kern_descrip.c revision 186601
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 186601 2008-12-30 12:51:56Z kib $"); 39 40#include "opt_compat.h" 41#include "opt_ddb.h" 42#include "opt_ktrace.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46 47#include <sys/conf.h> 48#include <sys/domain.h> 49#include <sys/fcntl.h> 50#include <sys/file.h> 51#include <sys/filedesc.h> 52#include <sys/filio.h> 53#include <sys/jail.h> 54#include <sys/kernel.h> 55#include <sys/limits.h> 56#include <sys/lock.h> 57#include <sys/malloc.h> 58#include <sys/mount.h> 59#include <sys/mqueue.h> 60#include <sys/mutex.h> 61#include <sys/namei.h> 62#include <sys/priv.h> 63#include <sys/proc.h> 64#include <sys/protosw.h> 65#include <sys/resourcevar.h> 66#include <sys/signalvar.h> 67#include <sys/socketvar.h> 68#include <sys/stat.h> 69#include <sys/sx.h> 70#include <sys/syscallsubr.h> 71#include <sys/sysctl.h> 72#include <sys/sysproto.h> 73#include <sys/tty.h> 74#include <sys/unistd.h> 75#include <sys/user.h> 76#include <sys/vnode.h> 77#ifdef KTRACE 78#include <sys/ktrace.h> 79#endif 80 81#include <security/audit/audit.h> 82 83#include <vm/uma.h> 84 85#include <ddb/ddb.h> 86 87static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 88static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 89 "file desc to leader structures"); 90static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 91 92static uma_zone_t file_zone; 93 94 95/* Flags for do_dup() */ 96#define DUP_FIXED 0x1 /* Force fixed allocation */ 97#define DUP_FCNTL 0x2 /* fcntl()-style errors */ 98 99static int do_dup(struct thread *td, int flags, int old, int new, 100 register_t *retval); 101static int fd_first_free(struct filedesc *, int, int); 102static int fd_last_used(struct filedesc *, int, int); 103static void fdgrowtable(struct filedesc *, int); 104static void fdunused(struct filedesc *fdp, int fd); 105static void fdused(struct filedesc *fdp, int fd); 106 107/* 108 * A process is initially started out with NDFILE descriptors stored within 109 * this structure, selected to be enough for typical applications based on 110 * the historical limit of 20 open files (and the usage of descriptors by 111 * shells). If these descriptors are exhausted, a larger descriptor table 112 * may be allocated, up to a process' resource limit; the internal arrays 113 * are then unused. 114 */ 115#define NDFILE 20 116#define NDSLOTSIZE sizeof(NDSLOTTYPE) 117#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 118#define NDSLOT(x) ((x) / NDENTRIES) 119#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 120#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 121 122/* 123 * Storage required per open file descriptor. 124 */ 125#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 126 127/* 128 * Basic allocation of descriptors: 129 * one of the above, plus arrays for NDFILE descriptors. 130 */ 131struct filedesc0 { 132 struct filedesc fd_fd; 133 /* 134 * These arrays are used when the number of open files is 135 * <= NDFILE, and are then pointed to by the pointers above. 136 */ 137 struct file *fd_dfiles[NDFILE]; 138 char fd_dfileflags[NDFILE]; 139 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 140}; 141 142/* 143 * Descriptor management. 144 */ 145volatile int openfiles; /* actual number of open files */ 146struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 147void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 148 149/* A mutex to protect the association between a proc and filedesc. */ 150static struct mtx fdesc_mtx; 151 152/* 153 * Find the first zero bit in the given bitmap, starting at low and not 154 * exceeding size - 1. 155 */ 156static int 157fd_first_free(struct filedesc *fdp, int low, int size) 158{ 159 NDSLOTTYPE *map = fdp->fd_map; 160 NDSLOTTYPE mask; 161 int off, maxoff; 162 163 if (low >= size) 164 return (low); 165 166 off = NDSLOT(low); 167 if (low % NDENTRIES) { 168 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 169 if ((mask &= ~map[off]) != 0UL) 170 return (off * NDENTRIES + ffsl(mask) - 1); 171 ++off; 172 } 173 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 174 if (map[off] != ~0UL) 175 return (off * NDENTRIES + ffsl(~map[off]) - 1); 176 return (size); 177} 178 179/* 180 * Find the highest non-zero bit in the given bitmap, starting at low and 181 * not exceeding size - 1. 182 */ 183static int 184fd_last_used(struct filedesc *fdp, int low, int size) 185{ 186 NDSLOTTYPE *map = fdp->fd_map; 187 NDSLOTTYPE mask; 188 int off, minoff; 189 190 if (low >= size) 191 return (-1); 192 193 off = NDSLOT(size); 194 if (size % NDENTRIES) { 195 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 196 if ((mask &= map[off]) != 0) 197 return (off * NDENTRIES + flsl(mask) - 1); 198 --off; 199 } 200 for (minoff = NDSLOT(low); off >= minoff; --off) 201 if (map[off] != 0) 202 return (off * NDENTRIES + flsl(map[off]) - 1); 203 return (low - 1); 204} 205 206static int 207fdisused(struct filedesc *fdp, int fd) 208{ 209 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 210 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 211 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 212} 213 214/* 215 * Mark a file descriptor as used. 216 */ 217static void 218fdused(struct filedesc *fdp, int fd) 219{ 220 221 FILEDESC_XLOCK_ASSERT(fdp); 222 KASSERT(!fdisused(fdp, fd), 223 ("fd already used")); 224 225 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 226 if (fd > fdp->fd_lastfile) 227 fdp->fd_lastfile = fd; 228 if (fd == fdp->fd_freefile) 229 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 230} 231 232/* 233 * Mark a file descriptor as unused. 234 */ 235static void 236fdunused(struct filedesc *fdp, int fd) 237{ 238 239 FILEDESC_XLOCK_ASSERT(fdp); 240 KASSERT(fdisused(fdp, fd), 241 ("fd is already unused")); 242 KASSERT(fdp->fd_ofiles[fd] == NULL, 243 ("fd is still in use")); 244 245 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 246 if (fd < fdp->fd_freefile) 247 fdp->fd_freefile = fd; 248 if (fd == fdp->fd_lastfile) 249 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 250} 251 252/* 253 * System calls on descriptors. 254 */ 255#ifndef _SYS_SYSPROTO_H_ 256struct getdtablesize_args { 257 int dummy; 258}; 259#endif 260/* ARGSUSED */ 261int 262getdtablesize(struct thread *td, struct getdtablesize_args *uap) 263{ 264 struct proc *p = td->td_proc; 265 266 PROC_LOCK(p); 267 td->td_retval[0] = 268 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 269 PROC_UNLOCK(p); 270 return (0); 271} 272 273/* 274 * Duplicate a file descriptor to a particular value. 275 * 276 * Note: keep in mind that a potential race condition exists when closing 277 * descriptors from a shared descriptor table (via rfork). 278 */ 279#ifndef _SYS_SYSPROTO_H_ 280struct dup2_args { 281 u_int from; 282 u_int to; 283}; 284#endif 285/* ARGSUSED */ 286int 287dup2(struct thread *td, struct dup2_args *uap) 288{ 289 290 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 291 td->td_retval)); 292} 293 294/* 295 * Duplicate a file descriptor. 296 */ 297#ifndef _SYS_SYSPROTO_H_ 298struct dup_args { 299 u_int fd; 300}; 301#endif 302/* ARGSUSED */ 303int 304dup(struct thread *td, struct dup_args *uap) 305{ 306 307 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 308} 309 310/* 311 * The file control system call. 312 */ 313#ifndef _SYS_SYSPROTO_H_ 314struct fcntl_args { 315 int fd; 316 int cmd; 317 long arg; 318}; 319#endif 320/* ARGSUSED */ 321int 322fcntl(struct thread *td, struct fcntl_args *uap) 323{ 324 struct flock fl; 325 struct oflock ofl; 326 intptr_t arg; 327 int error; 328 int cmd; 329 330 error = 0; 331 cmd = uap->cmd; 332 switch (uap->cmd) { 333 case F_OGETLK: 334 case F_OSETLK: 335 case F_OSETLKW: 336 /* 337 * Convert old flock structure to new. 338 */ 339 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 340 fl.l_start = ofl.l_start; 341 fl.l_len = ofl.l_len; 342 fl.l_pid = ofl.l_pid; 343 fl.l_type = ofl.l_type; 344 fl.l_whence = ofl.l_whence; 345 fl.l_sysid = 0; 346 347 switch (uap->cmd) { 348 case F_OGETLK: 349 cmd = F_GETLK; 350 break; 351 case F_OSETLK: 352 cmd = F_SETLK; 353 break; 354 case F_OSETLKW: 355 cmd = F_SETLKW; 356 break; 357 } 358 arg = (intptr_t)&fl; 359 break; 360 case F_GETLK: 361 case F_SETLK: 362 case F_SETLKW: 363 case F_SETLK_REMOTE: 364 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 365 arg = (intptr_t)&fl; 366 break; 367 default: 368 arg = uap->arg; 369 break; 370 } 371 if (error) 372 return (error); 373 error = kern_fcntl(td, uap->fd, cmd, arg); 374 if (error) 375 return (error); 376 if (uap->cmd == F_OGETLK) { 377 ofl.l_start = fl.l_start; 378 ofl.l_len = fl.l_len; 379 ofl.l_pid = fl.l_pid; 380 ofl.l_type = fl.l_type; 381 ofl.l_whence = fl.l_whence; 382 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 383 } else if (uap->cmd == F_GETLK) { 384 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 385 } 386 return (error); 387} 388 389static inline struct file * 390fdtofp(int fd, struct filedesc *fdp) 391{ 392 struct file *fp; 393 394 FILEDESC_LOCK_ASSERT(fdp); 395 if ((unsigned)fd >= fdp->fd_nfiles || 396 (fp = fdp->fd_ofiles[fd]) == NULL) 397 return (NULL); 398 return (fp); 399} 400 401int 402kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 403{ 404 struct filedesc *fdp; 405 struct flock *flp; 406 struct file *fp; 407 struct proc *p; 408 char *pop; 409 struct vnode *vp; 410 int error, flg, tmp; 411 int vfslocked; 412 413 vfslocked = 0; 414 error = 0; 415 flg = F_POSIX; 416 p = td->td_proc; 417 fdp = p->p_fd; 418 419 switch (cmd) { 420 case F_DUPFD: 421 tmp = arg; 422 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 423 break; 424 425 case F_DUP2FD: 426 tmp = arg; 427 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 428 break; 429 430 case F_GETFD: 431 FILEDESC_SLOCK(fdp); 432 if ((fp = fdtofp(fd, fdp)) == NULL) { 433 FILEDESC_SUNLOCK(fdp); 434 error = EBADF; 435 break; 436 } 437 pop = &fdp->fd_ofileflags[fd]; 438 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 439 FILEDESC_SUNLOCK(fdp); 440 break; 441 442 case F_SETFD: 443 FILEDESC_XLOCK(fdp); 444 if ((fp = fdtofp(fd, fdp)) == NULL) { 445 FILEDESC_XUNLOCK(fdp); 446 error = EBADF; 447 break; 448 } 449 pop = &fdp->fd_ofileflags[fd]; 450 *pop = (*pop &~ UF_EXCLOSE) | 451 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 452 FILEDESC_XUNLOCK(fdp); 453 break; 454 455 case F_GETFL: 456 FILEDESC_SLOCK(fdp); 457 if ((fp = fdtofp(fd, fdp)) == NULL) { 458 FILEDESC_SUNLOCK(fdp); 459 error = EBADF; 460 break; 461 } 462 td->td_retval[0] = OFLAGS(fp->f_flag); 463 FILEDESC_SUNLOCK(fdp); 464 break; 465 466 case F_SETFL: 467 FILEDESC_SLOCK(fdp); 468 if ((fp = fdtofp(fd, fdp)) == NULL) { 469 FILEDESC_SUNLOCK(fdp); 470 error = EBADF; 471 break; 472 } 473 fhold(fp); 474 FILEDESC_SUNLOCK(fdp); 475 do { 476 tmp = flg = fp->f_flag; 477 tmp &= ~FCNTLFLAGS; 478 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 479 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 480 tmp = fp->f_flag & FNONBLOCK; 481 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 482 if (error) { 483 fdrop(fp, td); 484 break; 485 } 486 tmp = fp->f_flag & FASYNC; 487 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 488 if (error == 0) { 489 fdrop(fp, td); 490 break; 491 } 492 atomic_clear_int(&fp->f_flag, FNONBLOCK); 493 tmp = 0; 494 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 495 fdrop(fp, td); 496 break; 497 498 case F_GETOWN: 499 FILEDESC_SLOCK(fdp); 500 if ((fp = fdtofp(fd, fdp)) == NULL) { 501 FILEDESC_SUNLOCK(fdp); 502 error = EBADF; 503 break; 504 } 505 fhold(fp); 506 FILEDESC_SUNLOCK(fdp); 507 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 508 if (error == 0) 509 td->td_retval[0] = tmp; 510 fdrop(fp, td); 511 break; 512 513 case F_SETOWN: 514 FILEDESC_SLOCK(fdp); 515 if ((fp = fdtofp(fd, fdp)) == NULL) { 516 FILEDESC_SUNLOCK(fdp); 517 error = EBADF; 518 break; 519 } 520 fhold(fp); 521 FILEDESC_SUNLOCK(fdp); 522 tmp = arg; 523 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 524 fdrop(fp, td); 525 break; 526 527 case F_SETLK_REMOTE: 528 error = priv_check(td, PRIV_NFS_LOCKD); 529 if (error) 530 return (error); 531 flg = F_REMOTE; 532 goto do_setlk; 533 534 case F_SETLKW: 535 flg |= F_WAIT; 536 /* FALLTHROUGH F_SETLK */ 537 538 case F_SETLK: 539 do_setlk: 540 FILEDESC_SLOCK(fdp); 541 if ((fp = fdtofp(fd, fdp)) == NULL) { 542 FILEDESC_SUNLOCK(fdp); 543 error = EBADF; 544 break; 545 } 546 if (fp->f_type != DTYPE_VNODE) { 547 FILEDESC_SUNLOCK(fdp); 548 error = EBADF; 549 break; 550 } 551 flp = (struct flock *)arg; 552 if (flp->l_whence == SEEK_CUR) { 553 if (fp->f_offset < 0 || 554 (flp->l_start > 0 && 555 fp->f_offset > OFF_MAX - flp->l_start)) { 556 FILEDESC_SUNLOCK(fdp); 557 error = EOVERFLOW; 558 break; 559 } 560 flp->l_start += fp->f_offset; 561 } 562 563 /* 564 * VOP_ADVLOCK() may block. 565 */ 566 fhold(fp); 567 FILEDESC_SUNLOCK(fdp); 568 vp = fp->f_vnode; 569 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 570 switch (flp->l_type) { 571 case F_RDLCK: 572 if ((fp->f_flag & FREAD) == 0) { 573 error = EBADF; 574 break; 575 } 576 PROC_LOCK(p->p_leader); 577 p->p_leader->p_flag |= P_ADVLOCK; 578 PROC_UNLOCK(p->p_leader); 579 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 580 flp, flg); 581 break; 582 case F_WRLCK: 583 if ((fp->f_flag & FWRITE) == 0) { 584 error = EBADF; 585 break; 586 } 587 PROC_LOCK(p->p_leader); 588 p->p_leader->p_flag |= P_ADVLOCK; 589 PROC_UNLOCK(p->p_leader); 590 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 591 flp, flg); 592 break; 593 case F_UNLCK: 594 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 595 flp, flg); 596 break; 597 case F_UNLCKSYS: 598 /* 599 * Temporary api for testing remote lock 600 * infrastructure. 601 */ 602 if (flg != F_REMOTE) { 603 error = EINVAL; 604 break; 605 } 606 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 607 F_UNLCKSYS, flp, flg); 608 break; 609 default: 610 error = EINVAL; 611 break; 612 } 613 VFS_UNLOCK_GIANT(vfslocked); 614 vfslocked = 0; 615 /* Check for race with close */ 616 FILEDESC_SLOCK(fdp); 617 if ((unsigned) fd >= fdp->fd_nfiles || 618 fp != fdp->fd_ofiles[fd]) { 619 FILEDESC_SUNLOCK(fdp); 620 flp->l_whence = SEEK_SET; 621 flp->l_start = 0; 622 flp->l_len = 0; 623 flp->l_type = F_UNLCK; 624 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 625 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 626 F_UNLCK, flp, F_POSIX); 627 VFS_UNLOCK_GIANT(vfslocked); 628 vfslocked = 0; 629 } else 630 FILEDESC_SUNLOCK(fdp); 631 fdrop(fp, td); 632 break; 633 634 case F_GETLK: 635 FILEDESC_SLOCK(fdp); 636 if ((fp = fdtofp(fd, fdp)) == NULL) { 637 FILEDESC_SUNLOCK(fdp); 638 error = EBADF; 639 break; 640 } 641 if (fp->f_type != DTYPE_VNODE) { 642 FILEDESC_SUNLOCK(fdp); 643 error = EBADF; 644 break; 645 } 646 flp = (struct flock *)arg; 647 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 648 flp->l_type != F_UNLCK) { 649 FILEDESC_SUNLOCK(fdp); 650 error = EINVAL; 651 break; 652 } 653 if (flp->l_whence == SEEK_CUR) { 654 if ((flp->l_start > 0 && 655 fp->f_offset > OFF_MAX - flp->l_start) || 656 (flp->l_start < 0 && 657 fp->f_offset < OFF_MIN - flp->l_start)) { 658 FILEDESC_SUNLOCK(fdp); 659 error = EOVERFLOW; 660 break; 661 } 662 flp->l_start += fp->f_offset; 663 } 664 /* 665 * VOP_ADVLOCK() may block. 666 */ 667 fhold(fp); 668 FILEDESC_SUNLOCK(fdp); 669 vp = fp->f_vnode; 670 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 671 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 672 F_POSIX); 673 VFS_UNLOCK_GIANT(vfslocked); 674 vfslocked = 0; 675 fdrop(fp, td); 676 break; 677 default: 678 error = EINVAL; 679 break; 680 } 681 VFS_UNLOCK_GIANT(vfslocked); 682 return (error); 683} 684 685/* 686 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 687 */ 688static int 689do_dup(struct thread *td, int flags, int old, int new, 690 register_t *retval) 691{ 692 struct filedesc *fdp; 693 struct proc *p; 694 struct file *fp; 695 struct file *delfp; 696 int error, holdleaders, maxfd; 697 698 p = td->td_proc; 699 fdp = p->p_fd; 700 701 /* 702 * Verify we have a valid descriptor to dup from and possibly to 703 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 704 * return EINVAL when the new descriptor is out of bounds. 705 */ 706 if (old < 0) 707 return (EBADF); 708 if (new < 0) 709 return (flags & DUP_FCNTL ? EINVAL : EBADF); 710 PROC_LOCK(p); 711 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 712 PROC_UNLOCK(p); 713 if (new >= maxfd) 714 return (flags & DUP_FCNTL ? EINVAL : EMFILE); 715 716 FILEDESC_XLOCK(fdp); 717 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 718 FILEDESC_XUNLOCK(fdp); 719 return (EBADF); 720 } 721 if (flags & DUP_FIXED && old == new) { 722 *retval = new; 723 FILEDESC_XUNLOCK(fdp); 724 return (0); 725 } 726 fp = fdp->fd_ofiles[old]; 727 fhold(fp); 728 729 /* 730 * If the caller specified a file descriptor, make sure the file 731 * table is large enough to hold it, and grab it. Otherwise, just 732 * allocate a new descriptor the usual way. Since the filedesc 733 * lock may be temporarily dropped in the process, we have to look 734 * out for a race. 735 */ 736 if (flags & DUP_FIXED) { 737 if (new >= fdp->fd_nfiles) 738 fdgrowtable(fdp, new + 1); 739 if (fdp->fd_ofiles[new] == NULL) 740 fdused(fdp, new); 741 } else { 742 if ((error = fdalloc(td, new, &new)) != 0) { 743 FILEDESC_XUNLOCK(fdp); 744 fdrop(fp, td); 745 return (error); 746 } 747 } 748 749 /* 750 * If the old file changed out from under us then treat it as a 751 * bad file descriptor. Userland should do its own locking to 752 * avoid this case. 753 */ 754 if (fdp->fd_ofiles[old] != fp) { 755 /* we've allocated a descriptor which we won't use */ 756 if (fdp->fd_ofiles[new] == NULL) 757 fdunused(fdp, new); 758 FILEDESC_XUNLOCK(fdp); 759 fdrop(fp, td); 760 return (EBADF); 761 } 762 KASSERT(old != new, 763 ("new fd is same as old")); 764 765 /* 766 * Save info on the descriptor being overwritten. We cannot close 767 * it without introducing an ownership race for the slot, since we 768 * need to drop the filedesc lock to call closef(). 769 * 770 * XXX this duplicates parts of close(). 771 */ 772 delfp = fdp->fd_ofiles[new]; 773 holdleaders = 0; 774 if (delfp != NULL) { 775 if (td->td_proc->p_fdtol != NULL) { 776 /* 777 * Ask fdfree() to sleep to ensure that all relevant 778 * process leaders can be traversed in closef(). 779 */ 780 fdp->fd_holdleaderscount++; 781 holdleaders = 1; 782 } 783 } 784 785 /* 786 * Duplicate the source descriptor 787 */ 788 fdp->fd_ofiles[new] = fp; 789 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 790 if (new > fdp->fd_lastfile) 791 fdp->fd_lastfile = new; 792 *retval = new; 793 794 /* 795 * If we dup'd over a valid file, we now own the reference to it 796 * and must dispose of it using closef() semantics (as if a 797 * close() were performed on it). 798 * 799 * XXX this duplicates parts of close(). 800 */ 801 if (delfp != NULL) { 802 knote_fdclose(td, new); 803 if (delfp->f_type == DTYPE_MQUEUE) 804 mq_fdclose(td, new, delfp); 805 FILEDESC_XUNLOCK(fdp); 806 (void) closef(delfp, td); 807 if (holdleaders) { 808 FILEDESC_XLOCK(fdp); 809 fdp->fd_holdleaderscount--; 810 if (fdp->fd_holdleaderscount == 0 && 811 fdp->fd_holdleaderswakeup != 0) { 812 fdp->fd_holdleaderswakeup = 0; 813 wakeup(&fdp->fd_holdleaderscount); 814 } 815 FILEDESC_XUNLOCK(fdp); 816 } 817 } else { 818 FILEDESC_XUNLOCK(fdp); 819 } 820 return (0); 821} 822 823/* 824 * If sigio is on the list associated with a process or process group, 825 * disable signalling from the device, remove sigio from the list and 826 * free sigio. 827 */ 828void 829funsetown(struct sigio **sigiop) 830{ 831 struct sigio *sigio; 832 833 SIGIO_LOCK(); 834 sigio = *sigiop; 835 if (sigio == NULL) { 836 SIGIO_UNLOCK(); 837 return; 838 } 839 *(sigio->sio_myref) = NULL; 840 if ((sigio)->sio_pgid < 0) { 841 struct pgrp *pg = (sigio)->sio_pgrp; 842 PGRP_LOCK(pg); 843 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 844 sigio, sio_pgsigio); 845 PGRP_UNLOCK(pg); 846 } else { 847 struct proc *p = (sigio)->sio_proc; 848 PROC_LOCK(p); 849 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 850 sigio, sio_pgsigio); 851 PROC_UNLOCK(p); 852 } 853 SIGIO_UNLOCK(); 854 crfree(sigio->sio_ucred); 855 free(sigio, M_SIGIO); 856} 857 858/* 859 * Free a list of sigio structures. 860 * We only need to lock the SIGIO_LOCK because we have made ourselves 861 * inaccessible to callers of fsetown and therefore do not need to lock 862 * the proc or pgrp struct for the list manipulation. 863 */ 864void 865funsetownlst(struct sigiolst *sigiolst) 866{ 867 struct proc *p; 868 struct pgrp *pg; 869 struct sigio *sigio; 870 871 sigio = SLIST_FIRST(sigiolst); 872 if (sigio == NULL) 873 return; 874 p = NULL; 875 pg = NULL; 876 877 /* 878 * Every entry of the list should belong 879 * to a single proc or pgrp. 880 */ 881 if (sigio->sio_pgid < 0) { 882 pg = sigio->sio_pgrp; 883 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 884 } else /* if (sigio->sio_pgid > 0) */ { 885 p = sigio->sio_proc; 886 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 887 } 888 889 SIGIO_LOCK(); 890 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 891 *(sigio->sio_myref) = NULL; 892 if (pg != NULL) { 893 KASSERT(sigio->sio_pgid < 0, 894 ("Proc sigio in pgrp sigio list")); 895 KASSERT(sigio->sio_pgrp == pg, 896 ("Bogus pgrp in sigio list")); 897 PGRP_LOCK(pg); 898 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 899 sio_pgsigio); 900 PGRP_UNLOCK(pg); 901 } else /* if (p != NULL) */ { 902 KASSERT(sigio->sio_pgid > 0, 903 ("Pgrp sigio in proc sigio list")); 904 KASSERT(sigio->sio_proc == p, 905 ("Bogus proc in sigio list")); 906 PROC_LOCK(p); 907 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 908 sio_pgsigio); 909 PROC_UNLOCK(p); 910 } 911 SIGIO_UNLOCK(); 912 crfree(sigio->sio_ucred); 913 free(sigio, M_SIGIO); 914 SIGIO_LOCK(); 915 } 916 SIGIO_UNLOCK(); 917} 918 919/* 920 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 921 * 922 * After permission checking, add a sigio structure to the sigio list for 923 * the process or process group. 924 */ 925int 926fsetown(pid_t pgid, struct sigio **sigiop) 927{ 928 struct proc *proc; 929 struct pgrp *pgrp; 930 struct sigio *sigio; 931 int ret; 932 933 if (pgid == 0) { 934 funsetown(sigiop); 935 return (0); 936 } 937 938 ret = 0; 939 940 /* Allocate and fill in the new sigio out of locks. */ 941 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 942 sigio->sio_pgid = pgid; 943 sigio->sio_ucred = crhold(curthread->td_ucred); 944 sigio->sio_myref = sigiop; 945 946 sx_slock(&proctree_lock); 947 if (pgid > 0) { 948 proc = pfind(pgid); 949 if (proc == NULL) { 950 ret = ESRCH; 951 goto fail; 952 } 953 954 /* 955 * Policy - Don't allow a process to FSETOWN a process 956 * in another session. 957 * 958 * Remove this test to allow maximum flexibility or 959 * restrict FSETOWN to the current process or process 960 * group for maximum safety. 961 */ 962 PROC_UNLOCK(proc); 963 if (proc->p_session != curthread->td_proc->p_session) { 964 ret = EPERM; 965 goto fail; 966 } 967 968 pgrp = NULL; 969 } else /* if (pgid < 0) */ { 970 pgrp = pgfind(-pgid); 971 if (pgrp == NULL) { 972 ret = ESRCH; 973 goto fail; 974 } 975 PGRP_UNLOCK(pgrp); 976 977 /* 978 * Policy - Don't allow a process to FSETOWN a process 979 * in another session. 980 * 981 * Remove this test to allow maximum flexibility or 982 * restrict FSETOWN to the current process or process 983 * group for maximum safety. 984 */ 985 if (pgrp->pg_session != curthread->td_proc->p_session) { 986 ret = EPERM; 987 goto fail; 988 } 989 990 proc = NULL; 991 } 992 funsetown(sigiop); 993 if (pgid > 0) { 994 PROC_LOCK(proc); 995 /* 996 * Since funsetownlst() is called without the proctree 997 * locked, we need to check for P_WEXIT. 998 * XXX: is ESRCH correct? 999 */ 1000 if ((proc->p_flag & P_WEXIT) != 0) { 1001 PROC_UNLOCK(proc); 1002 ret = ESRCH; 1003 goto fail; 1004 } 1005 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1006 sigio->sio_proc = proc; 1007 PROC_UNLOCK(proc); 1008 } else { 1009 PGRP_LOCK(pgrp); 1010 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1011 sigio->sio_pgrp = pgrp; 1012 PGRP_UNLOCK(pgrp); 1013 } 1014 sx_sunlock(&proctree_lock); 1015 SIGIO_LOCK(); 1016 *sigiop = sigio; 1017 SIGIO_UNLOCK(); 1018 return (0); 1019 1020fail: 1021 sx_sunlock(&proctree_lock); 1022 crfree(sigio->sio_ucred); 1023 free(sigio, M_SIGIO); 1024 return (ret); 1025} 1026 1027/* 1028 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1029 */ 1030pid_t 1031fgetown(sigiop) 1032 struct sigio **sigiop; 1033{ 1034 pid_t pgid; 1035 1036 SIGIO_LOCK(); 1037 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1038 SIGIO_UNLOCK(); 1039 return (pgid); 1040} 1041 1042/* 1043 * Close a file descriptor. 1044 */ 1045#ifndef _SYS_SYSPROTO_H_ 1046struct close_args { 1047 int fd; 1048}; 1049#endif 1050/* ARGSUSED */ 1051int 1052close(td, uap) 1053 struct thread *td; 1054 struct close_args *uap; 1055{ 1056 1057 return (kern_close(td, uap->fd)); 1058} 1059 1060int 1061kern_close(td, fd) 1062 struct thread *td; 1063 int fd; 1064{ 1065 struct filedesc *fdp; 1066 struct file *fp; 1067 int error; 1068 int holdleaders; 1069 1070 error = 0; 1071 holdleaders = 0; 1072 fdp = td->td_proc->p_fd; 1073 1074 AUDIT_SYSCLOSE(td, fd); 1075 1076 FILEDESC_XLOCK(fdp); 1077 if ((unsigned)fd >= fdp->fd_nfiles || 1078 (fp = fdp->fd_ofiles[fd]) == NULL) { 1079 FILEDESC_XUNLOCK(fdp); 1080 return (EBADF); 1081 } 1082 fdp->fd_ofiles[fd] = NULL; 1083 fdp->fd_ofileflags[fd] = 0; 1084 fdunused(fdp, fd); 1085 if (td->td_proc->p_fdtol != NULL) { 1086 /* 1087 * Ask fdfree() to sleep to ensure that all relevant 1088 * process leaders can be traversed in closef(). 1089 */ 1090 fdp->fd_holdleaderscount++; 1091 holdleaders = 1; 1092 } 1093 1094 /* 1095 * We now hold the fp reference that used to be owned by the 1096 * descriptor array. We have to unlock the FILEDESC *AFTER* 1097 * knote_fdclose to prevent a race of the fd getting opened, a knote 1098 * added, and deleteing a knote for the new fd. 1099 */ 1100 knote_fdclose(td, fd); 1101 if (fp->f_type == DTYPE_MQUEUE) 1102 mq_fdclose(td, fd, fp); 1103 FILEDESC_XUNLOCK(fdp); 1104 1105 error = closef(fp, td); 1106 if (holdleaders) { 1107 FILEDESC_XLOCK(fdp); 1108 fdp->fd_holdleaderscount--; 1109 if (fdp->fd_holdleaderscount == 0 && 1110 fdp->fd_holdleaderswakeup != 0) { 1111 fdp->fd_holdleaderswakeup = 0; 1112 wakeup(&fdp->fd_holdleaderscount); 1113 } 1114 FILEDESC_XUNLOCK(fdp); 1115 } 1116 return (error); 1117} 1118 1119#if defined(COMPAT_43) 1120/* 1121 * Return status information about a file descriptor. 1122 */ 1123#ifndef _SYS_SYSPROTO_H_ 1124struct ofstat_args { 1125 int fd; 1126 struct ostat *sb; 1127}; 1128#endif 1129/* ARGSUSED */ 1130int 1131ofstat(struct thread *td, struct ofstat_args *uap) 1132{ 1133 struct ostat oub; 1134 struct stat ub; 1135 int error; 1136 1137 error = kern_fstat(td, uap->fd, &ub); 1138 if (error == 0) { 1139 cvtstat(&ub, &oub); 1140 error = copyout(&oub, uap->sb, sizeof(oub)); 1141 } 1142 return (error); 1143} 1144#endif /* COMPAT_43 */ 1145 1146/* 1147 * Return status information about a file descriptor. 1148 */ 1149#ifndef _SYS_SYSPROTO_H_ 1150struct fstat_args { 1151 int fd; 1152 struct stat *sb; 1153}; 1154#endif 1155/* ARGSUSED */ 1156int 1157fstat(struct thread *td, struct fstat_args *uap) 1158{ 1159 struct stat ub; 1160 int error; 1161 1162 error = kern_fstat(td, uap->fd, &ub); 1163 if (error == 0) 1164 error = copyout(&ub, uap->sb, sizeof(ub)); 1165 return (error); 1166} 1167 1168int 1169kern_fstat(struct thread *td, int fd, struct stat *sbp) 1170{ 1171 struct file *fp; 1172 int error; 1173 1174 AUDIT_ARG(fd, fd); 1175 1176 if ((error = fget(td, fd, &fp)) != 0) 1177 return (error); 1178 1179 AUDIT_ARG(file, td->td_proc, fp); 1180 1181 error = fo_stat(fp, sbp, td->td_ucred, td); 1182 fdrop(fp, td); 1183#ifdef KTRACE 1184 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1185 ktrstat(sbp); 1186#endif 1187 return (error); 1188} 1189 1190/* 1191 * Return status information about a file descriptor. 1192 */ 1193#ifndef _SYS_SYSPROTO_H_ 1194struct nfstat_args { 1195 int fd; 1196 struct nstat *sb; 1197}; 1198#endif 1199/* ARGSUSED */ 1200int 1201nfstat(struct thread *td, struct nfstat_args *uap) 1202{ 1203 struct nstat nub; 1204 struct stat ub; 1205 int error; 1206 1207 error = kern_fstat(td, uap->fd, &ub); 1208 if (error == 0) { 1209 cvtnstat(&ub, &nub); 1210 error = copyout(&nub, uap->sb, sizeof(nub)); 1211 } 1212 return (error); 1213} 1214 1215/* 1216 * Return pathconf information about a file descriptor. 1217 */ 1218#ifndef _SYS_SYSPROTO_H_ 1219struct fpathconf_args { 1220 int fd; 1221 int name; 1222}; 1223#endif 1224/* ARGSUSED */ 1225int 1226fpathconf(struct thread *td, struct fpathconf_args *uap) 1227{ 1228 struct file *fp; 1229 struct vnode *vp; 1230 int error; 1231 1232 if ((error = fget(td, uap->fd, &fp)) != 0) 1233 return (error); 1234 1235 /* If asynchronous I/O is available, it works for all descriptors. */ 1236 if (uap->name == _PC_ASYNC_IO) { 1237 td->td_retval[0] = async_io_version; 1238 goto out; 1239 } 1240 vp = fp->f_vnode; 1241 if (vp != NULL) { 1242 int vfslocked; 1243 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1244 vn_lock(vp, LK_SHARED | LK_RETRY); 1245 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1246 VOP_UNLOCK(vp, 0); 1247 VFS_UNLOCK_GIANT(vfslocked); 1248 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1249 if (uap->name != _PC_PIPE_BUF) { 1250 error = EINVAL; 1251 } else { 1252 td->td_retval[0] = PIPE_BUF; 1253 error = 0; 1254 } 1255 } else { 1256 error = EOPNOTSUPP; 1257 } 1258out: 1259 fdrop(fp, td); 1260 return (error); 1261} 1262 1263/* 1264 * Grow the file table to accomodate (at least) nfd descriptors. This may 1265 * block and drop the filedesc lock, but it will reacquire it before 1266 * returning. 1267 */ 1268static void 1269fdgrowtable(struct filedesc *fdp, int nfd) 1270{ 1271 struct file **ntable; 1272 char *nfileflags; 1273 int nnfiles, onfiles; 1274 NDSLOTTYPE *nmap; 1275 1276 FILEDESC_XLOCK_ASSERT(fdp); 1277 1278 KASSERT(fdp->fd_nfiles > 0, 1279 ("zero-length file table")); 1280 1281 /* compute the size of the new table */ 1282 onfiles = fdp->fd_nfiles; 1283 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1284 if (nnfiles <= onfiles) 1285 /* the table is already large enough */ 1286 return; 1287 1288 /* allocate a new table and (if required) new bitmaps */ 1289 FILEDESC_XUNLOCK(fdp); 1290 ntable = malloc(nnfiles * OFILESIZE, 1291 M_FILEDESC, M_ZERO | M_WAITOK); 1292 nfileflags = (char *)&ntable[nnfiles]; 1293 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1294 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, 1295 M_FILEDESC, M_ZERO | M_WAITOK); 1296 else 1297 nmap = NULL; 1298 FILEDESC_XLOCK(fdp); 1299 1300 /* 1301 * We now have new tables ready to go. Since we dropped the 1302 * filedesc lock to call malloc(), watch out for a race. 1303 */ 1304 onfiles = fdp->fd_nfiles; 1305 if (onfiles >= nnfiles) { 1306 /* we lost the race, but that's OK */ 1307 free(ntable, M_FILEDESC); 1308 if (nmap != NULL) 1309 free(nmap, M_FILEDESC); 1310 return; 1311 } 1312 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1313 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1314 if (onfiles > NDFILE) 1315 free(fdp->fd_ofiles, M_FILEDESC); 1316 fdp->fd_ofiles = ntable; 1317 fdp->fd_ofileflags = nfileflags; 1318 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1319 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1320 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1321 free(fdp->fd_map, M_FILEDESC); 1322 fdp->fd_map = nmap; 1323 } 1324 fdp->fd_nfiles = nnfiles; 1325} 1326 1327/* 1328 * Allocate a file descriptor for the process. 1329 */ 1330int 1331fdalloc(struct thread *td, int minfd, int *result) 1332{ 1333 struct proc *p = td->td_proc; 1334 struct filedesc *fdp = p->p_fd; 1335 int fd = -1, maxfd; 1336 1337 FILEDESC_XLOCK_ASSERT(fdp); 1338 1339 if (fdp->fd_freefile > minfd) 1340 minfd = fdp->fd_freefile; 1341 1342 PROC_LOCK(p); 1343 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1344 PROC_UNLOCK(p); 1345 1346 /* 1347 * Search the bitmap for a free descriptor. If none is found, try 1348 * to grow the file table. Keep at it until we either get a file 1349 * descriptor or run into process or system limits; fdgrowtable() 1350 * may drop the filedesc lock, so we're in a race. 1351 */ 1352 for (;;) { 1353 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1354 if (fd >= maxfd) 1355 return (EMFILE); 1356 if (fd < fdp->fd_nfiles) 1357 break; 1358 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1359 } 1360 1361 /* 1362 * Perform some sanity checks, then mark the file descriptor as 1363 * used and return it to the caller. 1364 */ 1365 KASSERT(!fdisused(fdp, fd), 1366 ("fd_first_free() returned non-free descriptor")); 1367 KASSERT(fdp->fd_ofiles[fd] == NULL, 1368 ("free descriptor isn't")); 1369 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */ 1370 fdused(fdp, fd); 1371 *result = fd; 1372 return (0); 1373} 1374 1375/* 1376 * Check to see whether n user file descriptors are available to the process 1377 * p. 1378 */ 1379int 1380fdavail(struct thread *td, int n) 1381{ 1382 struct proc *p = td->td_proc; 1383 struct filedesc *fdp = td->td_proc->p_fd; 1384 struct file **fpp; 1385 int i, lim, last; 1386 1387 FILEDESC_LOCK_ASSERT(fdp); 1388 1389 PROC_LOCK(p); 1390 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1391 PROC_UNLOCK(p); 1392 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1393 return (1); 1394 last = min(fdp->fd_nfiles, lim); 1395 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 1396 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 1397 if (*fpp == NULL && --n <= 0) 1398 return (1); 1399 } 1400 return (0); 1401} 1402 1403/* 1404 * Create a new open file structure and allocate a file decriptor for the 1405 * process that refers to it. We add one reference to the file for the 1406 * descriptor table and one reference for resultfp. This is to prevent us 1407 * being preempted and the entry in the descriptor table closed after we 1408 * release the FILEDESC lock. 1409 */ 1410int 1411falloc(struct thread *td, struct file **resultfp, int *resultfd) 1412{ 1413 struct proc *p = td->td_proc; 1414 struct file *fp; 1415 int error, i; 1416 int maxuserfiles = maxfiles - (maxfiles / 20); 1417 static struct timeval lastfail; 1418 static int curfail; 1419 1420 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1421 if ((openfiles >= maxuserfiles && 1422 priv_check(td, PRIV_MAXFILES) != 0) || 1423 openfiles >= maxfiles) { 1424 if (ppsratecheck(&lastfail, &curfail, 1)) { 1425 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n", 1426 td->td_ucred->cr_ruid); 1427 } 1428 uma_zfree(file_zone, fp); 1429 return (ENFILE); 1430 } 1431 atomic_add_int(&openfiles, 1); 1432 1433 /* 1434 * If the process has file descriptor zero open, add the new file 1435 * descriptor to the list of open files at that point, otherwise 1436 * put it at the front of the list of open files. 1437 */ 1438 refcount_init(&fp->f_count, 1); 1439 if (resultfp) 1440 fhold(fp); 1441 fp->f_cred = crhold(td->td_ucred); 1442 fp->f_ops = &badfileops; 1443 fp->f_data = NULL; 1444 fp->f_vnode = NULL; 1445 FILEDESC_XLOCK(p->p_fd); 1446 if ((error = fdalloc(td, 0, &i))) { 1447 FILEDESC_XUNLOCK(p->p_fd); 1448 fdrop(fp, td); 1449 if (resultfp) 1450 fdrop(fp, td); 1451 return (error); 1452 } 1453 p->p_fd->fd_ofiles[i] = fp; 1454 FILEDESC_XUNLOCK(p->p_fd); 1455 if (resultfp) 1456 *resultfp = fp; 1457 if (resultfd) 1458 *resultfd = i; 1459 return (0); 1460} 1461 1462/* 1463 * Build a new filedesc structure from another. 1464 * Copy the current, root, and jail root vnode references. 1465 */ 1466struct filedesc * 1467fdinit(struct filedesc *fdp) 1468{ 1469 struct filedesc0 *newfdp; 1470 1471 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1472 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1473 if (fdp != NULL) { 1474 FILEDESC_XLOCK(fdp); 1475 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1476 if (newfdp->fd_fd.fd_cdir) 1477 VREF(newfdp->fd_fd.fd_cdir); 1478 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1479 if (newfdp->fd_fd.fd_rdir) 1480 VREF(newfdp->fd_fd.fd_rdir); 1481 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1482 if (newfdp->fd_fd.fd_jdir) 1483 VREF(newfdp->fd_fd.fd_jdir); 1484 FILEDESC_XUNLOCK(fdp); 1485 } 1486 1487 /* Create the file descriptor table. */ 1488 newfdp->fd_fd.fd_refcnt = 1; 1489 newfdp->fd_fd.fd_holdcnt = 1; 1490 newfdp->fd_fd.fd_cmask = CMASK; 1491 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1492 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1493 newfdp->fd_fd.fd_nfiles = NDFILE; 1494 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1495 newfdp->fd_fd.fd_lastfile = -1; 1496 return (&newfdp->fd_fd); 1497} 1498 1499static struct filedesc * 1500fdhold(struct proc *p) 1501{ 1502 struct filedesc *fdp; 1503 1504 mtx_lock(&fdesc_mtx); 1505 fdp = p->p_fd; 1506 if (fdp != NULL) 1507 fdp->fd_holdcnt++; 1508 mtx_unlock(&fdesc_mtx); 1509 return (fdp); 1510} 1511 1512static void 1513fddrop(struct filedesc *fdp) 1514{ 1515 int i; 1516 1517 mtx_lock(&fdesc_mtx); 1518 i = --fdp->fd_holdcnt; 1519 mtx_unlock(&fdesc_mtx); 1520 if (i > 0) 1521 return; 1522 1523 FILEDESC_LOCK_DESTROY(fdp); 1524 free(fdp, M_FILEDESC); 1525} 1526 1527/* 1528 * Share a filedesc structure. 1529 */ 1530struct filedesc * 1531fdshare(struct filedesc *fdp) 1532{ 1533 1534 FILEDESC_XLOCK(fdp); 1535 fdp->fd_refcnt++; 1536 FILEDESC_XUNLOCK(fdp); 1537 return (fdp); 1538} 1539 1540/* 1541 * Unshare a filedesc structure, if necessary by making a copy 1542 */ 1543void 1544fdunshare(struct proc *p, struct thread *td) 1545{ 1546 1547 FILEDESC_XLOCK(p->p_fd); 1548 if (p->p_fd->fd_refcnt > 1) { 1549 struct filedesc *tmp; 1550 1551 FILEDESC_XUNLOCK(p->p_fd); 1552 tmp = fdcopy(p->p_fd); 1553 fdfree(td); 1554 p->p_fd = tmp; 1555 } else 1556 FILEDESC_XUNLOCK(p->p_fd); 1557} 1558 1559/* 1560 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1561 * this is to ease callers, not catch errors. 1562 */ 1563struct filedesc * 1564fdcopy(struct filedesc *fdp) 1565{ 1566 struct filedesc *newfdp; 1567 int i; 1568 1569 /* Certain daemons might not have file descriptors. */ 1570 if (fdp == NULL) 1571 return (NULL); 1572 1573 newfdp = fdinit(fdp); 1574 FILEDESC_SLOCK(fdp); 1575 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1576 FILEDESC_SUNLOCK(fdp); 1577 FILEDESC_XLOCK(newfdp); 1578 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1579 FILEDESC_XUNLOCK(newfdp); 1580 FILEDESC_SLOCK(fdp); 1581 } 1582 /* copy everything except kqueue descriptors */ 1583 newfdp->fd_freefile = -1; 1584 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1585 if (fdisused(fdp, i) && 1586 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) { 1587 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1588 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1589 fhold(newfdp->fd_ofiles[i]); 1590 newfdp->fd_lastfile = i; 1591 } else { 1592 if (newfdp->fd_freefile == -1) 1593 newfdp->fd_freefile = i; 1594 } 1595 } 1596 FILEDESC_SUNLOCK(fdp); 1597 FILEDESC_XLOCK(newfdp); 1598 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1599 if (newfdp->fd_ofiles[i] != NULL) 1600 fdused(newfdp, i); 1601 FILEDESC_XUNLOCK(newfdp); 1602 FILEDESC_SLOCK(fdp); 1603 if (newfdp->fd_freefile == -1) 1604 newfdp->fd_freefile = i; 1605 newfdp->fd_cmask = fdp->fd_cmask; 1606 FILEDESC_SUNLOCK(fdp); 1607 return (newfdp); 1608} 1609 1610/* 1611 * Release a filedesc structure. 1612 */ 1613void 1614fdfree(struct thread *td) 1615{ 1616 struct filedesc *fdp; 1617 struct file **fpp; 1618 int i, locked; 1619 struct filedesc_to_leader *fdtol; 1620 struct file *fp; 1621 struct vnode *cdir, *jdir, *rdir, *vp; 1622 struct flock lf; 1623 1624 /* Certain daemons might not have file descriptors. */ 1625 fdp = td->td_proc->p_fd; 1626 if (fdp == NULL) 1627 return; 1628 1629 /* Check for special need to clear POSIX style locks */ 1630 fdtol = td->td_proc->p_fdtol; 1631 if (fdtol != NULL) { 1632 FILEDESC_XLOCK(fdp); 1633 KASSERT(fdtol->fdl_refcount > 0, 1634 ("filedesc_to_refcount botch: fdl_refcount=%d", 1635 fdtol->fdl_refcount)); 1636 if (fdtol->fdl_refcount == 1 && 1637 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1638 for (i = 0, fpp = fdp->fd_ofiles; 1639 i <= fdp->fd_lastfile; 1640 i++, fpp++) { 1641 if (*fpp == NULL || 1642 (*fpp)->f_type != DTYPE_VNODE) 1643 continue; 1644 fp = *fpp; 1645 fhold(fp); 1646 FILEDESC_XUNLOCK(fdp); 1647 lf.l_whence = SEEK_SET; 1648 lf.l_start = 0; 1649 lf.l_len = 0; 1650 lf.l_type = F_UNLCK; 1651 vp = fp->f_vnode; 1652 locked = VFS_LOCK_GIANT(vp->v_mount); 1653 (void) VOP_ADVLOCK(vp, 1654 (caddr_t)td->td_proc-> 1655 p_leader, 1656 F_UNLCK, 1657 &lf, 1658 F_POSIX); 1659 VFS_UNLOCK_GIANT(locked); 1660 FILEDESC_XLOCK(fdp); 1661 fdrop(fp, td); 1662 fpp = fdp->fd_ofiles + i; 1663 } 1664 } 1665 retry: 1666 if (fdtol->fdl_refcount == 1) { 1667 if (fdp->fd_holdleaderscount > 0 && 1668 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1669 /* 1670 * close() or do_dup() has cleared a reference 1671 * in a shared file descriptor table. 1672 */ 1673 fdp->fd_holdleaderswakeup = 1; 1674 sx_sleep(&fdp->fd_holdleaderscount, 1675 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1676 goto retry; 1677 } 1678 if (fdtol->fdl_holdcount > 0) { 1679 /* 1680 * Ensure that fdtol->fdl_leader remains 1681 * valid in closef(). 1682 */ 1683 fdtol->fdl_wakeup = 1; 1684 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1685 "fdlhold", 0); 1686 goto retry; 1687 } 1688 } 1689 fdtol->fdl_refcount--; 1690 if (fdtol->fdl_refcount == 0 && 1691 fdtol->fdl_holdcount == 0) { 1692 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1693 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1694 } else 1695 fdtol = NULL; 1696 td->td_proc->p_fdtol = NULL; 1697 FILEDESC_XUNLOCK(fdp); 1698 if (fdtol != NULL) 1699 free(fdtol, M_FILEDESC_TO_LEADER); 1700 } 1701 FILEDESC_XLOCK(fdp); 1702 i = --fdp->fd_refcnt; 1703 FILEDESC_XUNLOCK(fdp); 1704 if (i > 0) 1705 return; 1706 1707 fpp = fdp->fd_ofiles; 1708 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1709 if (*fpp) { 1710 FILEDESC_XLOCK(fdp); 1711 fp = *fpp; 1712 *fpp = NULL; 1713 FILEDESC_XUNLOCK(fdp); 1714 (void) closef(fp, td); 1715 } 1716 } 1717 FILEDESC_XLOCK(fdp); 1718 1719 /* XXX This should happen earlier. */ 1720 mtx_lock(&fdesc_mtx); 1721 td->td_proc->p_fd = NULL; 1722 mtx_unlock(&fdesc_mtx); 1723 1724 if (fdp->fd_nfiles > NDFILE) 1725 free(fdp->fd_ofiles, M_FILEDESC); 1726 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1727 free(fdp->fd_map, M_FILEDESC); 1728 1729 fdp->fd_nfiles = 0; 1730 1731 cdir = fdp->fd_cdir; 1732 fdp->fd_cdir = NULL; 1733 rdir = fdp->fd_rdir; 1734 fdp->fd_rdir = NULL; 1735 jdir = fdp->fd_jdir; 1736 fdp->fd_jdir = NULL; 1737 FILEDESC_XUNLOCK(fdp); 1738 1739 if (cdir) { 1740 locked = VFS_LOCK_GIANT(cdir->v_mount); 1741 vrele(cdir); 1742 VFS_UNLOCK_GIANT(locked); 1743 } 1744 if (rdir) { 1745 locked = VFS_LOCK_GIANT(rdir->v_mount); 1746 vrele(rdir); 1747 VFS_UNLOCK_GIANT(locked); 1748 } 1749 if (jdir) { 1750 locked = VFS_LOCK_GIANT(jdir->v_mount); 1751 vrele(jdir); 1752 VFS_UNLOCK_GIANT(locked); 1753 } 1754 1755 fddrop(fdp); 1756} 1757 1758/* 1759 * For setugid programs, we don't want to people to use that setugidness 1760 * to generate error messages which write to a file which otherwise would 1761 * otherwise be off-limits to the process. We check for filesystems where 1762 * the vnode can change out from under us after execve (like [lin]procfs). 1763 * 1764 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1765 * sufficient. We also don't check for setugidness since we know we are. 1766 */ 1767static int 1768is_unsafe(struct file *fp) 1769{ 1770 if (fp->f_type == DTYPE_VNODE) { 1771 struct vnode *vp = fp->f_vnode; 1772 1773 if ((vp->v_vflag & VV_PROCDEP) != 0) 1774 return (1); 1775 } 1776 return (0); 1777} 1778 1779/* 1780 * Make this setguid thing safe, if at all possible. 1781 */ 1782void 1783setugidsafety(struct thread *td) 1784{ 1785 struct filedesc *fdp; 1786 int i; 1787 1788 /* Certain daemons might not have file descriptors. */ 1789 fdp = td->td_proc->p_fd; 1790 if (fdp == NULL) 1791 return; 1792 1793 /* 1794 * Note: fdp->fd_ofiles may be reallocated out from under us while 1795 * we are blocked in a close. Be careful! 1796 */ 1797 FILEDESC_XLOCK(fdp); 1798 for (i = 0; i <= fdp->fd_lastfile; i++) { 1799 if (i > 2) 1800 break; 1801 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1802 struct file *fp; 1803 1804 knote_fdclose(td, i); 1805 /* 1806 * NULL-out descriptor prior to close to avoid 1807 * a race while close blocks. 1808 */ 1809 fp = fdp->fd_ofiles[i]; 1810 fdp->fd_ofiles[i] = NULL; 1811 fdp->fd_ofileflags[i] = 0; 1812 fdunused(fdp, i); 1813 FILEDESC_XUNLOCK(fdp); 1814 (void) closef(fp, td); 1815 FILEDESC_XLOCK(fdp); 1816 } 1817 } 1818 FILEDESC_XUNLOCK(fdp); 1819} 1820 1821/* 1822 * If a specific file object occupies a specific file descriptor, close the 1823 * file descriptor entry and drop a reference on the file object. This is a 1824 * convenience function to handle a subsequent error in a function that calls 1825 * falloc() that handles the race that another thread might have closed the 1826 * file descriptor out from under the thread creating the file object. 1827 */ 1828void 1829fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 1830{ 1831 1832 FILEDESC_XLOCK(fdp); 1833 if (fdp->fd_ofiles[idx] == fp) { 1834 fdp->fd_ofiles[idx] = NULL; 1835 fdunused(fdp, idx); 1836 FILEDESC_XUNLOCK(fdp); 1837 fdrop(fp, td); 1838 } else 1839 FILEDESC_XUNLOCK(fdp); 1840} 1841 1842/* 1843 * Close any files on exec? 1844 */ 1845void 1846fdcloseexec(struct thread *td) 1847{ 1848 struct filedesc *fdp; 1849 int i; 1850 1851 /* Certain daemons might not have file descriptors. */ 1852 fdp = td->td_proc->p_fd; 1853 if (fdp == NULL) 1854 return; 1855 1856 FILEDESC_XLOCK(fdp); 1857 1858 /* 1859 * We cannot cache fd_ofiles or fd_ofileflags since operations 1860 * may block and rip them out from under us. 1861 */ 1862 for (i = 0; i <= fdp->fd_lastfile; i++) { 1863 if (fdp->fd_ofiles[i] != NULL && 1864 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE || 1865 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 1866 struct file *fp; 1867 1868 knote_fdclose(td, i); 1869 /* 1870 * NULL-out descriptor prior to close to avoid 1871 * a race while close blocks. 1872 */ 1873 fp = fdp->fd_ofiles[i]; 1874 fdp->fd_ofiles[i] = NULL; 1875 fdp->fd_ofileflags[i] = 0; 1876 fdunused(fdp, i); 1877 if (fp->f_type == DTYPE_MQUEUE) 1878 mq_fdclose(td, i, fp); 1879 FILEDESC_XUNLOCK(fdp); 1880 (void) closef(fp, td); 1881 FILEDESC_XLOCK(fdp); 1882 } 1883 } 1884 FILEDESC_XUNLOCK(fdp); 1885} 1886 1887/* 1888 * It is unsafe for set[ug]id processes to be started with file 1889 * descriptors 0..2 closed, as these descriptors are given implicit 1890 * significance in the Standard C library. fdcheckstd() will create a 1891 * descriptor referencing /dev/null for each of stdin, stdout, and 1892 * stderr that is not already open. 1893 */ 1894int 1895fdcheckstd(struct thread *td) 1896{ 1897 struct filedesc *fdp; 1898 register_t retval, save; 1899 int i, error, devnull; 1900 1901 fdp = td->td_proc->p_fd; 1902 if (fdp == NULL) 1903 return (0); 1904 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 1905 devnull = -1; 1906 error = 0; 1907 for (i = 0; i < 3; i++) { 1908 if (fdp->fd_ofiles[i] != NULL) 1909 continue; 1910 if (devnull < 0) { 1911 save = td->td_retval[0]; 1912 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 1913 O_RDWR, 0); 1914 devnull = td->td_retval[0]; 1915 KASSERT(devnull == i, ("oof, we didn't get our fd")); 1916 td->td_retval[0] = save; 1917 if (error) 1918 break; 1919 } else { 1920 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 1921 if (error != 0) 1922 break; 1923 } 1924 } 1925 return (error); 1926} 1927 1928/* 1929 * Internal form of close. Decrement reference count on file structure. 1930 * Note: td may be NULL when closing a file that was being passed in a 1931 * message. 1932 * 1933 * XXXRW: Giant is not required for the caller, but often will be held; this 1934 * makes it moderately likely the Giant will be recursed in the VFS case. 1935 */ 1936int 1937closef(struct file *fp, struct thread *td) 1938{ 1939 struct vnode *vp; 1940 struct flock lf; 1941 struct filedesc_to_leader *fdtol; 1942 struct filedesc *fdp; 1943 1944 /* 1945 * POSIX record locking dictates that any close releases ALL 1946 * locks owned by this process. This is handled by setting 1947 * a flag in the unlock to free ONLY locks obeying POSIX 1948 * semantics, and not to free BSD-style file locks. 1949 * If the descriptor was in a message, POSIX-style locks 1950 * aren't passed with the descriptor, and the thread pointer 1951 * will be NULL. Callers should be careful only to pass a 1952 * NULL thread pointer when there really is no owning 1953 * context that might have locks, or the locks will be 1954 * leaked. 1955 */ 1956 if (fp->f_type == DTYPE_VNODE && td != NULL) { 1957 int vfslocked; 1958 1959 vp = fp->f_vnode; 1960 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1961 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1962 lf.l_whence = SEEK_SET; 1963 lf.l_start = 0; 1964 lf.l_len = 0; 1965 lf.l_type = F_UNLCK; 1966 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 1967 F_UNLCK, &lf, F_POSIX); 1968 } 1969 fdtol = td->td_proc->p_fdtol; 1970 if (fdtol != NULL) { 1971 /* 1972 * Handle special case where file descriptor table is 1973 * shared between multiple process leaders. 1974 */ 1975 fdp = td->td_proc->p_fd; 1976 FILEDESC_XLOCK(fdp); 1977 for (fdtol = fdtol->fdl_next; 1978 fdtol != td->td_proc->p_fdtol; 1979 fdtol = fdtol->fdl_next) { 1980 if ((fdtol->fdl_leader->p_flag & 1981 P_ADVLOCK) == 0) 1982 continue; 1983 fdtol->fdl_holdcount++; 1984 FILEDESC_XUNLOCK(fdp); 1985 lf.l_whence = SEEK_SET; 1986 lf.l_start = 0; 1987 lf.l_len = 0; 1988 lf.l_type = F_UNLCK; 1989 vp = fp->f_vnode; 1990 (void) VOP_ADVLOCK(vp, 1991 (caddr_t)fdtol->fdl_leader, 1992 F_UNLCK, &lf, F_POSIX); 1993 FILEDESC_XLOCK(fdp); 1994 fdtol->fdl_holdcount--; 1995 if (fdtol->fdl_holdcount == 0 && 1996 fdtol->fdl_wakeup != 0) { 1997 fdtol->fdl_wakeup = 0; 1998 wakeup(fdtol); 1999 } 2000 } 2001 FILEDESC_XUNLOCK(fdp); 2002 } 2003 VFS_UNLOCK_GIANT(vfslocked); 2004 } 2005 return (fdrop(fp, td)); 2006} 2007 2008/* 2009 * Initialize the file pointer with the specified properties. 2010 * 2011 * The ops are set with release semantics to be certain that the flags, type, 2012 * and data are visible when ops is. This is to prevent ops methods from being 2013 * called with bad data. 2014 */ 2015void 2016finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2017{ 2018 fp->f_data = data; 2019 fp->f_flag = flag; 2020 fp->f_type = type; 2021 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2022} 2023 2024 2025/* 2026 * Extract the file pointer associated with the specified descriptor for the 2027 * current user process. 2028 * 2029 * If the descriptor doesn't exist, EBADF is returned. 2030 * 2031 * If the descriptor exists but doesn't match 'flags' then return EBADF for 2032 * read attempts and EINVAL for write attempts. 2033 * 2034 * If 'hold' is set (non-zero) the file's refcount will be bumped on return. 2035 * It should be dropped with fdrop(). If it is not set, then the refcount 2036 * will not be bumped however the thread's filedesc struct will be returned 2037 * locked (for fgetsock). 2038 * 2039 * If an error occured the non-zero error is returned and *fpp is set to 2040 * NULL. Otherwise *fpp is set and zero is returned. 2041 */ 2042static __inline int 2043_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold) 2044{ 2045 struct filedesc *fdp; 2046 struct file *fp; 2047 2048 *fpp = NULL; 2049 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2050 return (EBADF); 2051 FILEDESC_SLOCK(fdp); 2052 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) { 2053 FILEDESC_SUNLOCK(fdp); 2054 return (EBADF); 2055 } 2056 2057 /* 2058 * FREAD and FWRITE failure return EBADF as per POSIX. 2059 * 2060 * Only one flag, or 0, may be specified. 2061 */ 2062 if (flags == FREAD && (fp->f_flag & FREAD) == 0) { 2063 FILEDESC_SUNLOCK(fdp); 2064 return (EBADF); 2065 } 2066 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) { 2067 FILEDESC_SUNLOCK(fdp); 2068 return (EBADF); 2069 } 2070 if (hold) { 2071 fhold(fp); 2072 FILEDESC_SUNLOCK(fdp); 2073 } 2074 *fpp = fp; 2075 return (0); 2076} 2077 2078int 2079fget(struct thread *td, int fd, struct file **fpp) 2080{ 2081 2082 return(_fget(td, fd, fpp, 0, 1)); 2083} 2084 2085int 2086fget_read(struct thread *td, int fd, struct file **fpp) 2087{ 2088 2089 return(_fget(td, fd, fpp, FREAD, 1)); 2090} 2091 2092int 2093fget_write(struct thread *td, int fd, struct file **fpp) 2094{ 2095 2096 return(_fget(td, fd, fpp, FWRITE, 1)); 2097} 2098 2099/* 2100 * Like fget() but loads the underlying vnode, or returns an error if the 2101 * descriptor does not represent a vnode. Note that pipes use vnodes but 2102 * never have VM objects. The returned vnode will be vref()'d. 2103 * 2104 * XXX: what about the unused flags ? 2105 */ 2106static __inline int 2107_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags) 2108{ 2109 struct file *fp; 2110 int error; 2111 2112 *vpp = NULL; 2113 if ((error = _fget(td, fd, &fp, flags, 0)) != 0) 2114 return (error); 2115 if (fp->f_vnode == NULL) { 2116 error = EINVAL; 2117 } else { 2118 *vpp = fp->f_vnode; 2119 vref(*vpp); 2120 } 2121 FILEDESC_SUNLOCK(td->td_proc->p_fd); 2122 return (error); 2123} 2124 2125int 2126fgetvp(struct thread *td, int fd, struct vnode **vpp) 2127{ 2128 2129 return (_fgetvp(td, fd, vpp, 0)); 2130} 2131 2132int 2133fgetvp_read(struct thread *td, int fd, struct vnode **vpp) 2134{ 2135 2136 return (_fgetvp(td, fd, vpp, FREAD)); 2137} 2138 2139#ifdef notyet 2140int 2141fgetvp_write(struct thread *td, int fd, struct vnode **vpp) 2142{ 2143 2144 return (_fgetvp(td, fd, vpp, FWRITE)); 2145} 2146#endif 2147 2148/* 2149 * Like fget() but loads the underlying socket, or returns an error if the 2150 * descriptor does not represent a socket. 2151 * 2152 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2153 * in the future. 2154 * 2155 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2156 * on their file descriptor reference to prevent the socket from being free'd 2157 * during use. 2158 */ 2159int 2160fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp) 2161{ 2162 struct file *fp; 2163 int error; 2164 2165 *spp = NULL; 2166 if (fflagp != NULL) 2167 *fflagp = 0; 2168 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2169 return (error); 2170 if (fp->f_type != DTYPE_SOCKET) { 2171 error = ENOTSOCK; 2172 } else { 2173 *spp = fp->f_data; 2174 if (fflagp) 2175 *fflagp = fp->f_flag; 2176 SOCK_LOCK(*spp); 2177 soref(*spp); 2178 SOCK_UNLOCK(*spp); 2179 } 2180 FILEDESC_SUNLOCK(td->td_proc->p_fd); 2181 return (error); 2182} 2183 2184/* 2185 * Drop the reference count on the socket and XXX release the SX lock in the 2186 * future. The last reference closes the socket. 2187 * 2188 * Note: fputsock() is deprecated, see comment for fgetsock(). 2189 */ 2190void 2191fputsock(struct socket *so) 2192{ 2193 2194 ACCEPT_LOCK(); 2195 SOCK_LOCK(so); 2196 sorele(so); 2197} 2198 2199/* 2200 * Handle the last reference to a file being closed. 2201 */ 2202int 2203_fdrop(struct file *fp, struct thread *td) 2204{ 2205 int error; 2206 2207 error = 0; 2208 if (fp->f_count != 0) 2209 panic("fdrop: count %d", fp->f_count); 2210 if (fp->f_ops != &badfileops) 2211 error = fo_close(fp, td); 2212 /* 2213 * The f_cdevpriv cannot be assigned non-NULL value while we 2214 * are destroying the file. 2215 */ 2216 if (fp->f_cdevpriv != NULL) 2217 devfs_fpdrop(fp); 2218 atomic_subtract_int(&openfiles, 1); 2219 crfree(fp->f_cred); 2220 uma_zfree(file_zone, fp); 2221 2222 return (error); 2223} 2224 2225/* 2226 * Apply an advisory lock on a file descriptor. 2227 * 2228 * Just attempt to get a record lock of the requested type on the entire file 2229 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2230 */ 2231#ifndef _SYS_SYSPROTO_H_ 2232struct flock_args { 2233 int fd; 2234 int how; 2235}; 2236#endif 2237/* ARGSUSED */ 2238int 2239flock(struct thread *td, struct flock_args *uap) 2240{ 2241 struct file *fp; 2242 struct vnode *vp; 2243 struct flock lf; 2244 int vfslocked; 2245 int error; 2246 2247 if ((error = fget(td, uap->fd, &fp)) != 0) 2248 return (error); 2249 if (fp->f_type != DTYPE_VNODE) { 2250 fdrop(fp, td); 2251 return (EOPNOTSUPP); 2252 } 2253 2254 vp = fp->f_vnode; 2255 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2256 lf.l_whence = SEEK_SET; 2257 lf.l_start = 0; 2258 lf.l_len = 0; 2259 if (uap->how & LOCK_UN) { 2260 lf.l_type = F_UNLCK; 2261 atomic_clear_int(&fp->f_flag, FHASLOCK); 2262 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2263 goto done2; 2264 } 2265 if (uap->how & LOCK_EX) 2266 lf.l_type = F_WRLCK; 2267 else if (uap->how & LOCK_SH) 2268 lf.l_type = F_RDLCK; 2269 else { 2270 error = EBADF; 2271 goto done2; 2272 } 2273 atomic_set_int(&fp->f_flag, FHASLOCK); 2274 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2275 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2276done2: 2277 fdrop(fp, td); 2278 VFS_UNLOCK_GIANT(vfslocked); 2279 return (error); 2280} 2281/* 2282 * Duplicate the specified descriptor to a free descriptor. 2283 */ 2284int 2285dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2286{ 2287 struct file *wfp; 2288 struct file *fp; 2289 2290 /* 2291 * If the to-be-dup'd fd number is greater than the allowed number 2292 * of file descriptors, or the fd to be dup'd has already been 2293 * closed, then reject. 2294 */ 2295 FILEDESC_XLOCK(fdp); 2296 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2297 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2298 FILEDESC_XUNLOCK(fdp); 2299 return (EBADF); 2300 } 2301 2302 /* 2303 * There are two cases of interest here. 2304 * 2305 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2306 * 2307 * For ENXIO steal away the file structure from (dfd) and store it in 2308 * (indx). (dfd) is effectively closed by this operation. 2309 * 2310 * Any other error code is just returned. 2311 */ 2312 switch (error) { 2313 case ENODEV: 2314 /* 2315 * Check that the mode the file is being opened for is a 2316 * subset of the mode of the existing descriptor. 2317 */ 2318 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2319 FILEDESC_XUNLOCK(fdp); 2320 return (EACCES); 2321 } 2322 fp = fdp->fd_ofiles[indx]; 2323 fdp->fd_ofiles[indx] = wfp; 2324 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2325 if (fp == NULL) 2326 fdused(fdp, indx); 2327 fhold(wfp); 2328 FILEDESC_XUNLOCK(fdp); 2329 if (fp != NULL) 2330 /* 2331 * We now own the reference to fp that the ofiles[] 2332 * array used to own. Release it. 2333 */ 2334 fdrop(fp, td); 2335 return (0); 2336 2337 case ENXIO: 2338 /* 2339 * Steal away the file pointer from dfd and stuff it into indx. 2340 */ 2341 fp = fdp->fd_ofiles[indx]; 2342 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2343 fdp->fd_ofiles[dfd] = NULL; 2344 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2345 fdp->fd_ofileflags[dfd] = 0; 2346 fdunused(fdp, dfd); 2347 if (fp == NULL) 2348 fdused(fdp, indx); 2349 FILEDESC_XUNLOCK(fdp); 2350 2351 /* 2352 * We now own the reference to fp that the ofiles[] array 2353 * used to own. Release it. 2354 */ 2355 if (fp != NULL) 2356 fdrop(fp, td); 2357 return (0); 2358 2359 default: 2360 FILEDESC_XUNLOCK(fdp); 2361 return (error); 2362 } 2363 /* NOTREACHED */ 2364} 2365 2366/* 2367 * Scan all active processes to see if any of them have a current or root 2368 * directory of `olddp'. If so, replace them with the new mount point. 2369 */ 2370void 2371mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2372{ 2373 struct filedesc *fdp; 2374 struct proc *p; 2375 int nrele; 2376 2377 if (vrefcnt(olddp) == 1) 2378 return; 2379 sx_slock(&allproc_lock); 2380 FOREACH_PROC_IN_SYSTEM(p) { 2381 fdp = fdhold(p); 2382 if (fdp == NULL) 2383 continue; 2384 nrele = 0; 2385 FILEDESC_XLOCK(fdp); 2386 if (fdp->fd_cdir == olddp) { 2387 vref(newdp); 2388 fdp->fd_cdir = newdp; 2389 nrele++; 2390 } 2391 if (fdp->fd_rdir == olddp) { 2392 vref(newdp); 2393 fdp->fd_rdir = newdp; 2394 nrele++; 2395 } 2396 FILEDESC_XUNLOCK(fdp); 2397 fddrop(fdp); 2398 while (nrele--) 2399 vrele(olddp); 2400 } 2401 sx_sunlock(&allproc_lock); 2402 if (rootvnode == olddp) { 2403 vrele(rootvnode); 2404 vref(newdp); 2405 rootvnode = newdp; 2406 } 2407} 2408 2409struct filedesc_to_leader * 2410filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2411{ 2412 struct filedesc_to_leader *fdtol; 2413 2414 fdtol = malloc(sizeof(struct filedesc_to_leader), 2415 M_FILEDESC_TO_LEADER, 2416 M_WAITOK); 2417 fdtol->fdl_refcount = 1; 2418 fdtol->fdl_holdcount = 0; 2419 fdtol->fdl_wakeup = 0; 2420 fdtol->fdl_leader = leader; 2421 if (old != NULL) { 2422 FILEDESC_XLOCK(fdp); 2423 fdtol->fdl_next = old->fdl_next; 2424 fdtol->fdl_prev = old; 2425 old->fdl_next = fdtol; 2426 fdtol->fdl_next->fdl_prev = fdtol; 2427 FILEDESC_XUNLOCK(fdp); 2428 } else { 2429 fdtol->fdl_next = fdtol; 2430 fdtol->fdl_prev = fdtol; 2431 } 2432 return (fdtol); 2433} 2434 2435/* 2436 * Get file structures globally. 2437 */ 2438static int 2439sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2440{ 2441 struct xfile xf; 2442 struct filedesc *fdp; 2443 struct file *fp; 2444 struct proc *p; 2445 int error, n; 2446 2447 error = sysctl_wire_old_buffer(req, 0); 2448 if (error != 0) 2449 return (error); 2450 if (req->oldptr == NULL) { 2451 n = 0; 2452 sx_slock(&allproc_lock); 2453 FOREACH_PROC_IN_SYSTEM(p) { 2454 if (p->p_state == PRS_NEW) 2455 continue; 2456 fdp = fdhold(p); 2457 if (fdp == NULL) 2458 continue; 2459 /* overestimates sparse tables. */ 2460 if (fdp->fd_lastfile > 0) 2461 n += fdp->fd_lastfile; 2462 fddrop(fdp); 2463 } 2464 sx_sunlock(&allproc_lock); 2465 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2466 } 2467 error = 0; 2468 bzero(&xf, sizeof(xf)); 2469 xf.xf_size = sizeof(xf); 2470 sx_slock(&allproc_lock); 2471 FOREACH_PROC_IN_SYSTEM(p) { 2472 if (p->p_state == PRS_NEW) 2473 continue; 2474 PROC_LOCK(p); 2475 if (p_cansee(req->td, p) != 0) { 2476 PROC_UNLOCK(p); 2477 continue; 2478 } 2479 xf.xf_pid = p->p_pid; 2480 xf.xf_uid = p->p_ucred->cr_uid; 2481 PROC_UNLOCK(p); 2482 fdp = fdhold(p); 2483 if (fdp == NULL) 2484 continue; 2485 FILEDESC_SLOCK(fdp); 2486 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2487 if ((fp = fdp->fd_ofiles[n]) == NULL) 2488 continue; 2489 xf.xf_fd = n; 2490 xf.xf_file = fp; 2491 xf.xf_data = fp->f_data; 2492 xf.xf_vnode = fp->f_vnode; 2493 xf.xf_type = fp->f_type; 2494 xf.xf_count = fp->f_count; 2495 xf.xf_msgcount = 0; 2496 xf.xf_offset = fp->f_offset; 2497 xf.xf_flag = fp->f_flag; 2498 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2499 if (error) 2500 break; 2501 } 2502 FILEDESC_SUNLOCK(fdp); 2503 fddrop(fdp); 2504 if (error) 2505 break; 2506 } 2507 sx_sunlock(&allproc_lock); 2508 return (error); 2509} 2510 2511SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2512 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2513 2514#ifdef KINFO_OFILE_SIZE 2515CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2516#endif 2517 2518#ifdef COMPAT_FREEBSD7 2519static int 2520export_vnode_for_osysctl(struct vnode *vp, int type, 2521 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2522{ 2523 int error; 2524 char *fullpath, *freepath; 2525 int vfslocked; 2526 2527 bzero(kif, sizeof(*kif)); 2528 kif->kf_structsize = sizeof(*kif); 2529 2530 vref(vp); 2531 kif->kf_fd = type; 2532 kif->kf_type = KF_TYPE_VNODE; 2533 /* This function only handles directories. */ 2534 KASSERT(vp->v_type == VDIR, ("export_vnode_for_osysctl: vnode not directory")); 2535 kif->kf_vnode_type = KF_VTYPE_VDIR; 2536 2537 /* 2538 * This is not a true file descriptor, so we set a bogus refcount 2539 * and offset to indicate these fields should be ignored. 2540 */ 2541 kif->kf_ref_count = -1; 2542 kif->kf_offset = -1; 2543 2544 freepath = NULL; 2545 fullpath = "-"; 2546 FILEDESC_SUNLOCK(fdp); 2547 vn_fullpath(curthread, vp, &fullpath, &freepath); 2548 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2549 vrele(vp); 2550 VFS_UNLOCK_GIANT(vfslocked); 2551 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2552 if (freepath != NULL) 2553 free(freepath, M_TEMP); 2554 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2555 FILEDESC_SLOCK(fdp); 2556 return (error); 2557} 2558 2559/* 2560 * Get per-process file descriptors for use by procstat(1), et al. 2561 */ 2562static int 2563sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2564{ 2565 char *fullpath, *freepath; 2566 struct kinfo_ofile *kif; 2567 struct filedesc *fdp; 2568 int error, i, *name; 2569 struct socket *so; 2570 struct vnode *vp; 2571 struct file *fp; 2572 struct proc *p; 2573 struct tty *tp; 2574 int vfslocked; 2575 2576 name = (int *)arg1; 2577 if ((p = pfind((pid_t)name[0])) == NULL) 2578 return (ESRCH); 2579 if ((error = p_candebug(curthread, p))) { 2580 PROC_UNLOCK(p); 2581 return (error); 2582 } 2583 fdp = fdhold(p); 2584 PROC_UNLOCK(p); 2585 if (fdp == NULL) 2586 return (ENOENT); 2587 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2588 FILEDESC_SLOCK(fdp); 2589 if (fdp->fd_cdir != NULL) 2590 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2591 fdp, req); 2592 if (fdp->fd_rdir != NULL) 2593 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2594 fdp, req); 2595 if (fdp->fd_jdir != NULL) 2596 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2597 fdp, req); 2598 for (i = 0; i < fdp->fd_nfiles; i++) { 2599 if ((fp = fdp->fd_ofiles[i]) == NULL) 2600 continue; 2601 bzero(kif, sizeof(*kif)); 2602 kif->kf_structsize = sizeof(*kif); 2603 vp = NULL; 2604 so = NULL; 2605 tp = NULL; 2606 kif->kf_fd = i; 2607 switch (fp->f_type) { 2608 case DTYPE_VNODE: 2609 kif->kf_type = KF_TYPE_VNODE; 2610 vp = fp->f_vnode; 2611 break; 2612 2613 case DTYPE_SOCKET: 2614 kif->kf_type = KF_TYPE_SOCKET; 2615 so = fp->f_data; 2616 break; 2617 2618 case DTYPE_PIPE: 2619 kif->kf_type = KF_TYPE_PIPE; 2620 break; 2621 2622 case DTYPE_FIFO: 2623 kif->kf_type = KF_TYPE_FIFO; 2624 vp = fp->f_vnode; 2625 vref(vp); 2626 break; 2627 2628 case DTYPE_KQUEUE: 2629 kif->kf_type = KF_TYPE_KQUEUE; 2630 break; 2631 2632 case DTYPE_CRYPTO: 2633 kif->kf_type = KF_TYPE_CRYPTO; 2634 break; 2635 2636 case DTYPE_MQUEUE: 2637 kif->kf_type = KF_TYPE_MQUEUE; 2638 break; 2639 2640 case DTYPE_SHM: 2641 kif->kf_type = KF_TYPE_SHM; 2642 break; 2643 2644 case DTYPE_SEM: 2645 kif->kf_type = KF_TYPE_SEM; 2646 break; 2647 2648 case DTYPE_PTS: 2649 kif->kf_type = KF_TYPE_PTS; 2650 tp = fp->f_data; 2651 break; 2652 2653 default: 2654 kif->kf_type = KF_TYPE_UNKNOWN; 2655 break; 2656 } 2657 kif->kf_ref_count = fp->f_count; 2658 if (fp->f_flag & FREAD) 2659 kif->kf_flags |= KF_FLAG_READ; 2660 if (fp->f_flag & FWRITE) 2661 kif->kf_flags |= KF_FLAG_WRITE; 2662 if (fp->f_flag & FAPPEND) 2663 kif->kf_flags |= KF_FLAG_APPEND; 2664 if (fp->f_flag & FASYNC) 2665 kif->kf_flags |= KF_FLAG_ASYNC; 2666 if (fp->f_flag & FFSYNC) 2667 kif->kf_flags |= KF_FLAG_FSYNC; 2668 if (fp->f_flag & FNONBLOCK) 2669 kif->kf_flags |= KF_FLAG_NONBLOCK; 2670 if (fp->f_flag & O_DIRECT) 2671 kif->kf_flags |= KF_FLAG_DIRECT; 2672 if (fp->f_flag & FHASLOCK) 2673 kif->kf_flags |= KF_FLAG_HASLOCK; 2674 kif->kf_offset = fp->f_offset; 2675 if (vp != NULL) { 2676 vref(vp); 2677 switch (vp->v_type) { 2678 case VNON: 2679 kif->kf_vnode_type = KF_VTYPE_VNON; 2680 break; 2681 case VREG: 2682 kif->kf_vnode_type = KF_VTYPE_VREG; 2683 break; 2684 case VDIR: 2685 kif->kf_vnode_type = KF_VTYPE_VDIR; 2686 break; 2687 case VBLK: 2688 kif->kf_vnode_type = KF_VTYPE_VBLK; 2689 break; 2690 case VCHR: 2691 kif->kf_vnode_type = KF_VTYPE_VCHR; 2692 break; 2693 case VLNK: 2694 kif->kf_vnode_type = KF_VTYPE_VLNK; 2695 break; 2696 case VSOCK: 2697 kif->kf_vnode_type = KF_VTYPE_VSOCK; 2698 break; 2699 case VFIFO: 2700 kif->kf_vnode_type = KF_VTYPE_VFIFO; 2701 break; 2702 case VBAD: 2703 kif->kf_vnode_type = KF_VTYPE_VBAD; 2704 break; 2705 default: 2706 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 2707 break; 2708 } 2709 /* 2710 * It is OK to drop the filedesc lock here as we will 2711 * re-validate and re-evaluate its properties when 2712 * the loop continues. 2713 */ 2714 freepath = NULL; 2715 fullpath = "-"; 2716 FILEDESC_SUNLOCK(fdp); 2717 vn_fullpath(curthread, vp, &fullpath, &freepath); 2718 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2719 vrele(vp); 2720 VFS_UNLOCK_GIANT(vfslocked); 2721 strlcpy(kif->kf_path, fullpath, 2722 sizeof(kif->kf_path)); 2723 if (freepath != NULL) 2724 free(freepath, M_TEMP); 2725 FILEDESC_SLOCK(fdp); 2726 } 2727 if (so != NULL) { 2728 struct sockaddr *sa; 2729 2730 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 2731 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 2732 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 2733 free(sa, M_SONAME); 2734 } 2735 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 2736 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 2737 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 2738 free(sa, M_SONAME); 2739 } 2740 kif->kf_sock_domain = 2741 so->so_proto->pr_domain->dom_family; 2742 kif->kf_sock_type = so->so_type; 2743 kif->kf_sock_protocol = so->so_proto->pr_protocol; 2744 } 2745 if (tp != NULL) { 2746 strlcpy(kif->kf_path, tty_devname(tp), 2747 sizeof(kif->kf_path)); 2748 } 2749 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2750 if (error) 2751 break; 2752 } 2753 FILEDESC_SUNLOCK(fdp); 2754 fddrop(fdp); 2755 free(kif, M_TEMP); 2756 return (0); 2757} 2758 2759static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 2760 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 2761#endif /* COMPAT_FREEBSD7 */ 2762 2763#ifdef KINFO_FILE_SIZE 2764CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2765#endif 2766 2767static int 2768export_vnode_for_sysctl(struct vnode *vp, int type, 2769 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req) 2770{ 2771 int error; 2772 char *fullpath, *freepath; 2773 int vfslocked; 2774 2775 bzero(kif, sizeof(*kif)); 2776 2777 vref(vp); 2778 kif->kf_fd = type; 2779 kif->kf_type = KF_TYPE_VNODE; 2780 /* This function only handles directories. */ 2781 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory")); 2782 kif->kf_vnode_type = KF_VTYPE_VDIR; 2783 2784 /* 2785 * This is not a true file descriptor, so we set a bogus refcount 2786 * and offset to indicate these fields should be ignored. 2787 */ 2788 kif->kf_ref_count = -1; 2789 kif->kf_offset = -1; 2790 2791 freepath = NULL; 2792 fullpath = "-"; 2793 FILEDESC_SUNLOCK(fdp); 2794 vn_fullpath(curthread, vp, &fullpath, &freepath); 2795 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2796 vrele(vp); 2797 VFS_UNLOCK_GIANT(vfslocked); 2798 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2799 if (freepath != NULL) 2800 free(freepath, M_TEMP); 2801 /* Pack record size down */ 2802 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 2803 strlen(kif->kf_path) + 1; 2804 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 2805 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 2806 FILEDESC_SLOCK(fdp); 2807 return (error); 2808} 2809 2810/* 2811 * Get per-process file descriptors for use by procstat(1), et al. 2812 */ 2813static int 2814sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 2815{ 2816 char *fullpath, *freepath; 2817 struct kinfo_file *kif; 2818 struct filedesc *fdp; 2819 int error, i, *name; 2820 struct socket *so; 2821 struct vnode *vp; 2822 struct file *fp; 2823 struct proc *p; 2824 struct tty *tp; 2825 int vfslocked; 2826 2827 name = (int *)arg1; 2828 if ((p = pfind((pid_t)name[0])) == NULL) 2829 return (ESRCH); 2830 if ((error = p_candebug(curthread, p))) { 2831 PROC_UNLOCK(p); 2832 return (error); 2833 } 2834 fdp = fdhold(p); 2835 PROC_UNLOCK(p); 2836 if (fdp == NULL) 2837 return (ENOENT); 2838 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2839 FILEDESC_SLOCK(fdp); 2840 if (fdp->fd_cdir != NULL) 2841 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2842 fdp, req); 2843 if (fdp->fd_rdir != NULL) 2844 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2845 fdp, req); 2846 if (fdp->fd_jdir != NULL) 2847 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2848 fdp, req); 2849 for (i = 0; i < fdp->fd_nfiles; i++) { 2850 if ((fp = fdp->fd_ofiles[i]) == NULL) 2851 continue; 2852 bzero(kif, sizeof(*kif)); 2853 vp = NULL; 2854 so = NULL; 2855 tp = NULL; 2856 kif->kf_fd = i; 2857 switch (fp->f_type) { 2858 case DTYPE_VNODE: 2859 kif->kf_type = KF_TYPE_VNODE; 2860 vp = fp->f_vnode; 2861 break; 2862 2863 case DTYPE_SOCKET: 2864 kif->kf_type = KF_TYPE_SOCKET; 2865 so = fp->f_data; 2866 break; 2867 2868 case DTYPE_PIPE: 2869 kif->kf_type = KF_TYPE_PIPE; 2870 break; 2871 2872 case DTYPE_FIFO: 2873 kif->kf_type = KF_TYPE_FIFO; 2874 vp = fp->f_vnode; 2875 vref(vp); 2876 break; 2877 2878 case DTYPE_KQUEUE: 2879 kif->kf_type = KF_TYPE_KQUEUE; 2880 break; 2881 2882 case DTYPE_CRYPTO: 2883 kif->kf_type = KF_TYPE_CRYPTO; 2884 break; 2885 2886 case DTYPE_MQUEUE: 2887 kif->kf_type = KF_TYPE_MQUEUE; 2888 break; 2889 2890 case DTYPE_SHM: 2891 kif->kf_type = KF_TYPE_SHM; 2892 break; 2893 2894 case DTYPE_SEM: 2895 kif->kf_type = KF_TYPE_SEM; 2896 break; 2897 2898 case DTYPE_PTS: 2899 kif->kf_type = KF_TYPE_PTS; 2900 tp = fp->f_data; 2901 break; 2902 2903 default: 2904 kif->kf_type = KF_TYPE_UNKNOWN; 2905 break; 2906 } 2907 kif->kf_ref_count = fp->f_count; 2908 if (fp->f_flag & FREAD) 2909 kif->kf_flags |= KF_FLAG_READ; 2910 if (fp->f_flag & FWRITE) 2911 kif->kf_flags |= KF_FLAG_WRITE; 2912 if (fp->f_flag & FAPPEND) 2913 kif->kf_flags |= KF_FLAG_APPEND; 2914 if (fp->f_flag & FASYNC) 2915 kif->kf_flags |= KF_FLAG_ASYNC; 2916 if (fp->f_flag & FFSYNC) 2917 kif->kf_flags |= KF_FLAG_FSYNC; 2918 if (fp->f_flag & FNONBLOCK) 2919 kif->kf_flags |= KF_FLAG_NONBLOCK; 2920 if (fp->f_flag & O_DIRECT) 2921 kif->kf_flags |= KF_FLAG_DIRECT; 2922 if (fp->f_flag & FHASLOCK) 2923 kif->kf_flags |= KF_FLAG_HASLOCK; 2924 kif->kf_offset = fp->f_offset; 2925 if (vp != NULL) { 2926 vref(vp); 2927 switch (vp->v_type) { 2928 case VNON: 2929 kif->kf_vnode_type = KF_VTYPE_VNON; 2930 break; 2931 case VREG: 2932 kif->kf_vnode_type = KF_VTYPE_VREG; 2933 break; 2934 case VDIR: 2935 kif->kf_vnode_type = KF_VTYPE_VDIR; 2936 break; 2937 case VBLK: 2938 kif->kf_vnode_type = KF_VTYPE_VBLK; 2939 break; 2940 case VCHR: 2941 kif->kf_vnode_type = KF_VTYPE_VCHR; 2942 break; 2943 case VLNK: 2944 kif->kf_vnode_type = KF_VTYPE_VLNK; 2945 break; 2946 case VSOCK: 2947 kif->kf_vnode_type = KF_VTYPE_VSOCK; 2948 break; 2949 case VFIFO: 2950 kif->kf_vnode_type = KF_VTYPE_VFIFO; 2951 break; 2952 case VBAD: 2953 kif->kf_vnode_type = KF_VTYPE_VBAD; 2954 break; 2955 default: 2956 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 2957 break; 2958 } 2959 /* 2960 * It is OK to drop the filedesc lock here as we will 2961 * re-validate and re-evaluate its properties when 2962 * the loop continues. 2963 */ 2964 freepath = NULL; 2965 fullpath = "-"; 2966 FILEDESC_SUNLOCK(fdp); 2967 vn_fullpath(curthread, vp, &fullpath, &freepath); 2968 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2969 vrele(vp); 2970 VFS_UNLOCK_GIANT(vfslocked); 2971 strlcpy(kif->kf_path, fullpath, 2972 sizeof(kif->kf_path)); 2973 if (freepath != NULL) 2974 free(freepath, M_TEMP); 2975 FILEDESC_SLOCK(fdp); 2976 } 2977 if (so != NULL) { 2978 struct sockaddr *sa; 2979 2980 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 2981 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 2982 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 2983 free(sa, M_SONAME); 2984 } 2985 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 2986 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 2987 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 2988 free(sa, M_SONAME); 2989 } 2990 kif->kf_sock_domain = 2991 so->so_proto->pr_domain->dom_family; 2992 kif->kf_sock_type = so->so_type; 2993 kif->kf_sock_protocol = so->so_proto->pr_protocol; 2994 } 2995 if (tp != NULL) { 2996 strlcpy(kif->kf_path, tty_devname(tp), 2997 sizeof(kif->kf_path)); 2998 } 2999 /* Pack record size down */ 3000 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3001 strlen(kif->kf_path) + 1; 3002 kif->kf_structsize = roundup(kif->kf_structsize, 3003 sizeof(uint64_t)); 3004 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 3005 if (error) 3006 break; 3007 } 3008 FILEDESC_SUNLOCK(fdp); 3009 fddrop(fdp); 3010 free(kif, M_TEMP); 3011 return (0); 3012} 3013 3014static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3015 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3016 3017#ifdef DDB 3018/* 3019 * For the purposes of debugging, generate a human-readable string for the 3020 * file type. 3021 */ 3022static const char * 3023file_type_to_name(short type) 3024{ 3025 3026 switch (type) { 3027 case 0: 3028 return ("zero"); 3029 case DTYPE_VNODE: 3030 return ("vnod"); 3031 case DTYPE_SOCKET: 3032 return ("sock"); 3033 case DTYPE_PIPE: 3034 return ("pipe"); 3035 case DTYPE_FIFO: 3036 return ("fifo"); 3037 case DTYPE_KQUEUE: 3038 return ("kque"); 3039 case DTYPE_CRYPTO: 3040 return ("crpt"); 3041 case DTYPE_MQUEUE: 3042 return ("mque"); 3043 case DTYPE_SHM: 3044 return ("shm"); 3045 case DTYPE_SEM: 3046 return ("ksem"); 3047 default: 3048 return ("unkn"); 3049 } 3050} 3051 3052/* 3053 * For the purposes of debugging, identify a process (if any, perhaps one of 3054 * many) that references the passed file in its file descriptor array. Return 3055 * NULL if none. 3056 */ 3057static struct proc * 3058file_to_first_proc(struct file *fp) 3059{ 3060 struct filedesc *fdp; 3061 struct proc *p; 3062 int n; 3063 3064 FOREACH_PROC_IN_SYSTEM(p) { 3065 if (p->p_state == PRS_NEW) 3066 continue; 3067 fdp = p->p_fd; 3068 if (fdp == NULL) 3069 continue; 3070 for (n = 0; n < fdp->fd_nfiles; n++) { 3071 if (fp == fdp->fd_ofiles[n]) 3072 return (p); 3073 } 3074 } 3075 return (NULL); 3076} 3077 3078static void 3079db_print_file(struct file *fp, int header) 3080{ 3081 struct proc *p; 3082 3083 if (header) 3084 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3085 "File", "Type", "Data", "Flag", "GCFl", "Count", 3086 "MCount", "Vnode", "FPID", "FCmd"); 3087 p = file_to_first_proc(fp); 3088 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3089 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3090 0, fp->f_count, 0, fp->f_vnode, 3091 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3092} 3093 3094DB_SHOW_COMMAND(file, db_show_file) 3095{ 3096 struct file *fp; 3097 3098 if (!have_addr) { 3099 db_printf("usage: show file <addr>\n"); 3100 return; 3101 } 3102 fp = (struct file *)addr; 3103 db_print_file(fp, 1); 3104} 3105 3106DB_SHOW_COMMAND(files, db_show_files) 3107{ 3108 struct filedesc *fdp; 3109 struct file *fp; 3110 struct proc *p; 3111 int header; 3112 int n; 3113 3114 header = 1; 3115 FOREACH_PROC_IN_SYSTEM(p) { 3116 if (p->p_state == PRS_NEW) 3117 continue; 3118 if ((fdp = p->p_fd) == NULL) 3119 continue; 3120 for (n = 0; n < fdp->fd_nfiles; ++n) { 3121 if ((fp = fdp->fd_ofiles[n]) == NULL) 3122 continue; 3123 db_print_file(fp, header); 3124 header = 0; 3125 } 3126 } 3127} 3128#endif 3129 3130SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3131 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3132 3133SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3134 &maxfiles, 0, "Maximum number of files"); 3135 3136SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3137 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3138 3139/* ARGSUSED*/ 3140static void 3141filelistinit(void *dummy) 3142{ 3143 3144 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3145 NULL, NULL, UMA_ALIGN_PTR, 0); 3146 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3147 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3148} 3149SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3150 3151/*-------------------------------------------------------------------*/ 3152 3153static int 3154badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) 3155{ 3156 3157 return (EBADF); 3158} 3159 3160static int 3161badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) 3162{ 3163 3164 return (EINVAL); 3165} 3166 3167static int 3168badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) 3169{ 3170 3171 return (EBADF); 3172} 3173 3174static int 3175badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) 3176{ 3177 3178 return (0); 3179} 3180 3181static int 3182badfo_kqfilter(struct file *fp, struct knote *kn) 3183{ 3184 3185 return (EBADF); 3186} 3187 3188static int 3189badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) 3190{ 3191 3192 return (EBADF); 3193} 3194 3195static int 3196badfo_close(struct file *fp, struct thread *td) 3197{ 3198 3199 return (EBADF); 3200} 3201 3202struct fileops badfileops = { 3203 .fo_read = badfo_readwrite, 3204 .fo_write = badfo_readwrite, 3205 .fo_truncate = badfo_truncate, 3206 .fo_ioctl = badfo_ioctl, 3207 .fo_poll = badfo_poll, 3208 .fo_kqfilter = badfo_kqfilter, 3209 .fo_stat = badfo_stat, 3210 .fo_close = badfo_close, 3211}; 3212 3213 3214/*-------------------------------------------------------------------*/ 3215 3216/* 3217 * File Descriptor pseudo-device driver (/dev/fd/). 3218 * 3219 * Opening minor device N dup()s the file (if any) connected to file 3220 * descriptor N belonging to the calling process. Note that this driver 3221 * consists of only the ``open()'' routine, because all subsequent 3222 * references to this file will be direct to the other driver. 3223 * 3224 * XXX: we could give this one a cloning event handler if necessary. 3225 */ 3226 3227/* ARGSUSED */ 3228static int 3229fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3230{ 3231 3232 /* 3233 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3234 * the file descriptor being sought for duplication. The error 3235 * return ensures that the vnode for this device will be released 3236 * by vn_open. Open will detect this special error and take the 3237 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3238 * will simply report the error. 3239 */ 3240 td->td_dupfd = dev2unit(dev); 3241 return (ENODEV); 3242} 3243 3244static struct cdevsw fildesc_cdevsw = { 3245 .d_version = D_VERSION, 3246 .d_open = fdopen, 3247 .d_name = "FD", 3248}; 3249 3250static void 3251fildesc_drvinit(void *unused) 3252{ 3253 struct cdev *dev; 3254 3255 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3256 make_dev_alias(dev, "stdin"); 3257 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3258 make_dev_alias(dev, "stdout"); 3259 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3260 make_dev_alias(dev, "stderr"); 3261} 3262 3263SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3264