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 * Initialize filecaps structure. 1469 */ 1470 void 1471 filecaps_init(struct filecaps *fcaps) 1472 { 1473 1474 bzero(fcaps, sizeof(*fcaps)); 1475 fcaps->fc_nioctls = -1; 1476 } 1477 1478 /* 1479 * Copy filecaps structure allocating memory for ioctls array if needed. 1480 * 1481 * The last parameter indicates whether the fdtable is locked. If it is not and 1482 * ioctls are encountered, copying fails and the caller must lock the table. 1483 * 1484 * Note that if the table was not locked, the caller has to check the relevant 1485 * sequence counter to determine whether the operation was successful. 1486 */ 1487 bool 1488 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1489 { 1490 size_t size; 1491 1492 if (src->fc_ioctls != NULL && !locked) 1493 return (false); 1494 memcpy(dst, src, sizeof(*src)); 1495 if (src->fc_ioctls == NULL) 1496 return (true); 1497 1498 KASSERT(src->fc_nioctls > 0, 1499 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1500 1501 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1502 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1503 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1504 return (true); 1505 } 1506 1507 static u_long * 1508 filecaps_copy_prep(const struct filecaps *src) 1509 { 1510 u_long *ioctls; 1511 size_t size; 1512 1513 if (__predict_true(src->fc_ioctls == NULL)) 1514 return (NULL); 1515 1516 KASSERT(src->fc_nioctls > 0, 1517 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1518 1519 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1520 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1521 return (ioctls); 1522 } 1523 1524 static void 1525 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1526 u_long *ioctls) 1527 { 1528 size_t size; 1529 1530 *dst = *src; 1531 if (__predict_true(src->fc_ioctls == NULL)) { 1532 MPASS(ioctls == NULL); 1533 return; 1534 } 1535 1536 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1537 dst->fc_ioctls = ioctls; 1538 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1539 } 1540 1541 /* 1542 * Move filecaps structure to the new place and clear the old place. 1543 */ 1544 void 1545 filecaps_move(struct filecaps *src, struct filecaps *dst) 1546 { 1547 1548 *dst = *src; 1549 bzero(src, sizeof(*src)); 1550 } 1551 1552 /* 1553 * Fill the given filecaps structure with full rights. 1554 */ 1555 static void 1556 filecaps_fill(struct filecaps *fcaps) 1557 { 1558 1559 CAP_ALL(&fcaps->fc_rights); 1560 fcaps->fc_ioctls = NULL; 1561 fcaps->fc_nioctls = -1; 1562 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1563 } 1564 1565 /* 1566 * Free memory allocated within filecaps structure. 1567 */ 1568 void 1569 filecaps_free(struct filecaps *fcaps) 1570 { 1571 1572 free(fcaps->fc_ioctls, M_FILECAPS); 1573 bzero(fcaps, sizeof(*fcaps)); 1574 } 1575 1576 static u_long * 1577 filecaps_free_prep(struct filecaps *fcaps) 1578 { 1579 u_long *ioctls; 1580 1581 ioctls = fcaps->fc_ioctls; 1582 bzero(fcaps, sizeof(*fcaps)); 1583 return (ioctls); 1584 } 1585 1586 static void 1587 filecaps_free_finish(u_long *ioctls) 1588 { 1589 1590 free(ioctls, M_FILECAPS); 1591 } 1592 1593 /* 1594 * Validate the given filecaps structure. 1595 */ 1596 static void 1597 filecaps_validate(const struct filecaps *fcaps, const char *func) 1598 { 1599 1600 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1601 ("%s: invalid rights", func)); 1602 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1603 ("%s: invalid fcntls", func)); 1604 KASSERT(fcaps->fc_fcntls == 0 || 1605 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1606 ("%s: fcntls without CAP_FCNTL", func)); 1607 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1608 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1609 ("%s: invalid ioctls", func)); 1610 KASSERT(fcaps->fc_nioctls == 0 || 1611 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1612 ("%s: ioctls without CAP_IOCTL", func)); 1613 } 1614 1615 static void 1616 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1617 { 1618 int nfd1; 1619 1620 FILEDESC_XLOCK_ASSERT(fdp); 1621 1622 nfd1 = fdp->fd_nfiles * 2; 1623 if (nfd1 < nfd) 1624 nfd1 = nfd; 1625 fdgrowtable(fdp, nfd1); 1626 } 1627 1628 /* 1629 * Grow the file table to accommodate (at least) nfd descriptors. 1630 */ 1631 static void 1632 fdgrowtable(struct filedesc *fdp, int nfd) 1633 { 1634 struct filedesc0 *fdp0; 1635 struct freetable *ft; 1636 struct fdescenttbl *ntable; 1637 struct fdescenttbl *otable; 1638 int nnfiles, onfiles; 1639 NDSLOTTYPE *nmap, *omap; 1640 1641 /* 1642 * If lastfile is -1 this struct filedesc was just allocated and we are 1643 * growing it to accommodate for the one we are going to copy from. There 1644 * is no need to have a lock on this one as it's not visible to anyone. 1645 */ 1646 if (fdp->fd_lastfile != -1) 1647 FILEDESC_XLOCK_ASSERT(fdp); 1648 1649 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1650 1651 /* save old values */ 1652 onfiles = fdp->fd_nfiles; 1653 otable = fdp->fd_files; 1654 omap = fdp->fd_map; 1655 1656 /* compute the size of the new table */ 1657 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1658 if (nnfiles <= onfiles) 1659 /* the table is already large enough */ 1660 return; 1661 1662 /* 1663 * Allocate a new table. We need enough space for the number of 1664 * entries, file entries themselves and the struct freetable we will use 1665 * when we decommission the table and place it on the freelist. 1666 * We place the struct freetable in the middle so we don't have 1667 * to worry about padding. 1668 */ 1669 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1670 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1671 sizeof(struct freetable), 1672 M_FILEDESC, M_ZERO | M_WAITOK); 1673 /* copy the old data */ 1674 ntable->fdt_nfiles = nnfiles; 1675 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1676 onfiles * sizeof(ntable->fdt_ofiles[0])); 1677 1678 /* 1679 * Allocate a new map only if the old is not large enough. It will 1680 * grow at a slower rate than the table as it can map more 1681 * entries than the table can hold. 1682 */ 1683 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1684 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1685 M_ZERO | M_WAITOK); 1686 /* copy over the old data and update the pointer */ 1687 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1688 fdp->fd_map = nmap; 1689 } 1690 1691 /* 1692 * Make sure that ntable is correctly initialized before we replace 1693 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1694 * data. 1695 */ 1696 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1697 1698 /* 1699 * Do not free the old file table, as some threads may still 1700 * reference entries within it. Instead, place it on a freelist 1701 * which will be processed when the struct filedesc is released. 1702 * 1703 * Note that if onfiles == NDFILE, we're dealing with the original 1704 * static allocation contained within (struct filedesc0 *)fdp, 1705 * which must not be freed. 1706 */ 1707 if (onfiles > NDFILE) { 1708 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1709 fdp0 = (struct filedesc0 *)fdp; 1710 ft->ft_table = otable; 1711 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1712 } 1713 /* 1714 * The map does not have the same possibility of threads still 1715 * holding references to it. So always free it as long as it 1716 * does not reference the original static allocation. 1717 */ 1718 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1719 free(omap, M_FILEDESC); 1720 } 1721 1722 /* 1723 * Allocate a file descriptor for the process. 1724 */ 1725 int 1726 fdalloc(struct thread *td, int minfd, int *result) 1727 { 1728 struct proc *p = td->td_proc; 1729 struct filedesc *fdp = p->p_fd; 1730 int fd, maxfd, allocfd; 1731 #ifdef RACCT 1732 int error; 1733 #endif 1734 1735 FILEDESC_XLOCK_ASSERT(fdp); 1736 1737 if (fdp->fd_freefile > minfd) 1738 minfd = fdp->fd_freefile; 1739 1740 maxfd = getmaxfd(td); 1741 1742 /* 1743 * Search the bitmap for a free descriptor starting at minfd. 1744 * If none is found, grow the file table. 1745 */ 1746 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1747 if (fd >= maxfd) 1748 return (EMFILE); 1749 if (fd >= fdp->fd_nfiles) { 1750 allocfd = min(fd * 2, maxfd); 1751 #ifdef RACCT 1752 if (RACCT_ENABLED()) { 1753 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 1754 if (error != 0) 1755 return (EMFILE); 1756 } 1757 #endif 1758 /* 1759 * fd is already equal to first free descriptor >= minfd, so 1760 * we only need to grow the table and we are done. 1761 */ 1762 fdgrowtable_exp(fdp, allocfd); 1763 } 1764 1765 /* 1766 * Perform some sanity checks, then mark the file descriptor as 1767 * used and return it to the caller. 1768 */ 1769 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1770 ("invalid descriptor %d", fd)); 1771 KASSERT(!fdisused(fdp, fd), 1772 ("fd_first_free() returned non-free descriptor")); 1773 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1774 ("file descriptor isn't free")); 1775 fdused(fdp, fd); 1776 *result = fd; 1777 return (0); 1778 } 1779 1780 /* 1781 * Allocate n file descriptors for the process. 1782 */ 1783 int 1784 fdallocn(struct thread *td, int minfd, int *fds, int n) 1785 { 1786 struct proc *p = td->td_proc; 1787 struct filedesc *fdp = p->p_fd; 1788 int i; 1789 1790 FILEDESC_XLOCK_ASSERT(fdp); 1791 1792 for (i = 0; i < n; i++) 1793 if (fdalloc(td, 0, &fds[i]) != 0) 1794 break; 1795 1796 if (i < n) { 1797 for (i--; i >= 0; i--) 1798 fdunused(fdp, fds[i]); 1799 return (EMFILE); 1800 } 1801 1802 return (0); 1803 } 1804 1805 /* 1806 * Create a new open file structure and allocate a file descriptor for the 1807 * process that refers to it. We add one reference to the file for the 1808 * descriptor table and one reference for resultfp. This is to prevent us 1809 * being preempted and the entry in the descriptor table closed after we 1810 * release the FILEDESC lock. 1811 */ 1812 int 1813 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1814 struct filecaps *fcaps) 1815 { 1816 struct file *fp; 1817 int error, fd; 1818 1819 error = falloc_noinstall(td, &fp); 1820 if (error) 1821 return (error); /* no reference held on error */ 1822 1823 error = finstall(td, fp, &fd, flags, fcaps); 1824 if (error) { 1825 fdrop(fp, td); /* one reference (fp only) */ 1826 return (error); 1827 } 1828 1829 if (resultfp != NULL) 1830 *resultfp = fp; /* copy out result */ 1831 else 1832 fdrop(fp, td); /* release local reference */ 1833 1834 if (resultfd != NULL) 1835 *resultfd = fd; 1836 1837 return (0); 1838 } 1839 1840 /* 1841 * Create a new open file structure without allocating a file descriptor. 1842 */ 1843 int 1844 falloc_noinstall(struct thread *td, struct file **resultfp) 1845 { 1846 struct file *fp; 1847 int maxuserfiles = maxfiles - (maxfiles / 20); 1848 int openfiles_new; 1849 static struct timeval lastfail; 1850 static int curfail; 1851 1852 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1853 1854 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 1855 if ((openfiles_new >= maxuserfiles && 1856 priv_check(td, PRIV_MAXFILES) != 0) || 1857 openfiles_new >= maxfiles) { 1858 atomic_subtract_int(&openfiles, 1); 1859 if (ppsratecheck(&lastfail, &curfail, 1)) { 1860 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 1861 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 1862 } 1863 return (ENFILE); 1864 } 1865 fp = uma_zalloc(file_zone, M_WAITOK); 1866 bzero(fp, sizeof(*fp)); 1867 refcount_init(&fp->f_count, 1); 1868 fp->f_cred = crhold(td->td_ucred); 1869 fp->f_ops = &badfileops; 1870 *resultfp = fp; 1871 return (0); 1872 } 1873 1874 /* 1875 * Install a file in a file descriptor table. 1876 */ 1877 void 1878 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1879 struct filecaps *fcaps) 1880 { 1881 struct filedescent *fde; 1882 1883 MPASS(fp != NULL); 1884 if (fcaps != NULL) 1885 filecaps_validate(fcaps, __func__); 1886 FILEDESC_XLOCK_ASSERT(fdp); 1887 1888 fde = &fdp->fd_ofiles[fd]; 1889 #ifdef CAPABILITIES 1890 seq_write_begin(&fde->fde_seq); 1891 #endif 1892 fde->fde_file = fp; 1893 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1894 if (fcaps != NULL) 1895 filecaps_move(fcaps, &fde->fde_caps); 1896 else 1897 filecaps_fill(&fde->fde_caps); 1898 #ifdef CAPABILITIES 1899 seq_write_end(&fde->fde_seq); 1900 #endif 1901 } 1902 1903 int 1904 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1905 struct filecaps *fcaps) 1906 { 1907 struct filedesc *fdp = td->td_proc->p_fd; 1908 int error; 1909 1910 MPASS(fd != NULL); 1911 1912 FILEDESC_XLOCK(fdp); 1913 if ((error = fdalloc(td, 0, fd))) { 1914 FILEDESC_XUNLOCK(fdp); 1915 return (error); 1916 } 1917 fhold(fp); 1918 _finstall(fdp, fp, *fd, flags, fcaps); 1919 FILEDESC_XUNLOCK(fdp); 1920 return (0); 1921 } 1922 1923 /* 1924 * Build a new filedesc structure from another. 1925 * Copy the current, root, and jail root vnode references. 1926 * 1927 * If fdp is not NULL, return with it shared locked. 1928 */ 1929 struct filedesc * 1930 fdinit(struct filedesc *fdp, bool prepfiles) 1931 { 1932 struct filedesc0 *newfdp0; 1933 struct filedesc *newfdp; 1934 1935 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1936 newfdp = &newfdp0->fd_fd; 1937 1938 /* Create the file descriptor table. */ 1939 FILEDESC_LOCK_INIT(newfdp); 1940 refcount_init(&newfdp->fd_refcnt, 1); 1941 refcount_init(&newfdp->fd_holdcnt, 1); 1942 newfdp->fd_cmask = CMASK; 1943 newfdp->fd_map = newfdp0->fd_dmap; 1944 newfdp->fd_lastfile = -1; 1945 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1946 newfdp->fd_files->fdt_nfiles = NDFILE; 1947 1948 if (fdp == NULL) 1949 return (newfdp); 1950 1951 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1952 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1953 1954 FILEDESC_SLOCK(fdp); 1955 newfdp->fd_cdir = fdp->fd_cdir; 1956 if (newfdp->fd_cdir) 1957 vrefact(newfdp->fd_cdir); 1958 newfdp->fd_rdir = fdp->fd_rdir; 1959 if (newfdp->fd_rdir) 1960 vrefact(newfdp->fd_rdir); 1961 newfdp->fd_jdir = fdp->fd_jdir; 1962 if (newfdp->fd_jdir) 1963 vrefact(newfdp->fd_jdir); 1964 1965 if (!prepfiles) { 1966 FILEDESC_SUNLOCK(fdp); 1967 } else { 1968 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1969 FILEDESC_SUNLOCK(fdp); 1970 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1971 FILEDESC_SLOCK(fdp); 1972 } 1973 } 1974 1975 return (newfdp); 1976 } 1977 1978 static struct filedesc * 1979 fdhold(struct proc *p) 1980 { 1981 struct filedesc *fdp; 1982 1983 PROC_LOCK_ASSERT(p, MA_OWNED); 1984 fdp = p->p_fd; 1985 if (fdp != NULL) 1986 refcount_acquire(&fdp->fd_holdcnt); 1987 return (fdp); 1988 } 1989 1990 static void 1991 fddrop(struct filedesc *fdp) 1992 { 1993 1994 if (fdp->fd_holdcnt > 1) { 1995 if (refcount_release(&fdp->fd_holdcnt) == 0) 1996 return; 1997 } 1998 1999 FILEDESC_LOCK_DESTROY(fdp); 2000 uma_zfree(filedesc0_zone, fdp); 2001 } 2002 2003 /* 2004 * Share a filedesc structure. 2005 */ 2006 struct filedesc * 2007 fdshare(struct filedesc *fdp) 2008 { 2009 2010 refcount_acquire(&fdp->fd_refcnt); 2011 return (fdp); 2012 } 2013 2014 /* 2015 * Unshare a filedesc structure, if necessary by making a copy 2016 */ 2017 void 2018 fdunshare(struct thread *td) 2019 { 2020 struct filedesc *tmp; 2021 struct proc *p = td->td_proc; 2022 2023 if (p->p_fd->fd_refcnt == 1) 2024 return; 2025 2026 tmp = fdcopy(p->p_fd); 2027 fdescfree(td); 2028 p->p_fd = tmp; 2029 } 2030 2031 void 2032 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 2033 { 2034 2035 fdescfree(td); 2036 td->td_proc->p_fd = fdp; 2037 } 2038 2039 /* 2040 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2041 * this is to ease callers, not catch errors. 2042 */ 2043 struct filedesc * 2044 fdcopy(struct filedesc *fdp) 2045 { 2046 struct filedesc *newfdp; 2047 struct filedescent *nfde, *ofde; 2048 int i; 2049 2050 MPASS(fdp != NULL); 2051 2052 newfdp = fdinit(fdp, true); 2053 /* copy all passable descriptors (i.e. not kqueue) */ 2054 newfdp->fd_freefile = -1; 2055 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2056 ofde = &fdp->fd_ofiles[i]; 2057 if (ofde->fde_file == NULL || 2058 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2059 if (newfdp->fd_freefile == -1) 2060 newfdp->fd_freefile = i; 2061 continue; 2062 } 2063 nfde = &newfdp->fd_ofiles[i]; 2064 *nfde = *ofde; 2065 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2066 fhold(nfde->fde_file); 2067 fdused_init(newfdp, i); 2068 newfdp->fd_lastfile = i; 2069 } 2070 if (newfdp->fd_freefile == -1) 2071 newfdp->fd_freefile = i; 2072 newfdp->fd_cmask = fdp->fd_cmask; 2073 FILEDESC_SUNLOCK(fdp); 2074 return (newfdp); 2075 } 2076 2077 /* 2078 * Copies a filedesc structure, while remapping all file descriptors 2079 * stored inside using a translation table. 2080 * 2081 * File descriptors are copied over to the new file descriptor table, 2082 * regardless of whether the close-on-exec flag is set. 2083 */ 2084 int 2085 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 2086 struct filedesc **ret) 2087 { 2088 struct filedesc *newfdp; 2089 struct filedescent *nfde, *ofde; 2090 int error, i; 2091 2092 MPASS(fdp != NULL); 2093 2094 newfdp = fdinit(fdp, true); 2095 if (nfds > fdp->fd_lastfile + 1) { 2096 /* New table cannot be larger than the old one. */ 2097 error = E2BIG; 2098 goto bad; 2099 } 2100 /* Copy all passable descriptors (i.e. not kqueue). */ 2101 newfdp->fd_freefile = nfds; 2102 for (i = 0; i < nfds; ++i) { 2103 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { 2104 /* File descriptor out of bounds. */ 2105 error = EBADF; 2106 goto bad; 2107 } 2108 ofde = &fdp->fd_ofiles[fds[i]]; 2109 if (ofde->fde_file == NULL) { 2110 /* Unused file descriptor. */ 2111 error = EBADF; 2112 goto bad; 2113 } 2114 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2115 /* File descriptor cannot be passed. */ 2116 error = EINVAL; 2117 goto bad; 2118 } 2119 nfde = &newfdp->fd_ofiles[i]; 2120 *nfde = *ofde; 2121 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2122 fhold(nfde->fde_file); 2123 fdused_init(newfdp, i); 2124 newfdp->fd_lastfile = i; 2125 } 2126 newfdp->fd_cmask = fdp->fd_cmask; 2127 FILEDESC_SUNLOCK(fdp); 2128 *ret = newfdp; 2129 return (0); 2130 bad: 2131 FILEDESC_SUNLOCK(fdp); 2132 fdescfree_remapped(newfdp); 2133 return (error); 2134 } 2135 2136 /* 2137 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2138 * one of processes using it exits) and the table used to be shared. 2139 */ 2140 static void 2141 fdclearlocks(struct thread *td) 2142 { 2143 struct filedesc *fdp; 2144 struct filedesc_to_leader *fdtol; 2145 struct flock lf; 2146 struct file *fp; 2147 struct proc *p; 2148 struct vnode *vp; 2149 int i; 2150 2151 p = td->td_proc; 2152 fdp = p->p_fd; 2153 fdtol = p->p_fdtol; 2154 MPASS(fdtol != NULL); 2155 2156 FILEDESC_XLOCK(fdp); 2157 KASSERT(fdtol->fdl_refcount > 0, 2158 ("filedesc_to_refcount botch: fdl_refcount=%d", 2159 fdtol->fdl_refcount)); 2160 if (fdtol->fdl_refcount == 1 && 2161 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2162 for (i = 0; i <= fdp->fd_lastfile; i++) { 2163 fp = fdp->fd_ofiles[i].fde_file; 2164 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2165 continue; 2166 fhold(fp); 2167 FILEDESC_XUNLOCK(fdp); 2168 lf.l_whence = SEEK_SET; 2169 lf.l_start = 0; 2170 lf.l_len = 0; 2171 lf.l_type = F_UNLCK; 2172 vp = fp->f_vnode; 2173 (void) VOP_ADVLOCK(vp, 2174 (caddr_t)p->p_leader, F_UNLCK, 2175 &lf, F_POSIX); 2176 FILEDESC_XLOCK(fdp); 2177 fdrop(fp, td); 2178 } 2179 } 2180 retry: 2181 if (fdtol->fdl_refcount == 1) { 2182 if (fdp->fd_holdleaderscount > 0 && 2183 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2184 /* 2185 * close() or kern_dup() has cleared a reference 2186 * in a shared file descriptor table. 2187 */ 2188 fdp->fd_holdleaderswakeup = 1; 2189 sx_sleep(&fdp->fd_holdleaderscount, 2190 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2191 goto retry; 2192 } 2193 if (fdtol->fdl_holdcount > 0) { 2194 /* 2195 * Ensure that fdtol->fdl_leader remains 2196 * valid in closef(). 2197 */ 2198 fdtol->fdl_wakeup = 1; 2199 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2200 "fdlhold", 0); 2201 goto retry; 2202 } 2203 } 2204 fdtol->fdl_refcount--; 2205 if (fdtol->fdl_refcount == 0 && 2206 fdtol->fdl_holdcount == 0) { 2207 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2208 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2209 } else 2210 fdtol = NULL; 2211 p->p_fdtol = NULL; 2212 FILEDESC_XUNLOCK(fdp); 2213 if (fdtol != NULL) 2214 free(fdtol, M_FILEDESC_TO_LEADER); 2215 } 2216 2217 /* 2218 * Release a filedesc structure. 2219 */ 2220 static void 2221 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2222 { 2223 struct filedesc0 *fdp0; 2224 struct freetable *ft, *tft; 2225 struct filedescent *fde; 2226 struct file *fp; 2227 int i; 2228 2229 for (i = 0; i <= fdp->fd_lastfile; i++) { 2230 fde = &fdp->fd_ofiles[i]; 2231 fp = fde->fde_file; 2232 if (fp != NULL) { 2233 fdefree_last(fde); 2234 if (needclose) 2235 (void) closef(fp, td); 2236 else 2237 fdrop(fp, td); 2238 } 2239 } 2240 2241 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2242 free(fdp->fd_map, M_FILEDESC); 2243 if (fdp->fd_nfiles > NDFILE) 2244 free(fdp->fd_files, M_FILEDESC); 2245 2246 fdp0 = (struct filedesc0 *)fdp; 2247 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2248 free(ft->ft_table, M_FILEDESC); 2249 2250 fddrop(fdp); 2251 } 2252 2253 void 2254 fdescfree(struct thread *td) 2255 { 2256 struct proc *p; 2257 struct filedesc *fdp; 2258 struct vnode *cdir, *jdir, *rdir; 2259 2260 p = td->td_proc; 2261 fdp = p->p_fd; 2262 MPASS(fdp != NULL); 2263 2264 #ifdef RACCT 2265 if (RACCT_ENABLED()) 2266 racct_set_unlocked(p, RACCT_NOFILE, 0); 2267 #endif 2268 2269 if (p->p_fdtol != NULL) 2270 fdclearlocks(td); 2271 2272 PROC_LOCK(p); 2273 p->p_fd = NULL; 2274 PROC_UNLOCK(p); 2275 2276 if (refcount_release(&fdp->fd_refcnt) == 0) 2277 return; 2278 2279 FILEDESC_XLOCK(fdp); 2280 cdir = fdp->fd_cdir; 2281 fdp->fd_cdir = NULL; 2282 rdir = fdp->fd_rdir; 2283 fdp->fd_rdir = NULL; 2284 jdir = fdp->fd_jdir; 2285 fdp->fd_jdir = NULL; 2286 FILEDESC_XUNLOCK(fdp); 2287 2288 if (cdir != NULL) 2289 vrele(cdir); 2290 if (rdir != NULL) 2291 vrele(rdir); 2292 if (jdir != NULL) 2293 vrele(jdir); 2294 2295 fdescfree_fds(td, fdp, 1); 2296 } 2297 2298 void 2299 fdescfree_remapped(struct filedesc *fdp) 2300 { 2301 2302 if (fdp->fd_cdir != NULL) 2303 vrele(fdp->fd_cdir); 2304 if (fdp->fd_rdir != NULL) 2305 vrele(fdp->fd_rdir); 2306 if (fdp->fd_jdir != NULL) 2307 vrele(fdp->fd_jdir); 2308 2309 fdescfree_fds(curthread, fdp, 0); 2310 } 2311 2312 /* 2313 * For setugid programs, we don't want to people to use that setugidness 2314 * to generate error messages which write to a file which otherwise would 2315 * otherwise be off-limits to the process. We check for filesystems where 2316 * the vnode can change out from under us after execve (like [lin]procfs). 2317 * 2318 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2319 * sufficient. We also don't check for setugidness since we know we are. 2320 */ 2321 static bool 2322 is_unsafe(struct file *fp) 2323 { 2324 struct vnode *vp; 2325 2326 if (fp->f_type != DTYPE_VNODE) 2327 return (false); 2328 2329 vp = fp->f_vnode; 2330 return ((vp->v_vflag & VV_PROCDEP) != 0); 2331 } 2332 2333 /* 2334 * Make this setguid thing safe, if at all possible. 2335 */ 2336 void 2337 fdsetugidsafety(struct thread *td) 2338 { 2339 struct filedesc *fdp; 2340 struct file *fp; 2341 int i; 2342 2343 fdp = td->td_proc->p_fd; 2344 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2345 MPASS(fdp->fd_nfiles >= 3); 2346 for (i = 0; i <= 2; i++) { 2347 fp = fdp->fd_ofiles[i].fde_file; 2348 if (fp != NULL && is_unsafe(fp)) { 2349 FILEDESC_XLOCK(fdp); 2350 knote_fdclose(td, i); 2351 /* 2352 * NULL-out descriptor prior to close to avoid 2353 * a race while close blocks. 2354 */ 2355 fdfree(fdp, i); 2356 FILEDESC_XUNLOCK(fdp); 2357 (void) closef(fp, td); 2358 } 2359 } 2360 } 2361 2362 /* 2363 * If a specific file object occupies a specific file descriptor, close the 2364 * file descriptor entry and drop a reference on the file object. This is a 2365 * convenience function to handle a subsequent error in a function that calls 2366 * falloc() that handles the race that another thread might have closed the 2367 * file descriptor out from under the thread creating the file object. 2368 */ 2369 void 2370 fdclose(struct thread *td, struct file *fp, int idx) 2371 { 2372 struct filedesc *fdp = td->td_proc->p_fd; 2373 2374 FILEDESC_XLOCK(fdp); 2375 if (fdp->fd_ofiles[idx].fde_file == fp) { 2376 fdfree(fdp, idx); 2377 FILEDESC_XUNLOCK(fdp); 2378 fdrop(fp, td); 2379 } else 2380 FILEDESC_XUNLOCK(fdp); 2381 } 2382 2383 /* 2384 * Close any files on exec? 2385 */ 2386 void 2387 fdcloseexec(struct thread *td) 2388 { 2389 struct filedesc *fdp; 2390 struct filedescent *fde; 2391 struct file *fp; 2392 int i; 2393 2394 fdp = td->td_proc->p_fd; 2395 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2396 for (i = 0; i <= fdp->fd_lastfile; i++) { 2397 fde = &fdp->fd_ofiles[i]; 2398 fp = fde->fde_file; 2399 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2400 (fde->fde_flags & UF_EXCLOSE))) { 2401 FILEDESC_XLOCK(fdp); 2402 fdfree(fdp, i); 2403 (void) closefp(fdp, i, fp, td, 0); 2404 FILEDESC_UNLOCK_ASSERT(fdp); 2405 } 2406 } 2407 } 2408 2409 /* 2410 * It is unsafe for set[ug]id processes to be started with file 2411 * descriptors 0..2 closed, as these descriptors are given implicit 2412 * significance in the Standard C library. fdcheckstd() will create a 2413 * descriptor referencing /dev/null for each of stdin, stdout, and 2414 * stderr that is not already open. 2415 */ 2416 int 2417 fdcheckstd(struct thread *td) 2418 { 2419 struct filedesc *fdp; 2420 register_t save; 2421 int i, error, devnull; 2422 2423 fdp = td->td_proc->p_fd; 2424 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2425 MPASS(fdp->fd_nfiles >= 3); 2426 devnull = -1; 2427 for (i = 0; i <= 2; i++) { 2428 if (fdp->fd_ofiles[i].fde_file != NULL) 2429 continue; 2430 2431 save = td->td_retval[0]; 2432 if (devnull != -1) { 2433 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2434 } else { 2435 error = kern_openat(td, AT_FDCWD, "/dev/null", 2436 UIO_SYSSPACE, O_RDWR, 0); 2437 if (error == 0) { 2438 devnull = td->td_retval[0]; 2439 KASSERT(devnull == i, ("we didn't get our fd")); 2440 } 2441 } 2442 td->td_retval[0] = save; 2443 if (error != 0) 2444 return (error); 2445 } 2446 return (0); 2447 } 2448 2449 /* 2450 * Internal form of close. Decrement reference count on file structure. 2451 * Note: td may be NULL when closing a file that was being passed in a 2452 * message. 2453 * 2454 * XXXRW: Giant is not required for the caller, but often will be held; this 2455 * makes it moderately likely the Giant will be recursed in the VFS case. 2456 */ 2457 int 2458 closef(struct file *fp, struct thread *td) 2459 { 2460 struct vnode *vp; 2461 struct flock lf; 2462 struct filedesc_to_leader *fdtol; 2463 struct filedesc *fdp; 2464 2465 /* 2466 * POSIX record locking dictates that any close releases ALL 2467 * locks owned by this process. This is handled by setting 2468 * a flag in the unlock to free ONLY locks obeying POSIX 2469 * semantics, and not to free BSD-style file locks. 2470 * If the descriptor was in a message, POSIX-style locks 2471 * aren't passed with the descriptor, and the thread pointer 2472 * will be NULL. Callers should be careful only to pass a 2473 * NULL thread pointer when there really is no owning 2474 * context that might have locks, or the locks will be 2475 * leaked. 2476 */ 2477 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2478 vp = fp->f_vnode; 2479 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2480 lf.l_whence = SEEK_SET; 2481 lf.l_start = 0; 2482 lf.l_len = 0; 2483 lf.l_type = F_UNLCK; 2484 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2485 F_UNLCK, &lf, F_POSIX); 2486 } 2487 fdtol = td->td_proc->p_fdtol; 2488 if (fdtol != NULL) { 2489 /* 2490 * Handle special case where file descriptor table is 2491 * shared between multiple process leaders. 2492 */ 2493 fdp = td->td_proc->p_fd; 2494 FILEDESC_XLOCK(fdp); 2495 for (fdtol = fdtol->fdl_next; 2496 fdtol != td->td_proc->p_fdtol; 2497 fdtol = fdtol->fdl_next) { 2498 if ((fdtol->fdl_leader->p_flag & 2499 P_ADVLOCK) == 0) 2500 continue; 2501 fdtol->fdl_holdcount++; 2502 FILEDESC_XUNLOCK(fdp); 2503 lf.l_whence = SEEK_SET; 2504 lf.l_start = 0; 2505 lf.l_len = 0; 2506 lf.l_type = F_UNLCK; 2507 vp = fp->f_vnode; 2508 (void) VOP_ADVLOCK(vp, 2509 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2510 F_POSIX); 2511 FILEDESC_XLOCK(fdp); 2512 fdtol->fdl_holdcount--; 2513 if (fdtol->fdl_holdcount == 0 && 2514 fdtol->fdl_wakeup != 0) { 2515 fdtol->fdl_wakeup = 0; 2516 wakeup(fdtol); 2517 } 2518 } 2519 FILEDESC_XUNLOCK(fdp); 2520 } 2521 } 2522 return (fdrop(fp, td)); 2523 } 2524 2525 /* 2526 * Initialize the file pointer with the specified properties. 2527 * 2528 * The ops are set with release semantics to be certain that the flags, type, 2529 * and data are visible when ops is. This is to prevent ops methods from being 2530 * called with bad data. 2531 */ 2532 void 2533 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2534 { 2535 fp->f_data = data; 2536 fp->f_flag = flag; 2537 fp->f_type = type; 2538 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2539 } 2540 2541 int 2542 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2543 struct file **fpp, struct filecaps *havecapsp) 2544 { 2545 struct filedescent *fde; 2546 int error; 2547 2548 FILEDESC_LOCK_ASSERT(fdp); 2549 2550 fde = fdeget_locked(fdp, fd); 2551 if (fde == NULL) { 2552 error = EBADF; 2553 goto out; 2554 } 2555 2556 #ifdef CAPABILITIES 2557 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 2558 if (error != 0) 2559 goto out; 2560 #endif 2561 2562 if (havecapsp != NULL) 2563 filecaps_copy(&fde->fde_caps, havecapsp, true); 2564 2565 *fpp = fde->fde_file; 2566 2567 error = 0; 2568 out: 2569 return (error); 2570 } 2571 2572 int 2573 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2574 struct file **fpp, struct filecaps *havecapsp) 2575 { 2576 struct filedesc *fdp = td->td_proc->p_fd; 2577 int error; 2578 #ifndef CAPABILITIES 2579 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL); 2580 if (error == 0 && havecapsp != NULL) 2581 filecaps_fill(havecapsp); 2582 #else 2583 struct file *fp; 2584 seq_t seq; 2585 2586 for (;;) { 2587 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq); 2588 if (error != 0) 2589 return (error); 2590 2591 if (havecapsp != NULL) { 2592 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2593 havecapsp, false)) { 2594 fdrop(fp, td); 2595 goto get_locked; 2596 } 2597 } 2598 2599 if (!fd_modified(fdp, fd, seq)) 2600 break; 2601 fdrop(fp, td); 2602 } 2603 2604 *fpp = fp; 2605 return (0); 2606 2607 get_locked: 2608 FILEDESC_SLOCK(fdp); 2609 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); 2610 if (error == 0) 2611 fhold(*fpp); 2612 FILEDESC_SUNLOCK(fdp); 2613 #endif 2614 return (error); 2615 } 2616 2617 int 2618 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2619 struct file **fpp, seq_t *seqp) 2620 { 2621 #ifdef CAPABILITIES 2622 const struct filedescent *fde; 2623 #endif 2624 const struct fdescenttbl *fdt; 2625 struct file *fp; 2626 u_int count; 2627 #ifdef CAPABILITIES 2628 seq_t seq; 2629 cap_rights_t haverights; 2630 int error; 2631 #endif 2632 2633 fdt = fdp->fd_files; 2634 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 2635 return (EBADF); 2636 /* 2637 * Fetch the descriptor locklessly. We avoid fdrop() races by 2638 * never raising a refcount above 0. To accomplish this we have 2639 * to use a cmpset loop rather than an atomic_add. The descriptor 2640 * must be re-verified once we acquire a reference to be certain 2641 * that the identity is still correct and we did not lose a race 2642 * due to preemption. 2643 */ 2644 for (;;) { 2645 #ifdef CAPABILITIES 2646 seq = seq_read(fd_seq(fdt, fd)); 2647 fde = &fdt->fdt_ofiles[fd]; 2648 haverights = *cap_rights_fde_inline(fde); 2649 fp = fde->fde_file; 2650 if (!seq_consistent(fd_seq(fdt, fd), seq)) 2651 continue; 2652 #else 2653 fp = fdt->fdt_ofiles[fd].fde_file; 2654 #endif 2655 if (fp == NULL) 2656 return (EBADF); 2657 #ifdef CAPABILITIES 2658 error = cap_check(&haverights, needrightsp); 2659 if (error != 0) 2660 return (error); 2661 #endif 2662 count = fp->f_count; 2663 retry: 2664 if (count == 0) { 2665 /* 2666 * Force a reload. Other thread could reallocate the 2667 * table before this fd was closed, so it possible that 2668 * there is a stale fp pointer in cached version. 2669 */ 2670 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files); 2671 continue; 2672 } 2673 /* 2674 * Use an acquire barrier to force re-reading of fdt so it is 2675 * refreshed for verification. 2676 */ 2677 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0) 2678 goto retry; 2679 fdt = fdp->fd_files; 2680 #ifdef CAPABILITIES 2681 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2682 #else 2683 if (fp == fdt->fdt_ofiles[fd].fde_file) 2684 #endif 2685 break; 2686 fdrop(fp, curthread); 2687 } 2688 *fpp = fp; 2689 if (seqp != NULL) { 2690 #ifdef CAPABILITIES 2691 *seqp = seq; 2692 #endif 2693 } 2694 return (0); 2695 } 2696 2697 /* 2698 * Extract the file pointer associated with the specified descriptor for the 2699 * current user process. 2700 * 2701 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2702 * returned. 2703 * 2704 * File's rights will be checked against the capability rights mask. 2705 * 2706 * If an error occurred the non-zero error is returned and *fpp is set to 2707 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2708 * responsible for fdrop(). 2709 */ 2710 static __inline int 2711 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2712 cap_rights_t *needrightsp, seq_t *seqp) 2713 { 2714 struct filedesc *fdp; 2715 struct file *fp; 2716 int error; 2717 2718 *fpp = NULL; 2719 fdp = td->td_proc->p_fd; 2720 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2721 if (error != 0) 2722 return (error); 2723 if (fp->f_ops == &badfileops) { 2724 fdrop(fp, td); 2725 return (EBADF); 2726 } 2727 2728 /* 2729 * FREAD and FWRITE failure return EBADF as per POSIX. 2730 */ 2731 error = 0; 2732 switch (flags) { 2733 case FREAD: 2734 case FWRITE: 2735 if ((fp->f_flag & flags) == 0) 2736 error = EBADF; 2737 break; 2738 case FEXEC: 2739 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2740 ((fp->f_flag & FWRITE) != 0)) 2741 error = EBADF; 2742 break; 2743 case 0: 2744 break; 2745 default: 2746 KASSERT(0, ("wrong flags")); 2747 } 2748 2749 if (error != 0) { 2750 fdrop(fp, td); 2751 return (error); 2752 } 2753 2754 *fpp = fp; 2755 return (0); 2756 } 2757 2758 int 2759 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2760 { 2761 2762 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2763 } 2764 2765 int 2766 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2767 struct file **fpp) 2768 { 2769 int error; 2770 #ifndef CAPABILITIES 2771 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2772 if (maxprotp != NULL) 2773 *maxprotp = VM_PROT_ALL; 2774 #else 2775 struct filedesc *fdp = td->td_proc->p_fd; 2776 seq_t seq; 2777 2778 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2779 for (;;) { 2780 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2781 if (error != 0) 2782 return (error); 2783 /* 2784 * If requested, convert capability rights to access flags. 2785 */ 2786 if (maxprotp != NULL) 2787 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2788 if (!fd_modified(fdp, fd, seq)) 2789 break; 2790 fdrop(*fpp, td); 2791 } 2792 #endif 2793 return (error); 2794 } 2795 2796 int 2797 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2798 { 2799 2800 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2801 } 2802 2803 int 2804 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2805 { 2806 2807 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2808 } 2809 2810 int 2811 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2812 struct file **fpp) 2813 { 2814 struct filedesc *fdp = td->td_proc->p_fd; 2815 #ifndef CAPABILITIES 2816 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2817 #else 2818 int error; 2819 seq_t seq; 2820 2821 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2822 for (;;) { 2823 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2824 if (error != 0) 2825 return (error); 2826 error = cap_fcntl_check(fdp, fd, needfcntl); 2827 if (!fd_modified(fdp, fd, seq)) 2828 break; 2829 fdrop(*fpp, td); 2830 } 2831 if (error != 0) { 2832 fdrop(*fpp, td); 2833 *fpp = NULL; 2834 } 2835 return (error); 2836 #endif 2837 } 2838 2839 /* 2840 * Like fget() but loads the underlying vnode, or returns an error if the 2841 * descriptor does not represent a vnode. Note that pipes use vnodes but 2842 * never have VM objects. The returned vnode will be vref()'d. 2843 * 2844 * XXX: what about the unused flags ? 2845 */ 2846 static __inline int 2847 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2848 struct vnode **vpp) 2849 { 2850 struct file *fp; 2851 int error; 2852 2853 *vpp = NULL; 2854 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2855 if (error != 0) 2856 return (error); 2857 if (fp->f_vnode == NULL) { 2858 error = EINVAL; 2859 } else { 2860 *vpp = fp->f_vnode; 2861 vrefact(*vpp); 2862 } 2863 fdrop(fp, td); 2864 2865 return (error); 2866 } 2867 2868 int 2869 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2870 { 2871 2872 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2873 } 2874 2875 int 2876 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2877 struct filecaps *havecaps, struct vnode **vpp) 2878 { 2879 struct filedesc *fdp; 2880 struct filecaps caps; 2881 struct file *fp; 2882 int error; 2883 2884 fdp = td->td_proc->p_fd; 2885 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps); 2886 if (error != 0) 2887 return (error); 2888 if (fp->f_ops == &badfileops) { 2889 error = EBADF; 2890 goto out; 2891 } 2892 if (fp->f_vnode == NULL) { 2893 error = EINVAL; 2894 goto out; 2895 } 2896 2897 *havecaps = caps; 2898 *vpp = fp->f_vnode; 2899 vrefact(*vpp); 2900 2901 return (0); 2902 out: 2903 filecaps_free(&caps); 2904 return (error); 2905 } 2906 2907 int 2908 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2909 { 2910 2911 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2912 } 2913 2914 int 2915 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2916 { 2917 2918 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2919 } 2920 2921 #ifdef notyet 2922 int 2923 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2924 struct vnode **vpp) 2925 { 2926 2927 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2928 } 2929 #endif 2930 2931 /* 2932 * Handle the last reference to a file being closed. 2933 * 2934 * Without the noinline attribute clang keeps inlining the func thorough this 2935 * file when fdrop is used. 2936 */ 2937 int __noinline 2938 _fdrop(struct file *fp, struct thread *td) 2939 { 2940 int error; 2941 2942 if (fp->f_count != 0) 2943 panic("fdrop: count %d", fp->f_count); 2944 error = fo_close(fp, td); 2945 atomic_subtract_int(&openfiles, 1); 2946 crfree(fp->f_cred); 2947 free(fp->f_advice, M_FADVISE); 2948 uma_zfree(file_zone, fp); 2949 2950 return (error); 2951 } 2952 2953 /* 2954 * Apply an advisory lock on a file descriptor. 2955 * 2956 * Just attempt to get a record lock of the requested type on the entire file 2957 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2958 */ 2959 #ifndef _SYS_SYSPROTO_H_ 2960 struct flock_args { 2961 int fd; 2962 int how; 2963 }; 2964 #endif 2965 /* ARGSUSED */ 2966 int 2967 sys_flock(struct thread *td, struct flock_args *uap) 2968 { 2969 struct file *fp; 2970 struct vnode *vp; 2971 struct flock lf; 2972 int error; 2973 2974 error = fget(td, uap->fd, &cap_flock_rights, &fp); 2975 if (error != 0) 2976 return (error); 2977 if (fp->f_type != DTYPE_VNODE) { 2978 fdrop(fp, td); 2979 return (EOPNOTSUPP); 2980 } 2981 2982 vp = fp->f_vnode; 2983 lf.l_whence = SEEK_SET; 2984 lf.l_start = 0; 2985 lf.l_len = 0; 2986 if (uap->how & LOCK_UN) { 2987 lf.l_type = F_UNLCK; 2988 atomic_clear_int(&fp->f_flag, FHASLOCK); 2989 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2990 goto done2; 2991 } 2992 if (uap->how & LOCK_EX) 2993 lf.l_type = F_WRLCK; 2994 else if (uap->how & LOCK_SH) 2995 lf.l_type = F_RDLCK; 2996 else { 2997 error = EBADF; 2998 goto done2; 2999 } 3000 atomic_set_int(&fp->f_flag, FHASLOCK); 3001 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3002 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3003 done2: 3004 fdrop(fp, td); 3005 return (error); 3006 } 3007 /* 3008 * Duplicate the specified descriptor to a free descriptor. 3009 */ 3010 int 3011 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3012 int openerror, int *indxp) 3013 { 3014 struct filedescent *newfde, *oldfde; 3015 struct file *fp; 3016 u_long *ioctls; 3017 int error, indx; 3018 3019 KASSERT(openerror == ENODEV || openerror == ENXIO, 3020 ("unexpected error %d in %s", openerror, __func__)); 3021 3022 /* 3023 * If the to-be-dup'd fd number is greater than the allowed number 3024 * of file descriptors, or the fd to be dup'd has already been 3025 * closed, then reject. 3026 */ 3027 FILEDESC_XLOCK(fdp); 3028 if ((fp = fget_locked(fdp, dfd)) == NULL) { 3029 FILEDESC_XUNLOCK(fdp); 3030 return (EBADF); 3031 } 3032 3033 error = fdalloc(td, 0, &indx); 3034 if (error != 0) { 3035 FILEDESC_XUNLOCK(fdp); 3036 return (error); 3037 } 3038 3039 /* 3040 * There are two cases of interest here. 3041 * 3042 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3043 * 3044 * For ENXIO steal away the file structure from (dfd) and store it in 3045 * (indx). (dfd) is effectively closed by this operation. 3046 */ 3047 switch (openerror) { 3048 case ENODEV: 3049 /* 3050 * Check that the mode the file is being opened for is a 3051 * subset of the mode of the existing descriptor. 3052 */ 3053 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3054 fdunused(fdp, indx); 3055 FILEDESC_XUNLOCK(fdp); 3056 return (EACCES); 3057 } 3058 fhold(fp); 3059 newfde = &fdp->fd_ofiles[indx]; 3060 oldfde = &fdp->fd_ofiles[dfd]; 3061 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3062 #ifdef CAPABILITIES 3063 seq_write_begin(&newfde->fde_seq); 3064 #endif 3065 memcpy(newfde, oldfde, fde_change_size); 3066 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3067 ioctls); 3068 #ifdef CAPABILITIES 3069 seq_write_end(&newfde->fde_seq); 3070 #endif 3071 break; 3072 case ENXIO: 3073 /* 3074 * Steal away the file pointer from dfd and stuff it into indx. 3075 */ 3076 newfde = &fdp->fd_ofiles[indx]; 3077 oldfde = &fdp->fd_ofiles[dfd]; 3078 #ifdef CAPABILITIES 3079 seq_write_begin(&newfde->fde_seq); 3080 #endif 3081 memcpy(newfde, oldfde, fde_change_size); 3082 oldfde->fde_file = NULL; 3083 fdunused(fdp, dfd); 3084 #ifdef CAPABILITIES 3085 seq_write_end(&newfde->fde_seq); 3086 #endif 3087 break; 3088 } 3089 FILEDESC_XUNLOCK(fdp); 3090 *indxp = indx; 3091 return (0); 3092 } 3093 3094 /* 3095 * This sysctl determines if we will allow a process to chroot(2) if it 3096 * has a directory open: 3097 * 0: disallowed for all processes. 3098 * 1: allowed for processes that were not already chroot(2)'ed. 3099 * 2: allowed for all processes. 3100 */ 3101 3102 static int chroot_allow_open_directories = 1; 3103 3104 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3105 &chroot_allow_open_directories, 0, 3106 "Allow a process to chroot(2) if it has a directory open"); 3107 3108 /* 3109 * Helper function for raised chroot(2) security function: Refuse if 3110 * any filedescriptors are open directories. 3111 */ 3112 static int 3113 chroot_refuse_vdir_fds(struct filedesc *fdp) 3114 { 3115 struct vnode *vp; 3116 struct file *fp; 3117 int fd; 3118 3119 FILEDESC_LOCK_ASSERT(fdp); 3120 3121 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 3122 fp = fget_locked(fdp, fd); 3123 if (fp == NULL) 3124 continue; 3125 if (fp->f_type == DTYPE_VNODE) { 3126 vp = fp->f_vnode; 3127 if (vp->v_type == VDIR) 3128 return (EPERM); 3129 } 3130 } 3131 return (0); 3132 } 3133 3134 /* 3135 * Common routine for kern_chroot() and jail_attach(). The caller is 3136 * responsible for invoking priv_check() and mac_vnode_check_chroot() to 3137 * authorize this operation. 3138 */ 3139 int 3140 pwd_chroot(struct thread *td, struct vnode *vp) 3141 { 3142 struct filedesc *fdp; 3143 struct vnode *oldvp; 3144 int error; 3145 3146 fdp = td->td_proc->p_fd; 3147 FILEDESC_XLOCK(fdp); 3148 if (chroot_allow_open_directories == 0 || 3149 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 3150 error = chroot_refuse_vdir_fds(fdp); 3151 if (error != 0) { 3152 FILEDESC_XUNLOCK(fdp); 3153 return (error); 3154 } 3155 } 3156 oldvp = fdp->fd_rdir; 3157 vrefact(vp); 3158 fdp->fd_rdir = vp; 3159 if (fdp->fd_jdir == NULL) { 3160 vrefact(vp); 3161 fdp->fd_jdir = vp; 3162 } 3163 FILEDESC_XUNLOCK(fdp); 3164 vrele(oldvp); 3165 return (0); 3166 } 3167 3168 void 3169 pwd_chdir(struct thread *td, struct vnode *vp) 3170 { 3171 struct filedesc *fdp; 3172 struct vnode *oldvp; 3173 3174 fdp = td->td_proc->p_fd; 3175 FILEDESC_XLOCK(fdp); 3176 VNASSERT(vp->v_usecount > 0, vp, 3177 ("chdir to a vnode with zero usecount")); 3178 oldvp = fdp->fd_cdir; 3179 fdp->fd_cdir = vp; 3180 FILEDESC_XUNLOCK(fdp); 3181 vrele(oldvp); 3182 } 3183 3184 /* 3185 * Scan all active processes and prisons to see if any of them have a current 3186 * or root directory of `olddp'. If so, replace them with the new mount point. 3187 */ 3188 void 3189 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3190 { 3191 struct filedesc *fdp; 3192 struct prison *pr; 3193 struct proc *p; 3194 int nrele; 3195 3196 if (vrefcnt(olddp) == 1) 3197 return; 3198 nrele = 0; 3199 sx_slock(&allproc_lock); 3200 FOREACH_PROC_IN_SYSTEM(p) { 3201 PROC_LOCK(p); 3202 fdp = fdhold(p); 3203 PROC_UNLOCK(p); 3204 if (fdp == NULL) 3205 continue; 3206 FILEDESC_XLOCK(fdp); 3207 if (fdp->fd_cdir == olddp) { 3208 vrefact(newdp); 3209 fdp->fd_cdir = newdp; 3210 nrele++; 3211 } 3212 if (fdp->fd_rdir == olddp) { 3213 vrefact(newdp); 3214 fdp->fd_rdir = newdp; 3215 nrele++; 3216 } 3217 if (fdp->fd_jdir == olddp) { 3218 vrefact(newdp); 3219 fdp->fd_jdir = newdp; 3220 nrele++; 3221 } 3222 FILEDESC_XUNLOCK(fdp); 3223 fddrop(fdp); 3224 } 3225 sx_sunlock(&allproc_lock); 3226 if (rootvnode == olddp) { 3227 vrefact(newdp); 3228 rootvnode = newdp; 3229 nrele++; 3230 } 3231 mtx_lock(&prison0.pr_mtx); 3232 if (prison0.pr_root == olddp) { 3233 vrefact(newdp); 3234 prison0.pr_root = newdp; 3235 nrele++; 3236 } 3237 mtx_unlock(&prison0.pr_mtx); 3238 sx_slock(&allprison_lock); 3239 TAILQ_FOREACH(pr, &allprison, pr_list) { 3240 mtx_lock(&pr->pr_mtx); 3241 if (pr->pr_root == olddp) { 3242 vrefact(newdp); 3243 pr->pr_root = newdp; 3244 nrele++; 3245 } 3246 mtx_unlock(&pr->pr_mtx); 3247 } 3248 sx_sunlock(&allprison_lock); 3249 while (nrele--) 3250 vrele(olddp); 3251 } 3252 3253 struct filedesc_to_leader * 3254 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 3255 { 3256 struct filedesc_to_leader *fdtol; 3257 3258 fdtol = malloc(sizeof(struct filedesc_to_leader), 3259 M_FILEDESC_TO_LEADER, M_WAITOK); 3260 fdtol->fdl_refcount = 1; 3261 fdtol->fdl_holdcount = 0; 3262 fdtol->fdl_wakeup = 0; 3263 fdtol->fdl_leader = leader; 3264 if (old != NULL) { 3265 FILEDESC_XLOCK(fdp); 3266 fdtol->fdl_next = old->fdl_next; 3267 fdtol->fdl_prev = old; 3268 old->fdl_next = fdtol; 3269 fdtol->fdl_next->fdl_prev = fdtol; 3270 FILEDESC_XUNLOCK(fdp); 3271 } else { 3272 fdtol->fdl_next = fdtol; 3273 fdtol->fdl_prev = fdtol; 3274 } 3275 return (fdtol); 3276 } 3277 3278 static int 3279 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 3280 { 3281 struct filedesc *fdp; 3282 int i, count, slots; 3283 3284 if (*(int *)arg1 != 0) 3285 return (EINVAL); 3286 3287 fdp = curproc->p_fd; 3288 count = 0; 3289 FILEDESC_SLOCK(fdp); 3290 slots = NDSLOTS(fdp->fd_lastfile + 1); 3291 for (i = 0; i < slots; i++) 3292 count += bitcountl(fdp->fd_map[i]); 3293 FILEDESC_SUNLOCK(fdp); 3294 3295 return (SYSCTL_OUT(req, &count, sizeof(count))); 3296 } 3297 3298 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 3299 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 3300 "Number of open file descriptors"); 3301 3302 /* 3303 * Get file structures globally. 3304 */ 3305 static int 3306 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 3307 { 3308 struct xfile xf; 3309 struct filedesc *fdp; 3310 struct file *fp; 3311 struct proc *p; 3312 int error, n; 3313 3314 error = sysctl_wire_old_buffer(req, 0); 3315 if (error != 0) 3316 return (error); 3317 if (req->oldptr == NULL) { 3318 n = 0; 3319 sx_slock(&allproc_lock); 3320 FOREACH_PROC_IN_SYSTEM(p) { 3321 PROC_LOCK(p); 3322 if (p->p_state == PRS_NEW) { 3323 PROC_UNLOCK(p); 3324 continue; 3325 } 3326 fdp = fdhold(p); 3327 PROC_UNLOCK(p); 3328 if (fdp == NULL) 3329 continue; 3330 /* overestimates sparse tables. */ 3331 if (fdp->fd_lastfile > 0) 3332 n += fdp->fd_lastfile; 3333 fddrop(fdp); 3334 } 3335 sx_sunlock(&allproc_lock); 3336 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3337 } 3338 error = 0; 3339 bzero(&xf, sizeof(xf)); 3340 xf.xf_size = sizeof(xf); 3341 sx_slock(&allproc_lock); 3342 FOREACH_PROC_IN_SYSTEM(p) { 3343 PROC_LOCK(p); 3344 if (p->p_state == PRS_NEW) { 3345 PROC_UNLOCK(p); 3346 continue; 3347 } 3348 if (p_cansee(req->td, p) != 0) { 3349 PROC_UNLOCK(p); 3350 continue; 3351 } 3352 xf.xf_pid = p->p_pid; 3353 xf.xf_uid = p->p_ucred->cr_uid; 3354 fdp = fdhold(p); 3355 PROC_UNLOCK(p); 3356 if (fdp == NULL) 3357 continue; 3358 FILEDESC_SLOCK(fdp); 3359 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3360 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3361 continue; 3362 xf.xf_fd = n; 3363 xf.xf_file = (uintptr_t)fp; 3364 xf.xf_data = (uintptr_t)fp->f_data; 3365 xf.xf_vnode = (uintptr_t)fp->f_vnode; 3366 xf.xf_type = (uintptr_t)fp->f_type; 3367 xf.xf_count = fp->f_count; 3368 xf.xf_msgcount = 0; 3369 xf.xf_offset = foffset_get(fp); 3370 xf.xf_flag = fp->f_flag; 3371 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3372 if (error) 3373 break; 3374 } 3375 FILEDESC_SUNLOCK(fdp); 3376 fddrop(fdp); 3377 if (error) 3378 break; 3379 } 3380 sx_sunlock(&allproc_lock); 3381 return (error); 3382 } 3383 3384 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3385 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3386 3387 #ifdef KINFO_FILE_SIZE 3388 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3389 #endif 3390 3391 static int 3392 xlate_fflags(int fflags) 3393 { 3394 static const struct { 3395 int fflag; 3396 int kf_fflag; 3397 } fflags_table[] = { 3398 { FAPPEND, KF_FLAG_APPEND }, 3399 { FASYNC, KF_FLAG_ASYNC }, 3400 { FFSYNC, KF_FLAG_FSYNC }, 3401 { FHASLOCK, KF_FLAG_HASLOCK }, 3402 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3403 { FREAD, KF_FLAG_READ }, 3404 { FWRITE, KF_FLAG_WRITE }, 3405 { O_CREAT, KF_FLAG_CREAT }, 3406 { O_DIRECT, KF_FLAG_DIRECT }, 3407 { O_EXCL, KF_FLAG_EXCL }, 3408 { O_EXEC, KF_FLAG_EXEC }, 3409 { O_EXLOCK, KF_FLAG_EXLOCK }, 3410 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3411 { O_SHLOCK, KF_FLAG_SHLOCK }, 3412 { O_TRUNC, KF_FLAG_TRUNC } 3413 }; 3414 unsigned int i; 3415 int kflags; 3416 3417 kflags = 0; 3418 for (i = 0; i < nitems(fflags_table); i++) 3419 if (fflags & fflags_table[i].fflag) 3420 kflags |= fflags_table[i].kf_fflag; 3421 return (kflags); 3422 } 3423 3424 /* Trim unused data from kf_path by truncating the structure size. */ 3425 static void 3426 pack_kinfo(struct kinfo_file *kif) 3427 { 3428 3429 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3430 strlen(kif->kf_path) + 1; 3431 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3432 } 3433 3434 static void 3435 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3436 struct kinfo_file *kif, struct filedesc *fdp, int flags) 3437 { 3438 int error; 3439 3440 bzero(kif, sizeof(*kif)); 3441 3442 /* Set a default type to allow for empty fill_kinfo() methods. */ 3443 kif->kf_type = KF_TYPE_UNKNOWN; 3444 kif->kf_flags = xlate_fflags(fp->f_flag); 3445 if (rightsp != NULL) 3446 kif->kf_cap_rights = *rightsp; 3447 else 3448 cap_rights_init(&kif->kf_cap_rights); 3449 kif->kf_fd = fd; 3450 kif->kf_ref_count = fp->f_count; 3451 kif->kf_offset = foffset_get(fp); 3452 3453 /* 3454 * This may drop the filedesc lock, so the 'fp' cannot be 3455 * accessed after this call. 3456 */ 3457 error = fo_fill_kinfo(fp, kif, fdp); 3458 if (error == 0) 3459 kif->kf_status |= KF_ATTR_VALID; 3460 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3461 pack_kinfo(kif); 3462 else 3463 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3464 } 3465 3466 static void 3467 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3468 struct kinfo_file *kif, int flags) 3469 { 3470 int error; 3471 3472 bzero(kif, sizeof(*kif)); 3473 3474 kif->kf_type = KF_TYPE_VNODE; 3475 error = vn_fill_kinfo_vnode(vp, kif); 3476 if (error == 0) 3477 kif->kf_status |= KF_ATTR_VALID; 3478 kif->kf_flags = xlate_fflags(fflags); 3479 cap_rights_init(&kif->kf_cap_rights); 3480 kif->kf_fd = fd; 3481 kif->kf_ref_count = -1; 3482 kif->kf_offset = -1; 3483 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3484 pack_kinfo(kif); 3485 else 3486 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3487 vrele(vp); 3488 } 3489 3490 struct export_fd_buf { 3491 struct filedesc *fdp; 3492 struct sbuf *sb; 3493 ssize_t remainder; 3494 struct kinfo_file kif; 3495 int flags; 3496 }; 3497 3498 static int 3499 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3500 { 3501 struct kinfo_file *kif; 3502 3503 kif = &efbuf->kif; 3504 if (efbuf->remainder != -1) { 3505 if (efbuf->remainder < kif->kf_structsize) { 3506 /* Terminate export. */ 3507 efbuf->remainder = 0; 3508 return (0); 3509 } 3510 efbuf->remainder -= kif->kf_structsize; 3511 } 3512 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3513 } 3514 3515 static int 3516 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3517 struct export_fd_buf *efbuf) 3518 { 3519 int error; 3520 3521 if (efbuf->remainder == 0) 3522 return (0); 3523 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 3524 efbuf->flags); 3525 FILEDESC_SUNLOCK(efbuf->fdp); 3526 error = export_kinfo_to_sb(efbuf); 3527 FILEDESC_SLOCK(efbuf->fdp); 3528 return (error); 3529 } 3530 3531 static int 3532 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3533 struct export_fd_buf *efbuf) 3534 { 3535 int error; 3536 3537 if (efbuf->remainder == 0) 3538 return (0); 3539 if (efbuf->fdp != NULL) 3540 FILEDESC_SUNLOCK(efbuf->fdp); 3541 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 3542 error = export_kinfo_to_sb(efbuf); 3543 if (efbuf->fdp != NULL) 3544 FILEDESC_SLOCK(efbuf->fdp); 3545 return (error); 3546 } 3547 3548 /* 3549 * Store a process file descriptor information to sbuf. 3550 * 3551 * Takes a locked proc as argument, and returns with the proc unlocked. 3552 */ 3553 int 3554 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 3555 int flags) 3556 { 3557 struct file *fp; 3558 struct filedesc *fdp; 3559 struct export_fd_buf *efbuf; 3560 struct vnode *cttyvp, *textvp, *tracevp; 3561 int error, i; 3562 cap_rights_t rights; 3563 3564 PROC_LOCK_ASSERT(p, MA_OWNED); 3565 3566 /* ktrace vnode */ 3567 tracevp = p->p_tracevp; 3568 if (tracevp != NULL) 3569 vrefact(tracevp); 3570 /* text vnode */ 3571 textvp = p->p_textvp; 3572 if (textvp != NULL) 3573 vrefact(textvp); 3574 /* Controlling tty. */ 3575 cttyvp = NULL; 3576 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3577 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3578 if (cttyvp != NULL) 3579 vrefact(cttyvp); 3580 } 3581 fdp = fdhold(p); 3582 PROC_UNLOCK(p); 3583 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3584 efbuf->fdp = NULL; 3585 efbuf->sb = sb; 3586 efbuf->remainder = maxlen; 3587 efbuf->flags = flags; 3588 if (tracevp != NULL) 3589 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3590 efbuf); 3591 if (textvp != NULL) 3592 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3593 if (cttyvp != NULL) 3594 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3595 efbuf); 3596 error = 0; 3597 if (fdp == NULL) 3598 goto fail; 3599 efbuf->fdp = fdp; 3600 FILEDESC_SLOCK(fdp); 3601 /* working directory */ 3602 if (fdp->fd_cdir != NULL) { 3603 vrefact(fdp->fd_cdir); 3604 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3605 } 3606 /* root directory */ 3607 if (fdp->fd_rdir != NULL) { 3608 vrefact(fdp->fd_rdir); 3609 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3610 } 3611 /* jail directory */ 3612 if (fdp->fd_jdir != NULL) { 3613 vrefact(fdp->fd_jdir); 3614 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3615 } 3616 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3617 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3618 continue; 3619 #ifdef CAPABILITIES 3620 rights = *cap_rights(fdp, i); 3621 #else /* !CAPABILITIES */ 3622 rights = cap_no_rights; 3623 #endif 3624 /* 3625 * Create sysctl entry. It is OK to drop the filedesc 3626 * lock inside of export_file_to_sb() as we will 3627 * re-validate and re-evaluate its properties when the 3628 * loop continues. 3629 */ 3630 error = export_file_to_sb(fp, i, &rights, efbuf); 3631 if (error != 0 || efbuf->remainder == 0) 3632 break; 3633 } 3634 FILEDESC_SUNLOCK(fdp); 3635 fddrop(fdp); 3636 fail: 3637 free(efbuf, M_TEMP); 3638 return (error); 3639 } 3640 3641 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3642 3643 /* 3644 * Get per-process file descriptors for use by procstat(1), et al. 3645 */ 3646 static int 3647 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3648 { 3649 struct sbuf sb; 3650 struct proc *p; 3651 ssize_t maxlen; 3652 int error, error2, *name; 3653 3654 name = (int *)arg1; 3655 3656 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3657 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3658 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3659 if (error != 0) { 3660 sbuf_delete(&sb); 3661 return (error); 3662 } 3663 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3664 error = kern_proc_filedesc_out(p, &sb, maxlen, 3665 KERN_FILEDESC_PACK_KINFO); 3666 error2 = sbuf_finish(&sb); 3667 sbuf_delete(&sb); 3668 return (error != 0 ? error : error2); 3669 } 3670 3671 #ifdef COMPAT_FREEBSD7 3672 #ifdef KINFO_OFILE_SIZE 3673 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3674 #endif 3675 3676 static void 3677 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3678 { 3679 3680 okif->kf_structsize = sizeof(*okif); 3681 okif->kf_type = kif->kf_type; 3682 okif->kf_fd = kif->kf_fd; 3683 okif->kf_ref_count = kif->kf_ref_count; 3684 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3685 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3686 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3687 okif->kf_offset = kif->kf_offset; 3688 if (kif->kf_type == KF_TYPE_VNODE) 3689 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 3690 else 3691 okif->kf_vnode_type = KF_VTYPE_VNON; 3692 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3693 if (kif->kf_type == KF_TYPE_SOCKET) { 3694 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 3695 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 3696 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 3697 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 3698 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 3699 } else { 3700 okif->kf_sa_local.ss_family = AF_UNSPEC; 3701 okif->kf_sa_peer.ss_family = AF_UNSPEC; 3702 } 3703 } 3704 3705 static int 3706 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3707 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3708 { 3709 int error; 3710 3711 vrefact(vp); 3712 FILEDESC_SUNLOCK(fdp); 3713 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 3714 kinfo_to_okinfo(kif, okif); 3715 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3716 FILEDESC_SLOCK(fdp); 3717 return (error); 3718 } 3719 3720 /* 3721 * Get per-process file descriptors for use by procstat(1), et al. 3722 */ 3723 static int 3724 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3725 { 3726 struct kinfo_ofile *okif; 3727 struct kinfo_file *kif; 3728 struct filedesc *fdp; 3729 int error, i, *name; 3730 struct file *fp; 3731 struct proc *p; 3732 3733 name = (int *)arg1; 3734 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3735 if (error != 0) 3736 return (error); 3737 fdp = fdhold(p); 3738 PROC_UNLOCK(p); 3739 if (fdp == NULL) 3740 return (ENOENT); 3741 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3742 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3743 FILEDESC_SLOCK(fdp); 3744 if (fdp->fd_cdir != NULL) 3745 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3746 okif, fdp, req); 3747 if (fdp->fd_rdir != NULL) 3748 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3749 okif, fdp, req); 3750 if (fdp->fd_jdir != NULL) 3751 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3752 okif, fdp, req); 3753 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3754 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3755 continue; 3756 export_file_to_kinfo(fp, i, NULL, kif, fdp, 3757 KERN_FILEDESC_PACK_KINFO); 3758 FILEDESC_SUNLOCK(fdp); 3759 kinfo_to_okinfo(kif, okif); 3760 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3761 FILEDESC_SLOCK(fdp); 3762 if (error) 3763 break; 3764 } 3765 FILEDESC_SUNLOCK(fdp); 3766 fddrop(fdp); 3767 free(kif, M_TEMP); 3768 free(okif, M_TEMP); 3769 return (0); 3770 } 3771 3772 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3773 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3774 "Process ofiledesc entries"); 3775 #endif /* COMPAT_FREEBSD7 */ 3776 3777 int 3778 vntype_to_kinfo(int vtype) 3779 { 3780 struct { 3781 int vtype; 3782 int kf_vtype; 3783 } vtypes_table[] = { 3784 { VBAD, KF_VTYPE_VBAD }, 3785 { VBLK, KF_VTYPE_VBLK }, 3786 { VCHR, KF_VTYPE_VCHR }, 3787 { VDIR, KF_VTYPE_VDIR }, 3788 { VFIFO, KF_VTYPE_VFIFO }, 3789 { VLNK, KF_VTYPE_VLNK }, 3790 { VNON, KF_VTYPE_VNON }, 3791 { VREG, KF_VTYPE_VREG }, 3792 { VSOCK, KF_VTYPE_VSOCK } 3793 }; 3794 unsigned int i; 3795 3796 /* 3797 * Perform vtype translation. 3798 */ 3799 for (i = 0; i < nitems(vtypes_table); i++) 3800 if (vtypes_table[i].vtype == vtype) 3801 return (vtypes_table[i].kf_vtype); 3802 3803 return (KF_VTYPE_UNKNOWN); 3804 } 3805 3806 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3807 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3808 "Process filedesc entries"); 3809 3810 /* 3811 * Store a process current working directory information to sbuf. 3812 * 3813 * Takes a locked proc as argument, and returns with the proc unlocked. 3814 */ 3815 int 3816 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3817 { 3818 struct filedesc *fdp; 3819 struct export_fd_buf *efbuf; 3820 int error; 3821 3822 PROC_LOCK_ASSERT(p, MA_OWNED); 3823 3824 fdp = fdhold(p); 3825 PROC_UNLOCK(p); 3826 if (fdp == NULL) 3827 return (EINVAL); 3828 3829 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3830 efbuf->fdp = fdp; 3831 efbuf->sb = sb; 3832 efbuf->remainder = maxlen; 3833 3834 FILEDESC_SLOCK(fdp); 3835 if (fdp->fd_cdir == NULL) 3836 error = EINVAL; 3837 else { 3838 vrefact(fdp->fd_cdir); 3839 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3840 FREAD, efbuf); 3841 } 3842 FILEDESC_SUNLOCK(fdp); 3843 fddrop(fdp); 3844 free(efbuf, M_TEMP); 3845 return (error); 3846 } 3847 3848 /* 3849 * Get per-process current working directory. 3850 */ 3851 static int 3852 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3853 { 3854 struct sbuf sb; 3855 struct proc *p; 3856 ssize_t maxlen; 3857 int error, error2, *name; 3858 3859 name = (int *)arg1; 3860 3861 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3862 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3863 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3864 if (error != 0) { 3865 sbuf_delete(&sb); 3866 return (error); 3867 } 3868 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3869 error = kern_proc_cwd_out(p, &sb, maxlen); 3870 error2 = sbuf_finish(&sb); 3871 sbuf_delete(&sb); 3872 return (error != 0 ? error : error2); 3873 } 3874 3875 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3876 sysctl_kern_proc_cwd, "Process current working directory"); 3877 3878 #ifdef DDB 3879 /* 3880 * For the purposes of debugging, generate a human-readable string for the 3881 * file type. 3882 */ 3883 static const char * 3884 file_type_to_name(short type) 3885 { 3886 3887 switch (type) { 3888 case 0: 3889 return ("zero"); 3890 case DTYPE_VNODE: 3891 return ("vnode"); 3892 case DTYPE_SOCKET: 3893 return ("socket"); 3894 case DTYPE_PIPE: 3895 return ("pipe"); 3896 case DTYPE_FIFO: 3897 return ("fifo"); 3898 case DTYPE_KQUEUE: 3899 return ("kqueue"); 3900 case DTYPE_CRYPTO: 3901 return ("crypto"); 3902 case DTYPE_MQUEUE: 3903 return ("mqueue"); 3904 case DTYPE_SHM: 3905 return ("shm"); 3906 case DTYPE_SEM: 3907 return ("ksem"); 3908 case DTYPE_PTS: 3909 return ("pts"); 3910 case DTYPE_DEV: 3911 return ("dev"); 3912 case DTYPE_PROCDESC: 3913 return ("proc"); 3914 case DTYPE_LINUXEFD: 3915 return ("levent"); 3916 case DTYPE_LINUXTFD: 3917 return ("ltimer"); 3918 default: 3919 return ("unkn"); 3920 } 3921 } 3922 3923 /* 3924 * For the purposes of debugging, identify a process (if any, perhaps one of 3925 * many) that references the passed file in its file descriptor array. Return 3926 * NULL if none. 3927 */ 3928 static struct proc * 3929 file_to_first_proc(struct file *fp) 3930 { 3931 struct filedesc *fdp; 3932 struct proc *p; 3933 int n; 3934 3935 FOREACH_PROC_IN_SYSTEM(p) { 3936 if (p->p_state == PRS_NEW) 3937 continue; 3938 fdp = p->p_fd; 3939 if (fdp == NULL) 3940 continue; 3941 for (n = 0; n <= fdp->fd_lastfile; n++) { 3942 if (fp == fdp->fd_ofiles[n].fde_file) 3943 return (p); 3944 } 3945 } 3946 return (NULL); 3947 } 3948 3949 static void 3950 db_print_file(struct file *fp, int header) 3951 { 3952 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 3953 struct proc *p; 3954 3955 if (header) 3956 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 3957 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 3958 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 3959 "FCmd"); 3960 p = file_to_first_proc(fp); 3961 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 3962 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 3963 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode, 3964 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3965 3966 #undef XPTRWIDTH 3967 } 3968 3969 DB_SHOW_COMMAND(file, db_show_file) 3970 { 3971 struct file *fp; 3972 3973 if (!have_addr) { 3974 db_printf("usage: show file <addr>\n"); 3975 return; 3976 } 3977 fp = (struct file *)addr; 3978 db_print_file(fp, 1); 3979 } 3980 3981 DB_SHOW_COMMAND(files, db_show_files) 3982 { 3983 struct filedesc *fdp; 3984 struct file *fp; 3985 struct proc *p; 3986 int header; 3987 int n; 3988 3989 header = 1; 3990 FOREACH_PROC_IN_SYSTEM(p) { 3991 if (p->p_state == PRS_NEW) 3992 continue; 3993 if ((fdp = p->p_fd) == NULL) 3994 continue; 3995 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3996 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3997 continue; 3998 db_print_file(fp, header); 3999 header = 0; 4000 } 4001 } 4002 } 4003 #endif 4004 4005 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 4006 &maxfilesperproc, 0, "Maximum files allowed open per process"); 4007 4008 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 4009 &maxfiles, 0, "Maximum number of files"); 4010 4011 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 4012 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 4013 4014 /* ARGSUSED*/ 4015 static void 4016 filelistinit(void *dummy) 4017 { 4018 4019 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 4020 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 4021 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 4022 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 4023 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 4024 } 4025 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 4026 4027 /*-------------------------------------------------------------------*/ 4028 4029 static int 4030 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 4031 int flags, struct thread *td) 4032 { 4033 4034 return (EBADF); 4035 } 4036 4037 static int 4038 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4039 struct thread *td) 4040 { 4041 4042 return (EINVAL); 4043 } 4044 4045 static int 4046 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 4047 struct thread *td) 4048 { 4049 4050 return (EBADF); 4051 } 4052 4053 static int 4054 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 4055 struct thread *td) 4056 { 4057 4058 return (0); 4059 } 4060 4061 static int 4062 badfo_kqfilter(struct file *fp, struct knote *kn) 4063 { 4064 4065 return (EBADF); 4066 } 4067 4068 static int 4069 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 4070 struct thread *td) 4071 { 4072 4073 return (EBADF); 4074 } 4075 4076 static int 4077 badfo_close(struct file *fp, struct thread *td) 4078 { 4079 4080 return (0); 4081 } 4082 4083 static int 4084 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4085 struct thread *td) 4086 { 4087 4088 return (EBADF); 4089 } 4090 4091 static int 4092 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4093 struct thread *td) 4094 { 4095 4096 return (EBADF); 4097 } 4098 4099 static int 4100 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4101 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4102 struct thread *td) 4103 { 4104 4105 return (EBADF); 4106 } 4107 4108 static int 4109 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 4110 { 4111 4112 return (0); 4113 } 4114 4115 struct fileops badfileops = { 4116 .fo_read = badfo_readwrite, 4117 .fo_write = badfo_readwrite, 4118 .fo_truncate = badfo_truncate, 4119 .fo_ioctl = badfo_ioctl, 4120 .fo_poll = badfo_poll, 4121 .fo_kqfilter = badfo_kqfilter, 4122 .fo_stat = badfo_stat, 4123 .fo_close = badfo_close, 4124 .fo_chmod = badfo_chmod, 4125 .fo_chown = badfo_chown, 4126 .fo_sendfile = badfo_sendfile, 4127 .fo_fill_kinfo = badfo_fill_kinfo, 4128 }; 4129 4130 int 4131 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 4132 int flags, struct thread *td) 4133 { 4134 4135 return (EOPNOTSUPP); 4136 } 4137 4138 int 4139 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4140 struct thread *td) 4141 { 4142 4143 return (EINVAL); 4144 } 4145 4146 int 4147 invfo_ioctl(struct file *fp, u_long com, void *data, 4148 struct ucred *active_cred, struct thread *td) 4149 { 4150 4151 return (ENOTTY); 4152 } 4153 4154 int 4155 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 4156 struct thread *td) 4157 { 4158 4159 return (poll_no_poll(events)); 4160 } 4161 4162 int 4163 invfo_kqfilter(struct file *fp, struct knote *kn) 4164 { 4165 4166 return (EINVAL); 4167 } 4168 4169 int 4170 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4171 struct thread *td) 4172 { 4173 4174 return (EINVAL); 4175 } 4176 4177 int 4178 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4179 struct thread *td) 4180 { 4181 4182 return (EINVAL); 4183 } 4184 4185 int 4186 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4187 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4188 struct thread *td) 4189 { 4190 4191 return (EINVAL); 4192 } 4193 4194 /*-------------------------------------------------------------------*/ 4195 4196 /* 4197 * File Descriptor pseudo-device driver (/dev/fd/). 4198 * 4199 * Opening minor device N dup()s the file (if any) connected to file 4200 * descriptor N belonging to the calling process. Note that this driver 4201 * consists of only the ``open()'' routine, because all subsequent 4202 * references to this file will be direct to the other driver. 4203 * 4204 * XXX: we could give this one a cloning event handler if necessary. 4205 */ 4206 4207 /* ARGSUSED */ 4208 static int 4209 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4210 { 4211 4212 /* 4213 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4214 * the file descriptor being sought for duplication. The error 4215 * return ensures that the vnode for this device will be released 4216 * by vn_open. Open will detect this special error and take the 4217 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4218 * will simply report the error. 4219 */ 4220 td->td_dupfd = dev2unit(dev); 4221 return (ENODEV); 4222 } 4223 4224 static struct cdevsw fildesc_cdevsw = { 4225 .d_version = D_VERSION, 4226 .d_open = fdopen, 4227 .d_name = "FD", 4228 }; 4229 4230 static void 4231 fildesc_drvinit(void *unused) 4232 { 4233 struct cdev *dev; 4234 4235 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4236 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4237 make_dev_alias(dev, "stdin"); 4238 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4239 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4240 make_dev_alias(dev, "stdout"); 4241 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4242 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4243 make_dev_alias(dev, "stderr"); 4244 } 4245 4246 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4247