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("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 #include "opt_ddb.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 46 #include <sys/conf.h> 47 #include <sys/fcntl.h> 48 #include <sys/file.h> 49 #include <sys/filedesc.h> 50 #include <sys/filio.h> 51 #include <sys/jail.h> 52 #include <sys/kernel.h> 53 #include <sys/limits.h> 54 #include <sys/lock.h> 55 #include <sys/malloc.h> 56 #include <sys/mount.h> 57 #include <sys/mqueue.h> 58 #include <sys/mutex.h> 59 #include <sys/namei.h> 60 #include <sys/proc.h> 61 #include <sys/resourcevar.h> 62 #include <sys/signalvar.h> 63 #include <sys/socketvar.h> 64 #include <sys/stat.h> 65 #include <sys/sx.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/sysctl.h> 68 #include <sys/sysproto.h> 69 #include <sys/unistd.h> 70 #include <sys/vnode.h> 71 72 #include <security/audit/audit.h> 73 74 #include <vm/uma.h> 75 76 #include <ddb/ddb.h> 77 78 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 79 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 80 "file desc to leader structures"); 81 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 82 83 static uma_zone_t file_zone; 84 85 86 /* How to treat 'new' parameter when allocating a fd for do_dup(). */ 87 enum dup_type { DUP_VARIABLE, DUP_FIXED }; 88 89 static int do_dup(struct thread *td, enum dup_type type, int old, int new, 90 register_t *retval); 91 static int fd_first_free(struct filedesc *, int, int); 92 static int fd_last_used(struct filedesc *, int, int); 93 static void fdgrowtable(struct filedesc *, int); 94 static int fdrop_locked(struct file *fp, struct thread *td); 95 static void fdunused(struct filedesc *fdp, int fd); 96 static void fdused(struct filedesc *fdp, int fd); 97 98 /* 99 * A process is initially started out with NDFILE descriptors stored within 100 * this structure, selected to be enough for typical applications based on 101 * the historical limit of 20 open files (and the usage of descriptors by 102 * shells). If these descriptors are exhausted, a larger descriptor table 103 * may be allocated, up to a process' resource limit; the internal arrays 104 * are then unused. 105 */ 106 #define NDFILE 20 107 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 108 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 109 #define NDSLOT(x) ((x) / NDENTRIES) 110 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 111 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 112 113 /* 114 * Storage required per open file descriptor. 115 */ 116 #define OFILESIZE (sizeof(struct file *) + sizeof(char)) 117 118 /* 119 * Basic allocation of descriptors: 120 * one of the above, plus arrays for NDFILE descriptors. 121 */ 122 struct filedesc0 { 123 struct filedesc fd_fd; 124 /* 125 * These arrays are used when the number of open files is 126 * <= NDFILE, and are then pointed to by the pointers above. 127 */ 128 struct file *fd_dfiles[NDFILE]; 129 char fd_dfileflags[NDFILE]; 130 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 131 }; 132 133 /* 134 * Descriptor management. 135 */ 136 struct filelist filehead; /* head of list of open files */ 137 int openfiles; /* actual number of open files */ 138 struct sx filelist_lock; /* sx to protect filelist */ 139 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 140 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 141 142 /* A mutex to protect the association between a proc and filedesc. */ 143 static struct mtx fdesc_mtx; 144 145 /* 146 * Find the first zero bit in the given bitmap, starting at low and not 147 * exceeding size - 1. 148 */ 149 static int 150 fd_first_free(struct filedesc *fdp, int low, int size) 151 { 152 NDSLOTTYPE *map = fdp->fd_map; 153 NDSLOTTYPE mask; 154 int off, maxoff; 155 156 if (low >= size) 157 return (low); 158 159 off = NDSLOT(low); 160 if (low % NDENTRIES) { 161 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 162 if ((mask &= ~map[off]) != 0UL) 163 return (off * NDENTRIES + ffsl(mask) - 1); 164 ++off; 165 } 166 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 167 if (map[off] != ~0UL) 168 return (off * NDENTRIES + ffsl(~map[off]) - 1); 169 return (size); 170 } 171 172 /* 173 * Find the highest non-zero bit in the given bitmap, starting at low and 174 * not exceeding size - 1. 175 */ 176 static int 177 fd_last_used(struct filedesc *fdp, int low, int size) 178 { 179 NDSLOTTYPE *map = fdp->fd_map; 180 NDSLOTTYPE mask; 181 int off, minoff; 182 183 if (low >= size) 184 return (-1); 185 186 off = NDSLOT(size); 187 if (size % NDENTRIES) { 188 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 189 if ((mask &= map[off]) != 0) 190 return (off * NDENTRIES + flsl(mask) - 1); 191 --off; 192 } 193 for (minoff = NDSLOT(low); off >= minoff; --off) 194 if (map[off] != 0) 195 return (off * NDENTRIES + flsl(map[off]) - 1); 196 return (low - 1); 197 } 198 199 static int 200 fdisused(struct filedesc *fdp, int fd) 201 { 202 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 203 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 204 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 205 } 206 207 /* 208 * Mark a file descriptor as used. 209 */ 210 static void 211 fdused(struct filedesc *fdp, int fd) 212 { 213 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 214 KASSERT(!fdisused(fdp, fd), 215 ("fd already used")); 216 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 217 if (fd > fdp->fd_lastfile) 218 fdp->fd_lastfile = fd; 219 if (fd == fdp->fd_freefile) 220 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 221 } 222 223 /* 224 * Mark a file descriptor as unused. 225 */ 226 static void 227 fdunused(struct filedesc *fdp, int fd) 228 { 229 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 230 KASSERT(fdisused(fdp, fd), 231 ("fd is already unused")); 232 KASSERT(fdp->fd_ofiles[fd] == NULL, 233 ("fd is still in use")); 234 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 235 if (fd < fdp->fd_freefile) 236 fdp->fd_freefile = fd; 237 if (fd == fdp->fd_lastfile) 238 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 239 } 240 241 /* 242 * System calls on descriptors. 243 */ 244 #ifndef _SYS_SYSPROTO_H_ 245 struct getdtablesize_args { 246 int dummy; 247 }; 248 #endif 249 /* 250 * MPSAFE 251 */ 252 /* ARGSUSED */ 253 int 254 getdtablesize(struct thread *td, struct getdtablesize_args *uap) 255 { 256 struct proc *p = td->td_proc; 257 258 PROC_LOCK(p); 259 td->td_retval[0] = 260 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 261 PROC_UNLOCK(p); 262 return (0); 263 } 264 265 /* 266 * Duplicate a file descriptor to a particular value. 267 * 268 * note: keep in mind that a potential race condition exists when closing 269 * descriptors from a shared descriptor table (via rfork). 270 */ 271 #ifndef _SYS_SYSPROTO_H_ 272 struct dup2_args { 273 u_int from; 274 u_int to; 275 }; 276 #endif 277 /* 278 * MPSAFE 279 */ 280 /* ARGSUSED */ 281 int 282 dup2(struct thread *td, struct dup2_args *uap) 283 { 284 285 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 286 td->td_retval)); 287 } 288 289 /* 290 * Duplicate a file descriptor. 291 */ 292 #ifndef _SYS_SYSPROTO_H_ 293 struct dup_args { 294 u_int fd; 295 }; 296 #endif 297 /* 298 * MPSAFE 299 */ 300 /* ARGSUSED */ 301 int 302 dup(struct thread *td, struct dup_args *uap) 303 { 304 305 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval)); 306 } 307 308 /* 309 * The file control system call. 310 */ 311 #ifndef _SYS_SYSPROTO_H_ 312 struct fcntl_args { 313 int fd; 314 int cmd; 315 long arg; 316 }; 317 #endif 318 /* 319 * MPSAFE 320 */ 321 /* ARGSUSED */ 322 int 323 fcntl(struct thread *td, struct fcntl_args *uap) 324 { 325 struct flock fl; 326 intptr_t arg; 327 int error; 328 329 error = 0; 330 switch (uap->cmd) { 331 case F_GETLK: 332 case F_SETLK: 333 case F_SETLKW: 334 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 335 arg = (intptr_t)&fl; 336 break; 337 default: 338 arg = uap->arg; 339 break; 340 } 341 if (error) 342 return (error); 343 error = kern_fcntl(td, uap->fd, uap->cmd, arg); 344 if (error) 345 return (error); 346 if (uap->cmd == F_GETLK) 347 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 348 return (error); 349 } 350 351 int 352 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 353 { 354 struct filedesc *fdp; 355 struct flock *flp; 356 struct file *fp; 357 struct proc *p; 358 char *pop; 359 struct vnode *vp; 360 u_int newmin; 361 int error, flg, tmp; 362 int giant_locked; 363 364 /* 365 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to 366 * avoid grabbing Giant for calls we know don't need it. 367 */ 368 switch (cmd) { 369 case F_DUPFD: 370 case F_GETFD: 371 case F_SETFD: 372 case F_GETFL: 373 giant_locked = 0; 374 break; 375 376 default: 377 giant_locked = 1; 378 mtx_lock(&Giant); 379 } 380 381 error = 0; 382 flg = F_POSIX; 383 p = td->td_proc; 384 fdp = p->p_fd; 385 FILEDESC_LOCK(fdp); 386 if ((unsigned)fd >= fdp->fd_nfiles || 387 (fp = fdp->fd_ofiles[fd]) == NULL) { 388 FILEDESC_UNLOCK(fdp); 389 error = EBADF; 390 goto done2; 391 } 392 pop = &fdp->fd_ofileflags[fd]; 393 394 switch (cmd) { 395 case F_DUPFD: 396 /* mtx_assert(&Giant, MA_NOTOWNED); */ 397 FILEDESC_UNLOCK(fdp); 398 newmin = arg; 399 PROC_LOCK(p); 400 if (newmin >= lim_cur(p, RLIMIT_NOFILE) || 401 newmin >= maxfilesperproc) { 402 PROC_UNLOCK(p); 403 error = EINVAL; 404 break; 405 } 406 PROC_UNLOCK(p); 407 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval); 408 break; 409 410 case F_GETFD: 411 /* mtx_assert(&Giant, MA_NOTOWNED); */ 412 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 413 FILEDESC_UNLOCK(fdp); 414 break; 415 416 case F_SETFD: 417 /* mtx_assert(&Giant, MA_NOTOWNED); */ 418 *pop = (*pop &~ UF_EXCLOSE) | 419 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 420 FILEDESC_UNLOCK(fdp); 421 break; 422 423 case F_GETFL: 424 /* mtx_assert(&Giant, MA_NOTOWNED); */ 425 FILE_LOCK(fp); 426 td->td_retval[0] = OFLAGS(fp->f_flag); 427 FILE_UNLOCK(fp); 428 FILEDESC_UNLOCK(fdp); 429 break; 430 431 case F_SETFL: 432 mtx_assert(&Giant, MA_OWNED); 433 FILE_LOCK(fp); 434 fhold_locked(fp); 435 fp->f_flag &= ~FCNTLFLAGS; 436 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 437 FILE_UNLOCK(fp); 438 FILEDESC_UNLOCK(fdp); 439 tmp = fp->f_flag & FNONBLOCK; 440 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 441 if (error) { 442 fdrop(fp, td); 443 break; 444 } 445 tmp = fp->f_flag & FASYNC; 446 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 447 if (error == 0) { 448 fdrop(fp, td); 449 break; 450 } 451 FILE_LOCK(fp); 452 fp->f_flag &= ~FNONBLOCK; 453 FILE_UNLOCK(fp); 454 tmp = 0; 455 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 456 fdrop(fp, td); 457 break; 458 459 case F_GETOWN: 460 mtx_assert(&Giant, MA_OWNED); 461 fhold(fp); 462 FILEDESC_UNLOCK(fdp); 463 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 464 if (error == 0) 465 td->td_retval[0] = tmp; 466 fdrop(fp, td); 467 break; 468 469 case F_SETOWN: 470 mtx_assert(&Giant, MA_OWNED); 471 fhold(fp); 472 FILEDESC_UNLOCK(fdp); 473 tmp = arg; 474 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 475 fdrop(fp, td); 476 break; 477 478 case F_SETLKW: 479 mtx_assert(&Giant, MA_OWNED); 480 flg |= F_WAIT; 481 /* FALLTHROUGH F_SETLK */ 482 483 case F_SETLK: 484 mtx_assert(&Giant, MA_OWNED); 485 if (fp->f_type != DTYPE_VNODE) { 486 FILEDESC_UNLOCK(fdp); 487 error = EBADF; 488 break; 489 } 490 491 flp = (struct flock *)arg; 492 if (flp->l_whence == SEEK_CUR) { 493 if (fp->f_offset < 0 || 494 (flp->l_start > 0 && 495 fp->f_offset > OFF_MAX - flp->l_start)) { 496 FILEDESC_UNLOCK(fdp); 497 error = EOVERFLOW; 498 break; 499 } 500 flp->l_start += fp->f_offset; 501 } 502 503 /* 504 * VOP_ADVLOCK() may block. 505 */ 506 fhold(fp); 507 FILEDESC_UNLOCK(fdp); 508 vp = fp->f_vnode; 509 510 switch (flp->l_type) { 511 case F_RDLCK: 512 if ((fp->f_flag & FREAD) == 0) { 513 error = EBADF; 514 break; 515 } 516 PROC_LOCK(p->p_leader); 517 p->p_leader->p_flag |= P_ADVLOCK; 518 PROC_UNLOCK(p->p_leader); 519 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 520 flp, flg); 521 break; 522 case F_WRLCK: 523 if ((fp->f_flag & FWRITE) == 0) { 524 error = EBADF; 525 break; 526 } 527 PROC_LOCK(p->p_leader); 528 p->p_leader->p_flag |= P_ADVLOCK; 529 PROC_UNLOCK(p->p_leader); 530 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 531 flp, flg); 532 break; 533 case F_UNLCK: 534 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 535 flp, F_POSIX); 536 break; 537 default: 538 error = EINVAL; 539 break; 540 } 541 /* Check for race with close */ 542 FILEDESC_LOCK_FAST(fdp); 543 if ((unsigned) fd >= fdp->fd_nfiles || 544 fp != fdp->fd_ofiles[fd]) { 545 FILEDESC_UNLOCK_FAST(fdp); 546 flp->l_whence = SEEK_SET; 547 flp->l_start = 0; 548 flp->l_len = 0; 549 flp->l_type = F_UNLCK; 550 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 551 F_UNLCK, flp, F_POSIX); 552 } else 553 FILEDESC_UNLOCK_FAST(fdp); 554 fdrop(fp, td); 555 break; 556 557 case F_GETLK: 558 mtx_assert(&Giant, MA_OWNED); 559 if (fp->f_type != DTYPE_VNODE) { 560 FILEDESC_UNLOCK(fdp); 561 error = EBADF; 562 break; 563 } 564 flp = (struct flock *)arg; 565 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 566 flp->l_type != F_UNLCK) { 567 FILEDESC_UNLOCK(fdp); 568 error = EINVAL; 569 break; 570 } 571 if (flp->l_whence == SEEK_CUR) { 572 if ((flp->l_start > 0 && 573 fp->f_offset > OFF_MAX - flp->l_start) || 574 (flp->l_start < 0 && 575 fp->f_offset < OFF_MIN - flp->l_start)) { 576 FILEDESC_UNLOCK(fdp); 577 error = EOVERFLOW; 578 break; 579 } 580 flp->l_start += fp->f_offset; 581 } 582 /* 583 * VOP_ADVLOCK() may block. 584 */ 585 fhold(fp); 586 FILEDESC_UNLOCK(fdp); 587 vp = fp->f_vnode; 588 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 589 F_POSIX); 590 fdrop(fp, td); 591 break; 592 default: 593 FILEDESC_UNLOCK(fdp); 594 error = EINVAL; 595 break; 596 } 597 done2: 598 if (giant_locked) 599 mtx_unlock(&Giant); 600 return (error); 601 } 602 603 /* 604 * Common code for dup, dup2, and fcntl(F_DUPFD). 605 */ 606 static int 607 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval) 608 { 609 struct filedesc *fdp; 610 struct proc *p; 611 struct file *fp; 612 struct file *delfp; 613 int error, holdleaders, maxfd; 614 615 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED), 616 ("invalid dup type %d", type)); 617 618 p = td->td_proc; 619 fdp = p->p_fd; 620 621 /* 622 * Verify we have a valid descriptor to dup from and possibly to 623 * dup to. 624 */ 625 if (old < 0 || new < 0) 626 return (EBADF); 627 PROC_LOCK(p); 628 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 629 PROC_UNLOCK(p); 630 if (new >= maxfd) 631 return (EMFILE); 632 633 FILEDESC_LOCK(fdp); 634 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 635 FILEDESC_UNLOCK(fdp); 636 return (EBADF); 637 } 638 if (type == DUP_FIXED && old == new) { 639 *retval = new; 640 FILEDESC_UNLOCK(fdp); 641 return (0); 642 } 643 fp = fdp->fd_ofiles[old]; 644 fhold(fp); 645 646 /* 647 * If the caller specified a file descriptor, make sure the file 648 * table is large enough to hold it, and grab it. Otherwise, just 649 * allocate a new descriptor the usual way. Since the filedesc 650 * lock may be temporarily dropped in the process, we have to look 651 * out for a race. 652 */ 653 if (type == DUP_FIXED) { 654 if (new >= fdp->fd_nfiles) 655 fdgrowtable(fdp, new + 1); 656 if (fdp->fd_ofiles[new] == NULL) 657 fdused(fdp, new); 658 } else { 659 if ((error = fdalloc(td, new, &new)) != 0) { 660 FILEDESC_UNLOCK(fdp); 661 fdrop(fp, td); 662 return (error); 663 } 664 } 665 666 /* 667 * If the old file changed out from under us then treat it as a 668 * bad file descriptor. Userland should do its own locking to 669 * avoid this case. 670 */ 671 if (fdp->fd_ofiles[old] != fp) { 672 /* we've allocated a descriptor which we won't use */ 673 if (fdp->fd_ofiles[new] == NULL) 674 fdunused(fdp, new); 675 FILEDESC_UNLOCK(fdp); 676 fdrop(fp, td); 677 return (EBADF); 678 } 679 KASSERT(old != new, 680 ("new fd is same as old")); 681 682 /* 683 * Save info on the descriptor being overwritten. We cannot close 684 * it without introducing an ownership race for the slot, since we 685 * need to drop the filedesc lock to call closef(). 686 * 687 * XXX this duplicates parts of close(). 688 */ 689 delfp = fdp->fd_ofiles[new]; 690 holdleaders = 0; 691 if (delfp != NULL) { 692 if (td->td_proc->p_fdtol != NULL) { 693 /* 694 * Ask fdfree() to sleep to ensure that all relevant 695 * process leaders can be traversed in closef(). 696 */ 697 fdp->fd_holdleaderscount++; 698 holdleaders = 1; 699 } 700 } 701 702 /* 703 * Duplicate the source descriptor 704 */ 705 fdp->fd_ofiles[new] = fp; 706 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 707 if (new > fdp->fd_lastfile) 708 fdp->fd_lastfile = new; 709 *retval = new; 710 711 /* 712 * If we dup'd over a valid file, we now own the reference to it 713 * and must dispose of it using closef() semantics (as if a 714 * close() were performed on it). 715 * 716 * XXX this duplicates parts of close(). 717 */ 718 if (delfp != NULL) { 719 knote_fdclose(td, new); 720 if (delfp->f_type == DTYPE_MQUEUE) 721 mq_fdclose(td, new, delfp); 722 FILEDESC_UNLOCK(fdp); 723 (void) closef(delfp, td); 724 if (holdleaders) { 725 FILEDESC_LOCK_FAST(fdp); 726 fdp->fd_holdleaderscount--; 727 if (fdp->fd_holdleaderscount == 0 && 728 fdp->fd_holdleaderswakeup != 0) { 729 fdp->fd_holdleaderswakeup = 0; 730 wakeup(&fdp->fd_holdleaderscount); 731 } 732 FILEDESC_UNLOCK_FAST(fdp); 733 } 734 } else { 735 FILEDESC_UNLOCK(fdp); 736 } 737 return (0); 738 } 739 740 /* 741 * If sigio is on the list associated with a process or process group, 742 * disable signalling from the device, remove sigio from the list and 743 * free sigio. 744 */ 745 void 746 funsetown(struct sigio **sigiop) 747 { 748 struct sigio *sigio; 749 750 SIGIO_LOCK(); 751 sigio = *sigiop; 752 if (sigio == NULL) { 753 SIGIO_UNLOCK(); 754 return; 755 } 756 *(sigio->sio_myref) = NULL; 757 if ((sigio)->sio_pgid < 0) { 758 struct pgrp *pg = (sigio)->sio_pgrp; 759 PGRP_LOCK(pg); 760 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 761 sigio, sio_pgsigio); 762 PGRP_UNLOCK(pg); 763 } else { 764 struct proc *p = (sigio)->sio_proc; 765 PROC_LOCK(p); 766 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 767 sigio, sio_pgsigio); 768 PROC_UNLOCK(p); 769 } 770 SIGIO_UNLOCK(); 771 crfree(sigio->sio_ucred); 772 FREE(sigio, M_SIGIO); 773 } 774 775 /* 776 * Free a list of sigio structures. 777 * We only need to lock the SIGIO_LOCK because we have made ourselves 778 * inaccessible to callers of fsetown and therefore do not need to lock 779 * the proc or pgrp struct for the list manipulation. 780 */ 781 void 782 funsetownlst(struct sigiolst *sigiolst) 783 { 784 struct proc *p; 785 struct pgrp *pg; 786 struct sigio *sigio; 787 788 sigio = SLIST_FIRST(sigiolst); 789 if (sigio == NULL) 790 return; 791 p = NULL; 792 pg = NULL; 793 794 /* 795 * Every entry of the list should belong 796 * to a single proc or pgrp. 797 */ 798 if (sigio->sio_pgid < 0) { 799 pg = sigio->sio_pgrp; 800 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 801 } else /* if (sigio->sio_pgid > 0) */ { 802 p = sigio->sio_proc; 803 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 804 } 805 806 SIGIO_LOCK(); 807 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 808 *(sigio->sio_myref) = NULL; 809 if (pg != NULL) { 810 KASSERT(sigio->sio_pgid < 0, 811 ("Proc sigio in pgrp sigio list")); 812 KASSERT(sigio->sio_pgrp == pg, 813 ("Bogus pgrp in sigio list")); 814 PGRP_LOCK(pg); 815 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 816 sio_pgsigio); 817 PGRP_UNLOCK(pg); 818 } else /* if (p != NULL) */ { 819 KASSERT(sigio->sio_pgid > 0, 820 ("Pgrp sigio in proc sigio list")); 821 KASSERT(sigio->sio_proc == p, 822 ("Bogus proc in sigio list")); 823 PROC_LOCK(p); 824 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 825 sio_pgsigio); 826 PROC_UNLOCK(p); 827 } 828 SIGIO_UNLOCK(); 829 crfree(sigio->sio_ucred); 830 FREE(sigio, M_SIGIO); 831 SIGIO_LOCK(); 832 } 833 SIGIO_UNLOCK(); 834 } 835 836 /* 837 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 838 * 839 * After permission checking, add a sigio structure to the sigio list for 840 * the process or process group. 841 */ 842 int 843 fsetown(pid_t pgid, struct sigio **sigiop) 844 { 845 struct proc *proc; 846 struct pgrp *pgrp; 847 struct sigio *sigio; 848 int ret; 849 850 if (pgid == 0) { 851 funsetown(sigiop); 852 return (0); 853 } 854 855 ret = 0; 856 857 /* Allocate and fill in the new sigio out of locks. */ 858 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK); 859 sigio->sio_pgid = pgid; 860 sigio->sio_ucred = crhold(curthread->td_ucred); 861 sigio->sio_myref = sigiop; 862 863 sx_slock(&proctree_lock); 864 if (pgid > 0) { 865 proc = pfind(pgid); 866 if (proc == NULL) { 867 ret = ESRCH; 868 goto fail; 869 } 870 871 /* 872 * Policy - Don't allow a process to FSETOWN a process 873 * in another session. 874 * 875 * Remove this test to allow maximum flexibility or 876 * restrict FSETOWN to the current process or process 877 * group for maximum safety. 878 */ 879 PROC_UNLOCK(proc); 880 if (proc->p_session != curthread->td_proc->p_session) { 881 ret = EPERM; 882 goto fail; 883 } 884 885 pgrp = NULL; 886 } else /* if (pgid < 0) */ { 887 pgrp = pgfind(-pgid); 888 if (pgrp == NULL) { 889 ret = ESRCH; 890 goto fail; 891 } 892 PGRP_UNLOCK(pgrp); 893 894 /* 895 * Policy - Don't allow a process to FSETOWN a process 896 * in another session. 897 * 898 * Remove this test to allow maximum flexibility or 899 * restrict FSETOWN to the current process or process 900 * group for maximum safety. 901 */ 902 if (pgrp->pg_session != curthread->td_proc->p_session) { 903 ret = EPERM; 904 goto fail; 905 } 906 907 proc = NULL; 908 } 909 funsetown(sigiop); 910 if (pgid > 0) { 911 PROC_LOCK(proc); 912 /* 913 * Since funsetownlst() is called without the proctree 914 * locked, we need to check for P_WEXIT. 915 * XXX: is ESRCH correct? 916 */ 917 if ((proc->p_flag & P_WEXIT) != 0) { 918 PROC_UNLOCK(proc); 919 ret = ESRCH; 920 goto fail; 921 } 922 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 923 sigio->sio_proc = proc; 924 PROC_UNLOCK(proc); 925 } else { 926 PGRP_LOCK(pgrp); 927 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 928 sigio->sio_pgrp = pgrp; 929 PGRP_UNLOCK(pgrp); 930 } 931 sx_sunlock(&proctree_lock); 932 SIGIO_LOCK(); 933 *sigiop = sigio; 934 SIGIO_UNLOCK(); 935 return (0); 936 937 fail: 938 sx_sunlock(&proctree_lock); 939 crfree(sigio->sio_ucred); 940 FREE(sigio, M_SIGIO); 941 return (ret); 942 } 943 944 /* 945 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 946 */ 947 pid_t 948 fgetown(sigiop) 949 struct sigio **sigiop; 950 { 951 pid_t pgid; 952 953 SIGIO_LOCK(); 954 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 955 SIGIO_UNLOCK(); 956 return (pgid); 957 } 958 959 /* 960 * Close a file descriptor. 961 */ 962 #ifndef _SYS_SYSPROTO_H_ 963 struct close_args { 964 int fd; 965 }; 966 #endif 967 /* 968 * MPSAFE 969 */ 970 /* ARGSUSED */ 971 int 972 close(td, uap) 973 struct thread *td; 974 struct close_args *uap; 975 { 976 struct filedesc *fdp; 977 struct file *fp; 978 int fd, error; 979 int holdleaders; 980 981 fd = uap->fd; 982 error = 0; 983 holdleaders = 0; 984 fdp = td->td_proc->p_fd; 985 986 AUDIT_SYSCLOSE(td, fd); 987 988 FILEDESC_LOCK(fdp); 989 if ((unsigned)fd >= fdp->fd_nfiles || 990 (fp = fdp->fd_ofiles[fd]) == NULL) { 991 FILEDESC_UNLOCK(fdp); 992 return (EBADF); 993 } 994 fdp->fd_ofiles[fd] = NULL; 995 fdp->fd_ofileflags[fd] = 0; 996 fdunused(fdp, fd); 997 if (td->td_proc->p_fdtol != NULL) { 998 /* 999 * Ask fdfree() to sleep to ensure that all relevant 1000 * process leaders can be traversed in closef(). 1001 */ 1002 fdp->fd_holdleaderscount++; 1003 holdleaders = 1; 1004 } 1005 1006 /* 1007 * We now hold the fp reference that used to be owned by the descriptor 1008 * array. 1009 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a 1010 * race of the fd getting opened, a knote added, and deleteing a knote 1011 * for the new fd. 1012 */ 1013 knote_fdclose(td, fd); 1014 if (fp->f_type == DTYPE_MQUEUE) 1015 mq_fdclose(td, fd, fp); 1016 FILEDESC_UNLOCK(fdp); 1017 1018 error = closef(fp, td); 1019 if (holdleaders) { 1020 FILEDESC_LOCK_FAST(fdp); 1021 fdp->fd_holdleaderscount--; 1022 if (fdp->fd_holdleaderscount == 0 && 1023 fdp->fd_holdleaderswakeup != 0) { 1024 fdp->fd_holdleaderswakeup = 0; 1025 wakeup(&fdp->fd_holdleaderscount); 1026 } 1027 FILEDESC_UNLOCK_FAST(fdp); 1028 } 1029 return (error); 1030 } 1031 1032 #if defined(COMPAT_43) 1033 /* 1034 * Return status information about a file descriptor. 1035 */ 1036 #ifndef _SYS_SYSPROTO_H_ 1037 struct ofstat_args { 1038 int fd; 1039 struct ostat *sb; 1040 }; 1041 #endif 1042 /* 1043 * MPSAFE 1044 */ 1045 /* ARGSUSED */ 1046 int 1047 ofstat(struct thread *td, struct ofstat_args *uap) 1048 { 1049 struct ostat oub; 1050 struct stat ub; 1051 int error; 1052 1053 error = kern_fstat(td, uap->fd, &ub); 1054 if (error == 0) { 1055 cvtstat(&ub, &oub); 1056 error = copyout(&oub, uap->sb, sizeof(oub)); 1057 } 1058 return (error); 1059 } 1060 #endif /* COMPAT_43 */ 1061 1062 /* 1063 * Return status information about a file descriptor. 1064 */ 1065 #ifndef _SYS_SYSPROTO_H_ 1066 struct fstat_args { 1067 int fd; 1068 struct stat *sb; 1069 }; 1070 #endif 1071 /* 1072 * MPSAFE 1073 */ 1074 /* ARGSUSED */ 1075 int 1076 fstat(struct thread *td, struct fstat_args *uap) 1077 { 1078 struct stat ub; 1079 int error; 1080 1081 error = kern_fstat(td, uap->fd, &ub); 1082 if (error == 0) 1083 error = copyout(&ub, uap->sb, sizeof(ub)); 1084 return (error); 1085 } 1086 1087 int 1088 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1089 { 1090 struct file *fp; 1091 int error; 1092 1093 AUDIT_ARG(fd, fd); 1094 1095 if ((error = fget(td, fd, &fp)) != 0) 1096 return (error); 1097 1098 AUDIT_ARG(file, td->td_proc, fp); 1099 1100 error = fo_stat(fp, sbp, td->td_ucred, td); 1101 fdrop(fp, td); 1102 return (error); 1103 } 1104 1105 /* 1106 * Return status information about a file descriptor. 1107 */ 1108 #ifndef _SYS_SYSPROTO_H_ 1109 struct nfstat_args { 1110 int fd; 1111 struct nstat *sb; 1112 }; 1113 #endif 1114 /* 1115 * MPSAFE 1116 */ 1117 /* ARGSUSED */ 1118 int 1119 nfstat(struct thread *td, struct nfstat_args *uap) 1120 { 1121 struct nstat nub; 1122 struct stat ub; 1123 int error; 1124 1125 error = kern_fstat(td, uap->fd, &ub); 1126 if (error == 0) { 1127 cvtnstat(&ub, &nub); 1128 error = copyout(&nub, uap->sb, sizeof(nub)); 1129 } 1130 return (error); 1131 } 1132 1133 /* 1134 * Return pathconf information about a file descriptor. 1135 */ 1136 #ifndef _SYS_SYSPROTO_H_ 1137 struct fpathconf_args { 1138 int fd; 1139 int name; 1140 }; 1141 #endif 1142 /* 1143 * MPSAFE 1144 */ 1145 /* ARGSUSED */ 1146 int 1147 fpathconf(struct thread *td, struct fpathconf_args *uap) 1148 { 1149 struct file *fp; 1150 struct vnode *vp; 1151 int error; 1152 1153 if ((error = fget(td, uap->fd, &fp)) != 0) 1154 return (error); 1155 1156 /* If asynchronous I/O is available, it works for all descriptors. */ 1157 if (uap->name == _PC_ASYNC_IO) { 1158 td->td_retval[0] = async_io_version; 1159 goto out; 1160 } 1161 vp = fp->f_vnode; 1162 if (vp != NULL) { 1163 int vfslocked; 1164 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1165 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1166 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1167 VOP_UNLOCK(vp, 0, td); 1168 VFS_UNLOCK_GIANT(vfslocked); 1169 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1170 if (uap->name != _PC_PIPE_BUF) { 1171 error = EINVAL; 1172 } else { 1173 td->td_retval[0] = PIPE_BUF; 1174 error = 0; 1175 } 1176 } else { 1177 error = EOPNOTSUPP; 1178 } 1179 out: 1180 fdrop(fp, td); 1181 return (error); 1182 } 1183 1184 /* 1185 * Grow the file table to accomodate (at least) nfd descriptors. This may 1186 * block and drop the filedesc lock, but it will reacquire it before 1187 * returning. 1188 */ 1189 static void 1190 fdgrowtable(struct filedesc *fdp, int nfd) 1191 { 1192 struct file **ntable; 1193 char *nfileflags; 1194 int nnfiles, onfiles; 1195 NDSLOTTYPE *nmap; 1196 1197 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1198 1199 KASSERT(fdp->fd_nfiles > 0, 1200 ("zero-length file table")); 1201 1202 /* compute the size of the new table */ 1203 onfiles = fdp->fd_nfiles; 1204 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1205 if (nnfiles <= onfiles) 1206 /* the table is already large enough */ 1207 return; 1208 1209 /* allocate a new table and (if required) new bitmaps */ 1210 FILEDESC_UNLOCK(fdp); 1211 MALLOC(ntable, struct file **, nnfiles * OFILESIZE, 1212 M_FILEDESC, M_ZERO | M_WAITOK); 1213 nfileflags = (char *)&ntable[nnfiles]; 1214 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1215 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE, 1216 M_FILEDESC, M_ZERO | M_WAITOK); 1217 else 1218 nmap = NULL; 1219 FILEDESC_LOCK(fdp); 1220 1221 /* 1222 * We now have new tables ready to go. Since we dropped the 1223 * filedesc lock to call malloc(), watch out for a race. 1224 */ 1225 onfiles = fdp->fd_nfiles; 1226 if (onfiles >= nnfiles) { 1227 /* we lost the race, but that's OK */ 1228 free(ntable, M_FILEDESC); 1229 if (nmap != NULL) 1230 free(nmap, M_FILEDESC); 1231 return; 1232 } 1233 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1234 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1235 if (onfiles > NDFILE) 1236 free(fdp->fd_ofiles, M_FILEDESC); 1237 fdp->fd_ofiles = ntable; 1238 fdp->fd_ofileflags = nfileflags; 1239 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1240 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1241 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1242 free(fdp->fd_map, M_FILEDESC); 1243 fdp->fd_map = nmap; 1244 } 1245 fdp->fd_nfiles = nnfiles; 1246 } 1247 1248 /* 1249 * Allocate a file descriptor for the process. 1250 */ 1251 int 1252 fdalloc(struct thread *td, int minfd, int *result) 1253 { 1254 struct proc *p = td->td_proc; 1255 struct filedesc *fdp = p->p_fd; 1256 int fd = -1, maxfd; 1257 1258 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1259 1260 if (fdp->fd_freefile > minfd) 1261 minfd = fdp->fd_freefile; 1262 1263 PROC_LOCK(p); 1264 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1265 PROC_UNLOCK(p); 1266 1267 /* 1268 * Search the bitmap for a free descriptor. If none is found, try 1269 * to grow the file table. Keep at it until we either get a file 1270 * descriptor or run into process or system limits; fdgrowtable() 1271 * may drop the filedesc lock, so we're in a race. 1272 */ 1273 for (;;) { 1274 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1275 if (fd >= maxfd) 1276 return (EMFILE); 1277 if (fd < fdp->fd_nfiles) 1278 break; 1279 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1280 } 1281 1282 /* 1283 * Perform some sanity checks, then mark the file descriptor as 1284 * used and return it to the caller. 1285 */ 1286 KASSERT(!fdisused(fdp, fd), 1287 ("fd_first_free() returned non-free descriptor")); 1288 KASSERT(fdp->fd_ofiles[fd] == NULL, 1289 ("free descriptor isn't")); 1290 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */ 1291 fdused(fdp, fd); 1292 *result = fd; 1293 return (0); 1294 } 1295 1296 /* 1297 * Check to see whether n user file descriptors 1298 * are available to the process p. 1299 */ 1300 int 1301 fdavail(struct thread *td, int n) 1302 { 1303 struct proc *p = td->td_proc; 1304 struct filedesc *fdp = td->td_proc->p_fd; 1305 struct file **fpp; 1306 int i, lim, last; 1307 1308 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED); 1309 1310 PROC_LOCK(p); 1311 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1312 PROC_UNLOCK(p); 1313 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1314 return (1); 1315 last = min(fdp->fd_nfiles, lim); 1316 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 1317 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) { 1318 if (*fpp == NULL && --n <= 0) 1319 return (1); 1320 } 1321 return (0); 1322 } 1323 1324 /* 1325 * Create a new open file structure and allocate 1326 * a file decriptor for the process that refers to it. 1327 * We add one reference to the file for the descriptor table 1328 * and one reference for resultfp. This is to prevent us being 1329 * preempted and the entry in the descriptor table closed after 1330 * we release the FILEDESC lock. 1331 */ 1332 int 1333 falloc(struct thread *td, struct file **resultfp, int *resultfd) 1334 { 1335 struct proc *p = td->td_proc; 1336 struct file *fp, *fq; 1337 int error, i; 1338 int maxuserfiles = maxfiles - (maxfiles / 20); 1339 static struct timeval lastfail; 1340 static int curfail; 1341 1342 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1343 sx_xlock(&filelist_lock); 1344 if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 || 1345 jailed(td->td_ucred))) || openfiles >= maxfiles) { 1346 if (ppsratecheck(&lastfail, &curfail, 1)) { 1347 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n", 1348 td->td_ucred->cr_ruid); 1349 } 1350 sx_xunlock(&filelist_lock); 1351 uma_zfree(file_zone, fp); 1352 return (ENFILE); 1353 } 1354 openfiles++; 1355 1356 /* 1357 * If the process has file descriptor zero open, add the new file 1358 * descriptor to the list of open files at that point, otherwise 1359 * put it at the front of the list of open files. 1360 */ 1361 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep); 1362 fp->f_count = 1; 1363 if (resultfp) 1364 fp->f_count++; 1365 fp->f_cred = crhold(td->td_ucred); 1366 fp->f_ops = &badfileops; 1367 fp->f_data = NULL; 1368 fp->f_vnode = NULL; 1369 FILEDESC_LOCK(p->p_fd); 1370 if ((fq = p->p_fd->fd_ofiles[0])) { 1371 LIST_INSERT_AFTER(fq, fp, f_list); 1372 } else { 1373 LIST_INSERT_HEAD(&filehead, fp, f_list); 1374 } 1375 sx_xunlock(&filelist_lock); 1376 if ((error = fdalloc(td, 0, &i))) { 1377 FILEDESC_UNLOCK(p->p_fd); 1378 fdrop(fp, td); 1379 if (resultfp) 1380 fdrop(fp, td); 1381 return (error); 1382 } 1383 p->p_fd->fd_ofiles[i] = fp; 1384 FILEDESC_UNLOCK(p->p_fd); 1385 if (resultfp) 1386 *resultfp = fp; 1387 if (resultfd) 1388 *resultfd = i; 1389 return (0); 1390 } 1391 1392 /* 1393 * Build a new filedesc structure from another. 1394 * Copy the current, root, and jail root vnode references. 1395 */ 1396 struct filedesc * 1397 fdinit(struct filedesc *fdp) 1398 { 1399 struct filedesc0 *newfdp; 1400 1401 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1402 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF); 1403 if (fdp != NULL) { 1404 FILEDESC_LOCK(fdp); 1405 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1406 if (newfdp->fd_fd.fd_cdir) 1407 VREF(newfdp->fd_fd.fd_cdir); 1408 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1409 if (newfdp->fd_fd.fd_rdir) 1410 VREF(newfdp->fd_fd.fd_rdir); 1411 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1412 if (newfdp->fd_fd.fd_jdir) 1413 VREF(newfdp->fd_fd.fd_jdir); 1414 FILEDESC_UNLOCK(fdp); 1415 } 1416 1417 /* Create the file descriptor table. */ 1418 newfdp->fd_fd.fd_refcnt = 1; 1419 newfdp->fd_fd.fd_holdcnt = 1; 1420 newfdp->fd_fd.fd_cmask = CMASK; 1421 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1422 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1423 newfdp->fd_fd.fd_nfiles = NDFILE; 1424 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1425 newfdp->fd_fd.fd_lastfile = -1; 1426 return (&newfdp->fd_fd); 1427 } 1428 1429 static struct filedesc * 1430 fdhold(struct proc *p) 1431 { 1432 struct filedesc *fdp; 1433 1434 mtx_lock(&fdesc_mtx); 1435 fdp = p->p_fd; 1436 if (fdp != NULL) 1437 fdp->fd_holdcnt++; 1438 mtx_unlock(&fdesc_mtx); 1439 return (fdp); 1440 } 1441 1442 static void 1443 fddrop(struct filedesc *fdp) 1444 { 1445 int i; 1446 1447 mtx_lock(&fdesc_mtx); 1448 i = --fdp->fd_holdcnt; 1449 mtx_unlock(&fdesc_mtx); 1450 if (i > 0) 1451 return; 1452 1453 mtx_destroy(&fdp->fd_mtx); 1454 FREE(fdp, M_FILEDESC); 1455 } 1456 1457 /* 1458 * Share a filedesc structure. 1459 */ 1460 struct filedesc * 1461 fdshare(struct filedesc *fdp) 1462 { 1463 FILEDESC_LOCK_FAST(fdp); 1464 fdp->fd_refcnt++; 1465 FILEDESC_UNLOCK_FAST(fdp); 1466 return (fdp); 1467 } 1468 1469 /* 1470 * Unshare a filedesc structure, if necessary by making a copy 1471 */ 1472 void 1473 fdunshare(struct proc *p, struct thread *td) 1474 { 1475 1476 FILEDESC_LOCK_FAST(p->p_fd); 1477 if (p->p_fd->fd_refcnt > 1) { 1478 struct filedesc *tmp; 1479 1480 FILEDESC_UNLOCK_FAST(p->p_fd); 1481 tmp = fdcopy(p->p_fd); 1482 fdfree(td); 1483 p->p_fd = tmp; 1484 } else 1485 FILEDESC_UNLOCK_FAST(p->p_fd); 1486 } 1487 1488 /* 1489 * Copy a filedesc structure. 1490 * A NULL pointer in returns a NULL reference, this is to ease callers, 1491 * not catch errors. 1492 */ 1493 struct filedesc * 1494 fdcopy(struct filedesc *fdp) 1495 { 1496 struct filedesc *newfdp; 1497 int i; 1498 1499 /* Certain daemons might not have file descriptors. */ 1500 if (fdp == NULL) 1501 return (NULL); 1502 1503 newfdp = fdinit(fdp); 1504 FILEDESC_LOCK_FAST(fdp); 1505 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1506 FILEDESC_UNLOCK_FAST(fdp); 1507 FILEDESC_LOCK(newfdp); 1508 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1509 FILEDESC_UNLOCK(newfdp); 1510 FILEDESC_LOCK_FAST(fdp); 1511 } 1512 /* copy everything except kqueue descriptors */ 1513 newfdp->fd_freefile = -1; 1514 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1515 if (fdisused(fdp, i) && 1516 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) { 1517 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1518 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1519 fhold(newfdp->fd_ofiles[i]); 1520 newfdp->fd_lastfile = i; 1521 } else { 1522 if (newfdp->fd_freefile == -1) 1523 newfdp->fd_freefile = i; 1524 } 1525 } 1526 FILEDESC_UNLOCK_FAST(fdp); 1527 FILEDESC_LOCK(newfdp); 1528 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1529 if (newfdp->fd_ofiles[i] != NULL) 1530 fdused(newfdp, i); 1531 FILEDESC_UNLOCK(newfdp); 1532 FILEDESC_LOCK_FAST(fdp); 1533 if (newfdp->fd_freefile == -1) 1534 newfdp->fd_freefile = i; 1535 newfdp->fd_cmask = fdp->fd_cmask; 1536 FILEDESC_UNLOCK_FAST(fdp); 1537 return (newfdp); 1538 } 1539 1540 /* 1541 * Release a filedesc structure. 1542 */ 1543 void 1544 fdfree(struct thread *td) 1545 { 1546 struct filedesc *fdp; 1547 struct file **fpp; 1548 int i, locked; 1549 struct filedesc_to_leader *fdtol; 1550 struct file *fp; 1551 struct vnode *cdir, *jdir, *rdir, *vp; 1552 struct flock lf; 1553 1554 /* Certain daemons might not have file descriptors. */ 1555 fdp = td->td_proc->p_fd; 1556 if (fdp == NULL) 1557 return; 1558 1559 /* Check for special need to clear POSIX style locks */ 1560 fdtol = td->td_proc->p_fdtol; 1561 if (fdtol != NULL) { 1562 FILEDESC_LOCK(fdp); 1563 KASSERT(fdtol->fdl_refcount > 0, 1564 ("filedesc_to_refcount botch: fdl_refcount=%d", 1565 fdtol->fdl_refcount)); 1566 if (fdtol->fdl_refcount == 1 && 1567 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1568 for (i = 0, fpp = fdp->fd_ofiles; 1569 i <= fdp->fd_lastfile; 1570 i++, fpp++) { 1571 if (*fpp == NULL || 1572 (*fpp)->f_type != DTYPE_VNODE) 1573 continue; 1574 fp = *fpp; 1575 fhold(fp); 1576 FILEDESC_UNLOCK(fdp); 1577 lf.l_whence = SEEK_SET; 1578 lf.l_start = 0; 1579 lf.l_len = 0; 1580 lf.l_type = F_UNLCK; 1581 vp = fp->f_vnode; 1582 locked = VFS_LOCK_GIANT(vp->v_mount); 1583 (void) VOP_ADVLOCK(vp, 1584 (caddr_t)td->td_proc-> 1585 p_leader, 1586 F_UNLCK, 1587 &lf, 1588 F_POSIX); 1589 VFS_UNLOCK_GIANT(locked); 1590 FILEDESC_LOCK(fdp); 1591 fdrop(fp, td); 1592 fpp = fdp->fd_ofiles + i; 1593 } 1594 } 1595 retry: 1596 if (fdtol->fdl_refcount == 1) { 1597 if (fdp->fd_holdleaderscount > 0 && 1598 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1599 /* 1600 * close() or do_dup() has cleared a reference 1601 * in a shared file descriptor table. 1602 */ 1603 fdp->fd_holdleaderswakeup = 1; 1604 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx, 1605 PLOCK, "fdlhold", 0); 1606 goto retry; 1607 } 1608 if (fdtol->fdl_holdcount > 0) { 1609 /* 1610 * Ensure that fdtol->fdl_leader 1611 * remains valid in closef(). 1612 */ 1613 fdtol->fdl_wakeup = 1; 1614 msleep(fdtol, &fdp->fd_mtx, 1615 PLOCK, "fdlhold", 0); 1616 goto retry; 1617 } 1618 } 1619 fdtol->fdl_refcount--; 1620 if (fdtol->fdl_refcount == 0 && 1621 fdtol->fdl_holdcount == 0) { 1622 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1623 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1624 } else 1625 fdtol = NULL; 1626 td->td_proc->p_fdtol = NULL; 1627 FILEDESC_UNLOCK(fdp); 1628 if (fdtol != NULL) 1629 FREE(fdtol, M_FILEDESC_TO_LEADER); 1630 } 1631 FILEDESC_LOCK_FAST(fdp); 1632 i = --fdp->fd_refcnt; 1633 FILEDESC_UNLOCK_FAST(fdp); 1634 if (i > 0) 1635 return; 1636 /* 1637 * We are the last reference to the structure, so we can 1638 * safely assume it will not change out from under us. 1639 */ 1640 fpp = fdp->fd_ofiles; 1641 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) { 1642 if (*fpp) 1643 (void) closef(*fpp, td); 1644 } 1645 FILEDESC_LOCK(fdp); 1646 1647 /* XXX This should happen earlier. */ 1648 mtx_lock(&fdesc_mtx); 1649 td->td_proc->p_fd = NULL; 1650 mtx_unlock(&fdesc_mtx); 1651 1652 if (fdp->fd_nfiles > NDFILE) 1653 FREE(fdp->fd_ofiles, M_FILEDESC); 1654 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1655 FREE(fdp->fd_map, M_FILEDESC); 1656 1657 fdp->fd_nfiles = 0; 1658 1659 cdir = fdp->fd_cdir; 1660 fdp->fd_cdir = NULL; 1661 rdir = fdp->fd_rdir; 1662 fdp->fd_rdir = NULL; 1663 jdir = fdp->fd_jdir; 1664 fdp->fd_jdir = NULL; 1665 FILEDESC_UNLOCK(fdp); 1666 1667 if (cdir) { 1668 locked = VFS_LOCK_GIANT(cdir->v_mount); 1669 vrele(cdir); 1670 VFS_UNLOCK_GIANT(locked); 1671 } 1672 if (rdir) { 1673 locked = VFS_LOCK_GIANT(rdir->v_mount); 1674 vrele(rdir); 1675 VFS_UNLOCK_GIANT(locked); 1676 } 1677 if (jdir) { 1678 locked = VFS_LOCK_GIANT(jdir->v_mount); 1679 vrele(jdir); 1680 VFS_UNLOCK_GIANT(locked); 1681 } 1682 1683 fddrop(fdp); 1684 } 1685 1686 /* 1687 * For setugid programs, we don't want to people to use that setugidness 1688 * to generate error messages which write to a file which otherwise would 1689 * otherwise be off-limits to the process. We check for filesystems where 1690 * the vnode can change out from under us after execve (like [lin]procfs). 1691 * 1692 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1693 * sufficient. We also don't check for setugidness since we know we are. 1694 */ 1695 static int 1696 is_unsafe(struct file *fp) 1697 { 1698 if (fp->f_type == DTYPE_VNODE) { 1699 struct vnode *vp = fp->f_vnode; 1700 1701 if ((vp->v_vflag & VV_PROCDEP) != 0) 1702 return (1); 1703 } 1704 return (0); 1705 } 1706 1707 /* 1708 * Make this setguid thing safe, if at all possible. 1709 */ 1710 void 1711 setugidsafety(struct thread *td) 1712 { 1713 struct filedesc *fdp; 1714 int i; 1715 1716 /* Certain daemons might not have file descriptors. */ 1717 fdp = td->td_proc->p_fd; 1718 if (fdp == NULL) 1719 return; 1720 1721 /* 1722 * Note: fdp->fd_ofiles may be reallocated out from under us while 1723 * we are blocked in a close. Be careful! 1724 */ 1725 FILEDESC_LOCK(fdp); 1726 for (i = 0; i <= fdp->fd_lastfile; i++) { 1727 if (i > 2) 1728 break; 1729 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1730 struct file *fp; 1731 1732 knote_fdclose(td, i); 1733 /* 1734 * NULL-out descriptor prior to close to avoid 1735 * a race while close blocks. 1736 */ 1737 fp = fdp->fd_ofiles[i]; 1738 fdp->fd_ofiles[i] = NULL; 1739 fdp->fd_ofileflags[i] = 0; 1740 fdunused(fdp, i); 1741 FILEDESC_UNLOCK(fdp); 1742 (void) closef(fp, td); 1743 FILEDESC_LOCK(fdp); 1744 } 1745 } 1746 FILEDESC_UNLOCK(fdp); 1747 } 1748 1749 void 1750 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 1751 { 1752 1753 FILEDESC_LOCK(fdp); 1754 if (fdp->fd_ofiles[idx] == fp) { 1755 fdp->fd_ofiles[idx] = NULL; 1756 fdunused(fdp, idx); 1757 FILEDESC_UNLOCK(fdp); 1758 fdrop(fp, td); 1759 } else { 1760 FILEDESC_UNLOCK(fdp); 1761 } 1762 } 1763 1764 /* 1765 * Close any files on exec? 1766 */ 1767 void 1768 fdcloseexec(struct thread *td) 1769 { 1770 struct filedesc *fdp; 1771 int i; 1772 1773 /* Certain daemons might not have file descriptors. */ 1774 fdp = td->td_proc->p_fd; 1775 if (fdp == NULL) 1776 return; 1777 1778 FILEDESC_LOCK(fdp); 1779 1780 /* 1781 * We cannot cache fd_ofiles or fd_ofileflags since operations 1782 * may block and rip them out from under us. 1783 */ 1784 for (i = 0; i <= fdp->fd_lastfile; i++) { 1785 if (fdp->fd_ofiles[i] != NULL && 1786 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE || 1787 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 1788 struct file *fp; 1789 1790 knote_fdclose(td, i); 1791 /* 1792 * NULL-out descriptor prior to close to avoid 1793 * a race while close blocks. 1794 */ 1795 fp = fdp->fd_ofiles[i]; 1796 fdp->fd_ofiles[i] = NULL; 1797 fdp->fd_ofileflags[i] = 0; 1798 fdunused(fdp, i); 1799 if (fp->f_type == DTYPE_MQUEUE) 1800 mq_fdclose(td, i, fp); 1801 FILEDESC_UNLOCK(fdp); 1802 (void) closef(fp, td); 1803 FILEDESC_LOCK(fdp); 1804 } 1805 } 1806 FILEDESC_UNLOCK(fdp); 1807 } 1808 1809 /* 1810 * It is unsafe for set[ug]id processes to be started with file 1811 * descriptors 0..2 closed, as these descriptors are given implicit 1812 * significance in the Standard C library. fdcheckstd() will create a 1813 * descriptor referencing /dev/null for each of stdin, stdout, and 1814 * stderr that is not already open. 1815 */ 1816 int 1817 fdcheckstd(struct thread *td) 1818 { 1819 struct nameidata nd; 1820 struct filedesc *fdp; 1821 struct file *fp; 1822 register_t retval; 1823 int fd, i, error, flags, devnull; 1824 1825 fdp = td->td_proc->p_fd; 1826 if (fdp == NULL) 1827 return (0); 1828 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 1829 devnull = -1; 1830 error = 0; 1831 for (i = 0; i < 3; i++) { 1832 if (fdp->fd_ofiles[i] != NULL) 1833 continue; 1834 if (devnull < 0) { 1835 int vfslocked; 1836 error = falloc(td, &fp, &fd); 1837 if (error != 0) 1838 break; 1839 /* Note extra ref on `fp' held for us by falloc(). */ 1840 KASSERT(fd == i, ("oof, we didn't get our fd")); 1841 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, 1842 "/dev/null", td); 1843 flags = FREAD | FWRITE; 1844 error = vn_open(&nd, &flags, 0, fd); 1845 if (error != 0) { 1846 /* 1847 * Someone may have closed the entry in the 1848 * file descriptor table, so check it hasn't 1849 * changed before dropping the reference count. 1850 */ 1851 FILEDESC_LOCK(fdp); 1852 KASSERT(fdp->fd_ofiles[fd] == fp, 1853 ("table not shared, how did it change?")); 1854 fdp->fd_ofiles[fd] = NULL; 1855 fdunused(fdp, fd); 1856 FILEDESC_UNLOCK(fdp); 1857 fdrop(fp, td); 1858 fdrop(fp, td); 1859 break; 1860 } 1861 vfslocked = NDHASGIANT(&nd); 1862 NDFREE(&nd, NDF_ONLY_PNBUF); 1863 fp->f_flag = flags; 1864 fp->f_vnode = nd.ni_vp; 1865 if (fp->f_data == NULL) 1866 fp->f_data = nd.ni_vp; 1867 if (fp->f_ops == &badfileops) 1868 fp->f_ops = &vnops; 1869 fp->f_type = DTYPE_VNODE; 1870 VOP_UNLOCK(nd.ni_vp, 0, td); 1871 VFS_UNLOCK_GIANT(vfslocked); 1872 devnull = fd; 1873 fdrop(fp, td); 1874 } else { 1875 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 1876 if (error != 0) 1877 break; 1878 } 1879 } 1880 return (error); 1881 } 1882 1883 /* 1884 * Internal form of close. 1885 * Decrement reference count on file structure. 1886 * Note: td may be NULL when closing a file that was being passed in a 1887 * message. 1888 * 1889 * XXXRW: Giant is not required for the caller, but often will be held; this 1890 * makes it moderately likely the Giant will be recursed in the VFS case. 1891 */ 1892 int 1893 closef(struct file *fp, struct thread *td) 1894 { 1895 struct vnode *vp; 1896 struct flock lf; 1897 struct filedesc_to_leader *fdtol; 1898 struct filedesc *fdp; 1899 1900 /* 1901 * POSIX record locking dictates that any close releases ALL 1902 * locks owned by this process. This is handled by setting 1903 * a flag in the unlock to free ONLY locks obeying POSIX 1904 * semantics, and not to free BSD-style file locks. 1905 * If the descriptor was in a message, POSIX-style locks 1906 * aren't passed with the descriptor, and the thread pointer 1907 * will be NULL. Callers should be careful only to pass a 1908 * NULL thread pointer when there really is no owning 1909 * context that might have locks, or the locks will be 1910 * leaked. 1911 */ 1912 if (fp->f_type == DTYPE_VNODE && td != NULL) { 1913 int vfslocked; 1914 1915 vp = fp->f_vnode; 1916 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1917 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1918 lf.l_whence = SEEK_SET; 1919 lf.l_start = 0; 1920 lf.l_len = 0; 1921 lf.l_type = F_UNLCK; 1922 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 1923 F_UNLCK, &lf, F_POSIX); 1924 } 1925 fdtol = td->td_proc->p_fdtol; 1926 if (fdtol != NULL) { 1927 /* 1928 * Handle special case where file descriptor table 1929 * is shared between multiple process leaders. 1930 */ 1931 fdp = td->td_proc->p_fd; 1932 FILEDESC_LOCK(fdp); 1933 for (fdtol = fdtol->fdl_next; 1934 fdtol != td->td_proc->p_fdtol; 1935 fdtol = fdtol->fdl_next) { 1936 if ((fdtol->fdl_leader->p_flag & 1937 P_ADVLOCK) == 0) 1938 continue; 1939 fdtol->fdl_holdcount++; 1940 FILEDESC_UNLOCK(fdp); 1941 lf.l_whence = SEEK_SET; 1942 lf.l_start = 0; 1943 lf.l_len = 0; 1944 lf.l_type = F_UNLCK; 1945 vp = fp->f_vnode; 1946 (void) VOP_ADVLOCK(vp, 1947 (caddr_t)fdtol->fdl_leader, 1948 F_UNLCK, &lf, F_POSIX); 1949 FILEDESC_LOCK(fdp); 1950 fdtol->fdl_holdcount--; 1951 if (fdtol->fdl_holdcount == 0 && 1952 fdtol->fdl_wakeup != 0) { 1953 fdtol->fdl_wakeup = 0; 1954 wakeup(fdtol); 1955 } 1956 } 1957 FILEDESC_UNLOCK(fdp); 1958 } 1959 VFS_UNLOCK_GIANT(vfslocked); 1960 } 1961 return (fdrop(fp, td)); 1962 } 1963 1964 /* 1965 * Extract the file pointer associated with the specified descriptor for 1966 * the current user process. 1967 * 1968 * If the descriptor doesn't exist, EBADF is returned. 1969 * 1970 * If the descriptor exists but doesn't match 'flags' then 1971 * return EBADF for read attempts and EINVAL for write attempts. 1972 * 1973 * If 'hold' is set (non-zero) the file's refcount will be bumped on return. 1974 * It should be dropped with fdrop(). 1975 * If it is not set, then the refcount will not be bumped however the 1976 * thread's filedesc struct will be returned locked (for fgetsock). 1977 * 1978 * If an error occured the non-zero error is returned and *fpp is set to NULL. 1979 * Otherwise *fpp is set and zero is returned. 1980 */ 1981 static __inline int 1982 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold) 1983 { 1984 struct filedesc *fdp; 1985 struct file *fp; 1986 1987 *fpp = NULL; 1988 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 1989 return (EBADF); 1990 FILEDESC_LOCK(fdp); 1991 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) { 1992 FILEDESC_UNLOCK(fdp); 1993 return (EBADF); 1994 } 1995 1996 /* 1997 * FREAD and FWRITE failure return EBADF as per POSIX. 1998 * 1999 * Only one flag, or 0, may be specified. 2000 */ 2001 if (flags == FREAD && (fp->f_flag & FREAD) == 0) { 2002 FILEDESC_UNLOCK(fdp); 2003 return (EBADF); 2004 } 2005 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) { 2006 FILEDESC_UNLOCK(fdp); 2007 return (EBADF); 2008 } 2009 if (hold) { 2010 fhold(fp); 2011 FILEDESC_UNLOCK(fdp); 2012 } 2013 *fpp = fp; 2014 return (0); 2015 } 2016 2017 int 2018 fget(struct thread *td, int fd, struct file **fpp) 2019 { 2020 2021 return(_fget(td, fd, fpp, 0, 1)); 2022 } 2023 2024 int 2025 fget_read(struct thread *td, int fd, struct file **fpp) 2026 { 2027 2028 return(_fget(td, fd, fpp, FREAD, 1)); 2029 } 2030 2031 int 2032 fget_write(struct thread *td, int fd, struct file **fpp) 2033 { 2034 2035 return(_fget(td, fd, fpp, FWRITE, 1)); 2036 } 2037 2038 /* 2039 * Like fget() but loads the underlying vnode, or returns an error if 2040 * the descriptor does not represent a vnode. Note that pipes use vnodes 2041 * but never have VM objects. The returned vnode will be vref()d. 2042 * 2043 * XXX: what about the unused flags ? 2044 */ 2045 static __inline int 2046 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags) 2047 { 2048 struct file *fp; 2049 int error; 2050 2051 *vpp = NULL; 2052 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2053 return (error); 2054 if (fp->f_vnode == NULL) { 2055 error = EINVAL; 2056 } else { 2057 *vpp = fp->f_vnode; 2058 vref(*vpp); 2059 } 2060 FILEDESC_UNLOCK(td->td_proc->p_fd); 2061 return (error); 2062 } 2063 2064 int 2065 fgetvp(struct thread *td, int fd, struct vnode **vpp) 2066 { 2067 2068 return (_fgetvp(td, fd, vpp, 0)); 2069 } 2070 2071 int 2072 fgetvp_read(struct thread *td, int fd, struct vnode **vpp) 2073 { 2074 2075 return (_fgetvp(td, fd, vpp, FREAD)); 2076 } 2077 2078 #ifdef notyet 2079 int 2080 fgetvp_write(struct thread *td, int fd, struct vnode **vpp) 2081 { 2082 2083 return (_fgetvp(td, fd, vpp, FWRITE)); 2084 } 2085 #endif 2086 2087 /* 2088 * Like fget() but loads the underlying socket, or returns an error if 2089 * the descriptor does not represent a socket. 2090 * 2091 * We bump the ref count on the returned socket. XXX Also obtain the SX 2092 * lock in the future. 2093 * 2094 * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely 2095 * on their file descriptor reference to prevent the socket from being 2096 * freed during use. 2097 */ 2098 int 2099 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp) 2100 { 2101 struct file *fp; 2102 int error; 2103 2104 NET_ASSERT_GIANT(); 2105 2106 *spp = NULL; 2107 if (fflagp != NULL) 2108 *fflagp = 0; 2109 if ((error = _fget(td, fd, &fp, 0, 0)) != 0) 2110 return (error); 2111 if (fp->f_type != DTYPE_SOCKET) { 2112 error = ENOTSOCK; 2113 } else { 2114 *spp = fp->f_data; 2115 if (fflagp) 2116 *fflagp = fp->f_flag; 2117 SOCK_LOCK(*spp); 2118 soref(*spp); 2119 SOCK_UNLOCK(*spp); 2120 } 2121 FILEDESC_UNLOCK(td->td_proc->p_fd); 2122 return (error); 2123 } 2124 2125 /* 2126 * Drop the reference count on the socket and XXX release the SX lock in the 2127 * future. The last reference closes the socket. 2128 * 2129 * XXXRW: fputsock() is deprecated, see comment for fgetsock(). 2130 */ 2131 void 2132 fputsock(struct socket *so) 2133 { 2134 2135 NET_ASSERT_GIANT(); 2136 ACCEPT_LOCK(); 2137 SOCK_LOCK(so); 2138 sorele(so); 2139 } 2140 2141 int 2142 fdrop(struct file *fp, struct thread *td) 2143 { 2144 2145 FILE_LOCK(fp); 2146 return (fdrop_locked(fp, td)); 2147 } 2148 2149 /* 2150 * Drop reference on struct file passed in, may call closef if the 2151 * reference hits zero. 2152 * Expects struct file locked, and will unlock it. 2153 */ 2154 static int 2155 fdrop_locked(struct file *fp, struct thread *td) 2156 { 2157 int error; 2158 2159 FILE_LOCK_ASSERT(fp, MA_OWNED); 2160 2161 if (--fp->f_count > 0) { 2162 FILE_UNLOCK(fp); 2163 return (0); 2164 } 2165 /* We have the last ref so we can proceed without the file lock. */ 2166 FILE_UNLOCK(fp); 2167 if (fp->f_count < 0) 2168 panic("fdrop: count < 0"); 2169 if (fp->f_ops != &badfileops) 2170 error = fo_close(fp, td); 2171 else 2172 error = 0; 2173 2174 sx_xlock(&filelist_lock); 2175 LIST_REMOVE(fp, f_list); 2176 openfiles--; 2177 sx_xunlock(&filelist_lock); 2178 crfree(fp->f_cred); 2179 uma_zfree(file_zone, fp); 2180 2181 return (error); 2182 } 2183 2184 /* 2185 * Apply an advisory lock on a file descriptor. 2186 * 2187 * Just attempt to get a record lock of the requested type on 2188 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2189 */ 2190 #ifndef _SYS_SYSPROTO_H_ 2191 struct flock_args { 2192 int fd; 2193 int how; 2194 }; 2195 #endif 2196 /* 2197 * MPSAFE 2198 */ 2199 /* ARGSUSED */ 2200 int 2201 flock(struct thread *td, struct flock_args *uap) 2202 { 2203 struct file *fp; 2204 struct vnode *vp; 2205 struct flock lf; 2206 int error; 2207 2208 if ((error = fget(td, uap->fd, &fp)) != 0) 2209 return (error); 2210 if (fp->f_type != DTYPE_VNODE) { 2211 fdrop(fp, td); 2212 return (EOPNOTSUPP); 2213 } 2214 2215 mtx_lock(&Giant); 2216 vp = fp->f_vnode; 2217 lf.l_whence = SEEK_SET; 2218 lf.l_start = 0; 2219 lf.l_len = 0; 2220 if (uap->how & LOCK_UN) { 2221 lf.l_type = F_UNLCK; 2222 FILE_LOCK(fp); 2223 fp->f_flag &= ~FHASLOCK; 2224 FILE_UNLOCK(fp); 2225 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2226 goto done2; 2227 } 2228 if (uap->how & LOCK_EX) 2229 lf.l_type = F_WRLCK; 2230 else if (uap->how & LOCK_SH) 2231 lf.l_type = F_RDLCK; 2232 else { 2233 error = EBADF; 2234 goto done2; 2235 } 2236 FILE_LOCK(fp); 2237 fp->f_flag |= FHASLOCK; 2238 FILE_UNLOCK(fp); 2239 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2240 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2241 done2: 2242 fdrop(fp, td); 2243 mtx_unlock(&Giant); 2244 return (error); 2245 } 2246 /* 2247 * Duplicate the specified descriptor to a free descriptor. 2248 */ 2249 int 2250 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2251 { 2252 struct file *wfp; 2253 struct file *fp; 2254 2255 /* 2256 * If the to-be-dup'd fd number is greater than the allowed number 2257 * of file descriptors, or the fd to be dup'd has already been 2258 * closed, then reject. 2259 */ 2260 FILEDESC_LOCK(fdp); 2261 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2262 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2263 FILEDESC_UNLOCK(fdp); 2264 return (EBADF); 2265 } 2266 2267 /* 2268 * There are two cases of interest here. 2269 * 2270 * For ENODEV simply dup (dfd) to file descriptor 2271 * (indx) and return. 2272 * 2273 * For ENXIO steal away the file structure from (dfd) and 2274 * store it in (indx). (dfd) is effectively closed by 2275 * this operation. 2276 * 2277 * Any other error code is just returned. 2278 */ 2279 switch (error) { 2280 case ENODEV: 2281 /* 2282 * Check that the mode the file is being opened for is a 2283 * subset of the mode of the existing descriptor. 2284 */ 2285 FILE_LOCK(wfp); 2286 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2287 FILE_UNLOCK(wfp); 2288 FILEDESC_UNLOCK(fdp); 2289 return (EACCES); 2290 } 2291 fp = fdp->fd_ofiles[indx]; 2292 fdp->fd_ofiles[indx] = wfp; 2293 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2294 if (fp == NULL) 2295 fdused(fdp, indx); 2296 fhold_locked(wfp); 2297 FILE_UNLOCK(wfp); 2298 FILEDESC_UNLOCK(fdp); 2299 if (fp != NULL) { 2300 /* 2301 * We now own the reference to fp that the ofiles[] 2302 * array used to own. Release it. 2303 */ 2304 FILE_LOCK(fp); 2305 fdrop_locked(fp, td); 2306 } 2307 return (0); 2308 2309 case ENXIO: 2310 /* 2311 * Steal away the file pointer from dfd and stuff it into indx. 2312 */ 2313 fp = fdp->fd_ofiles[indx]; 2314 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2315 fdp->fd_ofiles[dfd] = NULL; 2316 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2317 fdp->fd_ofileflags[dfd] = 0; 2318 fdunused(fdp, dfd); 2319 if (fp == NULL) 2320 fdused(fdp, indx); 2321 if (fp != NULL) 2322 FILE_LOCK(fp); 2323 2324 /* 2325 * We now own the reference to fp that the ofiles[] array 2326 * used to own. Release it. 2327 */ 2328 if (fp != NULL) 2329 fdrop_locked(fp, td); 2330 2331 FILEDESC_UNLOCK(fdp); 2332 2333 return (0); 2334 2335 default: 2336 FILEDESC_UNLOCK(fdp); 2337 return (error); 2338 } 2339 /* NOTREACHED */ 2340 } 2341 2342 /* 2343 * Scan all active processes to see if any of them have a current 2344 * or root directory of `olddp'. If so, replace them with the new 2345 * mount point. 2346 */ 2347 void 2348 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2349 { 2350 struct filedesc *fdp; 2351 struct proc *p; 2352 int nrele; 2353 2354 if (vrefcnt(olddp) == 1) 2355 return; 2356 sx_slock(&allproc_lock); 2357 LIST_FOREACH(p, &allproc, p_list) { 2358 fdp = fdhold(p); 2359 if (fdp == NULL) 2360 continue; 2361 nrele = 0; 2362 FILEDESC_LOCK_FAST(fdp); 2363 if (fdp->fd_cdir == olddp) { 2364 vref(newdp); 2365 fdp->fd_cdir = newdp; 2366 nrele++; 2367 } 2368 if (fdp->fd_rdir == olddp) { 2369 vref(newdp); 2370 fdp->fd_rdir = newdp; 2371 nrele++; 2372 } 2373 FILEDESC_UNLOCK_FAST(fdp); 2374 fddrop(fdp); 2375 while (nrele--) 2376 vrele(olddp); 2377 } 2378 sx_sunlock(&allproc_lock); 2379 if (rootvnode == olddp) { 2380 vrele(rootvnode); 2381 vref(newdp); 2382 rootvnode = newdp; 2383 } 2384 } 2385 2386 struct filedesc_to_leader * 2387 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2388 { 2389 struct filedesc_to_leader *fdtol; 2390 2391 MALLOC(fdtol, struct filedesc_to_leader *, 2392 sizeof(struct filedesc_to_leader), 2393 M_FILEDESC_TO_LEADER, 2394 M_WAITOK); 2395 fdtol->fdl_refcount = 1; 2396 fdtol->fdl_holdcount = 0; 2397 fdtol->fdl_wakeup = 0; 2398 fdtol->fdl_leader = leader; 2399 if (old != NULL) { 2400 FILEDESC_LOCK(fdp); 2401 fdtol->fdl_next = old->fdl_next; 2402 fdtol->fdl_prev = old; 2403 old->fdl_next = fdtol; 2404 fdtol->fdl_next->fdl_prev = fdtol; 2405 FILEDESC_UNLOCK(fdp); 2406 } else { 2407 fdtol->fdl_next = fdtol; 2408 fdtol->fdl_prev = fdtol; 2409 } 2410 return (fdtol); 2411 } 2412 2413 /* 2414 * Get file structures. 2415 */ 2416 static int 2417 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2418 { 2419 struct xfile xf; 2420 struct filedesc *fdp; 2421 struct file *fp; 2422 struct proc *p; 2423 int error, n; 2424 2425 /* 2426 * Note: because the number of file descriptors is calculated 2427 * in different ways for sizing vs returning the data, 2428 * there is information leakage from the first loop. However, 2429 * it is of a similar order of magnitude to the leakage from 2430 * global system statistics such as kern.openfiles. 2431 */ 2432 error = sysctl_wire_old_buffer(req, 0); 2433 if (error != 0) 2434 return (error); 2435 if (req->oldptr == NULL) { 2436 n = 16; /* A slight overestimate. */ 2437 sx_slock(&filelist_lock); 2438 LIST_FOREACH(fp, &filehead, f_list) { 2439 /* 2440 * We should grab the lock, but this is an 2441 * estimate, so does it really matter? 2442 */ 2443 /* mtx_lock(fp->f_mtxp); */ 2444 n += fp->f_count; 2445 /* mtx_unlock(f->f_mtxp); */ 2446 } 2447 sx_sunlock(&filelist_lock); 2448 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2449 } 2450 error = 0; 2451 bzero(&xf, sizeof(xf)); 2452 xf.xf_size = sizeof(xf); 2453 sx_slock(&allproc_lock); 2454 LIST_FOREACH(p, &allproc, p_list) { 2455 if (p->p_state == PRS_NEW) 2456 continue; 2457 PROC_LOCK(p); 2458 if (p_cansee(req->td, p) != 0) { 2459 PROC_UNLOCK(p); 2460 continue; 2461 } 2462 xf.xf_pid = p->p_pid; 2463 xf.xf_uid = p->p_ucred->cr_uid; 2464 PROC_UNLOCK(p); 2465 fdp = fdhold(p); 2466 if (fdp == NULL) 2467 continue; 2468 FILEDESC_LOCK_FAST(fdp); 2469 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2470 if ((fp = fdp->fd_ofiles[n]) == NULL) 2471 continue; 2472 xf.xf_fd = n; 2473 xf.xf_file = fp; 2474 xf.xf_data = fp->f_data; 2475 xf.xf_vnode = fp->f_vnode; 2476 xf.xf_type = fp->f_type; 2477 xf.xf_count = fp->f_count; 2478 xf.xf_msgcount = fp->f_msgcount; 2479 xf.xf_offset = fp->f_offset; 2480 xf.xf_flag = fp->f_flag; 2481 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2482 if (error) 2483 break; 2484 } 2485 FILEDESC_UNLOCK_FAST(fdp); 2486 fddrop(fdp); 2487 if (error) 2488 break; 2489 } 2490 sx_sunlock(&allproc_lock); 2491 return (error); 2492 } 2493 2494 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2495 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2496 2497 #ifdef DDB 2498 /* 2499 * For the purposes of debugging, generate a human-readable string for the 2500 * file type. 2501 */ 2502 static const char * 2503 file_type_to_name(short type) 2504 { 2505 2506 switch (type) { 2507 case 0: 2508 return ("zero"); 2509 case DTYPE_VNODE: 2510 return ("vnod"); 2511 case DTYPE_SOCKET: 2512 return ("sock"); 2513 case DTYPE_PIPE: 2514 return ("pipe"); 2515 case DTYPE_FIFO: 2516 return ("fifo"); 2517 case DTYPE_CRYPTO: 2518 return ("crpt"); 2519 default: 2520 return ("unkn"); 2521 } 2522 } 2523 2524 /* 2525 * For the purposes of debugging, identify a process (if any, perhaps one of 2526 * many) that references the passed file in its file descriptor array. Return 2527 * NULL if none. 2528 */ 2529 static struct proc * 2530 file_to_first_proc(struct file *fp) 2531 { 2532 struct filedesc *fdp; 2533 struct proc *p; 2534 int n; 2535 2536 LIST_FOREACH(p, &allproc, p_list) { 2537 if (p->p_state == PRS_NEW) 2538 continue; 2539 fdp = p->p_fd; 2540 if (fdp == NULL) 2541 continue; 2542 for (n = 0; n < fdp->fd_nfiles; n++) { 2543 if (fp == fdp->fd_ofiles[n]) 2544 return (p); 2545 } 2546 } 2547 return (NULL); 2548 } 2549 2550 DB_SHOW_COMMAND(files, db_show_files) 2551 { 2552 struct file *fp; 2553 struct proc *p; 2554 2555 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File", 2556 "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode", 2557 "FPID", "FCmd"); 2558 LIST_FOREACH(fp, &filehead, f_list) { 2559 p = file_to_first_proc(fp); 2560 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 2561 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 2562 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode, 2563 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 2564 } 2565 } 2566 #endif 2567 2568 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 2569 &maxfilesperproc, 0, "Maximum files allowed open per process"); 2570 2571 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 2572 &maxfiles, 0, "Maximum number of files"); 2573 2574 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 2575 &openfiles, 0, "System-wide number of open files"); 2576 2577 /* ARGSUSED*/ 2578 static void 2579 filelistinit(void *dummy) 2580 { 2581 2582 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 2583 NULL, NULL, UMA_ALIGN_PTR, 0); 2584 sx_init(&filelist_lock, "filelist lock"); 2585 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 2586 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 2587 } 2588 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL) 2589 2590 /*-------------------------------------------------------------------*/ 2591 2592 static int 2593 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) 2594 { 2595 2596 return (EBADF); 2597 } 2598 2599 static int 2600 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) 2601 { 2602 2603 return (EBADF); 2604 } 2605 2606 static int 2607 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) 2608 { 2609 2610 return (0); 2611 } 2612 2613 static int 2614 badfo_kqfilter(struct file *fp, struct knote *kn) 2615 { 2616 2617 return (0); 2618 } 2619 2620 static int 2621 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) 2622 { 2623 2624 return (EBADF); 2625 } 2626 2627 static int 2628 badfo_close(struct file *fp, struct thread *td) 2629 { 2630 2631 return (EBADF); 2632 } 2633 2634 struct fileops badfileops = { 2635 .fo_read = badfo_readwrite, 2636 .fo_write = badfo_readwrite, 2637 .fo_ioctl = badfo_ioctl, 2638 .fo_poll = badfo_poll, 2639 .fo_kqfilter = badfo_kqfilter, 2640 .fo_stat = badfo_stat, 2641 .fo_close = badfo_close, 2642 }; 2643 2644 2645 /*-------------------------------------------------------------------*/ 2646 2647 /* 2648 * File Descriptor pseudo-device driver (/dev/fd/). 2649 * 2650 * Opening minor device N dup()s the file (if any) connected to file 2651 * descriptor N belonging to the calling process. Note that this driver 2652 * consists of only the ``open()'' routine, because all subsequent 2653 * references to this file will be direct to the other driver. 2654 * 2655 * XXX: we could give this one a cloning event handler if necessary. 2656 */ 2657 2658 /* ARGSUSED */ 2659 static int 2660 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 2661 { 2662 2663 /* 2664 * XXX Kludge: set curthread->td_dupfd to contain the value of the 2665 * the file descriptor being sought for duplication. The error 2666 * return ensures that the vnode for this device will be released 2667 * by vn_open. Open will detect this special error and take the 2668 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 2669 * will simply report the error. 2670 */ 2671 td->td_dupfd = dev2unit(dev); 2672 return (ENODEV); 2673 } 2674 2675 static struct cdevsw fildesc_cdevsw = { 2676 .d_version = D_VERSION, 2677 .d_flags = D_NEEDGIANT, 2678 .d_open = fdopen, 2679 .d_name = "FD", 2680 }; 2681 2682 static void 2683 fildesc_drvinit(void *unused) 2684 { 2685 struct cdev *dev; 2686 2687 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0"); 2688 make_dev_alias(dev, "stdin"); 2689 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1"); 2690 make_dev_alias(dev, "stdout"); 2691 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2"); 2692 make_dev_alias(dev, "stderr"); 2693 } 2694 2695 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL) 2696