1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_capsicum.h" 43 #include "opt_ddb.h" 44 #include "opt_ktrace.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 49 #include <sys/capsicum.h> 50 #include <sys/conf.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/filio.h> 55 #include <sys/jail.h> 56 #include <sys/kernel.h> 57 #include <sys/limits.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mount.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/selinfo.h> 64 #include <sys/priv.h> 65 #include <sys/proc.h> 66 #include <sys/protosw.h> 67 #include <sys/racct.h> 68 #include <sys/resourcevar.h> 69 #include <sys/sbuf.h> 70 #include <sys/signalvar.h> 71 #include <sys/kdb.h> 72 #include <sys/stat.h> 73 #include <sys/sx.h> 74 #include <sys/syscallsubr.h> 75 #include <sys/sysctl.h> 76 #include <sys/sysproto.h> 77 #include <sys/unistd.h> 78 #include <sys/user.h> 79 #include <sys/vnode.h> 80 #ifdef KTRACE 81 #include <sys/ktrace.h> 82 #endif 83 84 #include <net/vnet.h> 85 86 #include <security/audit/audit.h> 87 88 #include <vm/uma.h> 89 #include <vm/vm.h> 90 91 #include <ddb/ddb.h> 92 93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 95 "file desc to leader structures"); 96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 98 99 MALLOC_DECLARE(M_FADVISE); 100 101 static __read_mostly uma_zone_t file_zone; 102 static __read_mostly uma_zone_t filedesc0_zone; 103 104 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 105 struct thread *td, int holdleaders); 106 static int fd_first_free(struct filedesc *fdp, int low, int size); 107 static int fd_last_used(struct filedesc *fdp, int size); 108 static void fdgrowtable(struct filedesc *fdp, int nfd); 109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 110 static void fdunused(struct filedesc *fdp, int fd); 111 static void fdused(struct filedesc *fdp, int fd); 112 static int getmaxfd(struct thread *td); 113 static u_long *filecaps_copy_prep(const struct filecaps *src); 114 static void filecaps_copy_finish(const struct filecaps *src, 115 struct filecaps *dst, u_long *ioctls); 116 static u_long *filecaps_free_prep(struct filecaps *fcaps); 117 static void filecaps_free_finish(u_long *ioctls); 118 119 /* 120 * Each process has: 121 * 122 * - An array of open file descriptors (fd_ofiles) 123 * - An array of file flags (fd_ofileflags) 124 * - A bitmap recording which descriptors are in use (fd_map) 125 * 126 * A process starts out with NDFILE descriptors. The value of NDFILE has 127 * been selected based the historical limit of 20 open files, and an 128 * assumption that the majority of processes, especially short-lived 129 * processes like shells, will never need more. 130 * 131 * If this initial allocation is exhausted, a larger descriptor table and 132 * map are allocated dynamically, and the pointers in the process's struct 133 * filedesc are updated to point to those. This is repeated every time 134 * the process runs out of file descriptors (provided it hasn't hit its 135 * resource limit). 136 * 137 * Since threads may hold references to individual descriptor table 138 * entries, the tables are never freed. Instead, they are placed on a 139 * linked list and freed only when the struct filedesc is released. 140 */ 141 #define NDFILE 20 142 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 144 #define NDSLOT(x) ((x) / NDENTRIES) 145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 147 148 /* 149 * SLIST entry used to keep track of ofiles which must be reclaimed when 150 * the process exits. 151 */ 152 struct freetable { 153 struct fdescenttbl *ft_table; 154 SLIST_ENTRY(freetable) ft_next; 155 }; 156 157 /* 158 * Initial allocation: a filedesc structure + the head of SLIST used to 159 * keep track of old ofiles + enough space for NDFILE descriptors. 160 */ 161 162 struct fdescenttbl0 { 163 int fdt_nfiles; 164 struct filedescent fdt_ofiles[NDFILE]; 165 }; 166 167 struct filedesc0 { 168 struct filedesc fd_fd; 169 SLIST_HEAD(, freetable) fd_free; 170 struct fdescenttbl0 fd_dfiles; 171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 172 }; 173 174 /* 175 * Descriptor management. 176 */ 177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */ 178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 180 181 /* 182 * If low >= size, just return low. Otherwise find the first zero bit in the 183 * given bitmap, starting at low and not exceeding size - 1. Return size if 184 * not found. 185 */ 186 static int 187 fd_first_free(struct filedesc *fdp, int low, int size) 188 { 189 NDSLOTTYPE *map = fdp->fd_map; 190 NDSLOTTYPE mask; 191 int off, maxoff; 192 193 if (low >= size) 194 return (low); 195 196 off = NDSLOT(low); 197 if (low % NDENTRIES) { 198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 199 if ((mask &= ~map[off]) != 0UL) 200 return (off * NDENTRIES + ffsl(mask) - 1); 201 ++off; 202 } 203 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 204 if (map[off] != ~0UL) 205 return (off * NDENTRIES + ffsl(~map[off]) - 1); 206 return (size); 207 } 208 209 /* 210 * Find the highest non-zero bit in the given bitmap, starting at 0 and 211 * not exceeding size - 1. Return -1 if not found. 212 */ 213 static int 214 fd_last_used(struct filedesc *fdp, int size) 215 { 216 NDSLOTTYPE *map = fdp->fd_map; 217 NDSLOTTYPE mask; 218 int off, minoff; 219 220 off = NDSLOT(size); 221 if (size % NDENTRIES) { 222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 223 if ((mask &= map[off]) != 0) 224 return (off * NDENTRIES + flsl(mask) - 1); 225 --off; 226 } 227 for (minoff = NDSLOT(0); off >= minoff; --off) 228 if (map[off] != 0) 229 return (off * NDENTRIES + flsl(map[off]) - 1); 230 return (-1); 231 } 232 233 static int 234 fdisused(struct filedesc *fdp, int fd) 235 { 236 237 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 238 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 239 240 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 241 } 242 243 /* 244 * Mark a file descriptor as used. 245 */ 246 static void 247 fdused_init(struct filedesc *fdp, int fd) 248 { 249 250 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 251 252 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 253 } 254 255 static void 256 fdused(struct filedesc *fdp, int fd) 257 { 258 259 FILEDESC_XLOCK_ASSERT(fdp); 260 261 fdused_init(fdp, fd); 262 if (fd > fdp->fd_lastfile) 263 fdp->fd_lastfile = fd; 264 if (fd == fdp->fd_freefile) 265 fdp->fd_freefile++; 266 } 267 268 /* 269 * Mark a file descriptor as unused. 270 */ 271 static void 272 fdunused(struct filedesc *fdp, int fd) 273 { 274 275 FILEDESC_XLOCK_ASSERT(fdp); 276 277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 279 ("fd=%d is still in use", fd)); 280 281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 282 if (fd < fdp->fd_freefile) 283 fdp->fd_freefile = fd; 284 if (fd == fdp->fd_lastfile) 285 fdp->fd_lastfile = fd_last_used(fdp, fd); 286 } 287 288 /* 289 * Free a file descriptor. 290 * 291 * Avoid some work if fdp is about to be destroyed. 292 */ 293 static inline void 294 fdefree_last(struct filedescent *fde) 295 { 296 297 filecaps_free(&fde->fde_caps); 298 } 299 300 static inline void 301 fdfree(struct filedesc *fdp, int fd) 302 { 303 struct filedescent *fde; 304 305 fde = &fdp->fd_ofiles[fd]; 306 #ifdef CAPABILITIES 307 seq_write_begin(&fde->fde_seq); 308 #endif 309 fde->fde_file = NULL; 310 #ifdef CAPABILITIES 311 seq_write_end(&fde->fde_seq); 312 #endif 313 fdefree_last(fde); 314 fdunused(fdp, fd); 315 } 316 317 void 318 pwd_ensure_dirs(void) 319 { 320 struct filedesc *fdp; 321 322 fdp = curproc->p_fd; 323 FILEDESC_XLOCK(fdp); 324 if (fdp->fd_cdir == NULL) { 325 fdp->fd_cdir = rootvnode; 326 vrefact(rootvnode); 327 } 328 if (fdp->fd_rdir == NULL) { 329 fdp->fd_rdir = rootvnode; 330 vrefact(rootvnode); 331 } 332 FILEDESC_XUNLOCK(fdp); 333 } 334 335 /* 336 * System calls on descriptors. 337 */ 338 #ifndef _SYS_SYSPROTO_H_ 339 struct getdtablesize_args { 340 int dummy; 341 }; 342 #endif 343 /* ARGSUSED */ 344 int 345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 346 { 347 #ifdef RACCT 348 uint64_t lim; 349 #endif 350 351 td->td_retval[0] = getmaxfd(td); 352 #ifdef RACCT 353 PROC_LOCK(td->td_proc); 354 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 355 PROC_UNLOCK(td->td_proc); 356 if (lim < td->td_retval[0]) 357 td->td_retval[0] = lim; 358 #endif 359 return (0); 360 } 361 362 /* 363 * Duplicate a file descriptor to a particular value. 364 * 365 * Note: keep in mind that a potential race condition exists when closing 366 * descriptors from a shared descriptor table (via rfork). 367 */ 368 #ifndef _SYS_SYSPROTO_H_ 369 struct dup2_args { 370 u_int from; 371 u_int to; 372 }; 373 #endif 374 /* ARGSUSED */ 375 int 376 sys_dup2(struct thread *td, struct dup2_args *uap) 377 { 378 379 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 380 } 381 382 /* 383 * Duplicate a file descriptor. 384 */ 385 #ifndef _SYS_SYSPROTO_H_ 386 struct dup_args { 387 u_int fd; 388 }; 389 #endif 390 /* ARGSUSED */ 391 int 392 sys_dup(struct thread *td, struct dup_args *uap) 393 { 394 395 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 396 } 397 398 /* 399 * The file control system call. 400 */ 401 #ifndef _SYS_SYSPROTO_H_ 402 struct fcntl_args { 403 int fd; 404 int cmd; 405 long arg; 406 }; 407 #endif 408 /* ARGSUSED */ 409 int 410 sys_fcntl(struct thread *td, struct fcntl_args *uap) 411 { 412 413 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 414 } 415 416 int 417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 418 { 419 struct flock fl; 420 struct __oflock ofl; 421 intptr_t arg1; 422 int error, newcmd; 423 424 error = 0; 425 newcmd = cmd; 426 switch (cmd) { 427 case F_OGETLK: 428 case F_OSETLK: 429 case F_OSETLKW: 430 /* 431 * Convert old flock structure to new. 432 */ 433 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 434 fl.l_start = ofl.l_start; 435 fl.l_len = ofl.l_len; 436 fl.l_pid = ofl.l_pid; 437 fl.l_type = ofl.l_type; 438 fl.l_whence = ofl.l_whence; 439 fl.l_sysid = 0; 440 441 switch (cmd) { 442 case F_OGETLK: 443 newcmd = F_GETLK; 444 break; 445 case F_OSETLK: 446 newcmd = F_SETLK; 447 break; 448 case F_OSETLKW: 449 newcmd = F_SETLKW; 450 break; 451 } 452 arg1 = (intptr_t)&fl; 453 break; 454 case F_GETLK: 455 case F_SETLK: 456 case F_SETLKW: 457 case F_SETLK_REMOTE: 458 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 459 arg1 = (intptr_t)&fl; 460 break; 461 default: 462 arg1 = arg; 463 break; 464 } 465 if (error) 466 return (error); 467 error = kern_fcntl(td, fd, newcmd, arg1); 468 if (error) 469 return (error); 470 if (cmd == F_OGETLK) { 471 ofl.l_start = fl.l_start; 472 ofl.l_len = fl.l_len; 473 ofl.l_pid = fl.l_pid; 474 ofl.l_type = fl.l_type; 475 ofl.l_whence = fl.l_whence; 476 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 477 } else if (cmd == F_GETLK) { 478 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 479 } 480 return (error); 481 } 482 483 int 484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 485 { 486 struct filedesc *fdp; 487 struct flock *flp; 488 struct file *fp, *fp2; 489 struct filedescent *fde; 490 struct proc *p; 491 struct vnode *vp; 492 int error, flg, tmp; 493 uint64_t bsize; 494 off_t foffset; 495 496 error = 0; 497 flg = F_POSIX; 498 p = td->td_proc; 499 fdp = p->p_fd; 500 501 AUDIT_ARG_FD(cmd); 502 AUDIT_ARG_CMD(cmd); 503 switch (cmd) { 504 case F_DUPFD: 505 tmp = arg; 506 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 507 break; 508 509 case F_DUPFD_CLOEXEC: 510 tmp = arg; 511 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 512 break; 513 514 case F_DUP2FD: 515 tmp = arg; 516 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 517 break; 518 519 case F_DUP2FD_CLOEXEC: 520 tmp = arg; 521 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 522 break; 523 524 case F_GETFD: 525 error = EBADF; 526 FILEDESC_SLOCK(fdp); 527 fde = fdeget_locked(fdp, fd); 528 if (fde != NULL) { 529 td->td_retval[0] = 530 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 531 error = 0; 532 } 533 FILEDESC_SUNLOCK(fdp); 534 break; 535 536 case F_SETFD: 537 error = EBADF; 538 FILEDESC_XLOCK(fdp); 539 fde = fdeget_locked(fdp, fd); 540 if (fde != NULL) { 541 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 542 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 543 error = 0; 544 } 545 FILEDESC_XUNLOCK(fdp); 546 break; 547 548 case F_GETFL: 549 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); 550 if (error != 0) 551 break; 552 td->td_retval[0] = OFLAGS(fp->f_flag); 553 fdrop(fp, td); 554 break; 555 556 case F_SETFL: 557 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); 558 if (error != 0) 559 break; 560 do { 561 tmp = flg = fp->f_flag; 562 tmp &= ~FCNTLFLAGS; 563 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 564 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 565 tmp = fp->f_flag & FNONBLOCK; 566 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 567 if (error != 0) { 568 fdrop(fp, td); 569 break; 570 } 571 tmp = fp->f_flag & FASYNC; 572 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 573 if (error == 0) { 574 fdrop(fp, td); 575 break; 576 } 577 atomic_clear_int(&fp->f_flag, FNONBLOCK); 578 tmp = 0; 579 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 580 fdrop(fp, td); 581 break; 582 583 case F_GETOWN: 584 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 585 if (error != 0) 586 break; 587 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 588 if (error == 0) 589 td->td_retval[0] = tmp; 590 fdrop(fp, td); 591 break; 592 593 case F_SETOWN: 594 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 595 if (error != 0) 596 break; 597 tmp = arg; 598 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 599 fdrop(fp, td); 600 break; 601 602 case F_SETLK_REMOTE: 603 error = priv_check(td, PRIV_NFS_LOCKD); 604 if (error) 605 return (error); 606 flg = F_REMOTE; 607 goto do_setlk; 608 609 case F_SETLKW: 610 flg |= F_WAIT; 611 /* FALLTHROUGH F_SETLK */ 612 613 case F_SETLK: 614 do_setlk: 615 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL); 616 if (error != 0) 617 break; 618 if (fp->f_type != DTYPE_VNODE) { 619 error = EBADF; 620 fdrop(fp, td); 621 break; 622 } 623 624 flp = (struct flock *)arg; 625 if (flp->l_whence == SEEK_CUR) { 626 foffset = foffset_get(fp); 627 if (foffset < 0 || 628 (flp->l_start > 0 && 629 foffset > OFF_MAX - flp->l_start)) { 630 error = EOVERFLOW; 631 fdrop(fp, td); 632 break; 633 } 634 flp->l_start += foffset; 635 } 636 637 vp = fp->f_vnode; 638 switch (flp->l_type) { 639 case F_RDLCK: 640 if ((fp->f_flag & FREAD) == 0) { 641 error = EBADF; 642 break; 643 } 644 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 645 PROC_LOCK(p->p_leader); 646 p->p_leader->p_flag |= P_ADVLOCK; 647 PROC_UNLOCK(p->p_leader); 648 } 649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 650 flp, flg); 651 break; 652 case F_WRLCK: 653 if ((fp->f_flag & FWRITE) == 0) { 654 error = EBADF; 655 break; 656 } 657 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 658 PROC_LOCK(p->p_leader); 659 p->p_leader->p_flag |= P_ADVLOCK; 660 PROC_UNLOCK(p->p_leader); 661 } 662 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 663 flp, flg); 664 break; 665 case F_UNLCK: 666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 667 flp, flg); 668 break; 669 case F_UNLCKSYS: 670 /* 671 * Temporary api for testing remote lock 672 * infrastructure. 673 */ 674 if (flg != F_REMOTE) { 675 error = EINVAL; 676 break; 677 } 678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 679 F_UNLCKSYS, flp, flg); 680 break; 681 default: 682 error = EINVAL; 683 break; 684 } 685 if (error != 0 || flp->l_type == F_UNLCK || 686 flp->l_type == F_UNLCKSYS) { 687 fdrop(fp, td); 688 break; 689 } 690 691 /* 692 * Check for a race with close. 693 * 694 * The vnode is now advisory locked (or unlocked, but this case 695 * is not really important) as the caller requested. 696 * We had to drop the filedesc lock, so we need to recheck if 697 * the descriptor is still valid, because if it was closed 698 * in the meantime we need to remove advisory lock from the 699 * vnode - close on any descriptor leading to an advisory 700 * locked vnode, removes that lock. 701 * We will return 0 on purpose in that case, as the result of 702 * successful advisory lock might have been externally visible 703 * already. This is fine - effectively we pretend to the caller 704 * that the closing thread was a bit slower and that the 705 * advisory lock succeeded before the close. 706 */ 707 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL); 708 if (error != 0) { 709 fdrop(fp, td); 710 break; 711 } 712 if (fp != fp2) { 713 flp->l_whence = SEEK_SET; 714 flp->l_start = 0; 715 flp->l_len = 0; 716 flp->l_type = F_UNLCK; 717 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 718 F_UNLCK, flp, F_POSIX); 719 } 720 fdrop(fp, td); 721 fdrop(fp2, td); 722 break; 723 724 case F_GETLK: 725 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL); 726 if (error != 0) 727 break; 728 if (fp->f_type != DTYPE_VNODE) { 729 error = EBADF; 730 fdrop(fp, td); 731 break; 732 } 733 flp = (struct flock *)arg; 734 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 735 flp->l_type != F_UNLCK) { 736 error = EINVAL; 737 fdrop(fp, td); 738 break; 739 } 740 if (flp->l_whence == SEEK_CUR) { 741 foffset = foffset_get(fp); 742 if ((flp->l_start > 0 && 743 foffset > OFF_MAX - flp->l_start) || 744 (flp->l_start < 0 && 745 foffset < OFF_MIN - flp->l_start)) { 746 error = EOVERFLOW; 747 fdrop(fp, td); 748 break; 749 } 750 flp->l_start += foffset; 751 } 752 vp = fp->f_vnode; 753 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 754 F_POSIX); 755 fdrop(fp, td); 756 break; 757 758 case F_RDAHEAD: 759 arg = arg ? 128 * 1024: 0; 760 /* FALLTHROUGH */ 761 case F_READAHEAD: 762 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL); 763 if (error != 0) 764 break; 765 if (fp->f_type != DTYPE_VNODE) { 766 fdrop(fp, td); 767 error = EBADF; 768 break; 769 } 770 vp = fp->f_vnode; 771 /* 772 * Exclusive lock synchronizes against f_seqcount reads and 773 * writes in sequential_heuristic(). 774 */ 775 error = vn_lock(vp, LK_EXCLUSIVE); 776 if (error != 0) { 777 fdrop(fp, td); 778 break; 779 } 780 if (arg >= 0) { 781 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 782 fp->f_seqcount = (arg + bsize - 1) / bsize; 783 atomic_set_int(&fp->f_flag, FRDAHEAD); 784 } else { 785 atomic_clear_int(&fp->f_flag, FRDAHEAD); 786 } 787 VOP_UNLOCK(vp, 0); 788 fdrop(fp, td); 789 break; 790 791 default: 792 error = EINVAL; 793 break; 794 } 795 return (error); 796 } 797 798 static int 799 getmaxfd(struct thread *td) 800 { 801 802 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 803 } 804 805 /* 806 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 807 */ 808 int 809 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 810 { 811 struct filedesc *fdp; 812 struct filedescent *oldfde, *newfde; 813 struct proc *p; 814 struct file *delfp; 815 u_long *oioctls, *nioctls; 816 int error, maxfd; 817 818 p = td->td_proc; 819 fdp = p->p_fd; 820 oioctls = NULL; 821 822 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 823 MPASS(mode < FDDUP_LASTMODE); 824 825 AUDIT_ARG_FD(old); 826 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 827 828 /* 829 * Verify we have a valid descriptor to dup from and possibly to 830 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 831 * return EINVAL when the new descriptor is out of bounds. 832 */ 833 if (old < 0) 834 return (EBADF); 835 if (new < 0) 836 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 837 maxfd = getmaxfd(td); 838 if (new >= maxfd) 839 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 840 841 error = EBADF; 842 FILEDESC_XLOCK(fdp); 843 if (fget_locked(fdp, old) == NULL) 844 goto unlock; 845 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 846 td->td_retval[0] = new; 847 if (flags & FDDUP_FLAG_CLOEXEC) 848 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 849 error = 0; 850 goto unlock; 851 } 852 853 /* 854 * If the caller specified a file descriptor, make sure the file 855 * table is large enough to hold it, and grab it. Otherwise, just 856 * allocate a new descriptor the usual way. 857 */ 858 switch (mode) { 859 case FDDUP_NORMAL: 860 case FDDUP_FCNTL: 861 if ((error = fdalloc(td, new, &new)) != 0) 862 goto unlock; 863 break; 864 case FDDUP_MUSTREPLACE: 865 /* Target file descriptor must exist. */ 866 if (fget_locked(fdp, new) == NULL) 867 goto unlock; 868 break; 869 case FDDUP_FIXED: 870 if (new >= fdp->fd_nfiles) { 871 /* 872 * The resource limits are here instead of e.g. 873 * fdalloc(), because the file descriptor table may be 874 * shared between processes, so we can't really use 875 * racct_add()/racct_sub(). Instead of counting the 876 * number of actually allocated descriptors, just put 877 * the limit on the size of the file descriptor table. 878 */ 879 #ifdef RACCT 880 if (RACCT_ENABLED()) { 881 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); 882 if (error != 0) { 883 error = EMFILE; 884 goto unlock; 885 } 886 } 887 #endif 888 fdgrowtable_exp(fdp, new + 1); 889 } 890 if (!fdisused(fdp, new)) 891 fdused(fdp, new); 892 break; 893 default: 894 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 895 } 896 897 KASSERT(old != new, ("new fd is same as old")); 898 899 oldfde = &fdp->fd_ofiles[old]; 900 fhold(oldfde->fde_file); 901 newfde = &fdp->fd_ofiles[new]; 902 delfp = newfde->fde_file; 903 904 oioctls = filecaps_free_prep(&newfde->fde_caps); 905 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 906 907 /* 908 * Duplicate the source descriptor. 909 */ 910 #ifdef CAPABILITIES 911 seq_write_begin(&newfde->fde_seq); 912 #endif 913 memcpy(newfde, oldfde, fde_change_size); 914 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 915 nioctls); 916 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 917 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 918 else 919 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 920 #ifdef CAPABILITIES 921 seq_write_end(&newfde->fde_seq); 922 #endif 923 td->td_retval[0] = new; 924 925 error = 0; 926 927 if (delfp != NULL) { 928 (void) closefp(fdp, new, delfp, td, 1); 929 FILEDESC_UNLOCK_ASSERT(fdp); 930 } else { 931 unlock: 932 FILEDESC_XUNLOCK(fdp); 933 } 934 935 filecaps_free_finish(oioctls); 936 return (error); 937 } 938 939 /* 940 * If sigio is on the list associated with a process or process group, 941 * disable signalling from the device, remove sigio from the list and 942 * free sigio. 943 */ 944 void 945 funsetown(struct sigio **sigiop) 946 { 947 struct sigio *sigio; 948 949 if (*sigiop == NULL) 950 return; 951 SIGIO_LOCK(); 952 sigio = *sigiop; 953 if (sigio == NULL) { 954 SIGIO_UNLOCK(); 955 return; 956 } 957 *(sigio->sio_myref) = NULL; 958 if ((sigio)->sio_pgid < 0) { 959 struct pgrp *pg = (sigio)->sio_pgrp; 960 PGRP_LOCK(pg); 961 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 962 sigio, sio_pgsigio); 963 PGRP_UNLOCK(pg); 964 } else { 965 struct proc *p = (sigio)->sio_proc; 966 PROC_LOCK(p); 967 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 968 sigio, sio_pgsigio); 969 PROC_UNLOCK(p); 970 } 971 SIGIO_UNLOCK(); 972 crfree(sigio->sio_ucred); 973 free(sigio, M_SIGIO); 974 } 975 976 /* 977 * Free a list of sigio structures. 978 * We only need to lock the SIGIO_LOCK because we have made ourselves 979 * inaccessible to callers of fsetown and therefore do not need to lock 980 * the proc or pgrp struct for the list manipulation. 981 */ 982 void 983 funsetownlst(struct sigiolst *sigiolst) 984 { 985 struct proc *p; 986 struct pgrp *pg; 987 struct sigio *sigio; 988 989 sigio = SLIST_FIRST(sigiolst); 990 if (sigio == NULL) 991 return; 992 p = NULL; 993 pg = NULL; 994 995 /* 996 * Every entry of the list should belong 997 * to a single proc or pgrp. 998 */ 999 if (sigio->sio_pgid < 0) { 1000 pg = sigio->sio_pgrp; 1001 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 1002 } else /* if (sigio->sio_pgid > 0) */ { 1003 p = sigio->sio_proc; 1004 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1005 } 1006 1007 SIGIO_LOCK(); 1008 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1009 *(sigio->sio_myref) = NULL; 1010 if (pg != NULL) { 1011 KASSERT(sigio->sio_pgid < 0, 1012 ("Proc sigio in pgrp sigio list")); 1013 KASSERT(sigio->sio_pgrp == pg, 1014 ("Bogus pgrp in sigio list")); 1015 PGRP_LOCK(pg); 1016 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1017 sio_pgsigio); 1018 PGRP_UNLOCK(pg); 1019 } else /* if (p != NULL) */ { 1020 KASSERT(sigio->sio_pgid > 0, 1021 ("Pgrp sigio in proc sigio list")); 1022 KASSERT(sigio->sio_proc == p, 1023 ("Bogus proc in sigio list")); 1024 PROC_LOCK(p); 1025 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1026 sio_pgsigio); 1027 PROC_UNLOCK(p); 1028 } 1029 SIGIO_UNLOCK(); 1030 crfree(sigio->sio_ucred); 1031 free(sigio, M_SIGIO); 1032 SIGIO_LOCK(); 1033 } 1034 SIGIO_UNLOCK(); 1035 } 1036 1037 /* 1038 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1039 * 1040 * After permission checking, add a sigio structure to the sigio list for 1041 * the process or process group. 1042 */ 1043 int 1044 fsetown(pid_t pgid, struct sigio **sigiop) 1045 { 1046 struct proc *proc; 1047 struct pgrp *pgrp; 1048 struct sigio *sigio; 1049 int ret; 1050 1051 if (pgid == 0) { 1052 funsetown(sigiop); 1053 return (0); 1054 } 1055 1056 ret = 0; 1057 1058 /* Allocate and fill in the new sigio out of locks. */ 1059 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1060 sigio->sio_pgid = pgid; 1061 sigio->sio_ucred = crhold(curthread->td_ucred); 1062 sigio->sio_myref = sigiop; 1063 1064 sx_slock(&proctree_lock); 1065 if (pgid > 0) { 1066 proc = pfind(pgid); 1067 if (proc == NULL) { 1068 ret = ESRCH; 1069 goto fail; 1070 } 1071 1072 /* 1073 * Policy - Don't allow a process to FSETOWN a process 1074 * in another session. 1075 * 1076 * Remove this test to allow maximum flexibility or 1077 * restrict FSETOWN to the current process or process 1078 * group for maximum safety. 1079 */ 1080 PROC_UNLOCK(proc); 1081 if (proc->p_session != curthread->td_proc->p_session) { 1082 ret = EPERM; 1083 goto fail; 1084 } 1085 1086 pgrp = NULL; 1087 } else /* if (pgid < 0) */ { 1088 pgrp = pgfind(-pgid); 1089 if (pgrp == NULL) { 1090 ret = ESRCH; 1091 goto fail; 1092 } 1093 PGRP_UNLOCK(pgrp); 1094 1095 /* 1096 * Policy - Don't allow a process to FSETOWN a process 1097 * in another session. 1098 * 1099 * Remove this test to allow maximum flexibility or 1100 * restrict FSETOWN to the current process or process 1101 * group for maximum safety. 1102 */ 1103 if (pgrp->pg_session != curthread->td_proc->p_session) { 1104 ret = EPERM; 1105 goto fail; 1106 } 1107 1108 proc = NULL; 1109 } 1110 funsetown(sigiop); 1111 if (pgid > 0) { 1112 PROC_LOCK(proc); 1113 /* 1114 * Since funsetownlst() is called without the proctree 1115 * locked, we need to check for P_WEXIT. 1116 * XXX: is ESRCH correct? 1117 */ 1118 if ((proc->p_flag & P_WEXIT) != 0) { 1119 PROC_UNLOCK(proc); 1120 ret = ESRCH; 1121 goto fail; 1122 } 1123 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1124 sigio->sio_proc = proc; 1125 PROC_UNLOCK(proc); 1126 } else { 1127 PGRP_LOCK(pgrp); 1128 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1129 sigio->sio_pgrp = pgrp; 1130 PGRP_UNLOCK(pgrp); 1131 } 1132 sx_sunlock(&proctree_lock); 1133 SIGIO_LOCK(); 1134 *sigiop = sigio; 1135 SIGIO_UNLOCK(); 1136 return (0); 1137 1138 fail: 1139 sx_sunlock(&proctree_lock); 1140 crfree(sigio->sio_ucred); 1141 free(sigio, M_SIGIO); 1142 return (ret); 1143 } 1144 1145 /* 1146 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1147 */ 1148 pid_t 1149 fgetown(struct sigio **sigiop) 1150 { 1151 pid_t pgid; 1152 1153 SIGIO_LOCK(); 1154 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1155 SIGIO_UNLOCK(); 1156 return (pgid); 1157 } 1158 1159 /* 1160 * Function drops the filedesc lock on return. 1161 */ 1162 static int 1163 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1164 int holdleaders) 1165 { 1166 int error; 1167 1168 FILEDESC_XLOCK_ASSERT(fdp); 1169 1170 if (holdleaders) { 1171 if (td->td_proc->p_fdtol != NULL) { 1172 /* 1173 * Ask fdfree() to sleep to ensure that all relevant 1174 * process leaders can be traversed in closef(). 1175 */ 1176 fdp->fd_holdleaderscount++; 1177 } else { 1178 holdleaders = 0; 1179 } 1180 } 1181 1182 /* 1183 * We now hold the fp reference that used to be owned by the 1184 * descriptor array. We have to unlock the FILEDESC *AFTER* 1185 * knote_fdclose to prevent a race of the fd getting opened, a knote 1186 * added, and deleteing a knote for the new fd. 1187 */ 1188 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1189 knote_fdclose(td, fd); 1190 1191 /* 1192 * We need to notify mqueue if the object is of type mqueue. 1193 */ 1194 if (__predict_false(fp->f_type == DTYPE_MQUEUE)) 1195 mq_fdclose(td, fd, fp); 1196 FILEDESC_XUNLOCK(fdp); 1197 1198 error = closef(fp, td); 1199 if (holdleaders) { 1200 FILEDESC_XLOCK(fdp); 1201 fdp->fd_holdleaderscount--; 1202 if (fdp->fd_holdleaderscount == 0 && 1203 fdp->fd_holdleaderswakeup != 0) { 1204 fdp->fd_holdleaderswakeup = 0; 1205 wakeup(&fdp->fd_holdleaderscount); 1206 } 1207 FILEDESC_XUNLOCK(fdp); 1208 } 1209 return (error); 1210 } 1211 1212 /* 1213 * Close a file descriptor. 1214 */ 1215 #ifndef _SYS_SYSPROTO_H_ 1216 struct close_args { 1217 int fd; 1218 }; 1219 #endif 1220 /* ARGSUSED */ 1221 int 1222 sys_close(struct thread *td, struct close_args *uap) 1223 { 1224 1225 return (kern_close(td, uap->fd)); 1226 } 1227 1228 int 1229 kern_close(struct thread *td, int fd) 1230 { 1231 struct filedesc *fdp; 1232 struct file *fp; 1233 1234 fdp = td->td_proc->p_fd; 1235 1236 AUDIT_SYSCLOSE(td, fd); 1237 1238 FILEDESC_XLOCK(fdp); 1239 if ((fp = fget_locked(fdp, fd)) == NULL) { 1240 FILEDESC_XUNLOCK(fdp); 1241 return (EBADF); 1242 } 1243 fdfree(fdp, fd); 1244 1245 /* closefp() drops the FILEDESC lock for us. */ 1246 return (closefp(fdp, fd, fp, td, 1)); 1247 } 1248 1249 /* 1250 * Close open file descriptors. 1251 */ 1252 #ifndef _SYS_SYSPROTO_H_ 1253 struct closefrom_args { 1254 int lowfd; 1255 }; 1256 #endif 1257 /* ARGSUSED */ 1258 int 1259 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1260 { 1261 struct filedesc *fdp; 1262 int fd; 1263 1264 fdp = td->td_proc->p_fd; 1265 AUDIT_ARG_FD(uap->lowfd); 1266 1267 /* 1268 * Treat negative starting file descriptor values identical to 1269 * closefrom(0) which closes all files. 1270 */ 1271 if (uap->lowfd < 0) 1272 uap->lowfd = 0; 1273 FILEDESC_SLOCK(fdp); 1274 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1275 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1276 FILEDESC_SUNLOCK(fdp); 1277 (void)kern_close(td, fd); 1278 FILEDESC_SLOCK(fdp); 1279 } 1280 } 1281 FILEDESC_SUNLOCK(fdp); 1282 return (0); 1283 } 1284 1285 #if defined(COMPAT_43) 1286 /* 1287 * Return status information about a file descriptor. 1288 */ 1289 #ifndef _SYS_SYSPROTO_H_ 1290 struct ofstat_args { 1291 int fd; 1292 struct ostat *sb; 1293 }; 1294 #endif 1295 /* ARGSUSED */ 1296 int 1297 ofstat(struct thread *td, struct ofstat_args *uap) 1298 { 1299 struct ostat oub; 1300 struct stat ub; 1301 int error; 1302 1303 error = kern_fstat(td, uap->fd, &ub); 1304 if (error == 0) { 1305 cvtstat(&ub, &oub); 1306 error = copyout(&oub, uap->sb, sizeof(oub)); 1307 } 1308 return (error); 1309 } 1310 #endif /* COMPAT_43 */ 1311 1312 #if defined(COMPAT_FREEBSD11) 1313 int 1314 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1315 { 1316 struct stat sb; 1317 struct freebsd11_stat osb; 1318 int error; 1319 1320 error = kern_fstat(td, uap->fd, &sb); 1321 if (error != 0) 1322 return (error); 1323 error = freebsd11_cvtstat(&sb, &osb); 1324 if (error == 0) 1325 error = copyout(&osb, uap->sb, sizeof(osb)); 1326 return (error); 1327 } 1328 #endif /* COMPAT_FREEBSD11 */ 1329 1330 /* 1331 * Return status information about a file descriptor. 1332 */ 1333 #ifndef _SYS_SYSPROTO_H_ 1334 struct fstat_args { 1335 int fd; 1336 struct stat *sb; 1337 }; 1338 #endif 1339 /* ARGSUSED */ 1340 int 1341 sys_fstat(struct thread *td, struct fstat_args *uap) 1342 { 1343 struct stat ub; 1344 int error; 1345 1346 error = kern_fstat(td, uap->fd, &ub); 1347 if (error == 0) 1348 error = copyout(&ub, uap->sb, sizeof(ub)); 1349 return (error); 1350 } 1351 1352 int 1353 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1354 { 1355 struct file *fp; 1356 int error; 1357 1358 AUDIT_ARG_FD(fd); 1359 1360 error = fget(td, fd, &cap_fstat_rights, &fp); 1361 if (error != 0) 1362 return (error); 1363 1364 AUDIT_ARG_FILE(td->td_proc, fp); 1365 1366 error = fo_stat(fp, sbp, td->td_ucred, td); 1367 fdrop(fp, td); 1368 #ifdef __STAT_TIME_T_EXT 1369 if (error == 0) { 1370 sbp->st_atim_ext = 0; 1371 sbp->st_mtim_ext = 0; 1372 sbp->st_ctim_ext = 0; 1373 sbp->st_btim_ext = 0; 1374 } 1375 #endif 1376 #ifdef KTRACE 1377 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1378 ktrstat(sbp); 1379 #endif 1380 return (error); 1381 } 1382 1383 #if defined(COMPAT_FREEBSD11) 1384 /* 1385 * Return status information about a file descriptor. 1386 */ 1387 #ifndef _SYS_SYSPROTO_H_ 1388 struct freebsd11_nfstat_args { 1389 int fd; 1390 struct nstat *sb; 1391 }; 1392 #endif 1393 /* ARGSUSED */ 1394 int 1395 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1396 { 1397 struct nstat nub; 1398 struct stat ub; 1399 int error; 1400 1401 error = kern_fstat(td, uap->fd, &ub); 1402 if (error == 0) { 1403 freebsd11_cvtnstat(&ub, &nub); 1404 error = copyout(&nub, uap->sb, sizeof(nub)); 1405 } 1406 return (error); 1407 } 1408 #endif /* COMPAT_FREEBSD11 */ 1409 1410 /* 1411 * Return pathconf information about a file descriptor. 1412 */ 1413 #ifndef _SYS_SYSPROTO_H_ 1414 struct fpathconf_args { 1415 int fd; 1416 int name; 1417 }; 1418 #endif 1419 /* ARGSUSED */ 1420 int 1421 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1422 { 1423 long value; 1424 int error; 1425 1426 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1427 if (error == 0) 1428 td->td_retval[0] = value; 1429 return (error); 1430 } 1431 1432 int 1433 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1434 { 1435 struct file *fp; 1436 struct vnode *vp; 1437 int error; 1438 1439 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1440 if (error != 0) 1441 return (error); 1442 1443 if (name == _PC_ASYNC_IO) { 1444 *valuep = _POSIX_ASYNCHRONOUS_IO; 1445 goto out; 1446 } 1447 vp = fp->f_vnode; 1448 if (vp != NULL) { 1449 vn_lock(vp, LK_SHARED | LK_RETRY); 1450 error = VOP_PATHCONF(vp, name, valuep); 1451 VOP_UNLOCK(vp, 0); 1452 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1453 if (name != _PC_PIPE_BUF) { 1454 error = EINVAL; 1455 } else { 1456 *valuep = PIPE_BUF; 1457 error = 0; 1458 } 1459 } else { 1460 error = EOPNOTSUPP; 1461 } 1462 out: 1463 fdrop(fp, td); 1464 return (error); 1465 } 1466 1467 /* 1468 * Copy filecaps structure allocating memory for ioctls array if needed. 1469 * 1470 * The last parameter indicates whether the fdtable is locked. If it is not and 1471 * ioctls are encountered, copying fails and the caller must lock the table. 1472 * 1473 * Note that if the table was not locked, the caller has to check the relevant 1474 * sequence counter to determine whether the operation was successful. 1475 */ 1476 bool 1477 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1478 { 1479 size_t size; 1480 1481 if (src->fc_ioctls != NULL && !locked) 1482 return (false); 1483 memcpy(dst, src, sizeof(*src)); 1484 if (src->fc_ioctls == NULL) 1485 return (true); 1486 1487 KASSERT(src->fc_nioctls > 0, 1488 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1489 1490 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1491 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1492 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1493 return (true); 1494 } 1495 1496 static u_long * 1497 filecaps_copy_prep(const struct filecaps *src) 1498 { 1499 u_long *ioctls; 1500 size_t size; 1501 1502 if (__predict_true(src->fc_ioctls == NULL)) 1503 return (NULL); 1504 1505 KASSERT(src->fc_nioctls > 0, 1506 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1507 1508 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1509 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1510 return (ioctls); 1511 } 1512 1513 static void 1514 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1515 u_long *ioctls) 1516 { 1517 size_t size; 1518 1519 *dst = *src; 1520 if (__predict_true(src->fc_ioctls == NULL)) { 1521 MPASS(ioctls == NULL); 1522 return; 1523 } 1524 1525 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1526 dst->fc_ioctls = ioctls; 1527 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1528 } 1529 1530 /* 1531 * Move filecaps structure to the new place and clear the old place. 1532 */ 1533 void 1534 filecaps_move(struct filecaps *src, struct filecaps *dst) 1535 { 1536 1537 *dst = *src; 1538 bzero(src, sizeof(*src)); 1539 } 1540 1541 /* 1542 * Fill the given filecaps structure with full rights. 1543 */ 1544 static void 1545 filecaps_fill(struct filecaps *fcaps) 1546 { 1547 1548 CAP_ALL(&fcaps->fc_rights); 1549 fcaps->fc_ioctls = NULL; 1550 fcaps->fc_nioctls = -1; 1551 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1552 } 1553 1554 /* 1555 * Free memory allocated within filecaps structure. 1556 */ 1557 void 1558 filecaps_free(struct filecaps *fcaps) 1559 { 1560 1561 free(fcaps->fc_ioctls, M_FILECAPS); 1562 bzero(fcaps, sizeof(*fcaps)); 1563 } 1564 1565 static u_long * 1566 filecaps_free_prep(struct filecaps *fcaps) 1567 { 1568 u_long *ioctls; 1569 1570 ioctls = fcaps->fc_ioctls; 1571 bzero(fcaps, sizeof(*fcaps)); 1572 return (ioctls); 1573 } 1574 1575 static void 1576 filecaps_free_finish(u_long *ioctls) 1577 { 1578 1579 free(ioctls, M_FILECAPS); 1580 } 1581 1582 /* 1583 * Validate the given filecaps structure. 1584 */ 1585 static void 1586 filecaps_validate(const struct filecaps *fcaps, const char *func) 1587 { 1588 1589 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1590 ("%s: invalid rights", func)); 1591 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1592 ("%s: invalid fcntls", func)); 1593 KASSERT(fcaps->fc_fcntls == 0 || 1594 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1595 ("%s: fcntls without CAP_FCNTL", func)); 1596 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1597 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1598 ("%s: invalid ioctls", func)); 1599 KASSERT(fcaps->fc_nioctls == 0 || 1600 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1601 ("%s: ioctls without CAP_IOCTL", func)); 1602 } 1603 1604 static void 1605 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1606 { 1607 int nfd1; 1608 1609 FILEDESC_XLOCK_ASSERT(fdp); 1610 1611 nfd1 = fdp->fd_nfiles * 2; 1612 if (nfd1 < nfd) 1613 nfd1 = nfd; 1614 fdgrowtable(fdp, nfd1); 1615 } 1616 1617 /* 1618 * Grow the file table to accommodate (at least) nfd descriptors. 1619 */ 1620 static void 1621 fdgrowtable(struct filedesc *fdp, int nfd) 1622 { 1623 struct filedesc0 *fdp0; 1624 struct freetable *ft; 1625 struct fdescenttbl *ntable; 1626 struct fdescenttbl *otable; 1627 int nnfiles, onfiles; 1628 NDSLOTTYPE *nmap, *omap; 1629 1630 /* 1631 * If lastfile is -1 this struct filedesc was just allocated and we are 1632 * growing it to accommodate for the one we are going to copy from. There 1633 * is no need to have a lock on this one as it's not visible to anyone. 1634 */ 1635 if (fdp->fd_lastfile != -1) 1636 FILEDESC_XLOCK_ASSERT(fdp); 1637 1638 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1639 1640 /* save old values */ 1641 onfiles = fdp->fd_nfiles; 1642 otable = fdp->fd_files; 1643 omap = fdp->fd_map; 1644 1645 /* compute the size of the new table */ 1646 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1647 if (nnfiles <= onfiles) 1648 /* the table is already large enough */ 1649 return; 1650 1651 /* 1652 * Allocate a new table. We need enough space for the number of 1653 * entries, file entries themselves and the struct freetable we will use 1654 * when we decommission the table and place it on the freelist. 1655 * We place the struct freetable in the middle so we don't have 1656 * to worry about padding. 1657 */ 1658 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1659 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1660 sizeof(struct freetable), 1661 M_FILEDESC, M_ZERO | M_WAITOK); 1662 /* copy the old data */ 1663 ntable->fdt_nfiles = nnfiles; 1664 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1665 onfiles * sizeof(ntable->fdt_ofiles[0])); 1666 1667 /* 1668 * Allocate a new map only if the old is not large enough. It will 1669 * grow at a slower rate than the table as it can map more 1670 * entries than the table can hold. 1671 */ 1672 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1673 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1674 M_ZERO | M_WAITOK); 1675 /* copy over the old data and update the pointer */ 1676 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1677 fdp->fd_map = nmap; 1678 } 1679 1680 /* 1681 * Make sure that ntable is correctly initialized before we replace 1682 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1683 * data. 1684 */ 1685 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1686 1687 /* 1688 * Do not free the old file table, as some threads may still 1689 * reference entries within it. Instead, place it on a freelist 1690 * which will be processed when the struct filedesc is released. 1691 * 1692 * Note that if onfiles == NDFILE, we're dealing with the original 1693 * static allocation contained within (struct filedesc0 *)fdp, 1694 * which must not be freed. 1695 */ 1696 if (onfiles > NDFILE) { 1697 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1698 fdp0 = (struct filedesc0 *)fdp; 1699 ft->ft_table = otable; 1700 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1701 } 1702 /* 1703 * The map does not have the same possibility of threads still 1704 * holding references to it. So always free it as long as it 1705 * does not reference the original static allocation. 1706 */ 1707 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1708 free(omap, M_FILEDESC); 1709 } 1710 1711 /* 1712 * Allocate a file descriptor for the process. 1713 */ 1714 int 1715 fdalloc(struct thread *td, int minfd, int *result) 1716 { 1717 struct proc *p = td->td_proc; 1718 struct filedesc *fdp = p->p_fd; 1719 int fd, maxfd, allocfd; 1720 #ifdef RACCT 1721 int error; 1722 #endif 1723 1724 FILEDESC_XLOCK_ASSERT(fdp); 1725 1726 if (fdp->fd_freefile > minfd) 1727 minfd = fdp->fd_freefile; 1728 1729 maxfd = getmaxfd(td); 1730 1731 /* 1732 * Search the bitmap for a free descriptor starting at minfd. 1733 * If none is found, grow the file table. 1734 */ 1735 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1736 if (fd >= maxfd) 1737 return (EMFILE); 1738 if (fd >= fdp->fd_nfiles) { 1739 allocfd = min(fd * 2, maxfd); 1740 #ifdef RACCT 1741 if (RACCT_ENABLED()) { 1742 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 1743 if (error != 0) 1744 return (EMFILE); 1745 } 1746 #endif 1747 /* 1748 * fd is already equal to first free descriptor >= minfd, so 1749 * we only need to grow the table and we are done. 1750 */ 1751 fdgrowtable_exp(fdp, allocfd); 1752 } 1753 1754 /* 1755 * Perform some sanity checks, then mark the file descriptor as 1756 * used and return it to the caller. 1757 */ 1758 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1759 ("invalid descriptor %d", fd)); 1760 KASSERT(!fdisused(fdp, fd), 1761 ("fd_first_free() returned non-free descriptor")); 1762 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1763 ("file descriptor isn't free")); 1764 fdused(fdp, fd); 1765 *result = fd; 1766 return (0); 1767 } 1768 1769 /* 1770 * Allocate n file descriptors for the process. 1771 */ 1772 int 1773 fdallocn(struct thread *td, int minfd, int *fds, int n) 1774 { 1775 struct proc *p = td->td_proc; 1776 struct filedesc *fdp = p->p_fd; 1777 int i; 1778 1779 FILEDESC_XLOCK_ASSERT(fdp); 1780 1781 for (i = 0; i < n; i++) 1782 if (fdalloc(td, 0, &fds[i]) != 0) 1783 break; 1784 1785 if (i < n) { 1786 for (i--; i >= 0; i--) 1787 fdunused(fdp, fds[i]); 1788 return (EMFILE); 1789 } 1790 1791 return (0); 1792 } 1793 1794 /* 1795 * Create a new open file structure and allocate a file descriptor for the 1796 * process that refers to it. We add one reference to the file for the 1797 * descriptor table and one reference for resultfp. This is to prevent us 1798 * being preempted and the entry in the descriptor table closed after we 1799 * release the FILEDESC lock. 1800 */ 1801 int 1802 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1803 struct filecaps *fcaps) 1804 { 1805 struct file *fp; 1806 int error, fd; 1807 1808 error = falloc_noinstall(td, &fp); 1809 if (error) 1810 return (error); /* no reference held on error */ 1811 1812 error = finstall(td, fp, &fd, flags, fcaps); 1813 if (error) { 1814 fdrop(fp, td); /* one reference (fp only) */ 1815 return (error); 1816 } 1817 1818 if (resultfp != NULL) 1819 *resultfp = fp; /* copy out result */ 1820 else 1821 fdrop(fp, td); /* release local reference */ 1822 1823 if (resultfd != NULL) 1824 *resultfd = fd; 1825 1826 return (0); 1827 } 1828 1829 /* 1830 * Create a new open file structure without allocating a file descriptor. 1831 */ 1832 int 1833 falloc_noinstall(struct thread *td, struct file **resultfp) 1834 { 1835 struct file *fp; 1836 int maxuserfiles = maxfiles - (maxfiles / 20); 1837 int openfiles_new; 1838 static struct timeval lastfail; 1839 static int curfail; 1840 1841 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1842 1843 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 1844 if ((openfiles_new >= maxuserfiles && 1845 priv_check(td, PRIV_MAXFILES) != 0) || 1846 openfiles_new >= maxfiles) { 1847 atomic_subtract_int(&openfiles, 1); 1848 if (ppsratecheck(&lastfail, &curfail, 1)) { 1849 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 1850 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 1851 } 1852 return (ENFILE); 1853 } 1854 fp = uma_zalloc(file_zone, M_WAITOK); 1855 bzero(fp, sizeof(*fp)); 1856 refcount_init(&fp->f_count, 1); 1857 fp->f_cred = crhold(td->td_ucred); 1858 fp->f_ops = &badfileops; 1859 *resultfp = fp; 1860 return (0); 1861 } 1862 1863 /* 1864 * Install a file in a file descriptor table. 1865 */ 1866 void 1867 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1868 struct filecaps *fcaps) 1869 { 1870 struct filedescent *fde; 1871 1872 MPASS(fp != NULL); 1873 if (fcaps != NULL) 1874 filecaps_validate(fcaps, __func__); 1875 FILEDESC_XLOCK_ASSERT(fdp); 1876 1877 fde = &fdp->fd_ofiles[fd]; 1878 #ifdef CAPABILITIES 1879 seq_write_begin(&fde->fde_seq); 1880 #endif 1881 fde->fde_file = fp; 1882 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1883 if (fcaps != NULL) 1884 filecaps_move(fcaps, &fde->fde_caps); 1885 else 1886 filecaps_fill(&fde->fde_caps); 1887 #ifdef CAPABILITIES 1888 seq_write_end(&fde->fde_seq); 1889 #endif 1890 } 1891 1892 int 1893 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1894 struct filecaps *fcaps) 1895 { 1896 struct filedesc *fdp = td->td_proc->p_fd; 1897 int error; 1898 1899 MPASS(fd != NULL); 1900 1901 FILEDESC_XLOCK(fdp); 1902 if ((error = fdalloc(td, 0, fd))) { 1903 FILEDESC_XUNLOCK(fdp); 1904 return (error); 1905 } 1906 fhold(fp); 1907 _finstall(fdp, fp, *fd, flags, fcaps); 1908 FILEDESC_XUNLOCK(fdp); 1909 return (0); 1910 } 1911 1912 /* 1913 * Build a new filedesc structure from another. 1914 * Copy the current, root, and jail root vnode references. 1915 * 1916 * If fdp is not NULL, return with it shared locked. 1917 */ 1918 struct filedesc * 1919 fdinit(struct filedesc *fdp, bool prepfiles) 1920 { 1921 struct filedesc0 *newfdp0; 1922 struct filedesc *newfdp; 1923 1924 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1925 newfdp = &newfdp0->fd_fd; 1926 1927 /* Create the file descriptor table. */ 1928 FILEDESC_LOCK_INIT(newfdp); 1929 refcount_init(&newfdp->fd_refcnt, 1); 1930 refcount_init(&newfdp->fd_holdcnt, 1); 1931 newfdp->fd_cmask = CMASK; 1932 newfdp->fd_map = newfdp0->fd_dmap; 1933 newfdp->fd_lastfile = -1; 1934 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1935 newfdp->fd_files->fdt_nfiles = NDFILE; 1936 1937 if (fdp == NULL) 1938 return (newfdp); 1939 1940 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1941 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1942 1943 FILEDESC_SLOCK(fdp); 1944 newfdp->fd_cdir = fdp->fd_cdir; 1945 if (newfdp->fd_cdir) 1946 vrefact(newfdp->fd_cdir); 1947 newfdp->fd_rdir = fdp->fd_rdir; 1948 if (newfdp->fd_rdir) 1949 vrefact(newfdp->fd_rdir); 1950 newfdp->fd_jdir = fdp->fd_jdir; 1951 if (newfdp->fd_jdir) 1952 vrefact(newfdp->fd_jdir); 1953 1954 if (!prepfiles) { 1955 FILEDESC_SUNLOCK(fdp); 1956 } else { 1957 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1958 FILEDESC_SUNLOCK(fdp); 1959 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1960 FILEDESC_SLOCK(fdp); 1961 } 1962 } 1963 1964 return (newfdp); 1965 } 1966 1967 static struct filedesc * 1968 fdhold(struct proc *p) 1969 { 1970 struct filedesc *fdp; 1971 1972 PROC_LOCK_ASSERT(p, MA_OWNED); 1973 fdp = p->p_fd; 1974 if (fdp != NULL) 1975 refcount_acquire(&fdp->fd_holdcnt); 1976 return (fdp); 1977 } 1978 1979 static void 1980 fddrop(struct filedesc *fdp) 1981 { 1982 1983 if (fdp->fd_holdcnt > 1) { 1984 if (refcount_release(&fdp->fd_holdcnt) == 0) 1985 return; 1986 } 1987 1988 FILEDESC_LOCK_DESTROY(fdp); 1989 uma_zfree(filedesc0_zone, fdp); 1990 } 1991 1992 /* 1993 * Share a filedesc structure. 1994 */ 1995 struct filedesc * 1996 fdshare(struct filedesc *fdp) 1997 { 1998 1999 refcount_acquire(&fdp->fd_refcnt); 2000 return (fdp); 2001 } 2002 2003 /* 2004 * Unshare a filedesc structure, if necessary by making a copy 2005 */ 2006 void 2007 fdunshare(struct thread *td) 2008 { 2009 struct filedesc *tmp; 2010 struct proc *p = td->td_proc; 2011 2012 if (p->p_fd->fd_refcnt == 1) 2013 return; 2014 2015 tmp = fdcopy(p->p_fd); 2016 fdescfree(td); 2017 p->p_fd = tmp; 2018 } 2019 2020 void 2021 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 2022 { 2023 2024 fdescfree(td); 2025 td->td_proc->p_fd = fdp; 2026 } 2027 2028 /* 2029 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2030 * this is to ease callers, not catch errors. 2031 */ 2032 struct filedesc * 2033 fdcopy(struct filedesc *fdp) 2034 { 2035 struct filedesc *newfdp; 2036 struct filedescent *nfde, *ofde; 2037 int i; 2038 2039 MPASS(fdp != NULL); 2040 2041 newfdp = fdinit(fdp, true); 2042 /* copy all passable descriptors (i.e. not kqueue) */ 2043 newfdp->fd_freefile = -1; 2044 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2045 ofde = &fdp->fd_ofiles[i]; 2046 if (ofde->fde_file == NULL || 2047 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2048 if (newfdp->fd_freefile == -1) 2049 newfdp->fd_freefile = i; 2050 continue; 2051 } 2052 nfde = &newfdp->fd_ofiles[i]; 2053 *nfde = *ofde; 2054 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2055 fhold(nfde->fde_file); 2056 fdused_init(newfdp, i); 2057 newfdp->fd_lastfile = i; 2058 } 2059 if (newfdp->fd_freefile == -1) 2060 newfdp->fd_freefile = i; 2061 newfdp->fd_cmask = fdp->fd_cmask; 2062 FILEDESC_SUNLOCK(fdp); 2063 return (newfdp); 2064 } 2065 2066 /* 2067 * Copies a filedesc structure, while remapping all file descriptors 2068 * stored inside using a translation table. 2069 * 2070 * File descriptors are copied over to the new file descriptor table, 2071 * regardless of whether the close-on-exec flag is set. 2072 */ 2073 int 2074 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 2075 struct filedesc **ret) 2076 { 2077 struct filedesc *newfdp; 2078 struct filedescent *nfde, *ofde; 2079 int error, i; 2080 2081 MPASS(fdp != NULL); 2082 2083 newfdp = fdinit(fdp, true); 2084 if (nfds > fdp->fd_lastfile + 1) { 2085 /* New table cannot be larger than the old one. */ 2086 error = E2BIG; 2087 goto bad; 2088 } 2089 /* Copy all passable descriptors (i.e. not kqueue). */ 2090 newfdp->fd_freefile = nfds; 2091 for (i = 0; i < nfds; ++i) { 2092 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { 2093 /* File descriptor out of bounds. */ 2094 error = EBADF; 2095 goto bad; 2096 } 2097 ofde = &fdp->fd_ofiles[fds[i]]; 2098 if (ofde->fde_file == NULL) { 2099 /* Unused file descriptor. */ 2100 error = EBADF; 2101 goto bad; 2102 } 2103 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2104 /* File descriptor cannot be passed. */ 2105 error = EINVAL; 2106 goto bad; 2107 } 2108 nfde = &newfdp->fd_ofiles[i]; 2109 *nfde = *ofde; 2110 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2111 fhold(nfde->fde_file); 2112 fdused_init(newfdp, i); 2113 newfdp->fd_lastfile = i; 2114 } 2115 newfdp->fd_cmask = fdp->fd_cmask; 2116 FILEDESC_SUNLOCK(fdp); 2117 *ret = newfdp; 2118 return (0); 2119 bad: 2120 FILEDESC_SUNLOCK(fdp); 2121 fdescfree_remapped(newfdp); 2122 return (error); 2123 } 2124 2125 /* 2126 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2127 * one of processes using it exits) and the table used to be shared. 2128 */ 2129 static void 2130 fdclearlocks(struct thread *td) 2131 { 2132 struct filedesc *fdp; 2133 struct filedesc_to_leader *fdtol; 2134 struct flock lf; 2135 struct file *fp; 2136 struct proc *p; 2137 struct vnode *vp; 2138 int i; 2139 2140 p = td->td_proc; 2141 fdp = p->p_fd; 2142 fdtol = p->p_fdtol; 2143 MPASS(fdtol != NULL); 2144 2145 FILEDESC_XLOCK(fdp); 2146 KASSERT(fdtol->fdl_refcount > 0, 2147 ("filedesc_to_refcount botch: fdl_refcount=%d", 2148 fdtol->fdl_refcount)); 2149 if (fdtol->fdl_refcount == 1 && 2150 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2151 for (i = 0; i <= fdp->fd_lastfile; i++) { 2152 fp = fdp->fd_ofiles[i].fde_file; 2153 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2154 continue; 2155 fhold(fp); 2156 FILEDESC_XUNLOCK(fdp); 2157 lf.l_whence = SEEK_SET; 2158 lf.l_start = 0; 2159 lf.l_len = 0; 2160 lf.l_type = F_UNLCK; 2161 vp = fp->f_vnode; 2162 (void) VOP_ADVLOCK(vp, 2163 (caddr_t)p->p_leader, F_UNLCK, 2164 &lf, F_POSIX); 2165 FILEDESC_XLOCK(fdp); 2166 fdrop(fp, td); 2167 } 2168 } 2169 retry: 2170 if (fdtol->fdl_refcount == 1) { 2171 if (fdp->fd_holdleaderscount > 0 && 2172 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2173 /* 2174 * close() or kern_dup() has cleared a reference 2175 * in a shared file descriptor table. 2176 */ 2177 fdp->fd_holdleaderswakeup = 1; 2178 sx_sleep(&fdp->fd_holdleaderscount, 2179 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2180 goto retry; 2181 } 2182 if (fdtol->fdl_holdcount > 0) { 2183 /* 2184 * Ensure that fdtol->fdl_leader remains 2185 * valid in closef(). 2186 */ 2187 fdtol->fdl_wakeup = 1; 2188 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2189 "fdlhold", 0); 2190 goto retry; 2191 } 2192 } 2193 fdtol->fdl_refcount--; 2194 if (fdtol->fdl_refcount == 0 && 2195 fdtol->fdl_holdcount == 0) { 2196 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2197 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2198 } else 2199 fdtol = NULL; 2200 p->p_fdtol = NULL; 2201 FILEDESC_XUNLOCK(fdp); 2202 if (fdtol != NULL) 2203 free(fdtol, M_FILEDESC_TO_LEADER); 2204 } 2205 2206 /* 2207 * Release a filedesc structure. 2208 */ 2209 static void 2210 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2211 { 2212 struct filedesc0 *fdp0; 2213 struct freetable *ft, *tft; 2214 struct filedescent *fde; 2215 struct file *fp; 2216 int i; 2217 2218 for (i = 0; i <= fdp->fd_lastfile; i++) { 2219 fde = &fdp->fd_ofiles[i]; 2220 fp = fde->fde_file; 2221 if (fp != NULL) { 2222 fdefree_last(fde); 2223 if (needclose) 2224 (void) closef(fp, td); 2225 else 2226 fdrop(fp, td); 2227 } 2228 } 2229 2230 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2231 free(fdp->fd_map, M_FILEDESC); 2232 if (fdp->fd_nfiles > NDFILE) 2233 free(fdp->fd_files, M_FILEDESC); 2234 2235 fdp0 = (struct filedesc0 *)fdp; 2236 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2237 free(ft->ft_table, M_FILEDESC); 2238 2239 fddrop(fdp); 2240 } 2241 2242 void 2243 fdescfree(struct thread *td) 2244 { 2245 struct proc *p; 2246 struct filedesc *fdp; 2247 struct vnode *cdir, *jdir, *rdir; 2248 2249 p = td->td_proc; 2250 fdp = p->p_fd; 2251 MPASS(fdp != NULL); 2252 2253 #ifdef RACCT 2254 if (RACCT_ENABLED()) 2255 racct_set_unlocked(p, RACCT_NOFILE, 0); 2256 #endif 2257 2258 if (p->p_fdtol != NULL) 2259 fdclearlocks(td); 2260 2261 PROC_LOCK(p); 2262 p->p_fd = NULL; 2263 PROC_UNLOCK(p); 2264 2265 if (refcount_release(&fdp->fd_refcnt) == 0) 2266 return; 2267 2268 FILEDESC_XLOCK(fdp); 2269 cdir = fdp->fd_cdir; 2270 fdp->fd_cdir = NULL; 2271 rdir = fdp->fd_rdir; 2272 fdp->fd_rdir = NULL; 2273 jdir = fdp->fd_jdir; 2274 fdp->fd_jdir = NULL; 2275 FILEDESC_XUNLOCK(fdp); 2276 2277 if (cdir != NULL) 2278 vrele(cdir); 2279 if (rdir != NULL) 2280 vrele(rdir); 2281 if (jdir != NULL) 2282 vrele(jdir); 2283 2284 fdescfree_fds(td, fdp, 1); 2285 } 2286 2287 void 2288 fdescfree_remapped(struct filedesc *fdp) 2289 { 2290 2291 if (fdp->fd_cdir != NULL) 2292 vrele(fdp->fd_cdir); 2293 if (fdp->fd_rdir != NULL) 2294 vrele(fdp->fd_rdir); 2295 if (fdp->fd_jdir != NULL) 2296 vrele(fdp->fd_jdir); 2297 2298 fdescfree_fds(curthread, fdp, 0); 2299 } 2300 2301 /* 2302 * For setugid programs, we don't want to people to use that setugidness 2303 * to generate error messages which write to a file which otherwise would 2304 * otherwise be off-limits to the process. We check for filesystems where 2305 * the vnode can change out from under us after execve (like [lin]procfs). 2306 * 2307 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2308 * sufficient. We also don't check for setugidness since we know we are. 2309 */ 2310 static bool 2311 is_unsafe(struct file *fp) 2312 { 2313 struct vnode *vp; 2314 2315 if (fp->f_type != DTYPE_VNODE) 2316 return (false); 2317 2318 vp = fp->f_vnode; 2319 return ((vp->v_vflag & VV_PROCDEP) != 0); 2320 } 2321 2322 /* 2323 * Make this setguid thing safe, if at all possible. 2324 */ 2325 void 2326 fdsetugidsafety(struct thread *td) 2327 { 2328 struct filedesc *fdp; 2329 struct file *fp; 2330 int i; 2331 2332 fdp = td->td_proc->p_fd; 2333 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2334 MPASS(fdp->fd_nfiles >= 3); 2335 for (i = 0; i <= 2; i++) { 2336 fp = fdp->fd_ofiles[i].fde_file; 2337 if (fp != NULL && is_unsafe(fp)) { 2338 FILEDESC_XLOCK(fdp); 2339 knote_fdclose(td, i); 2340 /* 2341 * NULL-out descriptor prior to close to avoid 2342 * a race while close blocks. 2343 */ 2344 fdfree(fdp, i); 2345 FILEDESC_XUNLOCK(fdp); 2346 (void) closef(fp, td); 2347 } 2348 } 2349 } 2350 2351 /* 2352 * If a specific file object occupies a specific file descriptor, close the 2353 * file descriptor entry and drop a reference on the file object. This is a 2354 * convenience function to handle a subsequent error in a function that calls 2355 * falloc() that handles the race that another thread might have closed the 2356 * file descriptor out from under the thread creating the file object. 2357 */ 2358 void 2359 fdclose(struct thread *td, struct file *fp, int idx) 2360 { 2361 struct filedesc *fdp = td->td_proc->p_fd; 2362 2363 FILEDESC_XLOCK(fdp); 2364 if (fdp->fd_ofiles[idx].fde_file == fp) { 2365 fdfree(fdp, idx); 2366 FILEDESC_XUNLOCK(fdp); 2367 fdrop(fp, td); 2368 } else 2369 FILEDESC_XUNLOCK(fdp); 2370 } 2371 2372 /* 2373 * Close any files on exec? 2374 */ 2375 void 2376 fdcloseexec(struct thread *td) 2377 { 2378 struct filedesc *fdp; 2379 struct filedescent *fde; 2380 struct file *fp; 2381 int i; 2382 2383 fdp = td->td_proc->p_fd; 2384 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2385 for (i = 0; i <= fdp->fd_lastfile; i++) { 2386 fde = &fdp->fd_ofiles[i]; 2387 fp = fde->fde_file; 2388 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2389 (fde->fde_flags & UF_EXCLOSE))) { 2390 FILEDESC_XLOCK(fdp); 2391 fdfree(fdp, i); 2392 (void) closefp(fdp, i, fp, td, 0); 2393 FILEDESC_UNLOCK_ASSERT(fdp); 2394 } 2395 } 2396 } 2397 2398 /* 2399 * It is unsafe for set[ug]id processes to be started with file 2400 * descriptors 0..2 closed, as these descriptors are given implicit 2401 * significance in the Standard C library. fdcheckstd() will create a 2402 * descriptor referencing /dev/null for each of stdin, stdout, and 2403 * stderr that is not already open. 2404 */ 2405 int 2406 fdcheckstd(struct thread *td) 2407 { 2408 struct filedesc *fdp; 2409 register_t save; 2410 int i, error, devnull; 2411 2412 fdp = td->td_proc->p_fd; 2413 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2414 MPASS(fdp->fd_nfiles >= 3); 2415 devnull = -1; 2416 for (i = 0; i <= 2; i++) { 2417 if (fdp->fd_ofiles[i].fde_file != NULL) 2418 continue; 2419 2420 save = td->td_retval[0]; 2421 if (devnull != -1) { 2422 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2423 } else { 2424 error = kern_openat(td, AT_FDCWD, "/dev/null", 2425 UIO_SYSSPACE, O_RDWR, 0); 2426 if (error == 0) { 2427 devnull = td->td_retval[0]; 2428 KASSERT(devnull == i, ("we didn't get our fd")); 2429 } 2430 } 2431 td->td_retval[0] = save; 2432 if (error != 0) 2433 return (error); 2434 } 2435 return (0); 2436 } 2437 2438 /* 2439 * Internal form of close. Decrement reference count on file structure. 2440 * Note: td may be NULL when closing a file that was being passed in a 2441 * message. 2442 */ 2443 int 2444 closef(struct file *fp, struct thread *td) 2445 { 2446 struct vnode *vp; 2447 struct flock lf; 2448 struct filedesc_to_leader *fdtol; 2449 struct filedesc *fdp; 2450 2451 /* 2452 * POSIX record locking dictates that any close releases ALL 2453 * locks owned by this process. This is handled by setting 2454 * a flag in the unlock to free ONLY locks obeying POSIX 2455 * semantics, and not to free BSD-style file locks. 2456 * If the descriptor was in a message, POSIX-style locks 2457 * aren't passed with the descriptor, and the thread pointer 2458 * will be NULL. Callers should be careful only to pass a 2459 * NULL thread pointer when there really is no owning 2460 * context that might have locks, or the locks will be 2461 * leaked. 2462 */ 2463 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2464 vp = fp->f_vnode; 2465 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2466 lf.l_whence = SEEK_SET; 2467 lf.l_start = 0; 2468 lf.l_len = 0; 2469 lf.l_type = F_UNLCK; 2470 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2471 F_UNLCK, &lf, F_POSIX); 2472 } 2473 fdtol = td->td_proc->p_fdtol; 2474 if (fdtol != NULL) { 2475 /* 2476 * Handle special case where file descriptor table is 2477 * shared between multiple process leaders. 2478 */ 2479 fdp = td->td_proc->p_fd; 2480 FILEDESC_XLOCK(fdp); 2481 for (fdtol = fdtol->fdl_next; 2482 fdtol != td->td_proc->p_fdtol; 2483 fdtol = fdtol->fdl_next) { 2484 if ((fdtol->fdl_leader->p_flag & 2485 P_ADVLOCK) == 0) 2486 continue; 2487 fdtol->fdl_holdcount++; 2488 FILEDESC_XUNLOCK(fdp); 2489 lf.l_whence = SEEK_SET; 2490 lf.l_start = 0; 2491 lf.l_len = 0; 2492 lf.l_type = F_UNLCK; 2493 vp = fp->f_vnode; 2494 (void) VOP_ADVLOCK(vp, 2495 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2496 F_POSIX); 2497 FILEDESC_XLOCK(fdp); 2498 fdtol->fdl_holdcount--; 2499 if (fdtol->fdl_holdcount == 0 && 2500 fdtol->fdl_wakeup != 0) { 2501 fdtol->fdl_wakeup = 0; 2502 wakeup(fdtol); 2503 } 2504 } 2505 FILEDESC_XUNLOCK(fdp); 2506 } 2507 } 2508 return (fdrop(fp, td)); 2509 } 2510 2511 /* 2512 * Initialize the file pointer with the specified properties. 2513 * 2514 * The ops are set with release semantics to be certain that the flags, type, 2515 * and data are visible when ops is. This is to prevent ops methods from being 2516 * called with bad data. 2517 */ 2518 void 2519 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2520 { 2521 fp->f_data = data; 2522 fp->f_flag = flag; 2523 fp->f_type = type; 2524 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2525 } 2526 2527 int 2528 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2529 struct file **fpp, struct filecaps *havecapsp) 2530 { 2531 struct filedescent *fde; 2532 int error; 2533 2534 FILEDESC_LOCK_ASSERT(fdp); 2535 2536 fde = fdeget_locked(fdp, fd); 2537 if (fde == NULL) { 2538 error = EBADF; 2539 goto out; 2540 } 2541 2542 #ifdef CAPABILITIES 2543 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 2544 if (error != 0) 2545 goto out; 2546 #endif 2547 2548 if (havecapsp != NULL) 2549 filecaps_copy(&fde->fde_caps, havecapsp, true); 2550 2551 *fpp = fde->fde_file; 2552 2553 error = 0; 2554 out: 2555 return (error); 2556 } 2557 2558 int 2559 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2560 struct file **fpp, struct filecaps *havecapsp) 2561 { 2562 struct filedesc *fdp = td->td_proc->p_fd; 2563 int error; 2564 #ifndef CAPABILITIES 2565 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL); 2566 if (error == 0 && havecapsp != NULL) 2567 filecaps_fill(havecapsp); 2568 #else 2569 struct file *fp; 2570 seq_t seq; 2571 2572 for (;;) { 2573 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq); 2574 if (error != 0) 2575 return (error); 2576 2577 if (havecapsp != NULL) { 2578 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2579 havecapsp, false)) { 2580 fdrop(fp, td); 2581 goto get_locked; 2582 } 2583 } 2584 2585 if (!fd_modified(fdp, fd, seq)) 2586 break; 2587 fdrop(fp, td); 2588 } 2589 2590 *fpp = fp; 2591 return (0); 2592 2593 get_locked: 2594 FILEDESC_SLOCK(fdp); 2595 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); 2596 if (error == 0) 2597 fhold(*fpp); 2598 FILEDESC_SUNLOCK(fdp); 2599 #endif 2600 return (error); 2601 } 2602 2603 int 2604 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2605 struct file **fpp, seq_t *seqp) 2606 { 2607 #ifdef CAPABILITIES 2608 const struct filedescent *fde; 2609 #endif 2610 const struct fdescenttbl *fdt; 2611 struct file *fp; 2612 u_int count; 2613 #ifdef CAPABILITIES 2614 seq_t seq; 2615 cap_rights_t haverights; 2616 int error; 2617 #endif 2618 2619 fdt = fdp->fd_files; 2620 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 2621 return (EBADF); 2622 /* 2623 * Fetch the descriptor locklessly. We avoid fdrop() races by 2624 * never raising a refcount above 0. To accomplish this we have 2625 * to use a cmpset loop rather than an atomic_add. The descriptor 2626 * must be re-verified once we acquire a reference to be certain 2627 * that the identity is still correct and we did not lose a race 2628 * due to preemption. 2629 */ 2630 for (;;) { 2631 #ifdef CAPABILITIES 2632 seq = seq_load(fd_seq(fdt, fd)); 2633 fde = &fdt->fdt_ofiles[fd]; 2634 haverights = *cap_rights_fde_inline(fde); 2635 fp = fde->fde_file; 2636 if (!seq_consistent(fd_seq(fdt, fd), seq)) 2637 continue; 2638 #else 2639 fp = fdt->fdt_ofiles[fd].fde_file; 2640 #endif 2641 if (fp == NULL) 2642 return (EBADF); 2643 #ifdef CAPABILITIES 2644 error = cap_check(&haverights, needrightsp); 2645 if (error != 0) 2646 return (error); 2647 #endif 2648 count = fp->f_count; 2649 retry: 2650 if (count == 0) { 2651 /* 2652 * Force a reload. Other thread could reallocate the 2653 * table before this fd was closed, so it possible that 2654 * there is a stale fp pointer in cached version. 2655 */ 2656 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files); 2657 continue; 2658 } 2659 /* 2660 * Use an acquire barrier to force re-reading of fdt so it is 2661 * refreshed for verification. 2662 */ 2663 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0) 2664 goto retry; 2665 fdt = fdp->fd_files; 2666 #ifdef CAPABILITIES 2667 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2668 #else 2669 if (fp == fdt->fdt_ofiles[fd].fde_file) 2670 #endif 2671 break; 2672 fdrop(fp, curthread); 2673 } 2674 *fpp = fp; 2675 if (seqp != NULL) { 2676 #ifdef CAPABILITIES 2677 *seqp = seq; 2678 #endif 2679 } 2680 return (0); 2681 } 2682 2683 /* 2684 * Extract the file pointer associated with the specified descriptor for the 2685 * current user process. 2686 * 2687 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2688 * returned. 2689 * 2690 * File's rights will be checked against the capability rights mask. 2691 * 2692 * If an error occurred the non-zero error is returned and *fpp is set to 2693 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2694 * responsible for fdrop(). 2695 */ 2696 static __inline int 2697 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2698 cap_rights_t *needrightsp, seq_t *seqp) 2699 { 2700 struct filedesc *fdp; 2701 struct file *fp; 2702 int error; 2703 2704 *fpp = NULL; 2705 fdp = td->td_proc->p_fd; 2706 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2707 if (error != 0) 2708 return (error); 2709 if (fp->f_ops == &badfileops) { 2710 fdrop(fp, td); 2711 return (EBADF); 2712 } 2713 2714 /* 2715 * FREAD and FWRITE failure return EBADF as per POSIX. 2716 */ 2717 error = 0; 2718 switch (flags) { 2719 case FREAD: 2720 case FWRITE: 2721 if ((fp->f_flag & flags) == 0) 2722 error = EBADF; 2723 break; 2724 case FEXEC: 2725 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2726 ((fp->f_flag & FWRITE) != 0)) 2727 error = EBADF; 2728 break; 2729 case 0: 2730 break; 2731 default: 2732 KASSERT(0, ("wrong flags")); 2733 } 2734 2735 if (error != 0) { 2736 fdrop(fp, td); 2737 return (error); 2738 } 2739 2740 *fpp = fp; 2741 return (0); 2742 } 2743 2744 int 2745 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2746 { 2747 2748 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2749 } 2750 2751 int 2752 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2753 struct file **fpp) 2754 { 2755 int error; 2756 #ifndef CAPABILITIES 2757 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2758 if (maxprotp != NULL) 2759 *maxprotp = VM_PROT_ALL; 2760 #else 2761 struct filedesc *fdp = td->td_proc->p_fd; 2762 seq_t seq; 2763 2764 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2765 for (;;) { 2766 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2767 if (error != 0) 2768 return (error); 2769 /* 2770 * If requested, convert capability rights to access flags. 2771 */ 2772 if (maxprotp != NULL) 2773 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2774 if (!fd_modified(fdp, fd, seq)) 2775 break; 2776 fdrop(*fpp, td); 2777 } 2778 #endif 2779 return (error); 2780 } 2781 2782 int 2783 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2784 { 2785 2786 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2787 } 2788 2789 int 2790 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2791 { 2792 2793 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2794 } 2795 2796 int 2797 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2798 struct file **fpp) 2799 { 2800 struct filedesc *fdp = td->td_proc->p_fd; 2801 #ifndef CAPABILITIES 2802 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2803 #else 2804 int error; 2805 seq_t seq; 2806 2807 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2808 for (;;) { 2809 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2810 if (error != 0) 2811 return (error); 2812 error = cap_fcntl_check(fdp, fd, needfcntl); 2813 if (!fd_modified(fdp, fd, seq)) 2814 break; 2815 fdrop(*fpp, td); 2816 } 2817 if (error != 0) { 2818 fdrop(*fpp, td); 2819 *fpp = NULL; 2820 } 2821 return (error); 2822 #endif 2823 } 2824 2825 /* 2826 * Like fget() but loads the underlying vnode, or returns an error if the 2827 * descriptor does not represent a vnode. Note that pipes use vnodes but 2828 * never have VM objects. The returned vnode will be vref()'d. 2829 * 2830 * XXX: what about the unused flags ? 2831 */ 2832 static __inline int 2833 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2834 struct vnode **vpp) 2835 { 2836 struct file *fp; 2837 int error; 2838 2839 *vpp = NULL; 2840 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2841 if (error != 0) 2842 return (error); 2843 if (fp->f_vnode == NULL) { 2844 error = EINVAL; 2845 } else { 2846 *vpp = fp->f_vnode; 2847 vrefact(*vpp); 2848 } 2849 fdrop(fp, td); 2850 2851 return (error); 2852 } 2853 2854 int 2855 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2856 { 2857 2858 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2859 } 2860 2861 int 2862 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2863 struct filecaps *havecaps, struct vnode **vpp) 2864 { 2865 struct filedesc *fdp; 2866 struct filecaps caps; 2867 struct file *fp; 2868 int error; 2869 2870 fdp = td->td_proc->p_fd; 2871 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps); 2872 if (error != 0) 2873 return (error); 2874 if (fp->f_ops == &badfileops) { 2875 error = EBADF; 2876 goto out; 2877 } 2878 if (fp->f_vnode == NULL) { 2879 error = EINVAL; 2880 goto out; 2881 } 2882 2883 *havecaps = caps; 2884 *vpp = fp->f_vnode; 2885 vrefact(*vpp); 2886 2887 return (0); 2888 out: 2889 filecaps_free(&caps); 2890 return (error); 2891 } 2892 2893 int 2894 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2895 { 2896 2897 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2898 } 2899 2900 int 2901 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2902 { 2903 2904 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2905 } 2906 2907 #ifdef notyet 2908 int 2909 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2910 struct vnode **vpp) 2911 { 2912 2913 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2914 } 2915 #endif 2916 2917 /* 2918 * Handle the last reference to a file being closed. 2919 * 2920 * Without the noinline attribute clang keeps inlining the func thorough this 2921 * file when fdrop is used. 2922 */ 2923 int __noinline 2924 _fdrop(struct file *fp, struct thread *td) 2925 { 2926 int error; 2927 2928 if (fp->f_count != 0) 2929 panic("fdrop: count %d", fp->f_count); 2930 error = fo_close(fp, td); 2931 atomic_subtract_int(&openfiles, 1); 2932 crfree(fp->f_cred); 2933 free(fp->f_advice, M_FADVISE); 2934 uma_zfree(file_zone, fp); 2935 2936 return (error); 2937 } 2938 2939 /* 2940 * Apply an advisory lock on a file descriptor. 2941 * 2942 * Just attempt to get a record lock of the requested type on the entire file 2943 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2944 */ 2945 #ifndef _SYS_SYSPROTO_H_ 2946 struct flock_args { 2947 int fd; 2948 int how; 2949 }; 2950 #endif 2951 /* ARGSUSED */ 2952 int 2953 sys_flock(struct thread *td, struct flock_args *uap) 2954 { 2955 struct file *fp; 2956 struct vnode *vp; 2957 struct flock lf; 2958 int error; 2959 2960 error = fget(td, uap->fd, &cap_flock_rights, &fp); 2961 if (error != 0) 2962 return (error); 2963 if (fp->f_type != DTYPE_VNODE) { 2964 fdrop(fp, td); 2965 return (EOPNOTSUPP); 2966 } 2967 2968 vp = fp->f_vnode; 2969 lf.l_whence = SEEK_SET; 2970 lf.l_start = 0; 2971 lf.l_len = 0; 2972 if (uap->how & LOCK_UN) { 2973 lf.l_type = F_UNLCK; 2974 atomic_clear_int(&fp->f_flag, FHASLOCK); 2975 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2976 goto done2; 2977 } 2978 if (uap->how & LOCK_EX) 2979 lf.l_type = F_WRLCK; 2980 else if (uap->how & LOCK_SH) 2981 lf.l_type = F_RDLCK; 2982 else { 2983 error = EBADF; 2984 goto done2; 2985 } 2986 atomic_set_int(&fp->f_flag, FHASLOCK); 2987 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2988 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2989 done2: 2990 fdrop(fp, td); 2991 return (error); 2992 } 2993 /* 2994 * Duplicate the specified descriptor to a free descriptor. 2995 */ 2996 int 2997 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2998 int openerror, int *indxp) 2999 { 3000 struct filedescent *newfde, *oldfde; 3001 struct file *fp; 3002 u_long *ioctls; 3003 int error, indx; 3004 3005 KASSERT(openerror == ENODEV || openerror == ENXIO, 3006 ("unexpected error %d in %s", openerror, __func__)); 3007 3008 /* 3009 * If the to-be-dup'd fd number is greater than the allowed number 3010 * of file descriptors, or the fd to be dup'd has already been 3011 * closed, then reject. 3012 */ 3013 FILEDESC_XLOCK(fdp); 3014 if ((fp = fget_locked(fdp, dfd)) == NULL) { 3015 FILEDESC_XUNLOCK(fdp); 3016 return (EBADF); 3017 } 3018 3019 error = fdalloc(td, 0, &indx); 3020 if (error != 0) { 3021 FILEDESC_XUNLOCK(fdp); 3022 return (error); 3023 } 3024 3025 /* 3026 * There are two cases of interest here. 3027 * 3028 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3029 * 3030 * For ENXIO steal away the file structure from (dfd) and store it in 3031 * (indx). (dfd) is effectively closed by this operation. 3032 */ 3033 switch (openerror) { 3034 case ENODEV: 3035 /* 3036 * Check that the mode the file is being opened for is a 3037 * subset of the mode of the existing descriptor. 3038 */ 3039 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3040 fdunused(fdp, indx); 3041 FILEDESC_XUNLOCK(fdp); 3042 return (EACCES); 3043 } 3044 fhold(fp); 3045 newfde = &fdp->fd_ofiles[indx]; 3046 oldfde = &fdp->fd_ofiles[dfd]; 3047 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3048 #ifdef CAPABILITIES 3049 seq_write_begin(&newfde->fde_seq); 3050 #endif 3051 memcpy(newfde, oldfde, fde_change_size); 3052 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3053 ioctls); 3054 #ifdef CAPABILITIES 3055 seq_write_end(&newfde->fde_seq); 3056 #endif 3057 break; 3058 case ENXIO: 3059 /* 3060 * Steal away the file pointer from dfd and stuff it into indx. 3061 */ 3062 newfde = &fdp->fd_ofiles[indx]; 3063 oldfde = &fdp->fd_ofiles[dfd]; 3064 #ifdef CAPABILITIES 3065 seq_write_begin(&newfde->fde_seq); 3066 #endif 3067 memcpy(newfde, oldfde, fde_change_size); 3068 oldfde->fde_file = NULL; 3069 fdunused(fdp, dfd); 3070 #ifdef CAPABILITIES 3071 seq_write_end(&newfde->fde_seq); 3072 #endif 3073 break; 3074 } 3075 FILEDESC_XUNLOCK(fdp); 3076 *indxp = indx; 3077 return (0); 3078 } 3079 3080 /* 3081 * This sysctl determines if we will allow a process to chroot(2) if it 3082 * has a directory open: 3083 * 0: disallowed for all processes. 3084 * 1: allowed for processes that were not already chroot(2)'ed. 3085 * 2: allowed for all processes. 3086 */ 3087 3088 static int chroot_allow_open_directories = 1; 3089 3090 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3091 &chroot_allow_open_directories, 0, 3092 "Allow a process to chroot(2) if it has a directory open"); 3093 3094 /* 3095 * Helper function for raised chroot(2) security function: Refuse if 3096 * any filedescriptors are open directories. 3097 */ 3098 static int 3099 chroot_refuse_vdir_fds(struct filedesc *fdp) 3100 { 3101 struct vnode *vp; 3102 struct file *fp; 3103 int fd; 3104 3105 FILEDESC_LOCK_ASSERT(fdp); 3106 3107 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 3108 fp = fget_locked(fdp, fd); 3109 if (fp == NULL) 3110 continue; 3111 if (fp->f_type == DTYPE_VNODE) { 3112 vp = fp->f_vnode; 3113 if (vp->v_type == VDIR) 3114 return (EPERM); 3115 } 3116 } 3117 return (0); 3118 } 3119 3120 /* 3121 * Common routine for kern_chroot() and jail_attach(). The caller is 3122 * responsible for invoking priv_check() and mac_vnode_check_chroot() to 3123 * authorize this operation. 3124 */ 3125 int 3126 pwd_chroot(struct thread *td, struct vnode *vp) 3127 { 3128 struct filedesc *fdp; 3129 struct vnode *oldvp; 3130 int error; 3131 3132 fdp = td->td_proc->p_fd; 3133 FILEDESC_XLOCK(fdp); 3134 if (chroot_allow_open_directories == 0 || 3135 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 3136 error = chroot_refuse_vdir_fds(fdp); 3137 if (error != 0) { 3138 FILEDESC_XUNLOCK(fdp); 3139 return (error); 3140 } 3141 } 3142 oldvp = fdp->fd_rdir; 3143 vrefact(vp); 3144 fdp->fd_rdir = vp; 3145 if (fdp->fd_jdir == NULL) { 3146 vrefact(vp); 3147 fdp->fd_jdir = vp; 3148 } 3149 FILEDESC_XUNLOCK(fdp); 3150 vrele(oldvp); 3151 return (0); 3152 } 3153 3154 void 3155 pwd_chdir(struct thread *td, struct vnode *vp) 3156 { 3157 struct filedesc *fdp; 3158 struct vnode *oldvp; 3159 3160 fdp = td->td_proc->p_fd; 3161 FILEDESC_XLOCK(fdp); 3162 VNASSERT(vp->v_usecount > 0, vp, 3163 ("chdir to a vnode with zero usecount")); 3164 oldvp = fdp->fd_cdir; 3165 fdp->fd_cdir = vp; 3166 FILEDESC_XUNLOCK(fdp); 3167 vrele(oldvp); 3168 } 3169 3170 /* 3171 * Scan all active processes and prisons to see if any of them have a current 3172 * or root directory of `olddp'. If so, replace them with the new mount point. 3173 */ 3174 void 3175 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3176 { 3177 struct filedesc *fdp; 3178 struct prison *pr; 3179 struct proc *p; 3180 int nrele; 3181 3182 if (vrefcnt(olddp) == 1) 3183 return; 3184 nrele = 0; 3185 sx_slock(&allproc_lock); 3186 FOREACH_PROC_IN_SYSTEM(p) { 3187 PROC_LOCK(p); 3188 fdp = fdhold(p); 3189 PROC_UNLOCK(p); 3190 if (fdp == NULL) 3191 continue; 3192 FILEDESC_XLOCK(fdp); 3193 if (fdp->fd_cdir == olddp) { 3194 vrefact(newdp); 3195 fdp->fd_cdir = newdp; 3196 nrele++; 3197 } 3198 if (fdp->fd_rdir == olddp) { 3199 vrefact(newdp); 3200 fdp->fd_rdir = newdp; 3201 nrele++; 3202 } 3203 if (fdp->fd_jdir == olddp) { 3204 vrefact(newdp); 3205 fdp->fd_jdir = newdp; 3206 nrele++; 3207 } 3208 FILEDESC_XUNLOCK(fdp); 3209 fddrop(fdp); 3210 } 3211 sx_sunlock(&allproc_lock); 3212 if (rootvnode == olddp) { 3213 vrefact(newdp); 3214 rootvnode = newdp; 3215 nrele++; 3216 } 3217 mtx_lock(&prison0.pr_mtx); 3218 if (prison0.pr_root == olddp) { 3219 vrefact(newdp); 3220 prison0.pr_root = newdp; 3221 nrele++; 3222 } 3223 mtx_unlock(&prison0.pr_mtx); 3224 sx_slock(&allprison_lock); 3225 TAILQ_FOREACH(pr, &allprison, pr_list) { 3226 mtx_lock(&pr->pr_mtx); 3227 if (pr->pr_root == olddp) { 3228 vrefact(newdp); 3229 pr->pr_root = newdp; 3230 nrele++; 3231 } 3232 mtx_unlock(&pr->pr_mtx); 3233 } 3234 sx_sunlock(&allprison_lock); 3235 while (nrele--) 3236 vrele(olddp); 3237 } 3238 3239 struct filedesc_to_leader * 3240 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 3241 { 3242 struct filedesc_to_leader *fdtol; 3243 3244 fdtol = malloc(sizeof(struct filedesc_to_leader), 3245 M_FILEDESC_TO_LEADER, M_WAITOK); 3246 fdtol->fdl_refcount = 1; 3247 fdtol->fdl_holdcount = 0; 3248 fdtol->fdl_wakeup = 0; 3249 fdtol->fdl_leader = leader; 3250 if (old != NULL) { 3251 FILEDESC_XLOCK(fdp); 3252 fdtol->fdl_next = old->fdl_next; 3253 fdtol->fdl_prev = old; 3254 old->fdl_next = fdtol; 3255 fdtol->fdl_next->fdl_prev = fdtol; 3256 FILEDESC_XUNLOCK(fdp); 3257 } else { 3258 fdtol->fdl_next = fdtol; 3259 fdtol->fdl_prev = fdtol; 3260 } 3261 return (fdtol); 3262 } 3263 3264 static int 3265 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 3266 { 3267 struct filedesc *fdp; 3268 int i, count, slots; 3269 3270 if (*(int *)arg1 != 0) 3271 return (EINVAL); 3272 3273 fdp = curproc->p_fd; 3274 count = 0; 3275 FILEDESC_SLOCK(fdp); 3276 slots = NDSLOTS(fdp->fd_lastfile + 1); 3277 for (i = 0; i < slots; i++) 3278 count += bitcountl(fdp->fd_map[i]); 3279 FILEDESC_SUNLOCK(fdp); 3280 3281 return (SYSCTL_OUT(req, &count, sizeof(count))); 3282 } 3283 3284 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 3285 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 3286 "Number of open file descriptors"); 3287 3288 /* 3289 * Get file structures globally. 3290 */ 3291 static int 3292 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 3293 { 3294 struct xfile xf; 3295 struct filedesc *fdp; 3296 struct file *fp; 3297 struct proc *p; 3298 int error, n; 3299 3300 error = sysctl_wire_old_buffer(req, 0); 3301 if (error != 0) 3302 return (error); 3303 if (req->oldptr == NULL) { 3304 n = 0; 3305 sx_slock(&allproc_lock); 3306 FOREACH_PROC_IN_SYSTEM(p) { 3307 PROC_LOCK(p); 3308 if (p->p_state == PRS_NEW) { 3309 PROC_UNLOCK(p); 3310 continue; 3311 } 3312 fdp = fdhold(p); 3313 PROC_UNLOCK(p); 3314 if (fdp == NULL) 3315 continue; 3316 /* overestimates sparse tables. */ 3317 if (fdp->fd_lastfile > 0) 3318 n += fdp->fd_lastfile; 3319 fddrop(fdp); 3320 } 3321 sx_sunlock(&allproc_lock); 3322 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3323 } 3324 error = 0; 3325 bzero(&xf, sizeof(xf)); 3326 xf.xf_size = sizeof(xf); 3327 sx_slock(&allproc_lock); 3328 FOREACH_PROC_IN_SYSTEM(p) { 3329 PROC_LOCK(p); 3330 if (p->p_state == PRS_NEW) { 3331 PROC_UNLOCK(p); 3332 continue; 3333 } 3334 if (p_cansee(req->td, p) != 0) { 3335 PROC_UNLOCK(p); 3336 continue; 3337 } 3338 xf.xf_pid = p->p_pid; 3339 xf.xf_uid = p->p_ucred->cr_uid; 3340 fdp = fdhold(p); 3341 PROC_UNLOCK(p); 3342 if (fdp == NULL) 3343 continue; 3344 FILEDESC_SLOCK(fdp); 3345 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3346 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3347 continue; 3348 xf.xf_fd = n; 3349 xf.xf_file = (uintptr_t)fp; 3350 xf.xf_data = (uintptr_t)fp->f_data; 3351 xf.xf_vnode = (uintptr_t)fp->f_vnode; 3352 xf.xf_type = (uintptr_t)fp->f_type; 3353 xf.xf_count = fp->f_count; 3354 xf.xf_msgcount = 0; 3355 xf.xf_offset = foffset_get(fp); 3356 xf.xf_flag = fp->f_flag; 3357 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3358 if (error) 3359 break; 3360 } 3361 FILEDESC_SUNLOCK(fdp); 3362 fddrop(fdp); 3363 if (error) 3364 break; 3365 } 3366 sx_sunlock(&allproc_lock); 3367 return (error); 3368 } 3369 3370 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3371 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3372 3373 #ifdef KINFO_FILE_SIZE 3374 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3375 #endif 3376 3377 static int 3378 xlate_fflags(int fflags) 3379 { 3380 static const struct { 3381 int fflag; 3382 int kf_fflag; 3383 } fflags_table[] = { 3384 { FAPPEND, KF_FLAG_APPEND }, 3385 { FASYNC, KF_FLAG_ASYNC }, 3386 { FFSYNC, KF_FLAG_FSYNC }, 3387 { FHASLOCK, KF_FLAG_HASLOCK }, 3388 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3389 { FREAD, KF_FLAG_READ }, 3390 { FWRITE, KF_FLAG_WRITE }, 3391 { O_CREAT, KF_FLAG_CREAT }, 3392 { O_DIRECT, KF_FLAG_DIRECT }, 3393 { O_EXCL, KF_FLAG_EXCL }, 3394 { O_EXEC, KF_FLAG_EXEC }, 3395 { O_EXLOCK, KF_FLAG_EXLOCK }, 3396 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3397 { O_SHLOCK, KF_FLAG_SHLOCK }, 3398 { O_TRUNC, KF_FLAG_TRUNC } 3399 }; 3400 unsigned int i; 3401 int kflags; 3402 3403 kflags = 0; 3404 for (i = 0; i < nitems(fflags_table); i++) 3405 if (fflags & fflags_table[i].fflag) 3406 kflags |= fflags_table[i].kf_fflag; 3407 return (kflags); 3408 } 3409 3410 /* Trim unused data from kf_path by truncating the structure size. */ 3411 static void 3412 pack_kinfo(struct kinfo_file *kif) 3413 { 3414 3415 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3416 strlen(kif->kf_path) + 1; 3417 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3418 } 3419 3420 static void 3421 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3422 struct kinfo_file *kif, struct filedesc *fdp, int flags) 3423 { 3424 int error; 3425 3426 bzero(kif, sizeof(*kif)); 3427 3428 /* Set a default type to allow for empty fill_kinfo() methods. */ 3429 kif->kf_type = KF_TYPE_UNKNOWN; 3430 kif->kf_flags = xlate_fflags(fp->f_flag); 3431 if (rightsp != NULL) 3432 kif->kf_cap_rights = *rightsp; 3433 else 3434 cap_rights_init(&kif->kf_cap_rights); 3435 kif->kf_fd = fd; 3436 kif->kf_ref_count = fp->f_count; 3437 kif->kf_offset = foffset_get(fp); 3438 3439 /* 3440 * This may drop the filedesc lock, so the 'fp' cannot be 3441 * accessed after this call. 3442 */ 3443 error = fo_fill_kinfo(fp, kif, fdp); 3444 if (error == 0) 3445 kif->kf_status |= KF_ATTR_VALID; 3446 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3447 pack_kinfo(kif); 3448 else 3449 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3450 } 3451 3452 static void 3453 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3454 struct kinfo_file *kif, int flags) 3455 { 3456 int error; 3457 3458 bzero(kif, sizeof(*kif)); 3459 3460 kif->kf_type = KF_TYPE_VNODE; 3461 error = vn_fill_kinfo_vnode(vp, kif); 3462 if (error == 0) 3463 kif->kf_status |= KF_ATTR_VALID; 3464 kif->kf_flags = xlate_fflags(fflags); 3465 cap_rights_init(&kif->kf_cap_rights); 3466 kif->kf_fd = fd; 3467 kif->kf_ref_count = -1; 3468 kif->kf_offset = -1; 3469 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3470 pack_kinfo(kif); 3471 else 3472 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3473 vrele(vp); 3474 } 3475 3476 struct export_fd_buf { 3477 struct filedesc *fdp; 3478 struct sbuf *sb; 3479 ssize_t remainder; 3480 struct kinfo_file kif; 3481 int flags; 3482 }; 3483 3484 static int 3485 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3486 { 3487 struct kinfo_file *kif; 3488 3489 kif = &efbuf->kif; 3490 if (efbuf->remainder != -1) { 3491 if (efbuf->remainder < kif->kf_structsize) { 3492 /* Terminate export. */ 3493 efbuf->remainder = 0; 3494 return (0); 3495 } 3496 efbuf->remainder -= kif->kf_structsize; 3497 } 3498 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3499 } 3500 3501 static int 3502 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3503 struct export_fd_buf *efbuf) 3504 { 3505 int error; 3506 3507 if (efbuf->remainder == 0) 3508 return (0); 3509 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 3510 efbuf->flags); 3511 FILEDESC_SUNLOCK(efbuf->fdp); 3512 error = export_kinfo_to_sb(efbuf); 3513 FILEDESC_SLOCK(efbuf->fdp); 3514 return (error); 3515 } 3516 3517 static int 3518 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3519 struct export_fd_buf *efbuf) 3520 { 3521 int error; 3522 3523 if (efbuf->remainder == 0) 3524 return (0); 3525 if (efbuf->fdp != NULL) 3526 FILEDESC_SUNLOCK(efbuf->fdp); 3527 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 3528 error = export_kinfo_to_sb(efbuf); 3529 if (efbuf->fdp != NULL) 3530 FILEDESC_SLOCK(efbuf->fdp); 3531 return (error); 3532 } 3533 3534 /* 3535 * Store a process file descriptor information to sbuf. 3536 * 3537 * Takes a locked proc as argument, and returns with the proc unlocked. 3538 */ 3539 int 3540 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 3541 int flags) 3542 { 3543 struct file *fp; 3544 struct filedesc *fdp; 3545 struct export_fd_buf *efbuf; 3546 struct vnode *cttyvp, *textvp, *tracevp; 3547 int error, i; 3548 cap_rights_t rights; 3549 3550 PROC_LOCK_ASSERT(p, MA_OWNED); 3551 3552 /* ktrace vnode */ 3553 tracevp = p->p_tracevp; 3554 if (tracevp != NULL) 3555 vrefact(tracevp); 3556 /* text vnode */ 3557 textvp = p->p_textvp; 3558 if (textvp != NULL) 3559 vrefact(textvp); 3560 /* Controlling tty. */ 3561 cttyvp = NULL; 3562 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3563 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3564 if (cttyvp != NULL) 3565 vrefact(cttyvp); 3566 } 3567 fdp = fdhold(p); 3568 PROC_UNLOCK(p); 3569 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3570 efbuf->fdp = NULL; 3571 efbuf->sb = sb; 3572 efbuf->remainder = maxlen; 3573 efbuf->flags = flags; 3574 if (tracevp != NULL) 3575 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3576 efbuf); 3577 if (textvp != NULL) 3578 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3579 if (cttyvp != NULL) 3580 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3581 efbuf); 3582 error = 0; 3583 if (fdp == NULL) 3584 goto fail; 3585 efbuf->fdp = fdp; 3586 FILEDESC_SLOCK(fdp); 3587 /* working directory */ 3588 if (fdp->fd_cdir != NULL) { 3589 vrefact(fdp->fd_cdir); 3590 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3591 } 3592 /* root directory */ 3593 if (fdp->fd_rdir != NULL) { 3594 vrefact(fdp->fd_rdir); 3595 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3596 } 3597 /* jail directory */ 3598 if (fdp->fd_jdir != NULL) { 3599 vrefact(fdp->fd_jdir); 3600 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3601 } 3602 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3603 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3604 continue; 3605 #ifdef CAPABILITIES 3606 rights = *cap_rights(fdp, i); 3607 #else /* !CAPABILITIES */ 3608 rights = cap_no_rights; 3609 #endif 3610 /* 3611 * Create sysctl entry. It is OK to drop the filedesc 3612 * lock inside of export_file_to_sb() as we will 3613 * re-validate and re-evaluate its properties when the 3614 * loop continues. 3615 */ 3616 error = export_file_to_sb(fp, i, &rights, efbuf); 3617 if (error != 0 || efbuf->remainder == 0) 3618 break; 3619 } 3620 FILEDESC_SUNLOCK(fdp); 3621 fddrop(fdp); 3622 fail: 3623 free(efbuf, M_TEMP); 3624 return (error); 3625 } 3626 3627 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3628 3629 /* 3630 * Get per-process file descriptors for use by procstat(1), et al. 3631 */ 3632 static int 3633 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3634 { 3635 struct sbuf sb; 3636 struct proc *p; 3637 ssize_t maxlen; 3638 int error, error2, *name; 3639 3640 name = (int *)arg1; 3641 3642 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3643 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3644 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3645 if (error != 0) { 3646 sbuf_delete(&sb); 3647 return (error); 3648 } 3649 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3650 error = kern_proc_filedesc_out(p, &sb, maxlen, 3651 KERN_FILEDESC_PACK_KINFO); 3652 error2 = sbuf_finish(&sb); 3653 sbuf_delete(&sb); 3654 return (error != 0 ? error : error2); 3655 } 3656 3657 #ifdef COMPAT_FREEBSD7 3658 #ifdef KINFO_OFILE_SIZE 3659 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3660 #endif 3661 3662 static void 3663 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3664 { 3665 3666 okif->kf_structsize = sizeof(*okif); 3667 okif->kf_type = kif->kf_type; 3668 okif->kf_fd = kif->kf_fd; 3669 okif->kf_ref_count = kif->kf_ref_count; 3670 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3671 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3672 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3673 okif->kf_offset = kif->kf_offset; 3674 if (kif->kf_type == KF_TYPE_VNODE) 3675 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 3676 else 3677 okif->kf_vnode_type = KF_VTYPE_VNON; 3678 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3679 if (kif->kf_type == KF_TYPE_SOCKET) { 3680 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 3681 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 3682 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 3683 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 3684 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 3685 } else { 3686 okif->kf_sa_local.ss_family = AF_UNSPEC; 3687 okif->kf_sa_peer.ss_family = AF_UNSPEC; 3688 } 3689 } 3690 3691 static int 3692 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3693 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3694 { 3695 int error; 3696 3697 vrefact(vp); 3698 FILEDESC_SUNLOCK(fdp); 3699 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 3700 kinfo_to_okinfo(kif, okif); 3701 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3702 FILEDESC_SLOCK(fdp); 3703 return (error); 3704 } 3705 3706 /* 3707 * Get per-process file descriptors for use by procstat(1), et al. 3708 */ 3709 static int 3710 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3711 { 3712 struct kinfo_ofile *okif; 3713 struct kinfo_file *kif; 3714 struct filedesc *fdp; 3715 int error, i, *name; 3716 struct file *fp; 3717 struct proc *p; 3718 3719 name = (int *)arg1; 3720 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3721 if (error != 0) 3722 return (error); 3723 fdp = fdhold(p); 3724 PROC_UNLOCK(p); 3725 if (fdp == NULL) 3726 return (ENOENT); 3727 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3728 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3729 FILEDESC_SLOCK(fdp); 3730 if (fdp->fd_cdir != NULL) 3731 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3732 okif, fdp, req); 3733 if (fdp->fd_rdir != NULL) 3734 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3735 okif, fdp, req); 3736 if (fdp->fd_jdir != NULL) 3737 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3738 okif, fdp, req); 3739 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3740 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3741 continue; 3742 export_file_to_kinfo(fp, i, NULL, kif, fdp, 3743 KERN_FILEDESC_PACK_KINFO); 3744 FILEDESC_SUNLOCK(fdp); 3745 kinfo_to_okinfo(kif, okif); 3746 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3747 FILEDESC_SLOCK(fdp); 3748 if (error) 3749 break; 3750 } 3751 FILEDESC_SUNLOCK(fdp); 3752 fddrop(fdp); 3753 free(kif, M_TEMP); 3754 free(okif, M_TEMP); 3755 return (0); 3756 } 3757 3758 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3759 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3760 "Process ofiledesc entries"); 3761 #endif /* COMPAT_FREEBSD7 */ 3762 3763 int 3764 vntype_to_kinfo(int vtype) 3765 { 3766 struct { 3767 int vtype; 3768 int kf_vtype; 3769 } vtypes_table[] = { 3770 { VBAD, KF_VTYPE_VBAD }, 3771 { VBLK, KF_VTYPE_VBLK }, 3772 { VCHR, KF_VTYPE_VCHR }, 3773 { VDIR, KF_VTYPE_VDIR }, 3774 { VFIFO, KF_VTYPE_VFIFO }, 3775 { VLNK, KF_VTYPE_VLNK }, 3776 { VNON, KF_VTYPE_VNON }, 3777 { VREG, KF_VTYPE_VREG }, 3778 { VSOCK, KF_VTYPE_VSOCK } 3779 }; 3780 unsigned int i; 3781 3782 /* 3783 * Perform vtype translation. 3784 */ 3785 for (i = 0; i < nitems(vtypes_table); i++) 3786 if (vtypes_table[i].vtype == vtype) 3787 return (vtypes_table[i].kf_vtype); 3788 3789 return (KF_VTYPE_UNKNOWN); 3790 } 3791 3792 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3793 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3794 "Process filedesc entries"); 3795 3796 /* 3797 * Store a process current working directory information to sbuf. 3798 * 3799 * Takes a locked proc as argument, and returns with the proc unlocked. 3800 */ 3801 int 3802 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3803 { 3804 struct filedesc *fdp; 3805 struct export_fd_buf *efbuf; 3806 int error; 3807 3808 PROC_LOCK_ASSERT(p, MA_OWNED); 3809 3810 fdp = fdhold(p); 3811 PROC_UNLOCK(p); 3812 if (fdp == NULL) 3813 return (EINVAL); 3814 3815 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3816 efbuf->fdp = fdp; 3817 efbuf->sb = sb; 3818 efbuf->remainder = maxlen; 3819 3820 FILEDESC_SLOCK(fdp); 3821 if (fdp->fd_cdir == NULL) 3822 error = EINVAL; 3823 else { 3824 vrefact(fdp->fd_cdir); 3825 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3826 FREAD, efbuf); 3827 } 3828 FILEDESC_SUNLOCK(fdp); 3829 fddrop(fdp); 3830 free(efbuf, M_TEMP); 3831 return (error); 3832 } 3833 3834 /* 3835 * Get per-process current working directory. 3836 */ 3837 static int 3838 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3839 { 3840 struct sbuf sb; 3841 struct proc *p; 3842 ssize_t maxlen; 3843 int error, error2, *name; 3844 3845 name = (int *)arg1; 3846 3847 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3848 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3849 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3850 if (error != 0) { 3851 sbuf_delete(&sb); 3852 return (error); 3853 } 3854 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3855 error = kern_proc_cwd_out(p, &sb, maxlen); 3856 error2 = sbuf_finish(&sb); 3857 sbuf_delete(&sb); 3858 return (error != 0 ? error : error2); 3859 } 3860 3861 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3862 sysctl_kern_proc_cwd, "Process current working directory"); 3863 3864 #ifdef DDB 3865 /* 3866 * For the purposes of debugging, generate a human-readable string for the 3867 * file type. 3868 */ 3869 static const char * 3870 file_type_to_name(short type) 3871 { 3872 3873 switch (type) { 3874 case 0: 3875 return ("zero"); 3876 case DTYPE_VNODE: 3877 return ("vnode"); 3878 case DTYPE_SOCKET: 3879 return ("socket"); 3880 case DTYPE_PIPE: 3881 return ("pipe"); 3882 case DTYPE_FIFO: 3883 return ("fifo"); 3884 case DTYPE_KQUEUE: 3885 return ("kqueue"); 3886 case DTYPE_CRYPTO: 3887 return ("crypto"); 3888 case DTYPE_MQUEUE: 3889 return ("mqueue"); 3890 case DTYPE_SHM: 3891 return ("shm"); 3892 case DTYPE_SEM: 3893 return ("ksem"); 3894 case DTYPE_PTS: 3895 return ("pts"); 3896 case DTYPE_DEV: 3897 return ("dev"); 3898 case DTYPE_PROCDESC: 3899 return ("proc"); 3900 case DTYPE_LINUXEFD: 3901 return ("levent"); 3902 case DTYPE_LINUXTFD: 3903 return ("ltimer"); 3904 default: 3905 return ("unkn"); 3906 } 3907 } 3908 3909 /* 3910 * For the purposes of debugging, identify a process (if any, perhaps one of 3911 * many) that references the passed file in its file descriptor array. Return 3912 * NULL if none. 3913 */ 3914 static struct proc * 3915 file_to_first_proc(struct file *fp) 3916 { 3917 struct filedesc *fdp; 3918 struct proc *p; 3919 int n; 3920 3921 FOREACH_PROC_IN_SYSTEM(p) { 3922 if (p->p_state == PRS_NEW) 3923 continue; 3924 fdp = p->p_fd; 3925 if (fdp == NULL) 3926 continue; 3927 for (n = 0; n <= fdp->fd_lastfile; n++) { 3928 if (fp == fdp->fd_ofiles[n].fde_file) 3929 return (p); 3930 } 3931 } 3932 return (NULL); 3933 } 3934 3935 static void 3936 db_print_file(struct file *fp, int header) 3937 { 3938 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 3939 struct proc *p; 3940 3941 if (header) 3942 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 3943 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 3944 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 3945 "FCmd"); 3946 p = file_to_first_proc(fp); 3947 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 3948 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 3949 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode, 3950 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3951 3952 #undef XPTRWIDTH 3953 } 3954 3955 DB_SHOW_COMMAND(file, db_show_file) 3956 { 3957 struct file *fp; 3958 3959 if (!have_addr) { 3960 db_printf("usage: show file <addr>\n"); 3961 return; 3962 } 3963 fp = (struct file *)addr; 3964 db_print_file(fp, 1); 3965 } 3966 3967 DB_SHOW_COMMAND(files, db_show_files) 3968 { 3969 struct filedesc *fdp; 3970 struct file *fp; 3971 struct proc *p; 3972 int header; 3973 int n; 3974 3975 header = 1; 3976 FOREACH_PROC_IN_SYSTEM(p) { 3977 if (p->p_state == PRS_NEW) 3978 continue; 3979 if ((fdp = p->p_fd) == NULL) 3980 continue; 3981 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3982 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3983 continue; 3984 db_print_file(fp, header); 3985 header = 0; 3986 } 3987 } 3988 } 3989 #endif 3990 3991 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3992 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3993 3994 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3995 &maxfiles, 0, "Maximum number of files"); 3996 3997 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3998 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3999 4000 /* ARGSUSED*/ 4001 static void 4002 filelistinit(void *dummy) 4003 { 4004 4005 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 4006 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 4007 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 4008 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 4009 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 4010 } 4011 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 4012 4013 /*-------------------------------------------------------------------*/ 4014 4015 static int 4016 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 4017 int flags, struct thread *td) 4018 { 4019 4020 return (EBADF); 4021 } 4022 4023 static int 4024 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4025 struct thread *td) 4026 { 4027 4028 return (EINVAL); 4029 } 4030 4031 static int 4032 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 4033 struct thread *td) 4034 { 4035 4036 return (EBADF); 4037 } 4038 4039 static int 4040 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 4041 struct thread *td) 4042 { 4043 4044 return (0); 4045 } 4046 4047 static int 4048 badfo_kqfilter(struct file *fp, struct knote *kn) 4049 { 4050 4051 return (EBADF); 4052 } 4053 4054 static int 4055 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 4056 struct thread *td) 4057 { 4058 4059 return (EBADF); 4060 } 4061 4062 static int 4063 badfo_close(struct file *fp, struct thread *td) 4064 { 4065 4066 return (0); 4067 } 4068 4069 static int 4070 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4071 struct thread *td) 4072 { 4073 4074 return (EBADF); 4075 } 4076 4077 static int 4078 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4079 struct thread *td) 4080 { 4081 4082 return (EBADF); 4083 } 4084 4085 static int 4086 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4087 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4088 struct thread *td) 4089 { 4090 4091 return (EBADF); 4092 } 4093 4094 static int 4095 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 4096 { 4097 4098 return (0); 4099 } 4100 4101 struct fileops badfileops = { 4102 .fo_read = badfo_readwrite, 4103 .fo_write = badfo_readwrite, 4104 .fo_truncate = badfo_truncate, 4105 .fo_ioctl = badfo_ioctl, 4106 .fo_poll = badfo_poll, 4107 .fo_kqfilter = badfo_kqfilter, 4108 .fo_stat = badfo_stat, 4109 .fo_close = badfo_close, 4110 .fo_chmod = badfo_chmod, 4111 .fo_chown = badfo_chown, 4112 .fo_sendfile = badfo_sendfile, 4113 .fo_fill_kinfo = badfo_fill_kinfo, 4114 }; 4115 4116 int 4117 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 4118 int flags, struct thread *td) 4119 { 4120 4121 return (EOPNOTSUPP); 4122 } 4123 4124 int 4125 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4126 struct thread *td) 4127 { 4128 4129 return (EINVAL); 4130 } 4131 4132 int 4133 invfo_ioctl(struct file *fp, u_long com, void *data, 4134 struct ucred *active_cred, struct thread *td) 4135 { 4136 4137 return (ENOTTY); 4138 } 4139 4140 int 4141 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 4142 struct thread *td) 4143 { 4144 4145 return (poll_no_poll(events)); 4146 } 4147 4148 int 4149 invfo_kqfilter(struct file *fp, struct knote *kn) 4150 { 4151 4152 return (EINVAL); 4153 } 4154 4155 int 4156 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4157 struct thread *td) 4158 { 4159 4160 return (EINVAL); 4161 } 4162 4163 int 4164 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4165 struct thread *td) 4166 { 4167 4168 return (EINVAL); 4169 } 4170 4171 int 4172 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4173 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4174 struct thread *td) 4175 { 4176 4177 return (EINVAL); 4178 } 4179 4180 /*-------------------------------------------------------------------*/ 4181 4182 /* 4183 * File Descriptor pseudo-device driver (/dev/fd/). 4184 * 4185 * Opening minor device N dup()s the file (if any) connected to file 4186 * descriptor N belonging to the calling process. Note that this driver 4187 * consists of only the ``open()'' routine, because all subsequent 4188 * references to this file will be direct to the other driver. 4189 * 4190 * XXX: we could give this one a cloning event handler if necessary. 4191 */ 4192 4193 /* ARGSUSED */ 4194 static int 4195 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4196 { 4197 4198 /* 4199 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4200 * the file descriptor being sought for duplication. The error 4201 * return ensures that the vnode for this device will be released 4202 * by vn_open. Open will detect this special error and take the 4203 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4204 * will simply report the error. 4205 */ 4206 td->td_dupfd = dev2unit(dev); 4207 return (ENODEV); 4208 } 4209 4210 static struct cdevsw fildesc_cdevsw = { 4211 .d_version = D_VERSION, 4212 .d_open = fdopen, 4213 .d_name = "FD", 4214 }; 4215 4216 static void 4217 fildesc_drvinit(void *unused) 4218 { 4219 struct cdev *dev; 4220 4221 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4222 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4223 make_dev_alias(dev, "stdin"); 4224 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4225 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4226 make_dev_alias(dev, "stdout"); 4227 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4228 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4229 make_dev_alias(dev, "stderr"); 4230 } 4231 4232 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4233