kern_descrip.c revision 139804
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 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/kern_descrip.c 139804 2005-01-06 23:35:40Z imp $"); 39 40#include "opt_compat.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44 45#include <sys/conf.h> 46#include <sys/fcntl.h> 47#include <sys/file.h> 48#include <sys/filedesc.h> 49#include <sys/filio.h> 50#include <sys/jail.h> 51#include <sys/kernel.h> 52#include <sys/limits.h> 53#include <sys/lock.h> 54#include <sys/malloc.h> 55#include <sys/mount.h> 56#include <sys/mutex.h> 57#include <sys/namei.h> 58#include <sys/proc.h> 59#include <sys/resourcevar.h> 60#include <sys/signalvar.h> 61#include <sys/socketvar.h> 62#include <sys/stat.h> 63#include <sys/sx.h> 64#include <sys/syscallsubr.h> 65#include <sys/sysctl.h> 66#include <sys/sysproto.h> 67#include <sys/unistd.h> 68#include <sys/vnode.h> 69 70#include <vm/uma.h> 71 72static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 73static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 74 "file desc to leader structures"); 75static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 76 77static uma_zone_t file_zone; 78 79 80/* How to treat 'new' parameter when allocating a fd for do_dup(). */ 81enum dup_type { DUP_VARIABLE, DUP_FIXED }; 82 83static int do_dup(struct thread *td, enum dup_type type, int old, int new, 84 register_t *retval); 85static int fd_first_free(struct filedesc *, int, int); 86static int fd_last_used(struct filedesc *, int, int); 87static void fdgrowtable(struct filedesc *, int); 88static void fdunused(struct filedesc *fdp, int fd); 89 90/* 91 * A process is initially started out with NDFILE descriptors stored within 92 * this structure, selected to be enough for typical applications based on 93 * the historical limit of 20 open files (and the usage of descriptors by 94 * shells). If these descriptors are exhausted, a larger descriptor table 95 * may be allocated, up to a process' resource limit; the internal arrays 96 * are then unused. 97 */ 98#define NDFILE 20 99#define NDSLOTSIZE sizeof(NDSLOTTYPE) 100#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 101#define NDSLOT(x) ((x) / NDENTRIES) 102#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 103#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 104 105/* 106 * Storage required per open file descriptor. 107 */ 108#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 109 110/* 111 * Basic allocation of descriptors: 112 * one of the above, plus arrays for NDFILE descriptors. 113 */ 114struct filedesc0 { 115 struct filedesc fd_fd; 116 /* 117 * These arrays are used when the number of open files is 118 * <= NDFILE, and are then pointed to by the pointers above. 119 */ 120 struct file *fd_dfiles[NDFILE]; 121 char fd_dfileflags[NDFILE]; 122 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 123}; 124 125/* 126 * Descriptor management. 127 */ 128struct filelist filehead; /* head of list of open files */ 129int openfiles; /* actual number of open files */ 130struct sx filelist_lock; /* sx to protect filelist */ 131struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 132 133/* A mutex to protect the association between a proc and filedesc. */ 134static struct mtx fdesc_mtx; 135 136/* 137 * Find the first zero bit in the given bitmap, starting at low and not 138 * exceeding size - 1. 139 */ 140static int 141fd_first_free(struct filedesc *fdp, int low, int size) 142{ 143 NDSLOTTYPE *map = fdp->fd_map; 144 NDSLOTTYPE mask; 145 int off, maxoff; 146 147 if (low >= size) 148 return (low); 149 150 off = NDSLOT(low); 151 if (low % NDENTRIES) { 152 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 153 if ((mask &= ~map[off]) != 0UL) 154 return (off * NDENTRIES + ffsl(mask) - 1); 155 ++off; 156 } 157 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 158 if (map[off] != ~0UL) 159 return (off * NDENTRIES + ffsl(~map[off]) - 1); 160 return (size); 161} 162 163/* 164 * Find the highest non-zero bit in the given bitmap, starting at low and 165 * not exceeding size - 1. 166 */ 167static int 168fd_last_used(struct filedesc *fdp, int low, int size) 169{ 170 NDSLOTTYPE *map = fdp->fd_map; 171 NDSLOTTYPE mask; 172 int off, minoff; 173 174 if (low >= size) 175 return (-1); 176 177 off = NDSLOT(size); 178 if (size % NDENTRIES) { 179 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 180 if ((mask &= map[off]) != 0) 181 return (off * NDENTRIES + flsl(mask) - 1); 182 --off; 183 } 184 for (minoff = NDSLOT(low); off >= minoff; --off) 185 if (map[off] != 0) 186 return (off * NDENTRIES + flsl(map[off]) - 1); 187 return (size - 1); 188} 189 190static int 191fdisused(struct filedesc *fdp, int fd) 192{ 193 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 194 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 195 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 196} 197 198/* 199 * Mark a file descriptor as used. 200 */ 201void 202fdused(struct filedesc *fdp, int fd) 203{ 204 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 205 KASSERT(!fdisused(fdp, fd), 206 ("fd already used")); 207 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 208 if (fd > fdp->fd_lastfile) 209 fdp->fd_lastfile = fd; 210 if (fd == fdp->fd_freefile) 211 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 212} 213 214/* 215 * Mark a file descriptor as unused. 216 */ 217static void 218fdunused(struct filedesc *fdp, int fd) 219{ 220 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 221 KASSERT(fdisused(fdp, fd), 222 ("fd is already unused")); 223 KASSERT(fdp->fd_ofiles[fd] == NULL, 224 ("fd is still in use")); 225 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 226 if (fd < fdp->fd_freefile) 227 fdp->fd_freefile = fd; 228 if (fd == fdp->fd_lastfile) 229 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 230} 231 232/* 233 * System calls on descriptors. 234 */ 235#ifndef _SYS_SYSPROTO_H_ 236struct getdtablesize_args { 237 int dummy; 238}; 239#endif 240/* 241 * MPSAFE 242 */ 243/* ARGSUSED */ 244int 245getdtablesize(struct thread *td, struct getdtablesize_args *uap) 246{ 247 struct proc *p = td->td_proc; 248 249 PROC_LOCK(p); 250 td->td_retval[0] = 251 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 252 PROC_UNLOCK(p); 253 return (0); 254} 255 256/* 257 * Duplicate a file descriptor to a particular value. 258 * 259 * note: keep in mind that a potential race condition exists when closing 260 * descriptors from a shared descriptor table (via rfork). 261 */ 262#ifndef _SYS_SYSPROTO_H_ 263struct dup2_args { 264 u_int from; 265 u_int to; 266}; 267#endif 268/* 269 * MPSAFE 270 */ 271/* ARGSUSED */ 272int 273dup2(struct thread *td, struct dup2_args *uap) 274{ 275 276 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 277 td->td_retval)); 278} 279 280/* 281 * Duplicate a file descriptor. 282 */ 283#ifndef _SYS_SYSPROTO_H_ 284struct dup_args { 285 u_int fd; 286}; 287#endif 288/* 289 * MPSAFE 290 */ 291/* ARGSUSED */ 292int 293dup(struct thread *td, struct dup_args *uap) 294{ 295 296 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval)); 297} 298 299/* 300 * The file control system call. 301 */ 302#ifndef _SYS_SYSPROTO_H_ 303struct fcntl_args { 304 int fd; 305 int cmd; 306 long arg; 307}; 308#endif 309/* 310 * MPSAFE 311 */ 312/* ARGSUSED */ 313int 314fcntl(struct thread *td, struct fcntl_args *uap) 315{ 316 struct flock fl; 317 intptr_t arg; 318 int error; 319 320 error = 0; 321 switch (uap->cmd) { 322 case F_GETLK: 323 case F_SETLK: 324 case F_SETLKW: 325 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 326 arg = (intptr_t)&fl; 327 break; 328 default: 329 arg = uap->arg; 330 break; 331 } 332 if (error) 333 return (error); 334 error = kern_fcntl(td, uap->fd, uap->cmd, arg); 335 if (error) 336 return (error); 337 if (uap->cmd == F_GETLK) 338 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 339 return (error); 340} 341 342int 343kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 344{ 345 struct filedesc *fdp; 346 struct flock *flp; 347 struct file *fp; 348 struct proc *p; 349 char *pop; 350 struct vnode *vp; 351 u_int newmin; 352 int error, flg, tmp; 353 int giant_locked; 354 355 /* 356 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to 357 * avoid grabbing Giant for calls we know don't need it. 358 */ 359 switch (cmd) { 360 case F_DUPFD: 361 case F_GETFD: 362 case F_SETFD: 363 case F_GETFL: 364 giant_locked = 0; 365 break; 366 367 default: 368 giant_locked = 1; 369 mtx_lock(&Giant); 370 } 371 372 error = 0; 373 flg = F_POSIX; 374 p = td->td_proc; 375 fdp = p->p_fd; 376 FILEDESC_LOCK(fdp); 377 if ((unsigned)fd >= fdp->fd_nfiles || 378 (fp = fdp->fd_ofiles[fd]) == NULL) { 379 FILEDESC_UNLOCK(fdp); 380 error = EBADF; 381 goto done2; 382 } 383 pop = &fdp->fd_ofileflags[fd]; 384 385 switch (cmd) { 386 case F_DUPFD: 387 /* mtx_assert(&Giant, MA_NOTOWNED); */ 388 FILEDESC_UNLOCK(fdp); 389 newmin = arg; 390 PROC_LOCK(p); 391 if (newmin >= lim_cur(p, RLIMIT_NOFILE) || 392 newmin >= maxfilesperproc) { 393 PROC_UNLOCK(p); 394 error = EINVAL; 395 break; 396 } 397 PROC_UNLOCK(p); 398 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval); 399 break; 400 401 case F_GETFD: 402 /* mtx_assert(&Giant, MA_NOTOWNED); */ 403 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 404 FILEDESC_UNLOCK(fdp); 405 break; 406 407 case F_SETFD: 408 /* mtx_assert(&Giant, MA_NOTOWNED); */ 409 *pop = (*pop &~ UF_EXCLOSE) | 410 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 411 FILEDESC_UNLOCK(fdp); 412 break; 413 414 case F_GETFL: 415 /* mtx_assert(&Giant, MA_NOTOWNED); */ 416 FILE_LOCK(fp); 417 td->td_retval[0] = OFLAGS(fp->f_flag); 418 FILE_UNLOCK(fp); 419 FILEDESC_UNLOCK(fdp); 420 break; 421 422 case F_SETFL: 423 mtx_assert(&Giant, MA_OWNED); 424 FILE_LOCK(fp); 425 fhold_locked(fp); 426 fp->f_flag &= ~FCNTLFLAGS; 427 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 428 FILE_UNLOCK(fp); 429 FILEDESC_UNLOCK(fdp); 430 tmp = fp->f_flag & FNONBLOCK; 431 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 432 if (error) { 433 fdrop(fp, td); 434 break; 435 } 436 tmp = fp->f_flag & FASYNC; 437 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 438 if (error == 0) { 439 fdrop(fp, td); 440 break; 441 } 442 FILE_LOCK(fp); 443 fp->f_flag &= ~FNONBLOCK; 444 FILE_UNLOCK(fp); 445 tmp = 0; 446 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 447 fdrop(fp, td); 448 break; 449 450 case F_GETOWN: 451 mtx_assert(&Giant, MA_OWNED); 452 fhold(fp); 453 FILEDESC_UNLOCK(fdp); 454 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 455 if (error == 0) 456 td->td_retval[0] = tmp; 457 fdrop(fp, td); 458 break; 459 460 case F_SETOWN: 461 mtx_assert(&Giant, MA_OWNED); 462 fhold(fp); 463 FILEDESC_UNLOCK(fdp); 464 tmp = arg; 465 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 466 fdrop(fp, td); 467 break; 468 469 case F_SETLKW: 470 mtx_assert(&Giant, MA_OWNED); 471 flg |= F_WAIT; 472 /* FALLTHROUGH F_SETLK */ 473 474 case F_SETLK: 475 mtx_assert(&Giant, MA_OWNED); 476 if (fp->f_type != DTYPE_VNODE) { 477 FILEDESC_UNLOCK(fdp); 478 error = EBADF; 479 break; 480 } 481 482 flp = (struct flock *)arg; 483 if (flp->l_whence == SEEK_CUR) { 484 if (fp->f_offset < 0 || 485 (flp->l_start > 0 && 486 fp->f_offset > OFF_MAX - flp->l_start)) { 487 FILEDESC_UNLOCK(fdp); 488 error = EOVERFLOW; 489 break; 490 } 491 flp->l_start += fp->f_offset; 492 } 493 494 /* 495 * VOP_ADVLOCK() may block. 496 */ 497 fhold(fp); 498 FILEDESC_UNLOCK(fdp); 499 vp = fp->f_vnode; 500 501 switch (flp->l_type) { 502 case F_RDLCK: 503 if ((fp->f_flag & FREAD) == 0) { 504 error = EBADF; 505 break; 506 } 507 PROC_LOCK(p->p_leader); 508 p->p_leader->p_flag |= P_ADVLOCK; 509 PROC_UNLOCK(p->p_leader); 510 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 511 flp, flg); 512 break; 513 case F_WRLCK: 514 if ((fp->f_flag & FWRITE) == 0) { 515 error = EBADF; 516 break; 517 } 518 PROC_LOCK(p->p_leader); 519 p->p_leader->p_flag |= P_ADVLOCK; 520 PROC_UNLOCK(p->p_leader); 521 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 522 flp, flg); 523 break; 524 case F_UNLCK: 525 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 526 flp, F_POSIX); 527 break; 528 default: 529 error = EINVAL; 530 break; 531 } 532 /* Check for race with close */ 533 FILEDESC_LOCK_FAST(fdp); 534 if ((unsigned) fd >= fdp->fd_nfiles || 535 fp != fdp->fd_ofiles[fd]) { 536 FILEDESC_UNLOCK_FAST(fdp); 537 flp->l_whence = SEEK_SET; 538 flp->l_start = 0; 539 flp->l_len = 0; 540 flp->l_type = F_UNLCK; 541 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 542 F_UNLCK, flp, F_POSIX); 543 } else 544 FILEDESC_UNLOCK_FAST(fdp); 545 fdrop(fp, td); 546 break; 547 548 case F_GETLK: 549 mtx_assert(&Giant, MA_OWNED); 550 if (fp->f_type != DTYPE_VNODE) { 551 FILEDESC_UNLOCK(fdp); 552 error = EBADF; 553 break; 554 } 555 flp = (struct flock *)arg; 556 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 557 flp->l_type != F_UNLCK) { 558 FILEDESC_UNLOCK(fdp); 559 error = EINVAL; 560 break; 561 } 562 if (flp->l_whence == SEEK_CUR) { 563 if ((flp->l_start > 0 && 564 fp->f_offset > OFF_MAX - flp->l_start) || 565 (flp->l_start < 0 && 566 fp->f_offset < OFF_MIN - flp->l_start)) { 567 FILEDESC_UNLOCK(fdp); 568 error = EOVERFLOW; 569 break; 570 } 571 flp->l_start += fp->f_offset; 572 } 573 /* 574 * VOP_ADVLOCK() may block. 575 */ 576 fhold(fp); 577 FILEDESC_UNLOCK(fdp); 578 vp = fp->f_vnode; 579 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 580 F_POSIX); 581 fdrop(fp, td); 582 break; 583 default: 584 FILEDESC_UNLOCK(fdp); 585 error = EINVAL; 586 break; 587 } 588done2: 589 if (giant_locked) 590 mtx_unlock(&Giant); 591 return (error); 592} 593 594/* 595 * Common code for dup, dup2, and fcntl(F_DUPFD). 596 */ 597static int 598do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval) 599{ 600 struct filedesc *fdp; 601 struct proc *p; 602 struct file *fp; 603 struct file *delfp; 604 int error, holdleaders, maxfd; 605 606 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED), 607 ("invalid dup type %d", type)); 608 609 p = td->td_proc; 610 fdp = p->p_fd; 611 612 /* 613 * Verify we have a valid descriptor to dup from and possibly to 614 * dup to. 615 */ 616 if (old < 0 || new < 0) 617 return (EBADF); 618 PROC_LOCK(p); 619 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 620 PROC_UNLOCK(p); 621 if (new >= maxfd) 622 return (EMFILE); 623 624 FILEDESC_LOCK(fdp); 625 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 626 FILEDESC_UNLOCK(fdp); 627 return (EBADF); 628 } 629 if (type == DUP_FIXED && old == new) { 630 *retval = new; 631 FILEDESC_UNLOCK(fdp); 632 return (0); 633 } 634 fp = fdp->fd_ofiles[old]; 635 fhold(fp); 636 637 /* 638 * If the caller specified a file descriptor, make sure the file 639 * table is large enough to hold it, and grab it. Otherwise, just 640 * allocate a new descriptor the usual way. Since the filedesc 641 * lock may be temporarily dropped in the process, we have to look 642 * out for a race. 643 */ 644 if (type == DUP_FIXED) { 645 if (new >= fdp->fd_nfiles) 646 fdgrowtable(fdp, new + 1); 647 if (fdp->fd_ofiles[new] == NULL) 648 fdused(fdp, new); 649 } else { 650 if ((error = fdalloc(td, new, &new)) != 0) { 651 FILEDESC_UNLOCK(fdp); 652 fdrop(fp, td); 653 return (error); 654 } 655 } 656 657 /* 658 * If the old file changed out from under us then treat it as a 659 * bad file descriptor. Userland should do its own locking to 660 * avoid this case. 661 */ 662 if (fdp->fd_ofiles[old] != fp) { 663 /* we've allocated a descriptor which we won't use */ 664 if (fdp->fd_ofiles[new] == NULL) 665 fdunused(fdp, new); 666 FILEDESC_UNLOCK(fdp); 667 fdrop(fp, td); 668 return (EBADF); 669 } 670 KASSERT(old != new, 671 ("new fd is same as old")); 672 673 /* 674 * Save info on the descriptor being overwritten. We cannot close 675 * it without introducing an ownership race for the slot, since we 676 * need to drop the filedesc lock to call closef(). 677 * 678 * XXX this duplicates parts of close(). 679 */ 680 delfp = fdp->fd_ofiles[new]; 681 holdleaders = 0; 682 if (delfp != NULL) { 683 if (td->td_proc->p_fdtol != NULL) { 684 /* 685 * Ask fdfree() to sleep to ensure that all relevant 686 * process leaders can be traversed in closef(). 687 */ 688 fdp->fd_holdleaderscount++; 689 holdleaders = 1; 690 } 691 } 692 693 /* 694 * Duplicate the source descriptor 695 */ 696 fdp->fd_ofiles[new] = fp; 697 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 698 if (new > fdp->fd_lastfile) 699 fdp->fd_lastfile = new; 700 *retval = new; 701 702 /* 703 * If we dup'd over a valid file, we now own the reference to it 704 * and must dispose of it using closef() semantics (as if a 705 * close() were performed on it). 706 * 707 * XXX this duplicates parts of close(). 708 */ 709 if (delfp != NULL) { 710 knote_fdclose(td, new); 711 FILEDESC_UNLOCK(fdp); 712 (void) closef(delfp, td); 713 if (holdleaders) { 714 FILEDESC_LOCK_FAST(fdp); 715 fdp->fd_holdleaderscount--; 716 if (fdp->fd_holdleaderscount == 0 && 717 fdp->fd_holdleaderswakeup != 0) { 718 fdp->fd_holdleaderswakeup = 0; 719 wakeup(&fdp->fd_holdleaderscount); 720 } 721 FILEDESC_UNLOCK_FAST(fdp); 722 } 723 } else { 724 FILEDESC_UNLOCK(fdp); 725 } 726 return (0); 727} 728 729/* 730 * If sigio is on the list associated with a process or process group, 731 * disable signalling from the device, remove sigio from the list and 732 * free sigio. 733 */ 734void 735funsetown(struct sigio **sigiop) 736{ 737 struct sigio *sigio; 738 739 SIGIO_LOCK(); 740 sigio = *sigiop; 741 if (sigio == NULL) { 742 SIGIO_UNLOCK(); 743 return; 744 } 745 *(sigio->sio_myref) = NULL; 746 if ((sigio)->sio_pgid < 0) { 747 struct pgrp *pg = (sigio)->sio_pgrp; 748 PGRP_LOCK(pg); 749 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 750 sigio, sio_pgsigio); 751 PGRP_UNLOCK(pg); 752 } else { 753 struct proc *p = (sigio)->sio_proc; 754 PROC_LOCK(p); 755 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 756 sigio, sio_pgsigio); 757 PROC_UNLOCK(p); 758 } 759 SIGIO_UNLOCK(); 760 crfree(sigio->sio_ucred); 761 FREE(sigio, M_SIGIO); 762} 763 764/* 765 * Free a list of sigio structures. 766 * We only need to lock the SIGIO_LOCK because we have made ourselves 767 * inaccessable to callers of fsetown and therefore do not need to lock 768 * the proc or pgrp struct for the list manipulation. 769 */ 770void 771funsetownlst(struct sigiolst *sigiolst) 772{ 773 struct proc *p; 774 struct pgrp *pg; 775 struct sigio *sigio; 776 777 sigio = SLIST_FIRST(sigiolst); 778 if (sigio == NULL) 779 return; 780 p = NULL; 781 pg = NULL; 782 783 /* 784 * Every entry of the list should belong 785 * to a single proc or pgrp. 786 */ 787 if (sigio->sio_pgid < 0) { 788 pg = sigio->sio_pgrp; 789 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 790 } else /* if (sigio->sio_pgid > 0) */ { 791 p = sigio->sio_proc; 792 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 793 } 794 795 SIGIO_LOCK(); 796 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 797 *(sigio->sio_myref) = NULL; 798 if (pg != NULL) { 799 KASSERT(sigio->sio_pgid < 0, 800 ("Proc sigio in pgrp sigio list")); 801 KASSERT(sigio->sio_pgrp == pg, 802 ("Bogus pgrp in sigio list")); 803 PGRP_LOCK(pg); 804 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 805 sio_pgsigio); 806 PGRP_UNLOCK(pg); 807 } else /* if (p != NULL) */ { 808 KASSERT(sigio->sio_pgid > 0, 809 ("Pgrp sigio in proc sigio list")); 810 KASSERT(sigio->sio_proc == p, 811 ("Bogus proc in sigio list")); 812 PROC_LOCK(p); 813 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 814 sio_pgsigio); 815 PROC_UNLOCK(p); 816 } 817 SIGIO_UNLOCK(); 818 crfree(sigio->sio_ucred); 819 FREE(sigio, M_SIGIO); 820 SIGIO_LOCK(); 821 } 822 SIGIO_UNLOCK(); 823} 824 825/* 826 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 827 * 828 * After permission checking, add a sigio structure to the sigio list for 829 * the process or process group. 830 */ 831int 832fsetown(pid_t pgid, struct sigio **sigiop) 833{ 834 struct proc *proc; 835 struct pgrp *pgrp; 836 struct sigio *sigio; 837 int ret; 838 839 if (pgid == 0) { 840 funsetown(sigiop); 841 return (0); 842 } 843 844 ret = 0; 845 846 /* Allocate and fill in the new sigio out of locks. */ 847 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK); 848 sigio->sio_pgid = pgid; 849 sigio->sio_ucred = crhold(curthread->td_ucred); 850 sigio->sio_myref = sigiop; 851 852 sx_slock(&proctree_lock); 853 if (pgid > 0) { 854 proc = pfind(pgid); 855 if (proc == NULL) { 856 ret = ESRCH; 857 goto fail; 858 } 859 860 /* 861 * Policy - Don't allow a process to FSETOWN a process 862 * in another session. 863 * 864 * Remove this test to allow maximum flexibility or 865 * restrict FSETOWN to the current process or process 866 * group for maximum safety. 867 */ 868 PROC_UNLOCK(proc); 869 if (proc->p_session != curthread->td_proc->p_session) { 870 ret = EPERM; 871 goto fail; 872 } 873 874 pgrp = NULL; 875 } else /* if (pgid < 0) */ { 876 pgrp = pgfind(-pgid); 877 if (pgrp == NULL) { 878 ret = ESRCH; 879 goto fail; 880 } 881 PGRP_UNLOCK(pgrp); 882 883 /* 884 * Policy - Don't allow a process to FSETOWN a process 885 * in another session. 886 * 887 * Remove this test to allow maximum flexibility or 888 * restrict FSETOWN to the current process or process 889 * group for maximum safety. 890 */ 891 if (pgrp->pg_session != curthread->td_proc->p_session) { 892 ret = EPERM; 893 goto fail; 894 } 895 896 proc = NULL; 897 } 898 funsetown(sigiop); 899 if (pgid > 0) { 900 PROC_LOCK(proc); 901 /* 902 * Since funsetownlst() is called without the proctree 903 * locked, we need to check for P_WEXIT. 904 * XXX: is ESRCH correct? 905 */ 906 if ((proc->p_flag & P_WEXIT) != 0) { 907 PROC_UNLOCK(proc); 908 ret = ESRCH; 909 goto fail; 910 } 911 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 912 sigio->sio_proc = proc; 913 PROC_UNLOCK(proc); 914 } else { 915 PGRP_LOCK(pgrp); 916 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 917 sigio->sio_pgrp = pgrp; 918 PGRP_UNLOCK(pgrp); 919 } 920 sx_sunlock(&proctree_lock); 921 SIGIO_LOCK(); 922 *sigiop = sigio; 923 SIGIO_UNLOCK(); 924 return (0); 925 926fail: 927 sx_sunlock(&proctree_lock); 928 crfree(sigio->sio_ucred); 929 FREE(sigio, M_SIGIO); 930 return (ret); 931} 932 933/* 934 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 935 */ 936pid_t 937fgetown(sigiop) 938 struct sigio **sigiop; 939{ 940 pid_t pgid; 941 942 SIGIO_LOCK(); 943 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 944 SIGIO_UNLOCK(); 945 return (pgid); 946} 947 948/* 949 * Close a file descriptor. 950 */ 951#ifndef _SYS_SYSPROTO_H_ 952struct close_args { 953 int fd; 954}; 955#endif 956/* 957 * MPSAFE 958 */ 959/* ARGSUSED */ 960int 961close(td, uap) 962 struct thread *td; 963 struct close_args *uap; 964{ 965 struct filedesc *fdp; 966 struct file *fp; 967 int fd, error; 968 int holdleaders; 969 970 fd = uap->fd; 971 error = 0; 972 holdleaders = 0; 973 fdp = td->td_proc->p_fd; 974 FILEDESC_LOCK(fdp); 975 if ((unsigned)fd >= fdp->fd_nfiles || 976 (fp = fdp->fd_ofiles[fd]) == NULL) { 977 FILEDESC_UNLOCK(fdp); 978 return (EBADF); 979 } 980 fdp->fd_ofiles[fd] = NULL; 981 fdp->fd_ofileflags[fd] = 0; 982 fdunused(fdp, fd); 983 if (td->td_proc->p_fdtol != NULL) { 984 /* 985 * Ask fdfree() to sleep to ensure that all relevant 986 * process leaders can be traversed in closef(). 987 */ 988 fdp->fd_holdleaderscount++; 989 holdleaders = 1; 990 } 991 992 /* 993 * we now hold the fp reference that used to be owned by the descriptor 994 * array. 995 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a 996 * race of the fd getting opened, a knote added, and deleteing a knote 997 * for the new fd. 998 */ 999 knote_fdclose(td, fd); 1000 FILEDESC_UNLOCK(fdp); 1001 1002 error = closef(fp, td); 1003 if (holdleaders) { 1004 FILEDESC_LOCK_FAST(fdp); 1005 fdp->fd_holdleaderscount--; 1006 if (fdp->fd_holdleaderscount == 0 && 1007 fdp->fd_holdleaderswakeup != 0) { 1008 fdp->fd_holdleaderswakeup = 0; 1009 wakeup(&fdp->fd_holdleaderscount); 1010 } 1011 FILEDESC_UNLOCK_FAST(fdp); 1012 } 1013 return (error); 1014} 1015 1016#if defined(COMPAT_43) 1017/* 1018 * Return status information about a file descriptor. 1019 */ 1020#ifndef _SYS_SYSPROTO_H_ 1021struct ofstat_args { 1022 int fd; 1023 struct ostat *sb; 1024}; 1025#endif 1026/* 1027 * MPSAFE 1028 */ 1029/* ARGSUSED */ 1030int 1031ofstat(struct thread *td, struct ofstat_args *uap) 1032{ 1033 struct file *fp; 1034 struct stat ub; 1035 struct ostat oub; 1036 int error; 1037 1038 if ((error = fget(td, uap->fd, &fp)) != 0) 1039 goto done2; 1040 error = fo_stat(fp, &ub, td->td_ucred, td); 1041 if (error == 0) { 1042 cvtstat(&ub, &oub); 1043 error = copyout(&oub, uap->sb, sizeof(oub)); 1044 } 1045 fdrop(fp, td); 1046done2: 1047 return (error); 1048} 1049#endif /* COMPAT_43 */ 1050 1051/* 1052 * Return status information about a file descriptor. 1053 */ 1054#ifndef _SYS_SYSPROTO_H_ 1055struct fstat_args { 1056 int fd; 1057 struct stat *sb; 1058}; 1059#endif 1060/* 1061 * MPSAFE 1062 */ 1063/* ARGSUSED */ 1064int 1065fstat(struct thread *td, struct fstat_args *uap) 1066{ 1067 struct file *fp; 1068 struct stat ub; 1069 int error; 1070 1071 if ((error = fget(td, uap->fd, &fp)) != 0) 1072 goto done2; 1073 error = fo_stat(fp, &ub, td->td_ucred, td); 1074 if (error == 0) 1075 error = copyout(&ub, uap->sb, sizeof(ub)); 1076 fdrop(fp, td); 1077done2: 1078 return (error); 1079} 1080 1081/* 1082 * Return status information about a file descriptor. 1083 */ 1084#ifndef _SYS_SYSPROTO_H_ 1085struct nfstat_args { 1086 int fd; 1087 struct nstat *sb; 1088}; 1089#endif 1090/* 1091 * MPSAFE 1092 */ 1093/* ARGSUSED */ 1094int 1095nfstat(struct thread *td, struct nfstat_args *uap) 1096{ 1097 struct file *fp; 1098 struct stat ub; 1099 struct nstat nub; 1100 int error; 1101 1102 if ((error = fget(td, uap->fd, &fp)) != 0) 1103 goto done2; 1104 error = fo_stat(fp, &ub, td->td_ucred, td); 1105 if (error == 0) { 1106 cvtnstat(&ub, &nub); 1107 error = copyout(&nub, uap->sb, sizeof(nub)); 1108 } 1109 fdrop(fp, td); 1110done2: 1111 return (error); 1112} 1113 1114/* 1115 * Return pathconf information about a file descriptor. 1116 */ 1117#ifndef _SYS_SYSPROTO_H_ 1118struct fpathconf_args { 1119 int fd; 1120 int name; 1121}; 1122#endif 1123/* 1124 * MPSAFE 1125 */ 1126/* ARGSUSED */ 1127int 1128fpathconf(struct thread *td, struct fpathconf_args *uap) 1129{ 1130 struct file *fp; 1131 struct vnode *vp; 1132 int error; 1133 1134 if ((error = fget(td, uap->fd, &fp)) != 0) 1135 return (error); 1136 1137 /* If asynchronous I/O is available, it works for all descriptors. */ 1138 if (uap->name == _PC_ASYNC_IO) { 1139 td->td_retval[0] = async_io_version; 1140 goto out; 1141 } 1142 vp = fp->f_vnode; 1143 if (vp != NULL) { 1144 mtx_lock(&Giant); 1145 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1146 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1147 VOP_UNLOCK(vp, 0, td); 1148 mtx_unlock(&Giant); 1149 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1150 if (uap->name != _PC_PIPE_BUF) { 1151 error = EINVAL; 1152 } else { 1153 td->td_retval[0] = PIPE_BUF; 1154 error = 0; 1155 } 1156 } else { 1157 error = EOPNOTSUPP; 1158 } 1159out: 1160 fdrop(fp, td); 1161 return (error); 1162} 1163 1164/* 1165 * Grow the file table to accomodate (at least) nfd descriptors. This may 1166 * block and drop the filedesc lock, but it will reacquire it before 1167 * returing. 1168 */ 1169static void 1170fdgrowtable(struct filedesc *fdp, int nfd) 1171{ 1172 struct file **ntable; 1173 char *nfileflags; 1174 int nnfiles, onfiles; 1175 NDSLOTTYPE *nmap; 1176 1177 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1178 1179 KASSERT(fdp->fd_nfiles > 0, 1180 ("zero-length file table")); 1181 1182 /* compute the size of the new table */ 1183 onfiles = fdp->fd_nfiles; 1184 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1185 if (nnfiles <= onfiles) 1186 /* the table is already large enough */ 1187 return; 1188 1189 /* allocate a new table and (if required) new bitmaps */ 1190 FILEDESC_UNLOCK(fdp); 1191 MALLOC(ntable, struct file **, nnfiles * OFILESIZE, 1192 M_FILEDESC, M_ZERO | M_WAITOK); 1193 nfileflags = (char *)&ntable[nnfiles]; 1194 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1195 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE, 1196 M_FILEDESC, M_ZERO | M_WAITOK); 1197 else 1198 nmap = NULL; 1199 FILEDESC_LOCK(fdp); 1200 1201 /* 1202 * We now have new tables ready to go. Since we dropped the 1203 * filedesc lock to call malloc(), watch out for a race. 1204 */ 1205 onfiles = fdp->fd_nfiles; 1206 if (onfiles >= nnfiles) { 1207 /* we lost the race, but that's OK */ 1208 free(ntable, M_FILEDESC); 1209 if (nmap != NULL) 1210 free(nmap, M_FILEDESC); 1211 return; 1212 } 1213 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1214 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1215 if (onfiles > NDFILE) 1216 free(fdp->fd_ofiles, M_FILEDESC); 1217 fdp->fd_ofiles = ntable; 1218 fdp->fd_ofileflags = nfileflags; 1219 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1220 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1221 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1222 free(fdp->fd_map, M_FILEDESC); 1223 fdp->fd_map = nmap; 1224 } 1225 fdp->fd_nfiles = nnfiles; 1226} 1227 1228/* 1229 * Allocate a file descriptor for the process. 1230 */ 1231int 1232fdalloc(struct thread *td, int minfd, int *result) 1233{ 1234 struct proc *p = td->td_proc; 1235 struct filedesc *fdp = p->p_fd; 1236 int fd = -1, maxfd; 1237 1238 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1239 1240 PROC_LOCK(p); 1241 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1242 PROC_UNLOCK(p); 1243 1244 /* 1245 * Search the bitmap for a free descriptor. If none is found, try 1246 * to grow the file table. Keep at it until we either get a file 1247 * descriptor or run into process or system limits; fdgrowtable() 1248 * may drop the filedesc lock, so we're in a race. 1249 */ 1250 for (;;) { 1251 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1252 if (fd >= maxfd) 1253 return (EMFILE); 1254 if (fd < fdp->fd_nfiles) 1255 break; 1256 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1257 } 1258 1259 /* 1260 * Perform some sanity checks, then mark the file descriptor as 1261 * used and return it to the caller. 1262 */ 1263 KASSERT(!fdisused(fdp, fd), 1264 ("fd_first_free() returned non-free descriptor")); 1265 KASSERT(fdp->fd_ofiles[fd] == NULL, 1266 ("free descriptor isn't")); 1267 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */ 1268 fdused(fdp, fd); 1269 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 1270 *result = fd; 1271 return (0); 1272} 1273 1274/* 1275 * Check to see whether n user file descriptors 1276 * are available to the process p. 1277 */ 1278int 1279fdavail(struct thread *td, int n) 1280{ 1281 struct proc *p = td->td_proc; 1282 struct filedesc *fdp = td->td_proc->p_fd; 1283 struct file **fpp; 1284 int i, lim, last; 1285 1286 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1287 1288 PROC_LOCK(p); 1289 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1290 PROC_UNLOCK(p); 1291 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1292 return (1); 1293 last = min(fdp->fd_nfiles, lim); 1294 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 1295 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 1296 if (*fpp == NULL && --n <= 0) 1297 return (1); 1298 } 1299 return (0); 1300} 1301 1302/* 1303 * Create a new open file structure and allocate 1304 * a file decriptor for the process that refers to it. 1305 * We add one reference to the file for the descriptor table 1306 * and one reference for resultfp. This is to prevent us being 1307 * prempted and the entry in the descriptor table closed after 1308 * we release the FILEDESC lock. 1309 */ 1310int 1311falloc(struct thread *td, struct file **resultfp, int *resultfd) 1312{ 1313 struct proc *p = td->td_proc; 1314 struct file *fp, *fq; 1315 int error, i; 1316 int maxuserfiles = maxfiles - (maxfiles / 20); 1317 static struct timeval lastfail; 1318 static int curfail; 1319 1320 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1321 sx_xlock(&filelist_lock); 1322 if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 || 1323 jailed(td->td_ucred))) || openfiles >= maxfiles) { 1324 if (ppsratecheck(&lastfail, &curfail, 1)) { 1325 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n", 1326 td->td_ucred->cr_ruid); 1327 } 1328 sx_xunlock(&filelist_lock); 1329 uma_zfree(file_zone, fp); 1330 return (ENFILE); 1331 } 1332 openfiles++; 1333 1334 /* 1335 * If the process has file descriptor zero open, add the new file 1336 * descriptor to the list of open files at that point, otherwise 1337 * put it at the front of the list of open files. 1338 */ 1339 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep); 1340 fp->f_count = 1; 1341 if (resultfp) 1342 fp->f_count++; 1343 fp->f_cred = crhold(td->td_ucred); 1344 fp->f_ops = &badfileops; 1345 fp->f_data = NULL; 1346 fp->f_vnode = NULL; 1347 FILEDESC_LOCK(p->p_fd); 1348 if ((fq = p->p_fd->fd_ofiles[0])) { 1349 LIST_INSERT_AFTER(fq, fp, f_list); 1350 } else { 1351 LIST_INSERT_HEAD(&filehead, fp, f_list); 1352 } 1353 sx_xunlock(&filelist_lock); 1354 if ((error = fdalloc(td, 0, &i))) { 1355 FILEDESC_UNLOCK(p->p_fd); 1356 fdrop(fp, td); 1357 if (resultfp) 1358 fdrop(fp, td); 1359 return (error); 1360 } 1361 p->p_fd->fd_ofiles[i] = fp; 1362 FILEDESC_UNLOCK(p->p_fd); 1363 if (resultfp) 1364 *resultfp = fp; 1365 if (resultfd) 1366 *resultfd = i; 1367 return (0); 1368} 1369 1370/* 1371 * Build a new filedesc structure from another. 1372 * Copy the current, root, and jail root vnode references. 1373 */ 1374struct filedesc * 1375fdinit(struct filedesc *fdp) 1376{ 1377 struct filedesc0 *newfdp; 1378 1379 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1380 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF); 1381 if (fdp != NULL) { 1382 FILEDESC_LOCK(fdp); 1383 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1384 if (newfdp->fd_fd.fd_cdir) 1385 VREF(newfdp->fd_fd.fd_cdir); 1386 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1387 if (newfdp->fd_fd.fd_rdir) 1388 VREF(newfdp->fd_fd.fd_rdir); 1389 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1390 if (newfdp->fd_fd.fd_jdir) 1391 VREF(newfdp->fd_fd.fd_jdir); 1392 FILEDESC_UNLOCK(fdp); 1393 } 1394 1395 /* Create the file descriptor table. */ 1396 newfdp->fd_fd.fd_refcnt = 1; 1397 newfdp->fd_fd.fd_holdcnt = 1; 1398 newfdp->fd_fd.fd_cmask = CMASK; 1399 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1400 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1401 newfdp->fd_fd.fd_nfiles = NDFILE; 1402 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1403 return (&newfdp->fd_fd); 1404} 1405 1406static struct filedesc * 1407fdhold(struct proc *p) 1408{ 1409 struct filedesc *fdp; 1410 1411 mtx_lock(&fdesc_mtx); 1412 fdp = p->p_fd; 1413 if (fdp != NULL) 1414 fdp->fd_holdcnt++; 1415 mtx_unlock(&fdesc_mtx); 1416 return (fdp); 1417} 1418 1419static void 1420fddrop(struct filedesc *fdp) 1421{ 1422 int i; 1423 1424 mtx_lock(&fdesc_mtx); 1425 i = --fdp->fd_holdcnt; 1426 mtx_unlock(&fdesc_mtx); 1427 if (i > 0) 1428 return; 1429 1430 mtx_destroy(&fdp->fd_mtx); 1431 FREE(fdp, M_FILEDESC); 1432} 1433 1434/* 1435 * Share a filedesc structure. 1436 */ 1437struct filedesc * 1438fdshare(struct filedesc *fdp) 1439{ 1440 FILEDESC_LOCK_FAST(fdp); 1441 fdp->fd_refcnt++; 1442 FILEDESC_UNLOCK_FAST(fdp); 1443 return (fdp); 1444} 1445 1446/* 1447 * Unshare a filedesc structure, if necessary by making a copy 1448 */ 1449void 1450fdunshare(struct proc *p, struct thread *td) 1451{ 1452 1453 FILEDESC_LOCK_FAST(p->p_fd); 1454 if (p->p_fd->fd_refcnt > 1) { 1455 struct filedesc *tmp; 1456 1457 FILEDESC_UNLOCK_FAST(p->p_fd); 1458 tmp = fdcopy(p->p_fd); 1459 fdfree(td); 1460 p->p_fd = tmp; 1461 } else 1462 FILEDESC_UNLOCK_FAST(p->p_fd); 1463} 1464 1465/* 1466 * Copy a filedesc structure. 1467 * A NULL pointer in returns a NULL reference, this is to ease callers, 1468 * not catch errors. 1469 */ 1470struct filedesc * 1471fdcopy(struct filedesc *fdp) 1472{ 1473 struct filedesc *newfdp; 1474 int i; 1475 1476 /* Certain daemons might not have file descriptors. */ 1477 if (fdp == NULL) 1478 return (NULL); 1479 1480 newfdp = fdinit(fdp); 1481 FILEDESC_LOCK_FAST(fdp); 1482 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1483 FILEDESC_UNLOCK_FAST(fdp); 1484 FILEDESC_LOCK(newfdp); 1485 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1486 FILEDESC_UNLOCK(newfdp); 1487 FILEDESC_LOCK_FAST(fdp); 1488 } 1489 /* copy everything except kqueue descriptors */ 1490 newfdp->fd_freefile = -1; 1491 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1492 if (fdisused(fdp, i) && 1493 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) { 1494 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1495 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1496 fhold(newfdp->fd_ofiles[i]); 1497 newfdp->fd_lastfile = i; 1498 } else { 1499 if (newfdp->fd_freefile == -1) 1500 newfdp->fd_freefile = i; 1501 } 1502 } 1503 FILEDESC_UNLOCK_FAST(fdp); 1504 FILEDESC_LOCK(newfdp); 1505 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1506 if (newfdp->fd_ofiles[i] != NULL) 1507 fdused(newfdp, i); 1508 FILEDESC_UNLOCK(newfdp); 1509 FILEDESC_LOCK_FAST(fdp); 1510 if (newfdp->fd_freefile == -1) 1511 newfdp->fd_freefile = i; 1512 newfdp->fd_cmask = fdp->fd_cmask; 1513 FILEDESC_UNLOCK_FAST(fdp); 1514 return (newfdp); 1515} 1516 1517/* 1518 * Release a filedesc structure. 1519 */ 1520void 1521fdfree(struct thread *td) 1522{ 1523 struct filedesc *fdp; 1524 struct file **fpp; 1525 int i; 1526 struct filedesc_to_leader *fdtol; 1527 struct file *fp; 1528 struct vnode *vp; 1529 struct flock lf; 1530 1531 GIANT_REQUIRED; /* VFS */ 1532 1533 /* Certain daemons might not have file descriptors. */ 1534 fdp = td->td_proc->p_fd; 1535 if (fdp == NULL) 1536 return; 1537 1538 /* Check for special need to clear POSIX style locks */ 1539 fdtol = td->td_proc->p_fdtol; 1540 if (fdtol != NULL) { 1541 FILEDESC_LOCK(fdp); 1542 KASSERT(fdtol->fdl_refcount > 0, 1543 ("filedesc_to_refcount botch: fdl_refcount=%d", 1544 fdtol->fdl_refcount)); 1545 if (fdtol->fdl_refcount == 1 && 1546 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1547 i = 0; 1548 fpp = fdp->fd_ofiles; 1549 for (i = 0, fpp = fdp->fd_ofiles; 1550 i <= fdp->fd_lastfile; 1551 i++, fpp++) { 1552 if (*fpp == NULL || 1553 (*fpp)->f_type != DTYPE_VNODE) 1554 continue; 1555 fp = *fpp; 1556 fhold(fp); 1557 FILEDESC_UNLOCK(fdp); 1558 lf.l_whence = SEEK_SET; 1559 lf.l_start = 0; 1560 lf.l_len = 0; 1561 lf.l_type = F_UNLCK; 1562 vp = fp->f_vnode; 1563 (void) VOP_ADVLOCK(vp, 1564 (caddr_t)td->td_proc-> 1565 p_leader, 1566 F_UNLCK, 1567 &lf, 1568 F_POSIX); 1569 FILEDESC_LOCK(fdp); 1570 fdrop(fp, td); 1571 fpp = fdp->fd_ofiles + i; 1572 } 1573 } 1574 retry: 1575 if (fdtol->fdl_refcount == 1) { 1576 if (fdp->fd_holdleaderscount > 0 && 1577 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1578 /* 1579 * close() or do_dup() has cleared a reference 1580 * in a shared file descriptor table. 1581 */ 1582 fdp->fd_holdleaderswakeup = 1; 1583 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx, 1584 PLOCK, "fdlhold", 0); 1585 goto retry; 1586 } 1587 if (fdtol->fdl_holdcount > 0) { 1588 /* 1589 * Ensure that fdtol->fdl_leader 1590 * remains valid in closef(). 1591 */ 1592 fdtol->fdl_wakeup = 1; 1593 msleep(fdtol, &fdp->fd_mtx, 1594 PLOCK, "fdlhold", 0); 1595 goto retry; 1596 } 1597 } 1598 fdtol->fdl_refcount--; 1599 if (fdtol->fdl_refcount == 0 && 1600 fdtol->fdl_holdcount == 0) { 1601 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1602 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1603 } else 1604 fdtol = NULL; 1605 td->td_proc->p_fdtol = NULL; 1606 FILEDESC_UNLOCK(fdp); 1607 if (fdtol != NULL) 1608 FREE(fdtol, M_FILEDESC_TO_LEADER); 1609 } 1610 FILEDESC_LOCK_FAST(fdp); 1611 i = --fdp->fd_refcnt; 1612 FILEDESC_UNLOCK_FAST(fdp); 1613 if (i > 0) 1614 return; 1615 /* 1616 * We are the last reference to the structure, so we can 1617 * safely assume it will not change out from under us. 1618 */ 1619 fpp = fdp->fd_ofiles; 1620 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1621 if (*fpp) 1622 (void) closef(*fpp, td); 1623 } 1624 FILEDESC_LOCK(fdp); 1625 1626 /* XXX This should happen earlier. */ 1627 mtx_lock(&fdesc_mtx); 1628 td->td_proc->p_fd = NULL; 1629 mtx_unlock(&fdesc_mtx); 1630 1631 if (fdp->fd_nfiles > NDFILE) 1632 FREE(fdp->fd_ofiles, M_FILEDESC); 1633 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1634 FREE(fdp->fd_map, M_FILEDESC); 1635 1636 fdp->fd_nfiles = 0; 1637 1638 if (fdp->fd_cdir) 1639 vrele(fdp->fd_cdir); 1640 fdp->fd_cdir = NULL; 1641 if (fdp->fd_rdir) 1642 vrele(fdp->fd_rdir); 1643 fdp->fd_rdir = NULL; 1644 if (fdp->fd_jdir) 1645 vrele(fdp->fd_jdir); 1646 fdp->fd_jdir = NULL; 1647 1648 FILEDESC_UNLOCK(fdp); 1649 1650 fddrop(fdp); 1651} 1652 1653/* 1654 * For setugid programs, we don't want to people to use that setugidness 1655 * to generate error messages which write to a file which otherwise would 1656 * otherwise be off-limits to the process. We check for filesystems where 1657 * the vnode can change out from under us after execve (like [lin]procfs). 1658 * 1659 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1660 * sufficient. We also don't for check setugidness since we know we are. 1661 */ 1662static int 1663is_unsafe(struct file *fp) 1664{ 1665 if (fp->f_type == DTYPE_VNODE) { 1666 struct vnode *vp = fp->f_vnode; 1667 1668 if ((vp->v_vflag & VV_PROCDEP) != 0) 1669 return (1); 1670 } 1671 return (0); 1672} 1673 1674/* 1675 * Make this setguid thing safe, if at all possible. 1676 */ 1677void 1678setugidsafety(struct thread *td) 1679{ 1680 struct filedesc *fdp; 1681 int i; 1682 1683 /* Certain daemons might not have file descriptors. */ 1684 fdp = td->td_proc->p_fd; 1685 if (fdp == NULL) 1686 return; 1687 1688 /* 1689 * Note: fdp->fd_ofiles may be reallocated out from under us while 1690 * we are blocked in a close. Be careful! 1691 */ 1692 FILEDESC_LOCK(fdp); 1693 for (i = 0; i <= fdp->fd_lastfile; i++) { 1694 if (i > 2) 1695 break; 1696 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1697 struct file *fp; 1698 1699 knote_fdclose(td, i); 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 1716void 1717fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 1718{ 1719 1720 FILEDESC_LOCK(fdp); 1721 if (fdp->fd_ofiles[idx] == fp) { 1722 fdp->fd_ofiles[idx] = NULL; 1723 fdunused(fdp, idx); 1724 FILEDESC_UNLOCK(fdp); 1725 fdrop(fp, td); 1726 } else { 1727 FILEDESC_UNLOCK(fdp); 1728 } 1729} 1730 1731/* 1732 * Close any files on exec? 1733 */ 1734void 1735fdcloseexec(struct thread *td) 1736{ 1737 struct filedesc *fdp; 1738 int i; 1739 1740 /* Certain daemons might not have file descriptors. */ 1741 fdp = td->td_proc->p_fd; 1742 if (fdp == NULL) 1743 return; 1744 1745 FILEDESC_LOCK(fdp); 1746 1747 /* 1748 * We cannot cache fd_ofiles or fd_ofileflags since operations 1749 * may block and rip them out from under us. 1750 */ 1751 for (i = 0; i <= fdp->fd_lastfile; i++) { 1752 if (fdp->fd_ofiles[i] != NULL && 1753 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) { 1754 struct file *fp; 1755 1756 knote_fdclose(td, i); 1757 /* 1758 * NULL-out descriptor prior to close to avoid 1759 * a race while close blocks. 1760 */ 1761 fp = fdp->fd_ofiles[i]; 1762 fdp->fd_ofiles[i] = NULL; 1763 fdp->fd_ofileflags[i] = 0; 1764 fdunused(fdp, i); 1765 FILEDESC_UNLOCK(fdp); 1766 (void) closef(fp, td); 1767 FILEDESC_LOCK(fdp); 1768 } 1769 } 1770 FILEDESC_UNLOCK(fdp); 1771} 1772 1773/* 1774 * It is unsafe for set[ug]id processes to be started with file 1775 * descriptors 0..2 closed, as these descriptors are given implicit 1776 * significance in the Standard C library. fdcheckstd() will create a 1777 * descriptor referencing /dev/null for each of stdin, stdout, and 1778 * stderr that is not already open. 1779 */ 1780int 1781fdcheckstd(struct thread *td) 1782{ 1783 struct nameidata nd; 1784 struct filedesc *fdp; 1785 struct file *fp; 1786 register_t retval; 1787 int fd, i, error, flags, devnull; 1788 1789 GIANT_REQUIRED; /* VFS */ 1790 1791 fdp = td->td_proc->p_fd; 1792 if (fdp == NULL) 1793 return (0); 1794 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 1795 devnull = -1; 1796 error = 0; 1797 for (i = 0; i < 3; i++) { 1798 if (fdp->fd_ofiles[i] != NULL) 1799 continue; 1800 if (devnull < 0) { 1801 error = falloc(td, &fp, &fd); 1802 if (error != 0) 1803 break; 1804 /* Note extra ref on `fp' held for us by falloc(). */ 1805 KASSERT(fd == i, ("oof, we didn't get our fd")); 1806 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null", 1807 td); 1808 flags = FREAD | FWRITE; 1809 error = vn_open(&nd, &flags, 0, -1); 1810 if (error != 0) { 1811 /* 1812 * Someone may have closed the entry in the 1813 * file descriptor table, so check it hasn't 1814 * changed before dropping the reference count. 1815 */ 1816 FILEDESC_LOCK(fdp); 1817 KASSERT(fdp->fd_ofiles[fd] == fp, 1818 ("table not shared, how did it change?")); 1819 fdp->fd_ofiles[fd] = NULL; 1820 fdunused(fdp, fd); 1821 FILEDESC_UNLOCK(fdp); 1822 fdrop(fp, td); 1823 fdrop(fp, td); 1824 break; 1825 } 1826 NDFREE(&nd, NDF_ONLY_PNBUF); 1827 fp->f_flag = flags; 1828 fp->f_vnode = nd.ni_vp; 1829 if (fp->f_data == NULL) 1830 fp->f_data = nd.ni_vp; 1831 if (fp->f_ops == &badfileops) 1832 fp->f_ops = &vnops; 1833 fp->f_type = DTYPE_VNODE; 1834 VOP_UNLOCK(nd.ni_vp, 0, td); 1835 devnull = fd; 1836 fdrop(fp, td); 1837 } else { 1838 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 1839 if (error != 0) 1840 break; 1841 } 1842 } 1843 return (error); 1844} 1845 1846/* 1847 * Internal form of close. 1848 * Decrement reference count on file structure. 1849 * Note: td may be NULL when closing a file that was being passed in a 1850 * message. 1851 * 1852 * XXXRW: Giant is not required for the caller, but often will be held; this 1853 * makes it moderately likely the Giant will be recursed in the VFS case. 1854 */ 1855int 1856closef(struct file *fp, struct thread *td) 1857{ 1858 struct vnode *vp; 1859 struct flock lf; 1860 struct filedesc_to_leader *fdtol; 1861 struct filedesc *fdp; 1862 1863 /* 1864 * POSIX record locking dictates that any close releases ALL 1865 * locks owned by this process. This is handled by setting 1866 * a flag in the unlock to free ONLY locks obeying POSIX 1867 * semantics, and not to free BSD-style file locks. 1868 * If the descriptor was in a message, POSIX-style locks 1869 * aren't passed with the descriptor. 1870 */ 1871 if (fp->f_type == DTYPE_VNODE) { 1872 mtx_lock(&Giant); 1873 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1874 lf.l_whence = SEEK_SET; 1875 lf.l_start = 0; 1876 lf.l_len = 0; 1877 lf.l_type = F_UNLCK; 1878 vp = fp->f_vnode; 1879 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 1880 F_UNLCK, &lf, F_POSIX); 1881 } 1882 fdtol = td->td_proc->p_fdtol; 1883 if (fdtol != NULL) { 1884 /* 1885 * Handle special case where file descriptor table 1886 * is shared between multiple process leaders. 1887 */ 1888 fdp = td->td_proc->p_fd; 1889 FILEDESC_LOCK(fdp); 1890 for (fdtol = fdtol->fdl_next; 1891 fdtol != td->td_proc->p_fdtol; 1892 fdtol = fdtol->fdl_next) { 1893 if ((fdtol->fdl_leader->p_flag & 1894 P_ADVLOCK) == 0) 1895 continue; 1896 fdtol->fdl_holdcount++; 1897 FILEDESC_UNLOCK(fdp); 1898 lf.l_whence = SEEK_SET; 1899 lf.l_start = 0; 1900 lf.l_len = 0; 1901 lf.l_type = F_UNLCK; 1902 vp = fp->f_vnode; 1903 (void) VOP_ADVLOCK(vp, 1904 (caddr_t)fdtol->fdl_leader, 1905 F_UNLCK, &lf, F_POSIX); 1906 FILEDESC_LOCK(fdp); 1907 fdtol->fdl_holdcount--; 1908 if (fdtol->fdl_holdcount == 0 && 1909 fdtol->fdl_wakeup != 0) { 1910 fdtol->fdl_wakeup = 0; 1911 wakeup(fdtol); 1912 } 1913 } 1914 FILEDESC_UNLOCK(fdp); 1915 } 1916 mtx_unlock(&Giant); 1917 } 1918 return (fdrop(fp, td)); 1919} 1920 1921/* 1922 * Extract the file pointer associated with the specified descriptor for 1923 * the current user process. 1924 * 1925 * If the descriptor doesn't exist, EBADF is returned. 1926 * 1927 * If the descriptor exists but doesn't match 'flags' then 1928 * return EBADF for read attempts and EINVAL for write attempts. 1929 * 1930 * If 'hold' is set (non-zero) the file's refcount will be bumped on return. 1931 * It should be droped with fdrop(). 1932 * If it is not set, then the refcount will not be bumped however the 1933 * thread's filedesc struct will be returned locked (for fgetsock). 1934 * 1935 * If an error occured the non-zero error is returned and *fpp is set to NULL. 1936 * Otherwise *fpp is set and zero is returned. 1937 */ 1938static __inline int 1939_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold) 1940{ 1941 struct filedesc *fdp; 1942 struct file *fp; 1943 1944 *fpp = NULL; 1945 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 1946 return (EBADF); 1947 FILEDESC_LOCK(fdp); 1948 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) { 1949 FILEDESC_UNLOCK(fdp); 1950 return (EBADF); 1951 } 1952 1953 /* 1954 * Note: FREAD failures returns EBADF to maintain backwards 1955 * compatibility with what routines returned before. 1956 * 1957 * Only one flag, or 0, may be specified. 1958 */ 1959 if (flags == FREAD && (fp->f_flag & FREAD) == 0) { 1960 FILEDESC_UNLOCK(fdp); 1961 return (EBADF); 1962 } 1963 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) { 1964 FILEDESC_UNLOCK(fdp); 1965 return (EINVAL); 1966 } 1967 if (hold) { 1968 fhold(fp); 1969 FILEDESC_UNLOCK(fdp); 1970 } 1971 *fpp = fp; 1972 return (0); 1973} 1974 1975int 1976fget(struct thread *td, int fd, struct file **fpp) 1977{ 1978 1979 return(_fget(td, fd, fpp, 0, 1)); 1980} 1981 1982int 1983fget_read(struct thread *td, int fd, struct file **fpp) 1984{ 1985 1986 return(_fget(td, fd, fpp, FREAD, 1)); 1987} 1988 1989int 1990fget_write(struct thread *td, int fd, struct file **fpp) 1991{ 1992 1993 return(_fget(td, fd, fpp, FWRITE, 1)); 1994} 1995 1996/* 1997 * Like fget() but loads the underlying vnode, or returns an error if 1998 * the descriptor does not represent a vnode. Note that pipes use vnodes 1999 * but never have VM objects (so VOP_GETVOBJECT() calls will return an 2000 * error). The returned vnode will be vref()d. 2001 * 2002 * XXX: what about the unused flags ? 2003 */ 2004static __inline int 2005_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags) 2006{ 2007 struct file *fp; 2008 int error; 2009 2010 GIANT_REQUIRED; /* VFS */ 2011 2012 *vpp = NULL; 2013 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2014 return (error); 2015 if (fp->f_vnode == NULL) { 2016 error = EINVAL; 2017 } else { 2018 *vpp = fp->f_vnode; 2019 vref(*vpp); 2020 } 2021 FILEDESC_UNLOCK(td->td_proc->p_fd); 2022 return (error); 2023} 2024 2025int 2026fgetvp(struct thread *td, int fd, struct vnode **vpp) 2027{ 2028 2029 return (_fgetvp(td, fd, vpp, 0)); 2030} 2031 2032int 2033fgetvp_read(struct thread *td, int fd, struct vnode **vpp) 2034{ 2035 2036 return (_fgetvp(td, fd, vpp, FREAD)); 2037} 2038 2039#ifdef notyet 2040int 2041fgetvp_write(struct thread *td, int fd, struct vnode **vpp) 2042{ 2043 2044 return (_fgetvp(td, fd, vpp, FWRITE)); 2045} 2046#endif 2047 2048/* 2049 * Like fget() but loads the underlying socket, or returns an error if 2050 * the descriptor does not represent a socket. 2051 * 2052 * We bump the ref count on the returned socket. XXX Also obtain the SX 2053 * lock in the future. 2054 */ 2055int 2056fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp) 2057{ 2058 struct file *fp; 2059 int error; 2060 2061 NET_ASSERT_GIANT(); 2062 2063 *spp = NULL; 2064 if (fflagp != NULL) 2065 *fflagp = 0; 2066 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2067 return (error); 2068 if (fp->f_type != DTYPE_SOCKET) { 2069 error = ENOTSOCK; 2070 } else { 2071 *spp = fp->f_data; 2072 if (fflagp) 2073 *fflagp = fp->f_flag; 2074 SOCK_LOCK(*spp); 2075 soref(*spp); 2076 SOCK_UNLOCK(*spp); 2077 } 2078 FILEDESC_UNLOCK(td->td_proc->p_fd); 2079 return (error); 2080} 2081 2082/* 2083 * Drop the reference count on the the socket and XXX release the SX lock in 2084 * the future. The last reference closes the socket. 2085 */ 2086void 2087fputsock(struct socket *so) 2088{ 2089 2090 NET_ASSERT_GIANT(); 2091 ACCEPT_LOCK(); 2092 SOCK_LOCK(so); 2093 sorele(so); 2094} 2095 2096int 2097fdrop(struct file *fp, struct thread *td) 2098{ 2099 2100 FILE_LOCK(fp); 2101 return (fdrop_locked(fp, td)); 2102} 2103 2104/* 2105 * Drop reference on struct file passed in, may call closef if the 2106 * reference hits zero. 2107 * Expects struct file locked, and will unlock it. 2108 */ 2109int 2110fdrop_locked(struct file *fp, struct thread *td) 2111{ 2112 int error; 2113 2114 FILE_LOCK_ASSERT(fp, MA_OWNED); 2115 2116 if (--fp->f_count > 0) { 2117 FILE_UNLOCK(fp); 2118 return (0); 2119 } 2120 /* We have the last ref so we can proceed without the file lock. */ 2121 FILE_UNLOCK(fp); 2122 if (fp->f_count < 0) 2123 panic("fdrop: count < 0"); 2124 if (fp->f_ops != &badfileops) 2125 error = fo_close(fp, td); 2126 else 2127 error = 0; 2128 2129 sx_xlock(&filelist_lock); 2130 LIST_REMOVE(fp, f_list); 2131 openfiles--; 2132 sx_xunlock(&filelist_lock); 2133 crfree(fp->f_cred); 2134 uma_zfree(file_zone, fp); 2135 2136 return (error); 2137} 2138 2139/* 2140 * Apply an advisory lock on a file descriptor. 2141 * 2142 * Just attempt to get a record lock of the requested type on 2143 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2144 */ 2145#ifndef _SYS_SYSPROTO_H_ 2146struct flock_args { 2147 int fd; 2148 int how; 2149}; 2150#endif 2151/* 2152 * MPSAFE 2153 */ 2154/* ARGSUSED */ 2155int 2156flock(struct thread *td, struct flock_args *uap) 2157{ 2158 struct file *fp; 2159 struct vnode *vp; 2160 struct flock lf; 2161 int error; 2162 2163 if ((error = fget(td, uap->fd, &fp)) != 0) 2164 return (error); 2165 if (fp->f_type != DTYPE_VNODE) { 2166 fdrop(fp, td); 2167 return (EOPNOTSUPP); 2168 } 2169 2170 mtx_lock(&Giant); 2171 vp = fp->f_vnode; 2172 lf.l_whence = SEEK_SET; 2173 lf.l_start = 0; 2174 lf.l_len = 0; 2175 if (uap->how & LOCK_UN) { 2176 lf.l_type = F_UNLCK; 2177 FILE_LOCK(fp); 2178 fp->f_flag &= ~FHASLOCK; 2179 FILE_UNLOCK(fp); 2180 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2181 goto done2; 2182 } 2183 if (uap->how & LOCK_EX) 2184 lf.l_type = F_WRLCK; 2185 else if (uap->how & LOCK_SH) 2186 lf.l_type = F_RDLCK; 2187 else { 2188 error = EBADF; 2189 goto done2; 2190 } 2191 FILE_LOCK(fp); 2192 fp->f_flag |= FHASLOCK; 2193 FILE_UNLOCK(fp); 2194 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2195 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2196done2: 2197 fdrop(fp, td); 2198 mtx_unlock(&Giant); 2199 return (error); 2200} 2201/* 2202 * Duplicate the specified descriptor to a free descriptor. 2203 */ 2204int 2205dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2206{ 2207 struct file *wfp; 2208 struct file *fp; 2209 2210 /* 2211 * If the to-be-dup'd fd number is greater than the allowed number 2212 * of file descriptors, or the fd to be dup'd has already been 2213 * closed, then reject. 2214 */ 2215 FILEDESC_LOCK(fdp); 2216 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2217 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2218 FILEDESC_UNLOCK(fdp); 2219 return (EBADF); 2220 } 2221 2222 /* 2223 * There are two cases of interest here. 2224 * 2225 * For ENODEV simply dup (dfd) to file descriptor 2226 * (indx) and return. 2227 * 2228 * For ENXIO steal away the file structure from (dfd) and 2229 * store it in (indx). (dfd) is effectively closed by 2230 * this operation. 2231 * 2232 * Any other error code is just returned. 2233 */ 2234 switch (error) { 2235 case ENODEV: 2236 /* 2237 * Check that the mode the file is being opened for is a 2238 * subset of the mode of the existing descriptor. 2239 */ 2240 FILE_LOCK(wfp); 2241 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2242 FILE_UNLOCK(wfp); 2243 FILEDESC_UNLOCK(fdp); 2244 return (EACCES); 2245 } 2246 fp = fdp->fd_ofiles[indx]; 2247 fdp->fd_ofiles[indx] = wfp; 2248 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2249 if (fp == NULL) 2250 fdused(fdp, indx); 2251 fhold_locked(wfp); 2252 FILE_UNLOCK(wfp); 2253 FILEDESC_UNLOCK(fdp); 2254 if (fp != NULL) { 2255 /* 2256 * We now own the reference to fp that the ofiles[] 2257 * array used to own. Release it. 2258 */ 2259 FILE_LOCK(fp); 2260 fdrop_locked(fp, td); 2261 } 2262 return (0); 2263 2264 case ENXIO: 2265 /* 2266 * Steal away the file pointer from dfd and stuff it into indx. 2267 */ 2268 fp = fdp->fd_ofiles[indx]; 2269 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2270 fdp->fd_ofiles[dfd] = NULL; 2271 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2272 fdp->fd_ofileflags[dfd] = 0; 2273 fdunused(fdp, dfd); 2274 if (fp == NULL) 2275 fdused(fdp, indx); 2276 if (fp != NULL) 2277 FILE_LOCK(fp); 2278 FILEDESC_UNLOCK(fdp); 2279 2280 /* 2281 * we now own the reference to fp that the ofiles[] array 2282 * used to own. Release it. 2283 */ 2284 if (fp != NULL) 2285 fdrop_locked(fp, td); 2286 return (0); 2287 2288 default: 2289 FILEDESC_UNLOCK(fdp); 2290 return (error); 2291 } 2292 /* NOTREACHED */ 2293} 2294 2295/* 2296 * Scan all active processes to see if any of them have a current 2297 * or root directory of `olddp'. If so, replace them with the new 2298 * mount point. 2299 */ 2300void 2301mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2302{ 2303 struct filedesc *fdp; 2304 struct proc *p; 2305 int nrele; 2306 2307 if (vrefcnt(olddp) == 1) 2308 return; 2309 sx_slock(&allproc_lock); 2310 LIST_FOREACH(p, &allproc, p_list) { 2311 fdp = fdhold(p); 2312 if (fdp == NULL) 2313 continue; 2314 nrele = 0; 2315 FILEDESC_LOCK_FAST(fdp); 2316 if (fdp->fd_cdir == olddp) { 2317 vref(newdp); 2318 fdp->fd_cdir = newdp; 2319 nrele++; 2320 } 2321 if (fdp->fd_rdir == olddp) { 2322 vref(newdp); 2323 fdp->fd_rdir = newdp; 2324 nrele++; 2325 } 2326 FILEDESC_UNLOCK_FAST(fdp); 2327 fddrop(fdp); 2328 while (nrele--) 2329 vrele(olddp); 2330 } 2331 sx_sunlock(&allproc_lock); 2332 if (rootvnode == olddp) { 2333 vrele(rootvnode); 2334 vref(newdp); 2335 rootvnode = newdp; 2336 } 2337} 2338 2339struct filedesc_to_leader * 2340filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2341{ 2342 struct filedesc_to_leader *fdtol; 2343 2344 MALLOC(fdtol, struct filedesc_to_leader *, 2345 sizeof(struct filedesc_to_leader), 2346 M_FILEDESC_TO_LEADER, 2347 M_WAITOK); 2348 fdtol->fdl_refcount = 1; 2349 fdtol->fdl_holdcount = 0; 2350 fdtol->fdl_wakeup = 0; 2351 fdtol->fdl_leader = leader; 2352 if (old != NULL) { 2353 FILEDESC_LOCK(fdp); 2354 fdtol->fdl_next = old->fdl_next; 2355 fdtol->fdl_prev = old; 2356 old->fdl_next = fdtol; 2357 fdtol->fdl_next->fdl_prev = fdtol; 2358 FILEDESC_UNLOCK(fdp); 2359 } else { 2360 fdtol->fdl_next = fdtol; 2361 fdtol->fdl_prev = fdtol; 2362 } 2363 return (fdtol); 2364} 2365 2366/* 2367 * Get file structures. 2368 */ 2369static int 2370sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2371{ 2372 struct xfile xf; 2373 struct filedesc *fdp; 2374 struct file *fp; 2375 struct proc *p; 2376 int error, n; 2377 2378 /* 2379 * Note: because the number of file descriptors is calculated 2380 * in different ways for sizing vs returning the data, 2381 * there is information leakage from the first loop. However, 2382 * it is of a similar order of magnitude to the leakage from 2383 * global system statistics such as kern.openfiles. 2384 */ 2385 error = sysctl_wire_old_buffer(req, 0); 2386 if (error != 0) 2387 return (error); 2388 if (req->oldptr == NULL) { 2389 n = 16; /* A slight overestimate. */ 2390 sx_slock(&filelist_lock); 2391 LIST_FOREACH(fp, &filehead, f_list) { 2392 /* 2393 * We should grab the lock, but this is an 2394 * estimate, so does it really matter? 2395 */ 2396 /* mtx_lock(fp->f_mtxp); */ 2397 n += fp->f_count; 2398 /* mtx_unlock(f->f_mtxp); */ 2399 } 2400 sx_sunlock(&filelist_lock); 2401 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2402 } 2403 error = 0; 2404 bzero(&xf, sizeof(xf)); 2405 xf.xf_size = sizeof(xf); 2406 sx_slock(&allproc_lock); 2407 LIST_FOREACH(p, &allproc, p_list) { 2408 if (p->p_state == PRS_NEW) 2409 continue; 2410 PROC_LOCK(p); 2411 if (p_cansee(req->td, p) != 0) { 2412 PROC_UNLOCK(p); 2413 continue; 2414 } 2415 xf.xf_pid = p->p_pid; 2416 xf.xf_uid = p->p_ucred->cr_uid; 2417 PROC_UNLOCK(p); 2418 fdp = fdhold(p); 2419 if (fdp == NULL) 2420 continue; 2421 FILEDESC_LOCK_FAST(fdp); 2422 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2423 if ((fp = fdp->fd_ofiles[n]) == NULL) 2424 continue; 2425 xf.xf_fd = n; 2426 xf.xf_file = fp; 2427 xf.xf_data = fp->f_data; 2428 xf.xf_vnode = fp->f_vnode; 2429 xf.xf_type = fp->f_type; 2430 xf.xf_count = fp->f_count; 2431 xf.xf_msgcount = fp->f_msgcount; 2432 xf.xf_offset = fp->f_offset; 2433 xf.xf_flag = fp->f_flag; 2434 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2435 if (error) 2436 break; 2437 } 2438 FILEDESC_UNLOCK_FAST(fdp); 2439 fddrop(fdp); 2440 if (error) 2441 break; 2442 } 2443 sx_sunlock(&allproc_lock); 2444 return (error); 2445} 2446 2447SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2448 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2449 2450SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2451 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2452 2453SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2454 &maxfiles, 0, "Maximum number of files"); 2455 2456SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2457 &openfiles, 0, "System-wide number of open files"); 2458 2459/* ARGSUSED*/ 2460static void 2461filelistinit(void *dummy) 2462{ 2463 2464 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 2465 NULL, NULL, UMA_ALIGN_PTR, 0); 2466 sx_init(&filelist_lock, "filelist lock"); 2467 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 2468 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 2469} 2470SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL) 2471 2472/*-------------------------------------------------------------------*/ 2473 2474static int 2475badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) 2476{ 2477 2478 return (EBADF); 2479} 2480 2481static int 2482badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) 2483{ 2484 2485 return (EBADF); 2486} 2487 2488static int 2489badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) 2490{ 2491 2492 return (0); 2493} 2494 2495static int 2496badfo_kqfilter(struct file *fp, struct knote *kn) 2497{ 2498 2499 return (0); 2500} 2501 2502static int 2503badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) 2504{ 2505 2506 return (EBADF); 2507} 2508 2509static int 2510badfo_close(struct file *fp, struct thread *td) 2511{ 2512 2513 return (EBADF); 2514} 2515 2516struct fileops badfileops = { 2517 .fo_read = badfo_readwrite, 2518 .fo_write = badfo_readwrite, 2519 .fo_ioctl = badfo_ioctl, 2520 .fo_poll = badfo_poll, 2521 .fo_kqfilter = badfo_kqfilter, 2522 .fo_stat = badfo_stat, 2523 .fo_close = badfo_close, 2524}; 2525 2526 2527/*-------------------------------------------------------------------*/ 2528 2529/* 2530 * File Descriptor pseudo-device driver (/dev/fd/). 2531 * 2532 * Opening minor device N dup()s the file (if any) connected to file 2533 * descriptor N belonging to the calling process. Note that this driver 2534 * consists of only the ``open()'' routine, because all subsequent 2535 * references to this file will be direct to the other driver. 2536 * 2537 * XXX: we could give this one a cloning event handler if necessary. 2538 */ 2539 2540/* ARGSUSED */ 2541static int 2542fdopen(struct cdev *dev, int mode, int type, struct thread *td) 2543{ 2544 2545 /* 2546 * XXX Kludge: set curthread->td_dupfd to contain the value of the 2547 * the file descriptor being sought for duplication. The error 2548 * return ensures that the vnode for this device will be released 2549 * by vn_open. Open will detect this special error and take the 2550 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2551 * will simply report the error. 2552 */ 2553 td->td_dupfd = dev2unit(dev); 2554 return (ENODEV); 2555} 2556 2557static struct cdevsw fildesc_cdevsw = { 2558 .d_version = D_VERSION, 2559 .d_flags = D_NEEDGIANT, 2560 .d_open = fdopen, 2561 .d_name = "FD", 2562}; 2563 2564static void 2565fildesc_drvinit(void *unused) 2566{ 2567 struct cdev *dev; 2568 2569 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0"); 2570 make_dev_alias(dev, "stdin"); 2571 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1"); 2572 make_dev_alias(dev, "stdout"); 2573 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2"); 2574 make_dev_alias(dev, "stderr"); 2575} 2576 2577SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL) 2578