, i, error, flags, devnull, extraref; 1732 1733 fdp = td->td_proc->p_fd; 1734 if (fdp == NULL) 1735 return (0); 1736 devnull = -1; 1737 error = 0; 1738 for (i = 0; i < 3; i++) { 1739 if (fdp->fd_ofiles[i] != NULL) 1740 continue; 1741 if (devnull < 0) { 1742 error = falloc(td, &fp, &fd); 1743 if (error != 0) 1744 break; 1745 /* Note extra ref on `fp' held for us by falloc(). */ 1746 KASSERT(fd == i, ("oof, we didn't get our fd")); 1747 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null", 1748 td); 1749 flags = FREAD | FWRITE; 1750 error = vn_open(&nd, &flags, 0, -1); 1751 if (error != 0) { 1752 /* 1753 * Someone may have closed the entry in the 1754 * file descriptor table, so check it hasn't 1755 * changed before dropping the reference count. 1756 */ 1757 extraref = 0; 1758 FILEDESC_LOCK(fdp); 1759 if (fdp->fd_ofiles[fd] == fp) { 1760 fdp->fd_ofiles[fd] = NULL; 1761 fdunused(fdp, fd); 1762 extraref = 1; 1763 } 1764 FILEDESC_UNLOCK(fdp); 1765 fdrop(fp, td); 1766 if (extraref) 1767 fdrop(fp, td); 1768 break; 1769 } 1770 NDFREE(&nd, NDF_ONLY_PNBUF); 1771 fp->f_vnode = nd.ni_vp; 1772 fp->f_data = nd.ni_vp; 1773 fp->f_flag = flags; 1774 fp->f_ops = &vnops; 1775 fp->f_type = DTYPE_VNODE; 1776 VOP_UNLOCK(nd.ni_vp, 0, td); 1777 devnull = fd; 1778 fdrop(fp, td); 1779 } else { 1780 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 1781 if (error != 0) 1782 break; 1783 } 1784 } 1785 return (error); 1786} 1787 1788/* 1789 * Internal form of close. 1790 * Decrement reference count on file structure. 1791 * Note: td may be NULL when closing a file 1792 * that was being passed in a message. 1793 */ 1794int 1795closef(fp, td) 1796 struct file *fp; 1797 struct thread *td; 1798{ 1799 struct vnode *vp; 1800 struct flock lf; 1801 struct filedesc_to_leader *fdtol; 1802 struct filedesc *fdp; 1803 1804 if (fp == NULL) 1805 return (0); 1806 /* 1807 * POSIX record locking dictates that any close releases ALL 1808 * locks owned by this process. This is handled by setting 1809 * a flag in the unlock to free ONLY locks obeying POSIX 1810 * semantics, and not to free BSD-style file locks. 1811 * If the descriptor was in a message, POSIX-style locks 1812 * aren't passed with the descriptor. 1813 */ 1814 if (td != NULL && 1815 fp->f_type == DTYPE_VNODE) { 1816 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1817 lf.l_whence = SEEK_SET; 1818 lf.l_start = 0; 1819 lf.l_len = 0; 1820 lf.l_type = F_UNLCK; 1821 vp = fp->f_vnode; 1822 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 1823 F_UNLCK, &lf, F_POSIX); 1824 } 1825 fdtol = td->td_proc->p_fdtol; 1826 if (fdtol != NULL) { 1827 /* 1828 * Handle special case where file descriptor table 1829 * is shared between multiple process leaders. 1830 */ 1831 fdp = td->td_proc->p_fd; 1832 FILEDESC_LOCK(fdp); 1833 for (fdtol = fdtol->fdl_next; 1834 fdtol != td->td_proc->p_fdtol; 1835 fdtol = fdtol->fdl_next) { 1836 if ((fdtol->fdl_leader->p_flag & 1837 P_ADVLOCK) == 0) 1838 continue; 1839 fdtol->fdl_holdcount++; 1840 FILEDESC_UNLOCK(fdp); 1841 lf.l_whence = SEEK_SET; 1842 lf.l_start = 0; 1843 lf.l_len = 0; 1844 lf.l_type = F_UNLCK; 1845 vp = fp->f_vnode; 1846 (void) VOP_ADVLOCK(vp, 1847 (caddr_t)fdtol->fdl_leader, 1848 F_UNLCK, &lf, F_POSIX); 1849 FILEDESC_LOCK(fdp); 1850 fdtol->fdl_holdcount--; 1851 if (fdtol->fdl_holdcount == 0 && 1852 fdtol->fdl_wakeup != 0) { 1853 fdtol->fdl_wakeup = 0; 1854 wakeup(fdtol); 1855 } 1856 } 1857 FILEDESC_UNLOCK(fdp); 1858 } 1859 } 1860 return (fdrop(fp, td)); 1861} 1862 1863/* 1864 * Drop reference on struct file passed in, may call closef if the 1865 * reference hits zero. 1866 */ 1867int 1868fdrop(fp, td) 1869 struct file *fp; 1870 struct thread *td; 1871{ 1872 1873 FILE_LOCK(fp); 1874 return (fdrop_locked(fp, td)); 1875} 1876 1877/* 1878 * Extract the file pointer associated with the specified descriptor for 1879 * the current user process. 1880 * 1881 * If the descriptor doesn't exist, EBADF is returned. 1882 * 1883 * If the descriptor exists but doesn't match 'flags' then 1884 * return EBADF for read attempts and EINVAL for write attempts. 1885 * 1886 * If 'hold' is set (non-zero) the file's refcount will be bumped on return. 1887 * It should be droped with fdrop(). 1888 * If it is not set, then the refcount will not be bumped however the 1889 * thread's filedesc struct will be returned locked (for fgetsock). 1890 * 1891 * If an error occured the non-zero error is returned and *fpp is set to NULL. 1892 * Otherwise *fpp is set and zero is returned. 1893 */ 1894static __inline int 1895_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold) 1896{ 1897 struct filedesc *fdp; 1898 struct file *fp; 1899 1900 *fpp = NULL; 1901 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 1902 return (EBADF); 1903 FILEDESC_LOCK(fdp); 1904 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) { 1905 FILEDESC_UNLOCK(fdp); 1906 return (EBADF); 1907 } 1908 1909 /* 1910 * Note: FREAD failures returns EBADF to maintain backwards 1911 * compatibility with what routines returned before. 1912 * 1913 * Only one flag, or 0, may be specified. 1914 */ 1915 if (flags == FREAD && (fp->f_flag & FREAD) == 0) { 1916 FILEDESC_UNLOCK(fdp); 1917 return (EBADF); 1918 } 1919 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) { 1920 FILEDESC_UNLOCK(fdp); 1921 return (EINVAL); 1922 } 1923 if (hold) { 1924 fhold(fp); 1925 FILEDESC_UNLOCK(fdp); 1926 } 1927 *fpp = fp; 1928 return (0); 1929} 1930 1931int 1932fget(struct thread *td, int fd, struct file **fpp) 1933{ 1934 1935 return(_fget(td, fd, fpp, 0, 1)); 1936} 1937 1938int 1939fget_read(struct thread *td, int fd, struct file **fpp) 1940{ 1941 1942 return(_fget(td, fd, fpp, FREAD, 1)); 1943} 1944 1945int 1946fget_write(struct thread *td, int fd, struct file **fpp) 1947{ 1948 1949 return(_fget(td, fd, fpp, FWRITE, 1)); 1950} 1951 1952/* 1953 * Like fget() but loads the underlying vnode, or returns an error if 1954 * the descriptor does not represent a vnode. Note that pipes use vnodes 1955 * but never have VM objects (so VOP_GETVOBJECT() calls will return an 1956 * error). The returned vnode will be vref()d. 1957 */ 1958static __inline int 1959_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags) 1960{ 1961 struct file *fp; 1962 int error; 1963 1964 *vpp = NULL; 1965 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 1966 return (error); 1967 if (fp->f_vnode == NULL) { 1968 error = EINVAL; 1969 } else { 1970 *vpp = fp->f_vnode; 1971 vref(*vpp); 1972 } 1973 FILEDESC_UNLOCK(td->td_proc->p_fd); 1974 return (error); 1975} 1976 1977int 1978fgetvp(struct thread *td, int fd, struct vnode **vpp) 1979{ 1980 1981 return (_fgetvp(td, fd, vpp, 0)); 1982} 1983 1984int 1985fgetvp_read(struct thread *td, int fd, struct vnode **vpp) 1986{ 1987 1988 return (_fgetvp(td, fd, vpp, FREAD)); 1989} 1990 1991int 1992fgetvp_write(struct thread *td, int fd, struct vnode **vpp) 1993{ 1994 1995 return (_fgetvp(td, fd, vpp, FWRITE)); 1996} 1997 1998/* 1999 * Like fget() but loads the underlying socket, or returns an error if 2000 * the descriptor does not represent a socket. 2001 * 2002 * We bump the ref count on the returned socket. XXX Also obtain the SX 2003 * lock in the future. 2004 */ 2005int 2006fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp) 2007{ 2008 struct file *fp; 2009 int error; 2010 2011 *spp = NULL; 2012 if (fflagp != NULL) 2013 *fflagp = 0; 2014 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2015 return (error); 2016 if (fp->f_type != DTYPE_SOCKET) { 2017 error = ENOTSOCK; 2018 } else { 2019 *spp = fp->f_data; 2020 if (fflagp) 2021 *fflagp = fp->f_flag; 2022 soref(*spp); 2023 } 2024 FILEDESC_UNLOCK(td->td_proc->p_fd); 2025 return (error); 2026} 2027 2028/* 2029 * Drop the reference count on the the socket and XXX release the SX lock in 2030 * the future. The last reference closes the socket. 2031 */ 2032void 2033fputsock(struct socket *so) 2034{ 2035 2036 sorele(so); 2037} 2038 2039/* 2040 * Drop reference on struct file passed in, may call closef if the 2041 * reference hits zero. 2042 * Expects struct file locked, and will unlock it. 2043 */ 2044int 2045fdrop_locked(fp, td) 2046 struct file *fp; 2047 struct thread *td; 2048{ 2049 struct flock lf; 2050 struct vnode *vp; 2051 int error; 2052 2053 FILE_LOCK_ASSERT(fp, MA_OWNED); 2054 2055 if (--fp->f_count > 0) { 2056 FILE_UNLOCK(fp); 2057 return (0); 2058 } 2059 /* We have the last ref so we can proceed without the file lock. */ 2060 FILE_UNLOCK(fp); 2061 mtx_lock(&Giant); 2062 if (fp->f_count < 0) 2063 panic("fdrop: count < 0"); 2064 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 2065 lf.l_whence = SEEK_SET; 2066 lf.l_start = 0; 2067 lf.l_len = 0; 2068 lf.l_type = F_UNLCK; 2069 vp = fp->f_vnode; 2070 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2071 } 2072 if (fp->f_ops != &badfileops) 2073 error = fo_close(fp, td); 2074 else 2075 error = 0; 2076 ffree(fp); 2077 mtx_unlock(&Giant); 2078 return (error); 2079} 2080 2081/* 2082 * Apply an advisory lock on a file descriptor. 2083 * 2084 * Just attempt to get a record lock of the requested type on 2085 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2086 */ 2087#ifndef _SYS_SYSPROTO_H_ 2088struct flock_args { 2089 int fd; 2090 int how; 2091}; 2092#endif 2093/* 2094 * MPSAFE 2095 */ 2096/* ARGSUSED */ 2097int 2098flock(td, uap) 2099 struct thread *td; 2100 struct flock_args *uap; 2101{ 2102 struct file *fp; 2103 struct vnode *vp; 2104 struct flock lf; 2105 int error; 2106 2107 if ((error = fget(td, uap->fd, &fp)) != 0) 2108 return (error); 2109 if (fp->f_type != DTYPE_VNODE) { 2110 fdrop(fp, td); 2111 return (EOPNOTSUPP); 2112 } 2113 2114 mtx_lock(&Giant); 2115 vp = fp->f_vnode; 2116 lf.l_whence = SEEK_SET; 2117 lf.l_start = 0; 2118 lf.l_len = 0; 2119 if (uap->how & LOCK_UN) { 2120 lf.l_type = F_UNLCK; 2121 FILE_LOCK(fp); 2122 fp->f_flag &= ~FHASLOCK; 2123 FILE_UNLOCK(fp); 2124 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2125 goto done2; 2126 } 2127 if (uap->how & LOCK_EX) 2128 lf.l_type = F_WRLCK; 2129 else if (uap->how & LOCK_SH) 2130 lf.l_type = F_RDLCK; 2131 else { 2132 error = EBADF; 2133 goto done2; 2134 } 2135 FILE_LOCK(fp); 2136 fp->f_flag |= FHASLOCK; 2137 FILE_UNLOCK(fp); 2138 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2139 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2140done2: 2141 fdrop(fp, td); 2142 mtx_unlock(&Giant); 2143 return (error); 2144} 2145 2146/* 2147 * File Descriptor pseudo-device driver (/dev/fd/). 2148 * 2149 * Opening minor device N dup()s the file (if any) connected to file 2150 * descriptor N belonging to the calling process. Note that this driver 2151 * consists of only the ``open()'' routine, because all subsequent 2152 * references to this file will be direct to the other driver. 2153 */ 2154/* ARGSUSED */ 2155static int 2156fdopen(dev, mode, type, td) 2157 dev_t dev; 2158 int mode, type; 2159 struct thread *td; 2160{ 2161 2162 /* 2163 * XXX Kludge: set curthread->td_dupfd to contain the value of the 2164 * the file descriptor being sought for duplication. The error 2165 * return ensures that the vnode for this device will be released 2166 * by vn_open. Open will detect this special error and take the 2167 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2168 * will simply report the error. 2169 */ 2170 td->td_dupfd = dev2unit(dev); 2171 return (ENODEV); 2172} 2173 2174/* 2175 * Duplicate the specified descriptor to a free descriptor. 2176 */ 2177int 2178dupfdopen(td, fdp, indx, dfd, mode, error) 2179 struct thread *td; 2180 struct filedesc *fdp; 2181 int indx, dfd; 2182 int mode; 2183 int error; 2184{ 2185 struct file *wfp; 2186 struct file *fp; 2187 2188 /* 2189 * If the to-be-dup'd fd number is greater than the allowed number 2190 * of file descriptors, or the fd to be dup'd has already been 2191 * closed, then reject. 2192 */ 2193 FILEDESC_LOCK(fdp); 2194 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2195 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2196 FILEDESC_UNLOCK(fdp); 2197 return (EBADF); 2198 } 2199 2200 /* 2201 * There are two cases of interest here. 2202 * 2203 * For ENODEV simply dup (dfd) to file descriptor 2204 * (indx) and return. 2205 * 2206 * For ENXIO steal away the file structure from (dfd) and 2207 * store it in (indx). (dfd) is effectively closed by 2208 * this operation. 2209 * 2210 * Any other error code is just returned. 2211 */ 2212 switch (error) { 2213 case ENODEV: 2214 /* 2215 * Check that the mode the file is being opened for is a 2216 * subset of the mode of the existing descriptor. 2217 */ 2218 FILE_LOCK(wfp); 2219 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2220 FILE_UNLOCK(wfp); 2221 FILEDESC_UNLOCK(fdp); 2222 return (EACCES); 2223 } 2224 fp = fdp->fd_ofiles[indx]; 2225 fdp->fd_ofiles[indx] = wfp; 2226 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2227 if (fp == NULL) 2228 fdused(fdp, indx); 2229 fhold_locked(wfp); 2230 FILE_UNLOCK(wfp); 2231 if (fp != NULL) 2232 FILE_LOCK(fp); 2233 FILEDESC_UNLOCK(fdp); 2234 /* 2235 * We now own the reference to fp that the ofiles[] array 2236 * used to own. Release it. 2237 */ 2238 if (fp != NULL) 2239 fdrop_locked(fp, td); 2240 return (0); 2241 2242 case ENXIO: 2243 /* 2244 * Steal away the file pointer from dfd and stuff it into indx. 2245 */ 2246 fp = fdp->fd_ofiles[indx]; 2247 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2248 fdp->fd_ofiles[dfd] = NULL; 2249 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2250 fdp->fd_ofileflags[dfd] = 0; 2251 fdunused(fdp, dfd); 2252 if (fp == NULL) 2253 fdused(fdp, indx); 2254 if (fp != NULL) 2255 FILE_LOCK(fp); 2256 FILEDESC_UNLOCK(fdp); 2257 2258 /* 2259 * we now own the reference to fp that the ofiles[] array 2260 * used to own. Release it. 2261 */ 2262 if (fp != NULL) 2263 fdrop_locked(fp, td); 2264 return (0); 2265 2266 default: 2267 FILEDESC_UNLOCK(fdp); 2268 return (error); 2269 } 2270 /* NOTREACHED */ 2271} 2272 2273struct filedesc_to_leader * 2274filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2275 struct filedesc *fdp, 2276 struct proc *leader) 2277{ 2278 struct filedesc_to_leader *fdtol; 2279 2280 MALLOC(fdtol, struct filedesc_to_leader *, 2281 sizeof(struct filedesc_to_leader), 2282 M_FILEDESC_TO_LEADER, 2283 M_WAITOK); 2284 fdtol->fdl_refcount = 1; 2285 fdtol->fdl_holdcount = 0; 2286 fdtol->fdl_wakeup = 0; 2287 fdtol->fdl_leader = leader; 2288 if (old != NULL) { 2289 FILEDESC_LOCK(fdp); 2290 fdtol->fdl_next = old->fdl_next; 2291 fdtol->fdl_prev = old; 2292 old->fdl_next = fdtol; 2293 fdtol->fdl_next->fdl_prev = fdtol; 2294 FILEDESC_UNLOCK(fdp); 2295 } else { 2296 fdtol->fdl_next = fdtol; 2297 fdtol->fdl_prev = fdtol; 2298 } 2299 return (fdtol); 2300} 2301 2302/* 2303 * Get file structures. 2304 */ 2305static int 2306sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2307{ 2308 struct xfile xf; 2309 struct filedesc *fdp; 2310 struct file *fp; 2311 struct proc *p; 2312 int error, n; 2313 2314 /* 2315 * Note: because the number of file descriptors is calculated 2316 * in different ways for sizing vs returning the data, 2317 * there is information leakage from the first loop. However, 2318 * it is of a similar order of magnitude to the leakage from 2319 * global system statistics such as kern.openfiles. 2320 */ 2321 sysctl_wire_old_buffer(req, 0); 2322 if (req->oldptr == NULL) { 2323 n = 16; /* A slight overestimate. */ 2324 sx_slock(&filelist_lock); 2325 LIST_FOREACH(fp, &filehead, f_list) { 2326 /* 2327 * We should grab the lock, but this is an 2328 * estimate, so does it really matter? 2329 */ 2330 /* mtx_lock(fp->f_mtxp); */ 2331 n += fp->f_count; 2332 /* mtx_unlock(f->f_mtxp); */ 2333 } 2334 sx_sunlock(&filelist_lock); 2335 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2336 } 2337 error = 0; 2338 bzero(&xf, sizeof(xf)); 2339 xf.xf_size = sizeof(xf); 2340 sx_slock(&allproc_lock); 2341 LIST_FOREACH(p, &allproc, p_list) { 2342 PROC_LOCK(p); 2343 if (p_cansee(req->td, p) != 0) { 2344 PROC_UNLOCK(p); 2345 continue; 2346 } 2347 xf.xf_pid = p->p_pid; 2348 xf.xf_uid = p->p_ucred->cr_uid; 2349 PROC_UNLOCK(p); 2350 mtx_lock(&fdesc_mtx); 2351 if ((fdp = p->p_fd) == NULL) { 2352 mtx_unlock(&fdesc_mtx); 2353 continue; 2354 } 2355 FILEDESC_LOCK(fdp); 2356 for (n = 0; n < fdp->fd_nfiles; ++n) { 2357 if ((fp = fdp->fd_ofiles[n]) == NULL) 2358 continue; 2359 xf.xf_fd = n; 2360 xf.xf_file = fp; 2361 xf.xf_data = fp->f_data; 2362 xf.xf_type = fp->f_type; 2363 xf.xf_count = fp->f_count; 2364 xf.xf_msgcount = fp->f_msgcount; 2365 xf.xf_offset = fp->f_offset; 2366 xf.xf_flag = fp->f_flag; 2367 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2368 if (error) 2369 break; 2370 } 2371 FILEDESC_UNLOCK(fdp); 2372 mtx_unlock(&fdesc_mtx); 2373 if (error) 2374 break; 2375 } 2376 sx_sunlock(&allproc_lock); 2377 return (error); 2378} 2379 2380SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2381 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2382 2383SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2384 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2385 2386SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2387 &maxfiles, 0, "Maximum number of files"); 2388 2389SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2390 &nfiles, 0, "System-wide number of open files"); 2391 2392static void 2393fildesc_drvinit(void *unused) 2394{ 2395 dev_t dev; 2396 2397 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0"); 2398 make_dev_alias(dev, "stdin"); 2399 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1"); 2400 make_dev_alias(dev, "stdout"); 2401 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2"); 2402 make_dev_alias(dev, "stderr"); 2403} 2404 2405static fo_rdwr_t badfo_readwrite; 2406static fo_ioctl_t badfo_ioctl; 2407static fo_poll_t badfo_poll; 2408static fo_kqfilter_t badfo_kqfilter; 2409static fo_stat_t badfo_stat; 2410static fo_close_t badfo_close; 2411 2412struct fileops badfileops = { 2413 .fo_read = badfo_readwrite, 2414 .fo_write = badfo_readwrite, 2415 .fo_ioctl = badfo_ioctl, 2416 .fo_poll = badfo_poll, 2417 .fo_kqfilter = badfo_kqfilter, 2418 .fo_stat = badfo_stat, 2419 .fo_close = badfo_close, 2420}; 2421 2422static int 2423badfo_readwrite(fp, uio, active_cred, flags, td) 2424 struct file *fp; 2425 struct uio *uio; 2426 struct ucred *active_cred; 2427 struct thread *td; 2428 int flags; 2429{ 2430 2431 return (EBADF); 2432} 2433 2434static int 2435badfo_ioctl(fp, com, data, active_cred, td) 2436 struct file *fp; 2437 u_long com; 2438 void *data; 2439 struct ucred *active_cred; 2440 struct thread *td; 2441{ 2442 2443 return (EBADF); 2444} 2445 2446static int 2447badfo_poll(fp, events, active_cred, td) 2448 struct file *fp; 2449 int events; 2450 struct ucred *active_cred; 2451 struct thread *td; 2452{ 2453 2454 return (0); 2455} 2456 2457static int 2458badfo_kqfilter(fp, kn) 2459 struct file *fp; 2460 struct knote *kn; 2461{ 2462 2463 return (0); 2464} 2465 2466static int 2467badfo_stat(fp, sb, active_cred, td) 2468 struct file *fp; 2469 struct stat *sb; 2470 struct ucred *active_cred; 2471 struct thread *td; 2472{ 2473 2474 return (EBADF); 2475} 2476 2477static int 2478badfo_close(fp, td) 2479 struct file *fp; 2480 struct thread *td; 2481{ 2482 2483 return (EBADF); 2484} 2485 2486SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2487 fildesc_drvinit,NULL) 2488 2489static void filelistinit(void *); 2490SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL) 2491 2492/* ARGSUSED*/ 2493static void 2494filelistinit(dummy) 2495 void *dummy; 2496{ 2497 2498 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 2499 NULL, NULL, UMA_ALIGN_PTR, 0); 2500 sx_init(&filelist_lock, "filelist lock"); 2501 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 2502} 2503