kern_descrip.c revision 137686
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 137686 2004-11-14 09:21:01Z phk $"); 39 40#include "opt_compat.h" 41 42#include <sys/param.h> 43#include <sys/limits.h> 44#include <sys/systm.h> 45#include <sys/syscallsubr.h> 46#include <sys/sysproto.h> 47#include <sys/conf.h> 48#include <sys/filedesc.h> 49#include <sys/lock.h> 50#include <sys/jail.h> 51#include <sys/kernel.h> 52#include <sys/limits.h> 53#include <sys/malloc.h> 54#include <sys/mutex.h> 55#include <sys/sysctl.h> 56#include <sys/vnode.h> 57#include <sys/mount.h> 58#include <sys/proc.h> 59#include <sys/namei.h> 60#include <sys/file.h> 61#include <sys/stat.h> 62#include <sys/filio.h> 63#include <sys/fcntl.h> 64#include <sys/unistd.h> 65#include <sys/resourcevar.h> 66#include <sys/event.h> 67#include <sys/sx.h> 68#include <sys/socketvar.h> 69#include <sys/signalvar.h> 70 71#include <vm/vm.h> 72#include <vm/vm_extern.h> 73#include <vm/uma.h> 74 75static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); 76static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", 77 "file desc to leader structures"); 78static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 79 80static uma_zone_t file_zone; 81 82static d_open_t fdopen; 83#define NUMFDESC 64 84 85#define CDEV_MAJOR 22 86static struct cdevsw fildesc_cdevsw = { 87 .d_version = D_VERSION, 88 .d_flags = D_NEEDGIANT, 89 .d_open = fdopen, 90 .d_name = "FD", 91 .d_maj = CDEV_MAJOR, 92}; 93 94/* How to treat 'new' parameter when allocating a fd for do_dup(). */ 95enum dup_type { DUP_VARIABLE, DUP_FIXED }; 96 97static int do_dup(struct thread *td, enum dup_type type, int old, int new, 98 register_t *retval); 99static int fd_first_free(struct filedesc *, int, int); 100static int fd_last_used(struct filedesc *, int, int); 101static void fdgrowtable(struct filedesc *, int); 102static void fdunused(struct filedesc *fdp, int fd); 103 104/* 105 * A process is initially started out with NDFILE descriptors stored within 106 * this structure, selected to be enough for typical applications based on 107 * the historical limit of 20 open files (and the usage of descriptors by 108 * shells). If these descriptors are exhausted, a larger descriptor table 109 * may be allocated, up to a process' resource limit; the internal arrays 110 * are then unused. 111 */ 112#define NDFILE 20 113#define NDSLOTSIZE sizeof(NDSLOTTYPE) 114#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 115#define NDSLOT(x) ((x) / NDENTRIES) 116#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 117#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 118 119/* 120 * Storage required per open file descriptor. 121 */ 122#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 123 124/* 125 * Basic allocation of descriptors: 126 * one of the above, plus arrays for NDFILE descriptors. 127 */ 128struct filedesc0 { 129 struct filedesc fd_fd; 130 /* 131 * These arrays are used when the number of open files is 132 * <= NDFILE, and are then pointed to by the pointers above. 133 */ 134 struct file *fd_dfiles[NDFILE]; 135 char fd_dfileflags[NDFILE]; 136 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 137}; 138 139/* 140 * Descriptor management. 141 */ 142struct filelist filehead; /* head of list of open files */ 143int nfiles; /* actual number of open files */ 144struct sx filelist_lock; /* sx to protect filelist */ 145struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 146 147/* 148 * Find the first zero bit in the given bitmap, starting at low and not 149 * exceeding size - 1. 150 */ 151static int 152fd_first_free(struct filedesc *fdp, int low, int size) 153{ 154 NDSLOTTYPE *map = fdp->fd_map; 155 NDSLOTTYPE mask; 156 int off, maxoff; 157 158 if (low >= size) 159 return (low); 160 161 off = NDSLOT(low); 162 if (low % NDENTRIES) { 163 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 164 if ((mask &= ~map[off]) != 0UL) 165 return (off * NDENTRIES + ffsl(mask) - 1); 166 ++off; 167 } 168 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 169 if (map[off] != ~0UL) 170 return (off * NDENTRIES + ffsl(~map[off]) - 1); 171 return (size); 172} 173 174/* 175 * Find the highest non-zero bit in the given bitmap, starting at low and 176 * not exceeding size - 1. 177 */ 178static int 179fd_last_used(struct filedesc *fdp, int low, int size) 180{ 181 NDSLOTTYPE *map = fdp->fd_map; 182 NDSLOTTYPE mask; 183 int off, minoff; 184 185 if (low >= size) 186 return (-1); 187 188 off = NDSLOT(size); 189 if (size % NDENTRIES) { 190 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 191 if ((mask &= map[off]) != 0) 192 return (off * NDENTRIES + flsl(mask) - 1); 193 --off; 194 } 195 for (minoff = NDSLOT(low); off >= minoff; --off) 196 if (map[off] != 0) 197 return (off * NDENTRIES + flsl(map[off]) - 1); 198 return (size - 1); 199} 200 201static int 202fdisused(struct filedesc *fdp, int fd) 203{ 204 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 205 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 206 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 207} 208 209/* 210 * Mark a file descriptor as used. 211 */ 212void 213fdused(struct filedesc *fdp, int fd) 214{ 215 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 216 KASSERT(!fdisused(fdp, fd), 217 ("fd already used")); 218 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 219 if (fd > fdp->fd_lastfile) 220 fdp->fd_lastfile = fd; 221 if (fd == fdp->fd_freefile) 222 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 223} 224 225/* 226 * Mark a file descriptor as unused. 227 */ 228static void 229fdunused(struct filedesc *fdp, int fd) 230{ 231 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 232 KASSERT(fdisused(fdp, fd), 233 ("fd is already unused")); 234 KASSERT(fdp->fd_ofiles[fd] == NULL, 235 ("fd is still in use")); 236 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 237 if (fd < fdp->fd_freefile) 238 fdp->fd_freefile = fd; 239 if (fd == fdp->fd_lastfile) 240 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 241} 242 243/* 244 * System calls on descriptors. 245 */ 246#ifndef _SYS_SYSPROTO_H_ 247struct getdtablesize_args { 248 int dummy; 249}; 250#endif 251/* 252 * MPSAFE 253 */ 254/* ARGSUSED */ 255int 256getdtablesize(td, uap) 257 struct thread *td; 258 struct getdtablesize_args *uap; 259{ 260 struct proc *p = td->td_proc; 261 262 PROC_LOCK(p); 263 td->td_retval[0] = 264 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 265 PROC_UNLOCK(p); 266 return (0); 267} 268 269/* 270 * Duplicate a file descriptor to a particular value. 271 * 272 * note: keep in mind that a potential race condition exists when closing 273 * descriptors from a shared descriptor table (via rfork). 274 */ 275#ifndef _SYS_SYSPROTO_H_ 276struct dup2_args { 277 u_int from; 278 u_int to; 279}; 280#endif 281/* 282 * MPSAFE 283 */ 284/* ARGSUSED */ 285int 286dup2(td, uap) 287 struct thread *td; 288 struct dup2_args *uap; 289{ 290 291 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 292 td->td_retval)); 293} 294 295/* 296 * Duplicate a file descriptor. 297 */ 298#ifndef _SYS_SYSPROTO_H_ 299struct dup_args { 300 u_int fd; 301}; 302#endif 303/* 304 * MPSAFE 305 */ 306/* ARGSUSED */ 307int 308dup(td, uap) 309 struct thread *td; 310 struct dup_args *uap; 311{ 312 313 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval)); 314} 315 316/* 317 * The file control system call. 318 */ 319#ifndef _SYS_SYSPROTO_H_ 320struct fcntl_args { 321 int fd; 322 int cmd; 323 long arg; 324}; 325#endif 326/* 327 * MPSAFE 328 */ 329/* ARGSUSED */ 330int 331fcntl(td, uap) 332 struct thread *td; 333 struct fcntl_args *uap; 334{ 335 struct flock fl; 336 intptr_t arg; 337 int error; 338 339 error = 0; 340 switch (uap->cmd) { 341 case F_GETLK: 342 case F_SETLK: 343 case F_SETLKW: 344 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 345 arg = (intptr_t)&fl; 346 break; 347 default: 348 arg = uap->arg; 349 break; 350 } 351 if (error) 352 return (error); 353 error = kern_fcntl(td, uap->fd, uap->cmd, arg); 354 if (error) 355 return (error); 356 if (uap->cmd == F_GETLK) 357 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 358 return (error); 359} 360 361int 362kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 363{ 364 struct filedesc *fdp; 365 struct flock *flp; 366 struct file *fp; 367 struct proc *p; 368 char *pop; 369 struct vnode *vp; 370 u_int newmin; 371 int error, flg, tmp; 372 int giant_locked; 373 374 /* 375 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to 376 * avoid grabbing Giant for calls we know don't need it. 377 */ 378 switch (cmd) { 379 case F_DUPFD: 380 case F_GETFD: 381 case F_SETFD: 382 case F_GETFL: 383 giant_locked = 0; 384 break; 385 386 default: 387 giant_locked = 1; 388 mtx_lock(&Giant); 389 } 390 391 error = 0; 392 flg = F_POSIX; 393 p = td->td_proc; 394 fdp = p->p_fd; 395 FILEDESC_LOCK(fdp); 396 if ((unsigned)fd >= fdp->fd_nfiles || 397 (fp = fdp->fd_ofiles[fd]) == NULL) { 398 FILEDESC_UNLOCK(fdp); 399 error = EBADF; 400 goto done2; 401 } 402 pop = &fdp->fd_ofileflags[fd]; 403 404 switch (cmd) { 405 case F_DUPFD: 406 /* mtx_assert(&Giant, MA_NOTOWNED); */ 407 FILEDESC_UNLOCK(fdp); 408 newmin = arg; 409 PROC_LOCK(p); 410 if (newmin >= lim_cur(p, RLIMIT_NOFILE) || 411 newmin >= maxfilesperproc) { 412 PROC_UNLOCK(p); 413 error = EINVAL; 414 break; 415 } 416 PROC_UNLOCK(p); 417 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval); 418 break; 419 420 case F_GETFD: 421 /* mtx_assert(&Giant, MA_NOTOWNED); */ 422 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 423 FILEDESC_UNLOCK(fdp); 424 break; 425 426 case F_SETFD: 427 /* mtx_assert(&Giant, MA_NOTOWNED); */ 428 *pop = (*pop &~ UF_EXCLOSE) | 429 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 430 FILEDESC_UNLOCK(fdp); 431 break; 432 433 case F_GETFL: 434 /* mtx_assert(&Giant, MA_NOTOWNED); */ 435 FILE_LOCK(fp); 436 FILEDESC_UNLOCK(fdp); 437 td->td_retval[0] = OFLAGS(fp->f_flag); 438 FILE_UNLOCK(fp); 439 break; 440 441 case F_SETFL: 442 mtx_assert(&Giant, MA_OWNED); 443 FILE_LOCK(fp); 444 FILEDESC_UNLOCK(fdp); 445 fhold_locked(fp); 446 fp->f_flag &= ~FCNTLFLAGS; 447 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 448 FILE_UNLOCK(fp); 449 tmp = fp->f_flag & FNONBLOCK; 450 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 451 if (error) { 452 fdrop(fp, td); 453 break; 454 } 455 tmp = fp->f_flag & FASYNC; 456 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 457 if (error == 0) { 458 fdrop(fp, td); 459 break; 460 } 461 FILE_LOCK(fp); 462 fp->f_flag &= ~FNONBLOCK; 463 FILE_UNLOCK(fp); 464 tmp = 0; 465 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 466 fdrop(fp, td); 467 break; 468 469 case F_GETOWN: 470 mtx_assert(&Giant, MA_OWNED); 471 fhold(fp); 472 FILEDESC_UNLOCK(fdp); 473 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 474 if (error == 0) 475 td->td_retval[0] = tmp; 476 fdrop(fp, td); 477 break; 478 479 case F_SETOWN: 480 mtx_assert(&Giant, MA_OWNED); 481 fhold(fp); 482 FILEDESC_UNLOCK(fdp); 483 tmp = arg; 484 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 485 fdrop(fp, td); 486 break; 487 488 case F_SETLKW: 489 mtx_assert(&Giant, MA_OWNED); 490 flg |= F_WAIT; 491 /* FALLTHROUGH F_SETLK */ 492 493 case F_SETLK: 494 mtx_assert(&Giant, MA_OWNED); 495 if (fp->f_type != DTYPE_VNODE) { 496 FILEDESC_UNLOCK(fdp); 497 error = EBADF; 498 break; 499 } 500 501 flp = (struct flock *)arg; 502 if (flp->l_whence == SEEK_CUR) { 503 if (fp->f_offset < 0 || 504 (flp->l_start > 0 && 505 fp->f_offset > OFF_MAX - flp->l_start)) { 506 FILEDESC_UNLOCK(fdp); 507 error = EOVERFLOW; 508 break; 509 } 510 flp->l_start += fp->f_offset; 511 } 512 513 /* 514 * VOP_ADVLOCK() may block. 515 */ 516 fhold(fp); 517 FILEDESC_UNLOCK(fdp); 518 vp = fp->f_vnode; 519 520 switch (flp->l_type) { 521 case F_RDLCK: 522 if ((fp->f_flag & FREAD) == 0) { 523 error = EBADF; 524 break; 525 } 526 PROC_LOCK(p->p_leader); 527 p->p_leader->p_flag |= P_ADVLOCK; 528 PROC_UNLOCK(p->p_leader); 529 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 530 flp, flg); 531 break; 532 case F_WRLCK: 533 if ((fp->f_flag & FWRITE) == 0) { 534 error = EBADF; 535 break; 536 } 537 PROC_LOCK(p->p_leader); 538 p->p_leader->p_flag |= P_ADVLOCK; 539 PROC_UNLOCK(p->p_leader); 540 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 541 flp, flg); 542 break; 543 case F_UNLCK: 544 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 545 flp, F_POSIX); 546 break; 547 default: 548 error = EINVAL; 549 break; 550 } 551 /* Check for race with close */ 552 FILEDESC_LOCK_FAST(fdp); 553 if ((unsigned) fd >= fdp->fd_nfiles || 554 fp != fdp->fd_ofiles[fd]) { 555 FILEDESC_UNLOCK_FAST(fdp); 556 flp->l_whence = SEEK_SET; 557 flp->l_start = 0; 558 flp->l_len = 0; 559 flp->l_type = F_UNLCK; 560 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 561 F_UNLCK, flp, F_POSIX); 562 } else 563 FILEDESC_UNLOCK_FAST(fdp); 564 fdrop(fp, td); 565 break; 566 567 case F_GETLK: 568 mtx_assert(&Giant, MA_OWNED); 569 if (fp->f_type != DTYPE_VNODE) { 570 FILEDESC_UNLOCK(fdp); 571 error = EBADF; 572 break; 573 } 574 flp = (struct flock *)arg; 575 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 576 flp->l_type != F_UNLCK) { 577 FILEDESC_UNLOCK(fdp); 578 error = EINVAL; 579 break; 580 } 581 if (flp->l_whence == SEEK_CUR) { 582 if ((flp->l_start > 0 && 583 fp->f_offset > OFF_MAX - flp->l_start) || 584 (flp->l_start < 0 && 585 fp->f_offset < OFF_MIN - flp->l_start)) { 586 FILEDESC_UNLOCK(fdp); 587 error = EOVERFLOW; 588 break; 589 } 590 flp->l_start += fp->f_offset; 591 } 592 /* 593 * VOP_ADVLOCK() may block. 594 */ 595 fhold(fp); 596 FILEDESC_UNLOCK(fdp); 597 vp = fp->f_vnode; 598 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 599 F_POSIX); 600 fdrop(fp, td); 601 break; 602 default: 603 FILEDESC_UNLOCK(fdp); 604 error = EINVAL; 605 break; 606 } 607done2: 608 if (giant_locked) 609 mtx_unlock(&Giant); 610 return (error); 611} 612 613/* 614 * Common code for dup, dup2, and fcntl(F_DUPFD). 615 */ 616static int 617do_dup(td, type, old, new, retval) 618 enum dup_type type; 619 int old, new; 620 register_t *retval; 621 struct thread *td; 622{ 623 struct filedesc *fdp; 624 struct proc *p; 625 struct file *fp; 626 struct file *delfp; 627 int error, holdleaders, maxfd; 628 629 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED), 630 ("invalid dup type %d", type)); 631 632 p = td->td_proc; 633 fdp = p->p_fd; 634 635 /* 636 * Verify we have a valid descriptor to dup from and possibly to 637 * dup to. 638 */ 639 if (old < 0 || new < 0) 640 return (EBADF); 641 PROC_LOCK(p); 642 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 643 PROC_UNLOCK(p); 644 if (new >= maxfd) 645 return (EMFILE); 646 647 FILEDESC_LOCK(fdp); 648 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 649 FILEDESC_UNLOCK(fdp); 650 return (EBADF); 651 } 652 if (type == DUP_FIXED && old == new) { 653 *retval = new; 654 FILEDESC_UNLOCK(fdp); 655 return (0); 656 } 657 fp = fdp->fd_ofiles[old]; 658 fhold(fp); 659 660 /* 661 * If the caller specified a file descriptor, make sure the file 662 * table is large enough to hold it, and grab it. Otherwise, just 663 * allocate a new descriptor the usual way. Since the filedesc 664 * lock may be temporarily dropped in the process, we have to look 665 * out for a race. 666 */ 667 if (type == DUP_FIXED) { 668 if (new >= fdp->fd_nfiles) 669 fdgrowtable(fdp, new + 1); 670 if (fdp->fd_ofiles[new] == NULL) 671 fdused(fdp, new); 672 } else { 673 if ((error = fdalloc(td, new, &new)) != 0) { 674 FILEDESC_UNLOCK(fdp); 675 fdrop(fp, td); 676 return (error); 677 } 678 } 679 680 /* 681 * If the old file changed out from under us then treat it as a 682 * bad file descriptor. Userland should do its own locking to 683 * avoid this case. 684 */ 685 if (fdp->fd_ofiles[old] != fp) { 686 /* we've allocated a descriptor which we won't use */ 687 if (fdp->fd_ofiles[new] == NULL) 688 fdunused(fdp, new); 689 FILEDESC_UNLOCK(fdp); 690 fdrop(fp, td); 691 return (EBADF); 692 } 693 KASSERT(old != new, 694 ("new fd is same as old")); 695 696 /* 697 * Save info on the descriptor being overwritten. We cannot close 698 * it without introducing an ownership race for the slot, since we 699 * need to drop the filedesc lock to call closef(). 700 * 701 * XXX this duplicates parts of close(). 702 */ 703 delfp = fdp->fd_ofiles[new]; 704 holdleaders = 0; 705 if (delfp != NULL) { 706 if (td->td_proc->p_fdtol != NULL) { 707 /* 708 * Ask fdfree() to sleep to ensure that all relevant 709 * process leaders can be traversed in closef(). 710 */ 711 fdp->fd_holdleaderscount++; 712 holdleaders = 1; 713 } 714 } 715 716 /* 717 * Duplicate the source descriptor 718 */ 719 fdp->fd_ofiles[new] = fp; 720 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 721 if (new > fdp->fd_lastfile) 722 fdp->fd_lastfile = new; 723 *retval = new; 724 725 /* 726 * If we dup'd over a valid file, we now own the reference to it 727 * and must dispose of it using closef() semantics (as if a 728 * close() were performed on it). 729 * 730 * XXX this duplicates parts of close(). 731 */ 732 if (delfp != NULL) { 733 knote_fdclose(td, new); 734 FILEDESC_UNLOCK(fdp); 735 mtx_lock(&Giant); 736 (void) closef(delfp, td); 737 mtx_unlock(&Giant); 738 if (holdleaders) { 739 FILEDESC_LOCK_FAST(fdp); 740 fdp->fd_holdleaderscount--; 741 if (fdp->fd_holdleaderscount == 0 && 742 fdp->fd_holdleaderswakeup != 0) { 743 fdp->fd_holdleaderswakeup = 0; 744 wakeup(&fdp->fd_holdleaderscount); 745 } 746 FILEDESC_UNLOCK_FAST(fdp); 747 } 748 } else { 749 FILEDESC_UNLOCK(fdp); 750 } 751 return (0); 752} 753 754/* 755 * If sigio is on the list associated with a process or process group, 756 * disable signalling from the device, remove sigio from the list and 757 * free sigio. 758 */ 759void 760funsetown(sigiop) 761 struct sigio **sigiop; 762{ 763 struct sigio *sigio; 764 765 SIGIO_LOCK(); 766 sigio = *sigiop; 767 if (sigio == NULL) { 768 SIGIO_UNLOCK(); 769 return; 770 } 771 *(sigio->sio_myref) = NULL; 772 if ((sigio)->sio_pgid < 0) { 773 struct pgrp *pg = (sigio)->sio_pgrp; 774 PGRP_LOCK(pg); 775 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 776 sigio, sio_pgsigio); 777 PGRP_UNLOCK(pg); 778 } else { 779 struct proc *p = (sigio)->sio_proc; 780 PROC_LOCK(p); 781 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 782 sigio, sio_pgsigio); 783 PROC_UNLOCK(p); 784 } 785 SIGIO_UNLOCK(); 786 crfree(sigio->sio_ucred); 787 FREE(sigio, M_SIGIO); 788} 789 790/* 791 * Free a list of sigio structures. 792 * We only need to lock the SIGIO_LOCK because we have made ourselves 793 * inaccessable to callers of fsetown and therefore do not need to lock 794 * the proc or pgrp struct for the list manipulation. 795 */ 796void 797funsetownlst(sigiolst) 798 struct sigiolst *sigiolst; 799{ 800 struct proc *p; 801 struct pgrp *pg; 802 struct sigio *sigio; 803 804 sigio = SLIST_FIRST(sigiolst); 805 if (sigio == NULL) 806 return; 807 p = NULL; 808 pg = NULL; 809 810 /* 811 * Every entry of the list should belong 812 * to a single proc or pgrp. 813 */ 814 if (sigio->sio_pgid < 0) { 815 pg = sigio->sio_pgrp; 816 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 817 } else /* if (sigio->sio_pgid > 0) */ { 818 p = sigio->sio_proc; 819 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 820 } 821 822 SIGIO_LOCK(); 823 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 824 *(sigio->sio_myref) = NULL; 825 if (pg != NULL) { 826 KASSERT(sigio->sio_pgid < 0, 827 ("Proc sigio in pgrp sigio list")); 828 KASSERT(sigio->sio_pgrp == pg, 829 ("Bogus pgrp in sigio list")); 830 PGRP_LOCK(pg); 831 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 832 sio_pgsigio); 833 PGRP_UNLOCK(pg); 834 } else /* if (p != NULL) */ { 835 KASSERT(sigio->sio_pgid > 0, 836 ("Pgrp sigio in proc sigio list")); 837 KASSERT(sigio->sio_proc == p, 838 ("Bogus proc in sigio list")); 839 PROC_LOCK(p); 840 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 841 sio_pgsigio); 842 PROC_UNLOCK(p); 843 } 844 SIGIO_UNLOCK(); 845 crfree(sigio->sio_ucred); 846 FREE(sigio, M_SIGIO); 847 SIGIO_LOCK(); 848 } 849 SIGIO_UNLOCK(); 850} 851 852/* 853 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 854 * 855 * After permission checking, add a sigio structure to the sigio list for 856 * the process or process group. 857 */ 858int 859fsetown(pgid, sigiop) 860 pid_t pgid; 861 struct sigio **sigiop; 862{ 863 struct proc *proc; 864 struct pgrp *pgrp; 865 struct sigio *sigio; 866 int ret; 867 868 if (pgid == 0) { 869 funsetown(sigiop); 870 return (0); 871 } 872 873 ret = 0; 874 875 /* Allocate and fill in the new sigio out of locks. */ 876 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK); 877 sigio->sio_pgid = pgid; 878 sigio->sio_ucred = crhold(curthread->td_ucred); 879 sigio->sio_myref = sigiop; 880 881 sx_slock(&proctree_lock); 882 if (pgid > 0) { 883 proc = pfind(pgid); 884 if (proc == NULL) { 885 ret = ESRCH; 886 goto fail; 887 } 888 889 /* 890 * Policy - Don't allow a process to FSETOWN a process 891 * in another session. 892 * 893 * Remove this test to allow maximum flexibility or 894 * restrict FSETOWN to the current process or process 895 * group for maximum safety. 896 */ 897 PROC_UNLOCK(proc); 898 if (proc->p_session != curthread->td_proc->p_session) { 899 ret = EPERM; 900 goto fail; 901 } 902 903 pgrp = NULL; 904 } else /* if (pgid < 0) */ { 905 pgrp = pgfind(-pgid); 906 if (pgrp == NULL) { 907 ret = ESRCH; 908 goto fail; 909 } 910 PGRP_UNLOCK(pgrp); 911 912 /* 913 * Policy - Don't allow a process to FSETOWN a process 914 * in another session. 915 * 916 * Remove this test to allow maximum flexibility or 917 * restrict FSETOWN to the current process or process 918 * group for maximum safety. 919 */ 920 if (pgrp->pg_session != curthread->td_proc->p_session) { 921 ret = EPERM; 922 goto fail; 923 } 924 925 proc = NULL; 926 } 927 funsetown(sigiop); 928 if (pgid > 0) { 929 PROC_LOCK(proc); 930 /* 931 * Since funsetownlst() is called without the proctree 932 * locked, we need to check for P_WEXIT. 933 * XXX: is ESRCH correct? 934 */ 935 if ((proc->p_flag & P_WEXIT) != 0) { 936 PROC_UNLOCK(proc); 937 ret = ESRCH; 938 goto fail; 939 } 940 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 941 sigio->sio_proc = proc; 942 PROC_UNLOCK(proc); 943 } else { 944 PGRP_LOCK(pgrp); 945 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 946 sigio->sio_pgrp = pgrp; 947 PGRP_UNLOCK(pgrp); 948 } 949 sx_sunlock(&proctree_lock); 950 SIGIO_LOCK(); 951 *sigiop = sigio; 952 SIGIO_UNLOCK(); 953 return (0); 954 955fail: 956 sx_sunlock(&proctree_lock); 957 crfree(sigio->sio_ucred); 958 FREE(sigio, M_SIGIO); 959 return (ret); 960} 961 962/* 963 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 964 */ 965pid_t 966fgetown(sigiop) 967 struct sigio **sigiop; 968{ 969 pid_t pgid; 970 971 SIGIO_LOCK(); 972 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 973 SIGIO_UNLOCK(); 974 return (pgid); 975} 976 977/* 978 * Close a file descriptor. 979 */ 980#ifndef _SYS_SYSPROTO_H_ 981struct close_args { 982 int fd; 983}; 984#endif 985/* 986 * MPSAFE 987 */ 988/* ARGSUSED */ 989int 990close(td, uap) 991 struct thread *td; 992 struct close_args *uap; 993{ 994 struct filedesc *fdp; 995 struct file *fp; 996 int fd, error; 997 int holdleaders; 998 999 fd = uap->fd; 1000 error = 0; 1001 holdleaders = 0; 1002 fdp = td->td_proc->p_fd; 1003 mtx_lock(&Giant); 1004 FILEDESC_LOCK(fdp); 1005 if ((unsigned)fd >= fdp->fd_nfiles || 1006 (fp = fdp->fd_ofiles[fd]) == NULL) { 1007 FILEDESC_UNLOCK(fdp); 1008 mtx_unlock(&Giant); 1009 return (EBADF); 1010 } 1011 fdp->fd_ofiles[fd] = NULL; 1012 fdp->fd_ofileflags[fd] = 0; 1013 fdunused(fdp, fd); 1014 if (td->td_proc->p_fdtol != NULL) { 1015 /* 1016 * Ask fdfree() to sleep to ensure that all relevant 1017 * process leaders can be traversed in closef(). 1018 */ 1019 fdp->fd_holdleaderscount++; 1020 holdleaders = 1; 1021 } 1022 1023 /* 1024 * we now hold the fp reference that used to be owned by the descriptor 1025 * array. 1026 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a 1027 * race of the fd getting opened, a knote added, and deleteing a knote 1028 * for the new fd. 1029 */ 1030 knote_fdclose(td, fd); 1031 FILEDESC_UNLOCK(fdp); 1032 1033 error = closef(fp, td); 1034 mtx_unlock(&Giant); 1035 if (holdleaders) { 1036 FILEDESC_LOCK_FAST(fdp); 1037 fdp->fd_holdleaderscount--; 1038 if (fdp->fd_holdleaderscount == 0 && 1039 fdp->fd_holdleaderswakeup != 0) { 1040 fdp->fd_holdleaderswakeup = 0; 1041 wakeup(&fdp->fd_holdleaderscount); 1042 } 1043 FILEDESC_UNLOCK_FAST(fdp); 1044 } 1045 return (error); 1046} 1047 1048#if defined(COMPAT_43) 1049/* 1050 * Return status information about a file descriptor. 1051 */ 1052#ifndef _SYS_SYSPROTO_H_ 1053struct ofstat_args { 1054 int fd; 1055 struct ostat *sb; 1056}; 1057#endif 1058/* 1059 * MPSAFE 1060 */ 1061/* ARGSUSED */ 1062int 1063ofstat(td, uap) 1064 struct thread *td; 1065 struct ofstat_args *uap; 1066{ 1067 struct file *fp; 1068 struct stat ub; 1069 struct ostat oub; 1070 int error; 1071 1072 if ((error = fget(td, uap->fd, &fp)) != 0) 1073 goto done2; 1074 error = fo_stat(fp, &ub, td->td_ucred, td); 1075 if (error == 0) { 1076 cvtstat(&ub, &oub); 1077 error = copyout(&oub, uap->sb, sizeof(oub)); 1078 } 1079 fdrop(fp, td); 1080done2: 1081 return (error); 1082} 1083#endif /* COMPAT_43 */ 1084 1085/* 1086 * Return status information about a file descriptor. 1087 */ 1088#ifndef _SYS_SYSPROTO_H_ 1089struct fstat_args { 1090 int fd; 1091 struct stat *sb; 1092}; 1093#endif 1094/* 1095 * MPSAFE 1096 */ 1097/* ARGSUSED */ 1098int 1099fstat(td, uap) 1100 struct thread *td; 1101 struct fstat_args *uap; 1102{ 1103 struct file *fp; 1104 struct stat ub; 1105 int error; 1106 1107 if ((error = fget(td, uap->fd, &fp)) != 0) 1108 goto done2; 1109 error = fo_stat(fp, &ub, td->td_ucred, td); 1110 if (error == 0) 1111 error = copyout(&ub, uap->sb, sizeof(ub)); 1112 fdrop(fp, td); 1113done2: 1114 return (error); 1115} 1116 1117/* 1118 * Return status information about a file descriptor. 1119 */ 1120#ifndef _SYS_SYSPROTO_H_ 1121struct nfstat_args { 1122 int fd; 1123 struct nstat *sb; 1124}; 1125#endif 1126/* 1127 * MPSAFE 1128 */ 1129/* ARGSUSED */ 1130int 1131nfstat(td, uap) 1132 struct thread *td; 1133 struct nfstat_args *uap; 1134{ 1135 struct file *fp; 1136 struct stat ub; 1137 struct nstat nub; 1138 int error; 1139 1140 if ((error = fget(td, uap->fd, &fp)) != 0) 1141 goto done2; 1142 error = fo_stat(fp, &ub, td->td_ucred, td); 1143 if (error == 0) { 1144 cvtnstat(&ub, &nub); 1145 error = copyout(&nub, uap->sb, sizeof(nub)); 1146 } 1147 fdrop(fp, td); 1148done2: 1149 return (error); 1150} 1151 1152/* 1153 * Return pathconf information about a file descriptor. 1154 */ 1155#ifndef _SYS_SYSPROTO_H_ 1156struct fpathconf_args { 1157 int fd; 1158 int name; 1159}; 1160#endif 1161/* 1162 * MPSAFE 1163 */ 1164/* ARGSUSED */ 1165int 1166fpathconf(td, uap) 1167 struct thread *td; 1168 struct fpathconf_args *uap; 1169{ 1170 struct file *fp; 1171 struct vnode *vp; 1172 int error; 1173 1174 if ((error = fget(td, uap->fd, &fp)) != 0) 1175 return (error); 1176 1177 /* If asynchronous I/O is available, it works for all descriptors. */ 1178 if (uap->name == _PC_ASYNC_IO) { 1179 td->td_retval[0] = async_io_version; 1180 goto out; 1181 } 1182 vp = fp->f_vnode; 1183 if (vp != NULL) { 1184 mtx_lock(&Giant); 1185 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1186 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1187 VOP_UNLOCK(vp, 0, td); 1188 mtx_unlock(&Giant); 1189 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1190 if (uap->name != _PC_PIPE_BUF) { 1191 error = EINVAL; 1192 } else { 1193 td->td_retval[0] = PIPE_BUF; 1194 error = 0; 1195 } 1196 } else { 1197 error = EOPNOTSUPP; 1198 } 1199out: 1200 fdrop(fp, td); 1201 return (error); 1202} 1203 1204/* 1205 * Grow the file table to accomodate (at least) nfd descriptors. This may 1206 * block and drop the filedesc lock, but it will reacquire it before 1207 * returing. 1208 */ 1209static void 1210fdgrowtable(struct filedesc *fdp, int nfd) 1211{ 1212 struct file **ntable; 1213 char *nfileflags; 1214 int nnfiles, onfiles; 1215 NDSLOTTYPE *nmap; 1216 1217 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1218 1219 KASSERT(fdp->fd_nfiles > 0, 1220 ("zero-length file table")); 1221 1222 /* compute the size of the new table */ 1223 onfiles = fdp->fd_nfiles; 1224 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1225 if (nnfiles <= onfiles) 1226 /* the table is already large enough */ 1227 return; 1228 1229 /* allocate a new table and (if required) new bitmaps */ 1230 FILEDESC_UNLOCK(fdp); 1231 MALLOC(ntable, struct file **, nnfiles * OFILESIZE, 1232 M_FILEDESC, M_ZERO | M_WAITOK); 1233 nfileflags = (char *)&ntable[nnfiles]; 1234 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1235 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE, 1236 M_FILEDESC, M_ZERO | M_WAITOK); 1237 else 1238 nmap = NULL; 1239 FILEDESC_LOCK(fdp); 1240 1241 /* 1242 * We now have new tables ready to go. Since we dropped the 1243 * filedesc lock to call malloc(), watch out for a race. 1244 */ 1245 onfiles = fdp->fd_nfiles; 1246 if (onfiles >= nnfiles) { 1247 /* we lost the race, but that's OK */ 1248 free(ntable, M_FILEDESC); 1249 if (nmap != NULL) 1250 free(nmap, M_FILEDESC); 1251 return; 1252 } 1253 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1254 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1255 if (onfiles > NDFILE) 1256 free(fdp->fd_ofiles, M_FILEDESC); 1257 fdp->fd_ofiles = ntable; 1258 fdp->fd_ofileflags = nfileflags; 1259 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1260 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1261 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1262 free(fdp->fd_map, M_FILEDESC); 1263 fdp->fd_map = nmap; 1264 } 1265 fdp->fd_nfiles = nnfiles; 1266} 1267 1268/* 1269 * Allocate a file descriptor for the process. 1270 */ 1271int 1272fdalloc(struct thread *td, int minfd, int *result) 1273{ 1274 struct proc *p = td->td_proc; 1275 struct filedesc *fdp = p->p_fd; 1276 int fd = -1, maxfd; 1277 1278 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1279 1280 PROC_LOCK(p); 1281 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1282 PROC_UNLOCK(p); 1283 1284 /* 1285 * Search the bitmap for a free descriptor. If none is found, try 1286 * to grow the file table. Keep at it until we either get a file 1287 * descriptor or run into process or system limits; fdgrowtable() 1288 * may drop the filedesc lock, so we're in a race. 1289 */ 1290 for (;;) { 1291 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1292 if (fd >= maxfd) 1293 return (EMFILE); 1294 if (fd < fdp->fd_nfiles) 1295 break; 1296 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1297 } 1298 1299 /* 1300 * Perform some sanity checks, then mark the file descriptor as 1301 * used and return it to the caller. 1302 */ 1303 KASSERT(!fdisused(fdp, fd), 1304 ("fd_first_free() returned non-free descriptor")); 1305 KASSERT(fdp->fd_ofiles[fd] == NULL, 1306 ("free descriptor isn't")); 1307 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */ 1308 fdused(fdp, fd); 1309 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 1310 *result = fd; 1311 return (0); 1312} 1313 1314/* 1315 * Check to see whether n user file descriptors 1316 * are available to the process p. 1317 */ 1318int 1319fdavail(td, n) 1320 struct thread *td; 1321 int n; 1322{ 1323 struct proc *p = td->td_proc; 1324 struct filedesc *fdp = td->td_proc->p_fd; 1325 struct file **fpp; 1326 int i, lim, last; 1327 1328 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1329 1330 PROC_LOCK(p); 1331 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1332 PROC_UNLOCK(p); 1333 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1334 return (1); 1335 last = min(fdp->fd_nfiles, lim); 1336 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 1337 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 1338 if (*fpp == NULL && --n <= 0) 1339 return (1); 1340 } 1341 return (0); 1342} 1343 1344/* 1345 * Create a new open file structure and allocate 1346 * a file decriptor for the process that refers to it. 1347 * We add one reference to the file for the descriptor table 1348 * and one reference for resultfp. This is to prevent us being 1349 * prempted and the entry in the descriptor table closed after 1350 * we release the FILEDESC lock. 1351 */ 1352int 1353falloc(td, resultfp, resultfd) 1354 struct thread *td; 1355 struct file **resultfp; 1356 int *resultfd; 1357{ 1358 struct proc *p = td->td_proc; 1359 struct file *fp, *fq; 1360 int error, i; 1361 int maxuserfiles = maxfiles - (maxfiles / 20); 1362 static struct timeval lastfail; 1363 static int curfail; 1364 1365 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1366 sx_xlock(&filelist_lock); 1367 if ((nfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 || 1368 jailed(td->td_ucred))) || nfiles >= maxfiles) { 1369 if (ppsratecheck(&lastfail, &curfail, 1)) { 1370 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n", 1371 td->td_ucred->cr_ruid); 1372 } 1373 sx_xunlock(&filelist_lock); 1374 uma_zfree(file_zone, fp); 1375 return (ENFILE); 1376 } 1377 nfiles++; 1378 1379 /* 1380 * If the process has file descriptor zero open, add the new file 1381 * descriptor to the list of open files at that point, otherwise 1382 * put it at the front of the list of open files. 1383 */ 1384 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep); 1385 fp->f_count = 1; 1386 if (resultfp) 1387 fp->f_count++; 1388 fp->f_cred = crhold(td->td_ucred); 1389 fp->f_ops = &badfileops; 1390 fp->f_data = NULL; 1391 fp->f_vnode = NULL; 1392 FILEDESC_LOCK(p->p_fd); 1393 if ((fq = p->p_fd->fd_ofiles[0])) { 1394 LIST_INSERT_AFTER(fq, fp, f_list); 1395 } else { 1396 LIST_INSERT_HEAD(&filehead, fp, f_list); 1397 } 1398 sx_xunlock(&filelist_lock); 1399 if ((error = fdalloc(td, 0, &i))) { 1400 FILEDESC_UNLOCK(p->p_fd); 1401 fdrop(fp, td); 1402 if (resultfp) 1403 fdrop(fp, td); 1404 return (error); 1405 } 1406 p->p_fd->fd_ofiles[i] = fp; 1407 FILEDESC_UNLOCK(p->p_fd); 1408 if (resultfp) 1409 *resultfp = fp; 1410 if (resultfd) 1411 *resultfd = i; 1412 return (0); 1413} 1414 1415/* 1416 * Free a file descriptor. 1417 */ 1418void 1419ffree(fp) 1420 struct file *fp; 1421{ 1422 1423 KASSERT(fp->f_count == 0, ("ffree: fp_fcount not 0!")); 1424 sx_xlock(&filelist_lock); 1425 LIST_REMOVE(fp, f_list); 1426 nfiles--; 1427 sx_xunlock(&filelist_lock); 1428 crfree(fp->f_cred); 1429 uma_zfree(file_zone, fp); 1430} 1431 1432/* 1433 * Build a new filedesc structure from another. 1434 * Copy the current, root, and jail root vnode references. 1435 */ 1436struct filedesc * 1437fdinit(fdp) 1438 struct filedesc *fdp; 1439{ 1440 struct filedesc0 *newfdp; 1441 1442 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1443 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF); 1444 if (fdp != NULL) { 1445 FILEDESC_LOCK(fdp); 1446 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1447 if (newfdp->fd_fd.fd_cdir) 1448 VREF(newfdp->fd_fd.fd_cdir); 1449 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1450 if (newfdp->fd_fd.fd_rdir) 1451 VREF(newfdp->fd_fd.fd_rdir); 1452 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1453 if (newfdp->fd_fd.fd_jdir) 1454 VREF(newfdp->fd_fd.fd_jdir); 1455 FILEDESC_UNLOCK(fdp); 1456 } 1457 1458 /* Create the file descriptor table. */ 1459 newfdp->fd_fd.fd_refcnt = 1; 1460 newfdp->fd_fd.fd_cmask = CMASK; 1461 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1462 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1463 newfdp->fd_fd.fd_nfiles = NDFILE; 1464 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1465 return (&newfdp->fd_fd); 1466} 1467 1468/* 1469 * Share a filedesc structure. 1470 */ 1471struct filedesc * 1472fdshare(fdp) 1473 struct filedesc *fdp; 1474{ 1475 FILEDESC_LOCK_FAST(fdp); 1476 fdp->fd_refcnt++; 1477 FILEDESC_UNLOCK_FAST(fdp); 1478 return (fdp); 1479} 1480 1481/* 1482 * Copy a filedesc structure. 1483 * A NULL pointer in returns a NULL reference, this is to ease callers, 1484 * not catch errors. 1485 */ 1486struct filedesc * 1487fdcopy(fdp) 1488 struct filedesc *fdp; 1489{ 1490 struct filedesc *newfdp; 1491 int i; 1492 1493 /* Certain daemons might not have file descriptors. */ 1494 if (fdp == NULL) 1495 return (NULL); 1496 1497 newfdp = fdinit(fdp); 1498 FILEDESC_LOCK_FAST(fdp); 1499 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1500 FILEDESC_UNLOCK_FAST(fdp); 1501 FILEDESC_LOCK(newfdp); 1502 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1503 FILEDESC_UNLOCK(newfdp); 1504 FILEDESC_LOCK_FAST(fdp); 1505 } 1506 /* copy everything except kqueue descriptors */ 1507 newfdp->fd_freefile = -1; 1508 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1509 if (fdisused(fdp, i) && 1510 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) { 1511 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1512 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1513 fhold(newfdp->fd_ofiles[i]); 1514 newfdp->fd_lastfile = i; 1515 } else { 1516 if (newfdp->fd_freefile == -1) 1517 newfdp->fd_freefile = i; 1518 } 1519 } 1520 FILEDESC_UNLOCK_FAST(fdp); 1521 FILEDESC_LOCK(newfdp); 1522 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1523 if (newfdp->fd_ofiles[i] != NULL) 1524 fdused(newfdp, i); 1525 FILEDESC_UNLOCK(newfdp); 1526 FILEDESC_LOCK_FAST(fdp); 1527 if (newfdp->fd_freefile == -1) 1528 newfdp->fd_freefile = i; 1529 newfdp->fd_cmask = fdp->fd_cmask; 1530 FILEDESC_UNLOCK_FAST(fdp); 1531 return (newfdp); 1532} 1533 1534/* A mutex to protect the association between a proc and filedesc. */ 1535struct mtx fdesc_mtx; 1536MTX_SYSINIT(fdesc, &fdesc_mtx, "fdesc", MTX_DEF); 1537 1538/* 1539 * Release a filedesc structure. 1540 */ 1541void 1542fdfree(td) 1543 struct thread *td; 1544{ 1545 struct filedesc *fdp; 1546 struct file **fpp; 1547 int i; 1548 struct filedesc_to_leader *fdtol; 1549 struct file *fp; 1550 struct vnode *vp; 1551 struct flock lf; 1552 1553 GIANT_REQUIRED; /* VFS */ 1554 1555 /* Certain daemons might not have file descriptors. */ 1556 fdp = td->td_proc->p_fd; 1557 if (fdp == NULL) 1558 return; 1559 1560 /* Check for special need to clear POSIX style locks */ 1561 fdtol = td->td_proc->p_fdtol; 1562 if (fdtol != NULL) { 1563 FILEDESC_LOCK(fdp); 1564 KASSERT(fdtol->fdl_refcount > 0, 1565 ("filedesc_to_refcount botch: fdl_refcount=%d", 1566 fdtol->fdl_refcount)); 1567 if (fdtol->fdl_refcount == 1 && 1568 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1569 i = 0; 1570 fpp = fdp->fd_ofiles; 1571 for (i = 0, fpp = fdp->fd_ofiles; 1572 i <= fdp->fd_lastfile; 1573 i++, fpp++) { 1574 if (*fpp == NULL || 1575 (*fpp)->f_type != DTYPE_VNODE) 1576 continue; 1577 fp = *fpp; 1578 fhold(fp); 1579 FILEDESC_UNLOCK(fdp); 1580 lf.l_whence = SEEK_SET; 1581 lf.l_start = 0; 1582 lf.l_len = 0; 1583 lf.l_type = F_UNLCK; 1584 vp = fp->f_vnode; 1585 (void) VOP_ADVLOCK(vp, 1586 (caddr_t)td->td_proc-> 1587 p_leader, 1588 F_UNLCK, 1589 &lf, 1590 F_POSIX); 1591 FILEDESC_LOCK(fdp); 1592 fdrop(fp, td); 1593 fpp = fdp->fd_ofiles + i; 1594 } 1595 } 1596 retry: 1597 if (fdtol->fdl_refcount == 1) { 1598 if (fdp->fd_holdleaderscount > 0 && 1599 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1600 /* 1601 * close() or do_dup() has cleared a reference 1602 * in a shared file descriptor table. 1603 */ 1604 fdp->fd_holdleaderswakeup = 1; 1605 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx, 1606 PLOCK, "fdlhold", 0); 1607 goto retry; 1608 } 1609 if (fdtol->fdl_holdcount > 0) { 1610 /* 1611 * Ensure that fdtol->fdl_leader 1612 * remains valid in closef(). 1613 */ 1614 fdtol->fdl_wakeup = 1; 1615 msleep(fdtol, &fdp->fd_mtx, 1616 PLOCK, "fdlhold", 0); 1617 goto retry; 1618 } 1619 } 1620 fdtol->fdl_refcount--; 1621 if (fdtol->fdl_refcount == 0 && 1622 fdtol->fdl_holdcount == 0) { 1623 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1624 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1625 } else 1626 fdtol = NULL; 1627 td->td_proc->p_fdtol = NULL; 1628 FILEDESC_UNLOCK(fdp); 1629 if (fdtol != NULL) 1630 FREE(fdtol, M_FILEDESC_TO_LEADER); 1631 } 1632 FILEDESC_LOCK(fdp); 1633 if (--fdp->fd_refcnt > 0) { 1634 FILEDESC_UNLOCK(fdp); 1635 return; 1636 } 1637 1638 /* 1639 * We are the last reference to the structure, so we can 1640 * safely assume it will not change out from under us. 1641 */ 1642 FILEDESC_UNLOCK(fdp); 1643 fpp = fdp->fd_ofiles; 1644 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1645 if (*fpp) 1646 (void) closef(*fpp, td); 1647 } 1648 1649 /* XXX This should happen earlier. */ 1650 mtx_lock(&fdesc_mtx); 1651 td->td_proc->p_fd = NULL; 1652 mtx_unlock(&fdesc_mtx); 1653 1654 if (fdp->fd_nfiles > NDFILE) 1655 FREE(fdp->fd_ofiles, M_FILEDESC); 1656 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1657 FREE(fdp->fd_map, M_FILEDESC); 1658 if (fdp->fd_cdir) 1659 vrele(fdp->fd_cdir); 1660 if (fdp->fd_rdir) 1661 vrele(fdp->fd_rdir); 1662 if (fdp->fd_jdir) 1663 vrele(fdp->fd_jdir); 1664 mtx_destroy(&fdp->fd_mtx); 1665 FREE(fdp, M_FILEDESC); 1666} 1667 1668/* 1669 * For setugid programs, we don't want to people to use that setugidness 1670 * to generate error messages which write to a file which otherwise would 1671 * otherwise be off-limits to the process. We check for filesystems where 1672 * the vnode can change out from under us after execve (like [lin]procfs). 1673 * 1674 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1675 * sufficient. We also don't for check setugidness since we know we are. 1676 */ 1677static int 1678is_unsafe(struct file *fp) 1679{ 1680 if (fp->f_type == DTYPE_VNODE) { 1681 struct vnode *vp = fp->f_vnode; 1682 1683 if ((vp->v_vflag & VV_PROCDEP) != 0) 1684 return (1); 1685 } 1686 return (0); 1687} 1688 1689/* 1690 * Make this setguid thing safe, if at all possible. 1691 */ 1692void 1693setugidsafety(td) 1694 struct thread *td; 1695{ 1696 struct filedesc *fdp; 1697 int i; 1698 1699 /* Certain daemons might not have file descriptors. */ 1700 fdp = td->td_proc->p_fd; 1701 if (fdp == NULL) 1702 return; 1703 1704 /* 1705 * Note: fdp->fd_ofiles may be reallocated out from under us while 1706 * we are blocked in a close. Be careful! 1707 */ 1708 FILEDESC_LOCK(fdp); 1709 for (i = 0; i <= fdp->fd_lastfile; i++) { 1710 if (i > 2) 1711 break; 1712 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1713 struct file *fp; 1714 1715 knote_fdclose(td, i); 1716 /* 1717 * NULL-out descriptor prior to close to avoid 1718 * a race while close blocks. 1719 */ 1720 fp = fdp->fd_ofiles[i]; 1721 fdp->fd_ofiles[i] = NULL; 1722 fdp->fd_ofileflags[i] = 0; 1723 fdunused(fdp, i); 1724 FILEDESC_UNLOCK(fdp); 1725 (void) closef(fp, td); 1726 FILEDESC_LOCK(fdp); 1727 } 1728 } 1729 FILEDESC_UNLOCK(fdp); 1730} 1731 1732void 1733fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 1734{ 1735 1736 FILEDESC_LOCK(fdp); 1737 if (fdp->fd_ofiles[idx] == fp) { 1738 fdp->fd_ofiles[idx] = NULL; 1739 fdunused(fdp, idx); 1740 FILEDESC_UNLOCK(fdp); 1741 fdrop(fp, td); 1742 } else { 1743 FILEDESC_UNLOCK(fdp); 1744 } 1745} 1746 1747/* 1748 * Close any files on exec? 1749 */ 1750void 1751fdcloseexec(td) 1752 struct thread *td; 1753{ 1754 struct filedesc *fdp; 1755 int i; 1756 1757 /* Certain daemons might not have file descriptors. */ 1758 fdp = td->td_proc->p_fd; 1759 if (fdp == NULL) 1760 return; 1761 1762 FILEDESC_LOCK(fdp); 1763 1764 /* 1765 * We cannot cache fd_ofiles or fd_ofileflags since operations 1766 * may block and rip them out from under us. 1767 */ 1768 for (i = 0; i <= fdp->fd_lastfile; i++) { 1769 if (fdp->fd_ofiles[i] != NULL && 1770 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) { 1771 struct file *fp; 1772 1773 knote_fdclose(td, i); 1774 /* 1775 * NULL-out descriptor prior to close to avoid 1776 * a race while close blocks. 1777 */ 1778 fp = fdp->fd_ofiles[i]; 1779 fdp->fd_ofiles[i] = NULL; 1780 fdp->fd_ofileflags[i] = 0; 1781 fdunused(fdp, i); 1782 FILEDESC_UNLOCK(fdp); 1783 (void) closef(fp, td); 1784 FILEDESC_LOCK(fdp); 1785 } 1786 } 1787 FILEDESC_UNLOCK(fdp); 1788} 1789 1790/* 1791 * It is unsafe for set[ug]id processes to be started with file 1792 * descriptors 0..2 closed, as these descriptors are given implicit 1793 * significance in the Standard C library. fdcheckstd() will create a 1794 * descriptor referencing /dev/null for each of stdin, stdout, and 1795 * stderr that is not already open. 1796 */ 1797int 1798fdcheckstd(td) 1799 struct thread *td; 1800{ 1801 struct nameidata nd; 1802 struct filedesc *fdp; 1803 struct file *fp; 1804 register_t retval; 1805 int fd, i, error, flags, devnull; 1806 1807 GIANT_REQUIRED; /* VFS */ 1808 1809 fdp = td->td_proc->p_fd; 1810 if (fdp == NULL) 1811 return (0); 1812 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 1813 devnull = -1; 1814 error = 0; 1815 for (i = 0; i < 3; i++) { 1816 if (fdp->fd_ofiles[i] != NULL) 1817 continue; 1818 if (devnull < 0) { 1819 error = falloc(td, &fp, &fd); 1820 if (error != 0) 1821 break; 1822 /* Note extra ref on `fp' held for us by falloc(). */ 1823 KASSERT(fd == i, ("oof, we didn't get our fd")); 1824 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null", 1825 td); 1826 flags = FREAD | FWRITE; 1827 error = vn_open(&nd, &flags, 0, -1); 1828 if (error != 0) { 1829 /* 1830 * Someone may have closed the entry in the 1831 * file descriptor table, so check it hasn't 1832 * changed before dropping the reference count. 1833 */ 1834 FILEDESC_LOCK(fdp); 1835 KASSERT(fdp->fd_ofiles[fd] == fp, 1836 ("table not shared, how did it change?")); 1837 fdp->fd_ofiles[fd] = NULL; 1838 fdunused(fdp, fd); 1839 FILEDESC_UNLOCK(fdp); 1840 fdrop(fp, td); 1841 fdrop(fp, td); 1842 break; 1843 } 1844 NDFREE(&nd, NDF_ONLY_PNBUF); 1845 fp->f_flag = flags; 1846 fp->f_vnode = nd.ni_vp; 1847 if (fp->f_data == NULL) 1848 fp->f_data = nd.ni_vp; 1849 if (fp->f_ops == &badfileops) 1850 fp->f_ops = &vnops; 1851 fp->f_type = DTYPE_VNODE; 1852 VOP_UNLOCK(nd.ni_vp, 0, td); 1853 devnull = fd; 1854 fdrop(fp, td); 1855 } else { 1856 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 1857 if (error != 0) 1858 break; 1859 } 1860 } 1861 return (error); 1862} 1863 1864/* 1865 * Internal form of close. 1866 * Decrement reference count on file structure. 1867 * Note: td may be NULL when closing a file 1868 * that was being passed in a message. 1869 */ 1870int 1871closef(fp, td) 1872 struct file *fp; 1873 struct thread *td; 1874{ 1875 struct vnode *vp; 1876 struct flock lf; 1877 struct filedesc_to_leader *fdtol; 1878 struct filedesc *fdp; 1879 1880 if (fp == NULL) 1881 return (0); 1882 /* 1883 * POSIX record locking dictates that any close releases ALL 1884 * locks owned by this process. This is handled by setting 1885 * a flag in the unlock to free ONLY locks obeying POSIX 1886 * semantics, and not to free BSD-style file locks. 1887 * If the descriptor was in a message, POSIX-style locks 1888 * aren't passed with the descriptor. 1889 */ 1890 if (td != NULL && 1891 fp->f_type == DTYPE_VNODE) { 1892 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1893 lf.l_whence = SEEK_SET; 1894 lf.l_start = 0; 1895 lf.l_len = 0; 1896 lf.l_type = F_UNLCK; 1897 vp = fp->f_vnode; 1898 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 1899 F_UNLCK, &lf, F_POSIX); 1900 } 1901 fdtol = td->td_proc->p_fdtol; 1902 if (fdtol != NULL) { 1903 /* 1904 * Handle special case where file descriptor table 1905 * is shared between multiple process leaders. 1906 */ 1907 fdp = td->td_proc->p_fd; 1908 FILEDESC_LOCK(fdp); 1909 for (fdtol = fdtol->fdl_next; 1910 fdtol != td->td_proc->p_fdtol; 1911 fdtol = fdtol->fdl_next) { 1912 if ((fdtol->fdl_leader->p_flag & 1913 P_ADVLOCK) == 0) 1914 continue; 1915 fdtol->fdl_holdcount++; 1916 FILEDESC_UNLOCK(fdp); 1917 lf.l_whence = SEEK_SET; 1918 lf.l_start = 0; 1919 lf.l_len = 0; 1920 lf.l_type = F_UNLCK; 1921 vp = fp->f_vnode; 1922 (void) VOP_ADVLOCK(vp, 1923 (caddr_t)fdtol->fdl_leader, 1924 F_UNLCK, &lf, F_POSIX); 1925 FILEDESC_LOCK(fdp); 1926 fdtol->fdl_holdcount--; 1927 if (fdtol->fdl_holdcount == 0 && 1928 fdtol->fdl_wakeup != 0) { 1929 fdtol->fdl_wakeup = 0; 1930 wakeup(fdtol); 1931 } 1932 } 1933 FILEDESC_UNLOCK(fdp); 1934 } 1935 } 1936 return (fdrop(fp, td)); 1937} 1938 1939/* 1940 * Drop reference on struct file passed in, may call closef if the 1941 * reference hits zero. 1942 */ 1943int 1944fdrop(fp, td) 1945 struct file *fp; 1946 struct thread *td; 1947{ 1948 1949 FILE_LOCK(fp); 1950 return (fdrop_locked(fp, td)); 1951} 1952 1953/* 1954 * Extract the file pointer associated with the specified descriptor for 1955 * the current user process. 1956 * 1957 * If the descriptor doesn't exist, EBADF is returned. 1958 * 1959 * If the descriptor exists but doesn't match 'flags' then 1960 * return EBADF for read attempts and EINVAL for write attempts. 1961 * 1962 * If 'hold' is set (non-zero) the file's refcount will be bumped on return. 1963 * It should be droped with fdrop(). 1964 * If it is not set, then the refcount will not be bumped however the 1965 * thread's filedesc struct will be returned locked (for fgetsock). 1966 * 1967 * If an error occured the non-zero error is returned and *fpp is set to NULL. 1968 * Otherwise *fpp is set and zero is returned. 1969 */ 1970static __inline int 1971_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold) 1972{ 1973 struct filedesc *fdp; 1974 struct file *fp; 1975 1976 *fpp = NULL; 1977 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 1978 return (EBADF); 1979 FILEDESC_LOCK(fdp); 1980 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) { 1981 FILEDESC_UNLOCK(fdp); 1982 return (EBADF); 1983 } 1984 1985 /* 1986 * Note: FREAD failures returns EBADF to maintain backwards 1987 * compatibility with what routines returned before. 1988 * 1989 * Only one flag, or 0, may be specified. 1990 */ 1991 if (flags == FREAD && (fp->f_flag & FREAD) == 0) { 1992 FILEDESC_UNLOCK(fdp); 1993 return (EBADF); 1994 } 1995 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) { 1996 FILEDESC_UNLOCK(fdp); 1997 return (EINVAL); 1998 } 1999 if (hold) { 2000 fhold(fp); 2001 FILEDESC_UNLOCK(fdp); 2002 } 2003 *fpp = fp; 2004 return (0); 2005} 2006 2007int 2008fget(struct thread *td, int fd, struct file **fpp) 2009{ 2010 2011 return(_fget(td, fd, fpp, 0, 1)); 2012} 2013 2014int 2015fget_read(struct thread *td, int fd, struct file **fpp) 2016{ 2017 2018 return(_fget(td, fd, fpp, FREAD, 1)); 2019} 2020 2021int 2022fget_write(struct thread *td, int fd, struct file **fpp) 2023{ 2024 2025 return(_fget(td, fd, fpp, FWRITE, 1)); 2026} 2027 2028/* 2029 * Like fget() but loads the underlying vnode, or returns an error if 2030 * the descriptor does not represent a vnode. Note that pipes use vnodes 2031 * but never have VM objects (so VOP_GETVOBJECT() calls will return an 2032 * error). The returned vnode will be vref()d. 2033 */ 2034static __inline int 2035_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags) 2036{ 2037 struct file *fp; 2038 int error; 2039 2040 GIANT_REQUIRED; /* VFS */ 2041 2042 *vpp = NULL; 2043 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2044 return (error); 2045 if (fp->f_vnode == NULL) { 2046 error = EINVAL; 2047 } else { 2048 *vpp = fp->f_vnode; 2049 vref(*vpp); 2050 } 2051 FILEDESC_UNLOCK(td->td_proc->p_fd); 2052 return (error); 2053} 2054 2055int 2056fgetvp(struct thread *td, int fd, struct vnode **vpp) 2057{ 2058 2059 return (_fgetvp(td, fd, vpp, 0)); 2060} 2061 2062int 2063fgetvp_read(struct thread *td, int fd, struct vnode **vpp) 2064{ 2065 2066 return (_fgetvp(td, fd, vpp, FREAD)); 2067} 2068 2069int 2070fgetvp_write(struct thread *td, int fd, struct vnode **vpp) 2071{ 2072 2073 return (_fgetvp(td, fd, vpp, FWRITE)); 2074} 2075 2076/* 2077 * Like fget() but loads the underlying socket, or returns an error if 2078 * the descriptor does not represent a socket. 2079 * 2080 * We bump the ref count on the returned socket. XXX Also obtain the SX 2081 * lock in the future. 2082 */ 2083int 2084fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp) 2085{ 2086 struct file *fp; 2087 int error; 2088 2089 NET_ASSERT_GIANT(); 2090 2091 *spp = NULL; 2092 if (fflagp != NULL) 2093 *fflagp = 0; 2094 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2095 return (error); 2096 if (fp->f_type != DTYPE_SOCKET) { 2097 error = ENOTSOCK; 2098 } else { 2099 *spp = fp->f_data; 2100 if (fflagp) 2101 *fflagp = fp->f_flag; 2102 SOCK_LOCK(*spp); 2103 soref(*spp); 2104 SOCK_UNLOCK(*spp); 2105 } 2106 FILEDESC_UNLOCK(td->td_proc->p_fd); 2107 return (error); 2108} 2109 2110/* 2111 * Drop the reference count on the the socket and XXX release the SX lock in 2112 * the future. The last reference closes the socket. 2113 */ 2114void 2115fputsock(struct socket *so) 2116{ 2117 2118 NET_ASSERT_GIANT(); 2119 ACCEPT_LOCK(); 2120 SOCK_LOCK(so); 2121 sorele(so); 2122} 2123 2124/* 2125 * Drop reference on struct file passed in, may call closef if the 2126 * reference hits zero. 2127 * Expects struct file locked, and will unlock it. 2128 */ 2129int 2130fdrop_locked(fp, td) 2131 struct file *fp; 2132 struct thread *td; 2133{ 2134 int error; 2135 2136 FILE_LOCK_ASSERT(fp, MA_OWNED); 2137 2138 if (--fp->f_count > 0) { 2139 FILE_UNLOCK(fp); 2140 return (0); 2141 } 2142 /* We have the last ref so we can proceed without the file lock. */ 2143 FILE_UNLOCK(fp); 2144 if (fp->f_count < 0) 2145 panic("fdrop: count < 0"); 2146 if (fp->f_ops != &badfileops) 2147 error = fo_close(fp, td); 2148 else 2149 error = 0; 2150 ffree(fp); 2151 return (error); 2152} 2153 2154/* 2155 * Apply an advisory lock on a file descriptor. 2156 * 2157 * Just attempt to get a record lock of the requested type on 2158 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2159 */ 2160#ifndef _SYS_SYSPROTO_H_ 2161struct flock_args { 2162 int fd; 2163 int how; 2164}; 2165#endif 2166/* 2167 * MPSAFE 2168 */ 2169/* ARGSUSED */ 2170int 2171flock(td, uap) 2172 struct thread *td; 2173 struct flock_args *uap; 2174{ 2175 struct file *fp; 2176 struct vnode *vp; 2177 struct flock lf; 2178 int error; 2179 2180 if ((error = fget(td, uap->fd, &fp)) != 0) 2181 return (error); 2182 if (fp->f_type != DTYPE_VNODE) { 2183 fdrop(fp, td); 2184 return (EOPNOTSUPP); 2185 } 2186 2187 mtx_lock(&Giant); 2188 vp = fp->f_vnode; 2189 lf.l_whence = SEEK_SET; 2190 lf.l_start = 0; 2191 lf.l_len = 0; 2192 if (uap->how & LOCK_UN) { 2193 lf.l_type = F_UNLCK; 2194 FILE_LOCK(fp); 2195 fp->f_flag &= ~FHASLOCK; 2196 FILE_UNLOCK(fp); 2197 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2198 goto done2; 2199 } 2200 if (uap->how & LOCK_EX) 2201 lf.l_type = F_WRLCK; 2202 else if (uap->how & LOCK_SH) 2203 lf.l_type = F_RDLCK; 2204 else { 2205 error = EBADF; 2206 goto done2; 2207 } 2208 FILE_LOCK(fp); 2209 fp->f_flag |= FHASLOCK; 2210 FILE_UNLOCK(fp); 2211 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2212 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2213done2: 2214 fdrop(fp, td); 2215 mtx_unlock(&Giant); 2216 return (error); 2217} 2218 2219/* 2220 * File Descriptor pseudo-device driver (/dev/fd/). 2221 * 2222 * Opening minor device N dup()s the file (if any) connected to file 2223 * descriptor N belonging to the calling process. Note that this driver 2224 * consists of only the ``open()'' routine, because all subsequent 2225 * references to this file will be direct to the other driver. 2226 */ 2227/* ARGSUSED */ 2228static int 2229fdopen(dev, mode, type, td) 2230 struct cdev *dev; 2231 int mode, type; 2232 struct thread *td; 2233{ 2234 2235 /* 2236 * XXX Kludge: set curthread->td_dupfd to contain the value of the 2237 * the file descriptor being sought for duplication. The error 2238 * return ensures that the vnode for this device will be released 2239 * by vn_open. Open will detect this special error and take the 2240 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2241 * will simply report the error. 2242 */ 2243 td->td_dupfd = dev2unit(dev); 2244 return (ENODEV); 2245} 2246 2247/* 2248 * Duplicate the specified descriptor to a free descriptor. 2249 */ 2250int 2251dupfdopen(td, fdp, indx, dfd, mode, error) 2252 struct thread *td; 2253 struct filedesc *fdp; 2254 int indx, dfd; 2255 int mode; 2256 int error; 2257{ 2258 struct file *wfp; 2259 struct file *fp; 2260 2261 /* 2262 * If the to-be-dup'd fd number is greater than the allowed number 2263 * of file descriptors, or the fd to be dup'd has already been 2264 * closed, then reject. 2265 */ 2266 FILEDESC_LOCK(fdp); 2267 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2268 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2269 FILEDESC_UNLOCK(fdp); 2270 return (EBADF); 2271 } 2272 2273 /* 2274 * There are two cases of interest here. 2275 * 2276 * For ENODEV simply dup (dfd) to file descriptor 2277 * (indx) and return. 2278 * 2279 * For ENXIO steal away the file structure from (dfd) and 2280 * store it in (indx). (dfd) is effectively closed by 2281 * this operation. 2282 * 2283 * Any other error code is just returned. 2284 */ 2285 switch (error) { 2286 case ENODEV: 2287 /* 2288 * Check that the mode the file is being opened for is a 2289 * subset of the mode of the existing descriptor. 2290 */ 2291 FILE_LOCK(wfp); 2292 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2293 FILE_UNLOCK(wfp); 2294 FILEDESC_UNLOCK(fdp); 2295 return (EACCES); 2296 } 2297 fp = fdp->fd_ofiles[indx]; 2298 fdp->fd_ofiles[indx] = wfp; 2299 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2300 if (fp == NULL) 2301 fdused(fdp, indx); 2302 fhold_locked(wfp); 2303 FILE_UNLOCK(wfp); 2304 if (fp != NULL) 2305 FILE_LOCK(fp); 2306 FILEDESC_UNLOCK(fdp); 2307 /* 2308 * We now own the reference to fp that the ofiles[] array 2309 * used to own. Release it. 2310 */ 2311 if (fp != NULL) 2312 fdrop_locked(fp, td); 2313 return (0); 2314 2315 case ENXIO: 2316 /* 2317 * Steal away the file pointer from dfd and stuff it into indx. 2318 */ 2319 fp = fdp->fd_ofiles[indx]; 2320 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2321 fdp->fd_ofiles[dfd] = NULL; 2322 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2323 fdp->fd_ofileflags[dfd] = 0; 2324 fdunused(fdp, dfd); 2325 if (fp == NULL) 2326 fdused(fdp, indx); 2327 if (fp != NULL) 2328 FILE_LOCK(fp); 2329 FILEDESC_UNLOCK(fdp); 2330 2331 /* 2332 * we now own the reference to fp that the ofiles[] array 2333 * used to own. Release it. 2334 */ 2335 if (fp != NULL) 2336 fdrop_locked(fp, td); 2337 return (0); 2338 2339 default: 2340 FILEDESC_UNLOCK(fdp); 2341 return (error); 2342 } 2343 /* NOTREACHED */ 2344} 2345 2346struct filedesc_to_leader * 2347filedesc_to_leader_alloc(struct filedesc_to_leader *old, 2348 struct filedesc *fdp, 2349 struct proc *leader) 2350{ 2351 struct filedesc_to_leader *fdtol; 2352 2353 MALLOC(fdtol, struct filedesc_to_leader *, 2354 sizeof(struct filedesc_to_leader), 2355 M_FILEDESC_TO_LEADER, 2356 M_WAITOK); 2357 fdtol->fdl_refcount = 1; 2358 fdtol->fdl_holdcount = 0; 2359 fdtol->fdl_wakeup = 0; 2360 fdtol->fdl_leader = leader; 2361 if (old != NULL) { 2362 FILEDESC_LOCK(fdp); 2363 fdtol->fdl_next = old->fdl_next; 2364 fdtol->fdl_prev = old; 2365 old->fdl_next = fdtol; 2366 fdtol->fdl_next->fdl_prev = fdtol; 2367 FILEDESC_UNLOCK(fdp); 2368 } else { 2369 fdtol->fdl_next = fdtol; 2370 fdtol->fdl_prev = fdtol; 2371 } 2372 return (fdtol); 2373} 2374 2375/* 2376 * Get file structures. 2377 */ 2378static int 2379sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2380{ 2381 struct xfile xf; 2382 struct filedesc *fdp; 2383 struct file *fp; 2384 struct proc *p; 2385 int error, n; 2386 2387 /* 2388 * Note: because the number of file descriptors is calculated 2389 * in different ways for sizing vs returning the data, 2390 * there is information leakage from the first loop. However, 2391 * it is of a similar order of magnitude to the leakage from 2392 * global system statistics such as kern.openfiles. 2393 */ 2394 error = sysctl_wire_old_buffer(req, 0); 2395 if (error != 0) 2396 return (error); 2397 if (req->oldptr == NULL) { 2398 n = 16; /* A slight overestimate. */ 2399 sx_slock(&filelist_lock); 2400 LIST_FOREACH(fp, &filehead, f_list) { 2401 /* 2402 * We should grab the lock, but this is an 2403 * estimate, so does it really matter? 2404 */ 2405 /* mtx_lock(fp->f_mtxp); */ 2406 n += fp->f_count; 2407 /* mtx_unlock(f->f_mtxp); */ 2408 } 2409 sx_sunlock(&filelist_lock); 2410 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2411 } 2412 error = 0; 2413 bzero(&xf, sizeof(xf)); 2414 xf.xf_size = sizeof(xf); 2415 sx_slock(&allproc_lock); 2416 LIST_FOREACH(p, &allproc, p_list) { 2417 if (p->p_state == PRS_NEW) 2418 continue; 2419 PROC_LOCK(p); 2420 if (p_cansee(req->td, p) != 0) { 2421 PROC_UNLOCK(p); 2422 continue; 2423 } 2424 xf.xf_pid = p->p_pid; 2425 xf.xf_uid = p->p_ucred->cr_uid; 2426 PROC_UNLOCK(p); 2427 mtx_lock(&fdesc_mtx); 2428 if ((fdp = p->p_fd) == NULL) { 2429 mtx_unlock(&fdesc_mtx); 2430 continue; 2431 } 2432 FILEDESC_LOCK(fdp); 2433 for (n = 0; n < fdp->fd_nfiles; ++n) { 2434 if ((fp = fdp->fd_ofiles[n]) == NULL) 2435 continue; 2436 xf.xf_fd = n; 2437 xf.xf_file = fp; 2438 xf.xf_data = fp->f_data; 2439 xf.xf_vnode = fp->f_vnode; 2440 xf.xf_type = fp->f_type; 2441 xf.xf_count = fp->f_count; 2442 xf.xf_msgcount = fp->f_msgcount; 2443 xf.xf_offset = fp->f_offset; 2444 xf.xf_flag = fp->f_flag; 2445 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2446 if (error) 2447 break; 2448 } 2449 FILEDESC_UNLOCK(fdp); 2450 mtx_unlock(&fdesc_mtx); 2451 if (error) 2452 break; 2453 } 2454 sx_sunlock(&allproc_lock); 2455 return (error); 2456} 2457 2458SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2459 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2460 2461SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2462 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2463 2464SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2465 &maxfiles, 0, "Maximum number of files"); 2466 2467SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2468 &nfiles, 0, "System-wide number of open files"); 2469 2470static void 2471fildesc_drvinit(void *unused) 2472{ 2473 struct cdev *dev; 2474 2475 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0"); 2476 make_dev_alias(dev, "stdin"); 2477 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1"); 2478 make_dev_alias(dev, "stdout"); 2479 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2"); 2480 make_dev_alias(dev, "stderr"); 2481} 2482 2483static fo_rdwr_t badfo_readwrite; 2484static fo_ioctl_t badfo_ioctl; 2485static fo_poll_t badfo_poll; 2486static fo_kqfilter_t badfo_kqfilter; 2487static fo_stat_t badfo_stat; 2488static fo_close_t badfo_close; 2489 2490struct fileops badfileops = { 2491 .fo_read = badfo_readwrite, 2492 .fo_write = badfo_readwrite, 2493 .fo_ioctl = badfo_ioctl, 2494 .fo_poll = badfo_poll, 2495 .fo_kqfilter = badfo_kqfilter, 2496 .fo_stat = badfo_stat, 2497 .fo_close = badfo_close, 2498}; 2499 2500static int 2501badfo_readwrite(fp, uio, active_cred, flags, td) 2502 struct file *fp; 2503 struct uio *uio; 2504 struct ucred *active_cred; 2505 struct thread *td; 2506 int flags; 2507{ 2508 2509 return (EBADF); 2510} 2511 2512static int 2513badfo_ioctl(fp, com, data, active_cred, td) 2514 struct file *fp; 2515 u_long com; 2516 void *data; 2517 struct ucred *active_cred; 2518 struct thread *td; 2519{ 2520 2521 return (EBADF); 2522} 2523 2524static int 2525badfo_poll(fp, events, active_cred, td) 2526 struct file *fp; 2527 int events; 2528 struct ucred *active_cred; 2529 struct thread *td; 2530{ 2531 2532 return (0); 2533} 2534 2535static int 2536badfo_kqfilter(fp, kn) 2537 struct file *fp; 2538 struct knote *kn; 2539{ 2540 2541 return (0); 2542} 2543 2544static int 2545badfo_stat(fp, sb, active_cred, td) 2546 struct file *fp; 2547 struct stat *sb; 2548 struct ucred *active_cred; 2549 struct thread *td; 2550{ 2551 2552 return (EBADF); 2553} 2554 2555static int 2556badfo_close(fp, td) 2557 struct file *fp; 2558 struct thread *td; 2559{ 2560 2561 return (EBADF); 2562} 2563 2564SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, 2565 fildesc_drvinit,NULL) 2566 2567static void filelistinit(void *); 2568SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL) 2569 2570/* ARGSUSED*/ 2571static void 2572filelistinit(dummy) 2573 void *dummy; 2574{ 2575 2576 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 2577 NULL, NULL, UMA_ALIGN_PTR, 0); 2578 sx_init(&filelist_lock, "filelist lock"); 2579 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 2580} 2581