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 = fd_first_free(fdp, fd, fdp->fd_nfiles); 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] = 352 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); 353 #ifdef RACCT 354 PROC_LOCK(td->td_proc); 355 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 356 PROC_UNLOCK(td->td_proc); 357 if (lim < td->td_retval[0]) 358 td->td_retval[0] = lim; 359 #endif 360 return (0); 361 } 362 363 /* 364 * Duplicate a file descriptor to a particular value. 365 * 366 * Note: keep in mind that a potential race condition exists when closing 367 * descriptors from a shared descriptor table (via rfork). 368 */ 369 #ifndef _SYS_SYSPROTO_H_ 370 struct dup2_args { 371 u_int from; 372 u_int to; 373 }; 374 #endif 375 /* ARGSUSED */ 376 int 377 sys_dup2(struct thread *td, struct dup2_args *uap) 378 { 379 380 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 381 } 382 383 /* 384 * Duplicate a file descriptor. 385 */ 386 #ifndef _SYS_SYSPROTO_H_ 387 struct dup_args { 388 u_int fd; 389 }; 390 #endif 391 /* ARGSUSED */ 392 int 393 sys_dup(struct thread *td, struct dup_args *uap) 394 { 395 396 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 397 } 398 399 /* 400 * The file control system call. 401 */ 402 #ifndef _SYS_SYSPROTO_H_ 403 struct fcntl_args { 404 int fd; 405 int cmd; 406 long arg; 407 }; 408 #endif 409 /* ARGSUSED */ 410 int 411 sys_fcntl(struct thread *td, struct fcntl_args *uap) 412 { 413 414 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 415 } 416 417 int 418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 419 { 420 struct flock fl; 421 struct __oflock ofl; 422 intptr_t arg1; 423 int error, newcmd; 424 425 error = 0; 426 newcmd = cmd; 427 switch (cmd) { 428 case F_OGETLK: 429 case F_OSETLK: 430 case F_OSETLKW: 431 /* 432 * Convert old flock structure to new. 433 */ 434 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 435 fl.l_start = ofl.l_start; 436 fl.l_len = ofl.l_len; 437 fl.l_pid = ofl.l_pid; 438 fl.l_type = ofl.l_type; 439 fl.l_whence = ofl.l_whence; 440 fl.l_sysid = 0; 441 442 switch (cmd) { 443 case F_OGETLK: 444 newcmd = F_GETLK; 445 break; 446 case F_OSETLK: 447 newcmd = F_SETLK; 448 break; 449 case F_OSETLKW: 450 newcmd = F_SETLKW; 451 break; 452 } 453 arg1 = (intptr_t)&fl; 454 break; 455 case F_GETLK: 456 case F_SETLK: 457 case F_SETLKW: 458 case F_SETLK_REMOTE: 459 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 460 arg1 = (intptr_t)&fl; 461 break; 462 default: 463 arg1 = arg; 464 break; 465 } 466 if (error) 467 return (error); 468 error = kern_fcntl(td, fd, newcmd, arg1); 469 if (error) 470 return (error); 471 if (cmd == F_OGETLK) { 472 ofl.l_start = fl.l_start; 473 ofl.l_len = fl.l_len; 474 ofl.l_pid = fl.l_pid; 475 ofl.l_type = fl.l_type; 476 ofl.l_whence = fl.l_whence; 477 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 478 } else if (cmd == F_GETLK) { 479 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 480 } 481 return (error); 482 } 483 484 int 485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 486 { 487 struct filedesc *fdp; 488 struct flock *flp; 489 struct file *fp, *fp2; 490 struct filedescent *fde; 491 struct proc *p; 492 struct vnode *vp; 493 cap_rights_t rights; 494 int error, flg, tmp; 495 uint64_t bsize; 496 off_t foffset; 497 498 error = 0; 499 flg = F_POSIX; 500 p = td->td_proc; 501 fdp = p->p_fd; 502 503 AUDIT_ARG_FD(cmd); 504 AUDIT_ARG_CMD(cmd); 505 switch (cmd) { 506 case F_DUPFD: 507 tmp = arg; 508 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 509 break; 510 511 case F_DUPFD_CLOEXEC: 512 tmp = arg; 513 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 514 break; 515 516 case F_DUP2FD: 517 tmp = arg; 518 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 519 break; 520 521 case F_DUP2FD_CLOEXEC: 522 tmp = arg; 523 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 524 break; 525 526 case F_GETFD: 527 error = EBADF; 528 FILEDESC_SLOCK(fdp); 529 fde = fdeget_locked(fdp, fd); 530 if (fde != NULL) { 531 td->td_retval[0] = 532 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 533 error = 0; 534 } 535 FILEDESC_SUNLOCK(fdp); 536 break; 537 538 case F_SETFD: 539 error = EBADF; 540 FILEDESC_XLOCK(fdp); 541 fde = fdeget_locked(fdp, fd); 542 if (fde != NULL) { 543 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 544 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 545 error = 0; 546 } 547 FILEDESC_XUNLOCK(fdp); 548 break; 549 550 case F_GETFL: 551 error = fget_fcntl(td, fd, 552 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 553 if (error != 0) 554 break; 555 td->td_retval[0] = OFLAGS(fp->f_flag); 556 fdrop(fp, td); 557 break; 558 559 case F_SETFL: 560 error = fget_fcntl(td, fd, 561 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 562 if (error != 0) 563 break; 564 do { 565 tmp = flg = fp->f_flag; 566 tmp &= ~FCNTLFLAGS; 567 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 568 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 569 tmp = fp->f_flag & FNONBLOCK; 570 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 571 if (error != 0) { 572 fdrop(fp, td); 573 break; 574 } 575 tmp = fp->f_flag & FASYNC; 576 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 577 if (error == 0) { 578 fdrop(fp, td); 579 break; 580 } 581 atomic_clear_int(&fp->f_flag, FNONBLOCK); 582 tmp = 0; 583 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 584 fdrop(fp, td); 585 break; 586 587 case F_GETOWN: 588 error = fget_fcntl(td, fd, 589 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 590 if (error != 0) 591 break; 592 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 593 if (error == 0) 594 td->td_retval[0] = tmp; 595 fdrop(fp, td); 596 break; 597 598 case F_SETOWN: 599 error = fget_fcntl(td, fd, 600 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 601 if (error != 0) 602 break; 603 tmp = arg; 604 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 605 fdrop(fp, td); 606 break; 607 608 case F_SETLK_REMOTE: 609 error = priv_check(td, PRIV_NFS_LOCKD); 610 if (error) 611 return (error); 612 flg = F_REMOTE; 613 goto do_setlk; 614 615 case F_SETLKW: 616 flg |= F_WAIT; 617 /* FALLTHROUGH F_SETLK */ 618 619 case F_SETLK: 620 do_setlk: 621 cap_rights_init(&rights, CAP_FLOCK); 622 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 623 if (error != 0) 624 break; 625 if (fp->f_type != DTYPE_VNODE) { 626 error = EBADF; 627 fdrop(fp, td); 628 break; 629 } 630 631 flp = (struct flock *)arg; 632 if (flp->l_whence == SEEK_CUR) { 633 foffset = foffset_get(fp); 634 if (foffset < 0 || 635 (flp->l_start > 0 && 636 foffset > OFF_MAX - flp->l_start)) { 637 error = EOVERFLOW; 638 fdrop(fp, td); 639 break; 640 } 641 flp->l_start += foffset; 642 } 643 644 vp = fp->f_vnode; 645 switch (flp->l_type) { 646 case F_RDLCK: 647 if ((fp->f_flag & FREAD) == 0) { 648 error = EBADF; 649 break; 650 } 651 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 652 PROC_LOCK(p->p_leader); 653 p->p_leader->p_flag |= P_ADVLOCK; 654 PROC_UNLOCK(p->p_leader); 655 } 656 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 657 flp, flg); 658 break; 659 case F_WRLCK: 660 if ((fp->f_flag & FWRITE) == 0) { 661 error = EBADF; 662 break; 663 } 664 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 665 PROC_LOCK(p->p_leader); 666 p->p_leader->p_flag |= P_ADVLOCK; 667 PROC_UNLOCK(p->p_leader); 668 } 669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 670 flp, flg); 671 break; 672 case F_UNLCK: 673 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 674 flp, flg); 675 break; 676 case F_UNLCKSYS: 677 /* 678 * Temporary api for testing remote lock 679 * infrastructure. 680 */ 681 if (flg != F_REMOTE) { 682 error = EINVAL; 683 break; 684 } 685 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 686 F_UNLCKSYS, flp, flg); 687 break; 688 default: 689 error = EINVAL; 690 break; 691 } 692 if (error != 0 || flp->l_type == F_UNLCK || 693 flp->l_type == F_UNLCKSYS) { 694 fdrop(fp, td); 695 break; 696 } 697 698 /* 699 * Check for a race with close. 700 * 701 * The vnode is now advisory locked (or unlocked, but this case 702 * is not really important) as the caller requested. 703 * We had to drop the filedesc lock, so we need to recheck if 704 * the descriptor is still valid, because if it was closed 705 * in the meantime we need to remove advisory lock from the 706 * vnode - close on any descriptor leading to an advisory 707 * locked vnode, removes that lock. 708 * We will return 0 on purpose in that case, as the result of 709 * successful advisory lock might have been externally visible 710 * already. This is fine - effectively we pretend to the caller 711 * that the closing thread was a bit slower and that the 712 * advisory lock succeeded before the close. 713 */ 714 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 715 if (error != 0) { 716 fdrop(fp, td); 717 break; 718 } 719 if (fp != fp2) { 720 flp->l_whence = SEEK_SET; 721 flp->l_start = 0; 722 flp->l_len = 0; 723 flp->l_type = F_UNLCK; 724 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 725 F_UNLCK, flp, F_POSIX); 726 } 727 fdrop(fp, td); 728 fdrop(fp2, td); 729 break; 730 731 case F_GETLK: 732 error = fget_unlocked(fdp, fd, 733 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 734 if (error != 0) 735 break; 736 if (fp->f_type != DTYPE_VNODE) { 737 error = EBADF; 738 fdrop(fp, td); 739 break; 740 } 741 flp = (struct flock *)arg; 742 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 743 flp->l_type != F_UNLCK) { 744 error = EINVAL; 745 fdrop(fp, td); 746 break; 747 } 748 if (flp->l_whence == SEEK_CUR) { 749 foffset = foffset_get(fp); 750 if ((flp->l_start > 0 && 751 foffset > OFF_MAX - flp->l_start) || 752 (flp->l_start < 0 && 753 foffset < OFF_MIN - flp->l_start)) { 754 error = EOVERFLOW; 755 fdrop(fp, td); 756 break; 757 } 758 flp->l_start += foffset; 759 } 760 vp = fp->f_vnode; 761 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 762 F_POSIX); 763 fdrop(fp, td); 764 break; 765 766 case F_RDAHEAD: 767 arg = arg ? 128 * 1024: 0; 768 /* FALLTHROUGH */ 769 case F_READAHEAD: 770 error = fget_unlocked(fdp, fd, 771 cap_rights_init(&rights), &fp, NULL); 772 if (error != 0) 773 break; 774 if (fp->f_type != DTYPE_VNODE) { 775 fdrop(fp, td); 776 error = EBADF; 777 break; 778 } 779 vp = fp->f_vnode; 780 /* 781 * Exclusive lock synchronizes against f_seqcount reads and 782 * writes in sequential_heuristic(). 783 */ 784 error = vn_lock(vp, LK_EXCLUSIVE); 785 if (error != 0) { 786 fdrop(fp, td); 787 break; 788 } 789 if (arg >= 0) { 790 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 791 fp->f_seqcount = (arg + bsize - 1) / bsize; 792 atomic_set_int(&fp->f_flag, FRDAHEAD); 793 } else { 794 atomic_clear_int(&fp->f_flag, FRDAHEAD); 795 } 796 VOP_UNLOCK(vp, 0); 797 fdrop(fp, td); 798 break; 799 800 default: 801 error = EINVAL; 802 break; 803 } 804 return (error); 805 } 806 807 static int 808 getmaxfd(struct thread *td) 809 { 810 811 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 812 } 813 814 /* 815 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 816 */ 817 int 818 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 819 { 820 struct filedesc *fdp; 821 struct filedescent *oldfde, *newfde; 822 struct proc *p; 823 struct file *delfp; 824 u_long *oioctls, *nioctls; 825 int error, maxfd; 826 827 p = td->td_proc; 828 fdp = p->p_fd; 829 830 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 831 MPASS(mode < FDDUP_LASTMODE); 832 833 AUDIT_ARG_FD(old); 834 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 835 836 /* 837 * Verify we have a valid descriptor to dup from and possibly to 838 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 839 * return EINVAL when the new descriptor is out of bounds. 840 */ 841 if (old < 0) 842 return (EBADF); 843 if (new < 0) 844 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 845 maxfd = getmaxfd(td); 846 if (new >= maxfd) 847 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 848 849 error = EBADF; 850 FILEDESC_XLOCK(fdp); 851 if (fget_locked(fdp, old) == NULL) 852 goto unlock; 853 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 854 td->td_retval[0] = new; 855 if (flags & FDDUP_FLAG_CLOEXEC) 856 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 857 error = 0; 858 goto unlock; 859 } 860 861 /* 862 * If the caller specified a file descriptor, make sure the file 863 * table is large enough to hold it, and grab it. Otherwise, just 864 * allocate a new descriptor the usual way. 865 */ 866 switch (mode) { 867 case FDDUP_NORMAL: 868 case FDDUP_FCNTL: 869 if ((error = fdalloc(td, new, &new)) != 0) 870 goto unlock; 871 break; 872 case FDDUP_MUSTREPLACE: 873 /* Target file descriptor must exist. */ 874 if (fget_locked(fdp, new) == NULL) 875 goto unlock; 876 break; 877 case FDDUP_FIXED: 878 if (new >= fdp->fd_nfiles) { 879 /* 880 * The resource limits are here instead of e.g. 881 * fdalloc(), because the file descriptor table may be 882 * shared between processes, so we can't really use 883 * racct_add()/racct_sub(). Instead of counting the 884 * number of actually allocated descriptors, just put 885 * the limit on the size of the file descriptor table. 886 */ 887 #ifdef RACCT 888 if (racct_enable) { 889 PROC_LOCK(p); 890 error = racct_set(p, RACCT_NOFILE, new + 1); 891 PROC_UNLOCK(p); 892 if (error != 0) { 893 error = EMFILE; 894 goto unlock; 895 } 896 } 897 #endif 898 fdgrowtable_exp(fdp, new + 1); 899 } 900 if (!fdisused(fdp, new)) 901 fdused(fdp, new); 902 break; 903 default: 904 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 905 } 906 907 KASSERT(old != new, ("new fd is same as old")); 908 909 oldfde = &fdp->fd_ofiles[old]; 910 fhold(oldfde->fde_file); 911 newfde = &fdp->fd_ofiles[new]; 912 delfp = newfde->fde_file; 913 914 oioctls = filecaps_free_prep(&newfde->fde_caps); 915 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 916 917 /* 918 * Duplicate the source descriptor. 919 */ 920 #ifdef CAPABILITIES 921 seq_write_begin(&newfde->fde_seq); 922 #endif 923 memcpy(newfde, oldfde, fde_change_size); 924 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 925 nioctls); 926 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 927 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 928 else 929 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 930 #ifdef CAPABILITIES 931 seq_write_end(&newfde->fde_seq); 932 #endif 933 filecaps_free_finish(oioctls); 934 td->td_retval[0] = new; 935 936 error = 0; 937 938 if (delfp != NULL) { 939 (void) closefp(fdp, new, delfp, td, 1); 940 FILEDESC_UNLOCK_ASSERT(fdp); 941 } else { 942 unlock: 943 FILEDESC_XUNLOCK(fdp); 944 } 945 946 return (error); 947 } 948 949 /* 950 * If sigio is on the list associated with a process or process group, 951 * disable signalling from the device, remove sigio from the list and 952 * free sigio. 953 */ 954 void 955 funsetown(struct sigio **sigiop) 956 { 957 struct sigio *sigio; 958 959 if (*sigiop == NULL) 960 return; 961 SIGIO_LOCK(); 962 sigio = *sigiop; 963 if (sigio == NULL) { 964 SIGIO_UNLOCK(); 965 return; 966 } 967 *(sigio->sio_myref) = NULL; 968 if ((sigio)->sio_pgid < 0) { 969 struct pgrp *pg = (sigio)->sio_pgrp; 970 PGRP_LOCK(pg); 971 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 972 sigio, sio_pgsigio); 973 PGRP_UNLOCK(pg); 974 } else { 975 struct proc *p = (sigio)->sio_proc; 976 PROC_LOCK(p); 977 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 978 sigio, sio_pgsigio); 979 PROC_UNLOCK(p); 980 } 981 SIGIO_UNLOCK(); 982 crfree(sigio->sio_ucred); 983 free(sigio, M_SIGIO); 984 } 985 986 /* 987 * Free a list of sigio structures. 988 * We only need to lock the SIGIO_LOCK because we have made ourselves 989 * inaccessible to callers of fsetown and therefore do not need to lock 990 * the proc or pgrp struct for the list manipulation. 991 */ 992 void 993 funsetownlst(struct sigiolst *sigiolst) 994 { 995 struct proc *p; 996 struct pgrp *pg; 997 struct sigio *sigio; 998 999 sigio = SLIST_FIRST(sigiolst); 1000 if (sigio == NULL) 1001 return; 1002 p = NULL; 1003 pg = NULL; 1004 1005 /* 1006 * Every entry of the list should belong 1007 * to a single proc or pgrp. 1008 */ 1009 if (sigio->sio_pgid < 0) { 1010 pg = sigio->sio_pgrp; 1011 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 1012 } else /* if (sigio->sio_pgid > 0) */ { 1013 p = sigio->sio_proc; 1014 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1015 } 1016 1017 SIGIO_LOCK(); 1018 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1019 *(sigio->sio_myref) = NULL; 1020 if (pg != NULL) { 1021 KASSERT(sigio->sio_pgid < 0, 1022 ("Proc sigio in pgrp sigio list")); 1023 KASSERT(sigio->sio_pgrp == pg, 1024 ("Bogus pgrp in sigio list")); 1025 PGRP_LOCK(pg); 1026 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1027 sio_pgsigio); 1028 PGRP_UNLOCK(pg); 1029 } else /* if (p != NULL) */ { 1030 KASSERT(sigio->sio_pgid > 0, 1031 ("Pgrp sigio in proc sigio list")); 1032 KASSERT(sigio->sio_proc == p, 1033 ("Bogus proc in sigio list")); 1034 PROC_LOCK(p); 1035 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1036 sio_pgsigio); 1037 PROC_UNLOCK(p); 1038 } 1039 SIGIO_UNLOCK(); 1040 crfree(sigio->sio_ucred); 1041 free(sigio, M_SIGIO); 1042 SIGIO_LOCK(); 1043 } 1044 SIGIO_UNLOCK(); 1045 } 1046 1047 /* 1048 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1049 * 1050 * After permission checking, add a sigio structure to the sigio list for 1051 * the process or process group. 1052 */ 1053 int 1054 fsetown(pid_t pgid, struct sigio **sigiop) 1055 { 1056 struct proc *proc; 1057 struct pgrp *pgrp; 1058 struct sigio *sigio; 1059 int ret; 1060 1061 if (pgid == 0) { 1062 funsetown(sigiop); 1063 return (0); 1064 } 1065 1066 ret = 0; 1067 1068 /* Allocate and fill in the new sigio out of locks. */ 1069 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1070 sigio->sio_pgid = pgid; 1071 sigio->sio_ucred = crhold(curthread->td_ucred); 1072 sigio->sio_myref = sigiop; 1073 1074 sx_slock(&proctree_lock); 1075 if (pgid > 0) { 1076 proc = pfind(pgid); 1077 if (proc == NULL) { 1078 ret = ESRCH; 1079 goto fail; 1080 } 1081 1082 /* 1083 * Policy - Don't allow a process to FSETOWN a process 1084 * in another session. 1085 * 1086 * Remove this test to allow maximum flexibility or 1087 * restrict FSETOWN to the current process or process 1088 * group for maximum safety. 1089 */ 1090 PROC_UNLOCK(proc); 1091 if (proc->p_session != curthread->td_proc->p_session) { 1092 ret = EPERM; 1093 goto fail; 1094 } 1095 1096 pgrp = NULL; 1097 } else /* if (pgid < 0) */ { 1098 pgrp = pgfind(-pgid); 1099 if (pgrp == NULL) { 1100 ret = ESRCH; 1101 goto fail; 1102 } 1103 PGRP_UNLOCK(pgrp); 1104 1105 /* 1106 * Policy - Don't allow a process to FSETOWN a process 1107 * in another session. 1108 * 1109 * Remove this test to allow maximum flexibility or 1110 * restrict FSETOWN to the current process or process 1111 * group for maximum safety. 1112 */ 1113 if (pgrp->pg_session != curthread->td_proc->p_session) { 1114 ret = EPERM; 1115 goto fail; 1116 } 1117 1118 proc = NULL; 1119 } 1120 funsetown(sigiop); 1121 if (pgid > 0) { 1122 PROC_LOCK(proc); 1123 /* 1124 * Since funsetownlst() is called without the proctree 1125 * locked, we need to check for P_WEXIT. 1126 * XXX: is ESRCH correct? 1127 */ 1128 if ((proc->p_flag & P_WEXIT) != 0) { 1129 PROC_UNLOCK(proc); 1130 ret = ESRCH; 1131 goto fail; 1132 } 1133 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1134 sigio->sio_proc = proc; 1135 PROC_UNLOCK(proc); 1136 } else { 1137 PGRP_LOCK(pgrp); 1138 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1139 sigio->sio_pgrp = pgrp; 1140 PGRP_UNLOCK(pgrp); 1141 } 1142 sx_sunlock(&proctree_lock); 1143 SIGIO_LOCK(); 1144 *sigiop = sigio; 1145 SIGIO_UNLOCK(); 1146 return (0); 1147 1148 fail: 1149 sx_sunlock(&proctree_lock); 1150 crfree(sigio->sio_ucred); 1151 free(sigio, M_SIGIO); 1152 return (ret); 1153 } 1154 1155 /* 1156 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1157 */ 1158 pid_t 1159 fgetown(sigiop) 1160 struct sigio **sigiop; 1161 { 1162 pid_t pgid; 1163 1164 SIGIO_LOCK(); 1165 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1166 SIGIO_UNLOCK(); 1167 return (pgid); 1168 } 1169 1170 /* 1171 * Function drops the filedesc lock on return. 1172 */ 1173 static int 1174 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1175 int holdleaders) 1176 { 1177 int error; 1178 1179 FILEDESC_XLOCK_ASSERT(fdp); 1180 1181 if (holdleaders) { 1182 if (td->td_proc->p_fdtol != NULL) { 1183 /* 1184 * Ask fdfree() to sleep to ensure that all relevant 1185 * process leaders can be traversed in closef(). 1186 */ 1187 fdp->fd_holdleaderscount++; 1188 } else { 1189 holdleaders = 0; 1190 } 1191 } 1192 1193 /* 1194 * We now hold the fp reference that used to be owned by the 1195 * descriptor array. We have to unlock the FILEDESC *AFTER* 1196 * knote_fdclose to prevent a race of the fd getting opened, a knote 1197 * added, and deleteing a knote for the new fd. 1198 */ 1199 knote_fdclose(td, fd); 1200 1201 /* 1202 * We need to notify mqueue if the object is of type mqueue. 1203 */ 1204 if (fp->f_type == DTYPE_MQUEUE) 1205 mq_fdclose(td, fd, fp); 1206 FILEDESC_XUNLOCK(fdp); 1207 1208 error = closef(fp, td); 1209 if (holdleaders) { 1210 FILEDESC_XLOCK(fdp); 1211 fdp->fd_holdleaderscount--; 1212 if (fdp->fd_holdleaderscount == 0 && 1213 fdp->fd_holdleaderswakeup != 0) { 1214 fdp->fd_holdleaderswakeup = 0; 1215 wakeup(&fdp->fd_holdleaderscount); 1216 } 1217 FILEDESC_XUNLOCK(fdp); 1218 } 1219 return (error); 1220 } 1221 1222 /* 1223 * Close a file descriptor. 1224 */ 1225 #ifndef _SYS_SYSPROTO_H_ 1226 struct close_args { 1227 int fd; 1228 }; 1229 #endif 1230 /* ARGSUSED */ 1231 int 1232 sys_close(struct thread *td, struct close_args *uap) 1233 { 1234 1235 return (kern_close(td, uap->fd)); 1236 } 1237 1238 int 1239 kern_close(struct thread *td, int fd) 1240 { 1241 struct filedesc *fdp; 1242 struct file *fp; 1243 1244 fdp = td->td_proc->p_fd; 1245 1246 AUDIT_SYSCLOSE(td, fd); 1247 1248 FILEDESC_XLOCK(fdp); 1249 if ((fp = fget_locked(fdp, fd)) == NULL) { 1250 FILEDESC_XUNLOCK(fdp); 1251 return (EBADF); 1252 } 1253 fdfree(fdp, fd); 1254 1255 /* closefp() drops the FILEDESC lock for us. */ 1256 return (closefp(fdp, fd, fp, td, 1)); 1257 } 1258 1259 /* 1260 * Close open file descriptors. 1261 */ 1262 #ifndef _SYS_SYSPROTO_H_ 1263 struct closefrom_args { 1264 int lowfd; 1265 }; 1266 #endif 1267 /* ARGSUSED */ 1268 int 1269 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1270 { 1271 struct filedesc *fdp; 1272 int fd; 1273 1274 fdp = td->td_proc->p_fd; 1275 AUDIT_ARG_FD(uap->lowfd); 1276 1277 /* 1278 * Treat negative starting file descriptor values identical to 1279 * closefrom(0) which closes all files. 1280 */ 1281 if (uap->lowfd < 0) 1282 uap->lowfd = 0; 1283 FILEDESC_SLOCK(fdp); 1284 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1285 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1286 FILEDESC_SUNLOCK(fdp); 1287 (void)kern_close(td, fd); 1288 FILEDESC_SLOCK(fdp); 1289 } 1290 } 1291 FILEDESC_SUNLOCK(fdp); 1292 return (0); 1293 } 1294 1295 #if defined(COMPAT_43) 1296 /* 1297 * Return status information about a file descriptor. 1298 */ 1299 #ifndef _SYS_SYSPROTO_H_ 1300 struct ofstat_args { 1301 int fd; 1302 struct ostat *sb; 1303 }; 1304 #endif 1305 /* ARGSUSED */ 1306 int 1307 ofstat(struct thread *td, struct ofstat_args *uap) 1308 { 1309 struct ostat oub; 1310 struct stat ub; 1311 int error; 1312 1313 error = kern_fstat(td, uap->fd, &ub); 1314 if (error == 0) { 1315 cvtstat(&ub, &oub); 1316 error = copyout(&oub, uap->sb, sizeof(oub)); 1317 } 1318 return (error); 1319 } 1320 #endif /* COMPAT_43 */ 1321 1322 #if defined(COMPAT_FREEBSD11) 1323 int 1324 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1325 { 1326 struct stat sb; 1327 struct freebsd11_stat osb; 1328 int error; 1329 1330 error = kern_fstat(td, uap->fd, &sb); 1331 if (error != 0) 1332 return (error); 1333 error = freebsd11_cvtstat(&sb, &osb); 1334 if (error == 0) 1335 error = copyout(&osb, uap->sb, sizeof(osb)); 1336 return (error); 1337 } 1338 #endif /* COMPAT_FREEBSD11 */ 1339 1340 /* 1341 * Return status information about a file descriptor. 1342 */ 1343 #ifndef _SYS_SYSPROTO_H_ 1344 struct fstat_args { 1345 int fd; 1346 struct stat *sb; 1347 }; 1348 #endif 1349 /* ARGSUSED */ 1350 int 1351 sys_fstat(struct thread *td, struct fstat_args *uap) 1352 { 1353 struct stat ub; 1354 int error; 1355 1356 error = kern_fstat(td, uap->fd, &ub); 1357 if (error == 0) 1358 error = copyout(&ub, uap->sb, sizeof(ub)); 1359 return (error); 1360 } 1361 1362 int 1363 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1364 { 1365 struct file *fp; 1366 cap_rights_t rights; 1367 int error; 1368 1369 AUDIT_ARG_FD(fd); 1370 1371 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1372 if (error != 0) 1373 return (error); 1374 1375 AUDIT_ARG_FILE(td->td_proc, fp); 1376 1377 error = fo_stat(fp, sbp, td->td_ucred, td); 1378 fdrop(fp, td); 1379 #ifdef __STAT_TIME_T_EXT 1380 if (error == 0) { 1381 sbp->st_atim_ext = 0; 1382 sbp->st_mtim_ext = 0; 1383 sbp->st_ctim_ext = 0; 1384 sbp->st_btim_ext = 0; 1385 } 1386 #endif 1387 #ifdef KTRACE 1388 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1389 ktrstat(sbp); 1390 #endif 1391 return (error); 1392 } 1393 1394 #if defined(COMPAT_FREEBSD11) 1395 /* 1396 * Return status information about a file descriptor. 1397 */ 1398 #ifndef _SYS_SYSPROTO_H_ 1399 struct freebsd11_nfstat_args { 1400 int fd; 1401 struct nstat *sb; 1402 }; 1403 #endif 1404 /* ARGSUSED */ 1405 int 1406 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1407 { 1408 struct nstat nub; 1409 struct stat ub; 1410 int error; 1411 1412 error = kern_fstat(td, uap->fd, &ub); 1413 if (error == 0) { 1414 freebsd11_cvtnstat(&ub, &nub); 1415 error = copyout(&nub, uap->sb, sizeof(nub)); 1416 } 1417 return (error); 1418 } 1419 #endif /* COMPAT_FREEBSD11 */ 1420 1421 /* 1422 * Return pathconf information about a file descriptor. 1423 */ 1424 #ifndef _SYS_SYSPROTO_H_ 1425 struct fpathconf_args { 1426 int fd; 1427 int name; 1428 }; 1429 #endif 1430 /* ARGSUSED */ 1431 int 1432 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1433 { 1434 long value; 1435 int error; 1436 1437 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1438 if (error == 0) 1439 td->td_retval[0] = value; 1440 return (error); 1441 } 1442 1443 int 1444 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1445 { 1446 struct file *fp; 1447 struct vnode *vp; 1448 cap_rights_t rights; 1449 int error; 1450 1451 error = fget(td, fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1452 if (error != 0) 1453 return (error); 1454 1455 if (name == _PC_ASYNC_IO) { 1456 *valuep = _POSIX_ASYNCHRONOUS_IO; 1457 goto out; 1458 } 1459 vp = fp->f_vnode; 1460 if (vp != NULL) { 1461 vn_lock(vp, LK_SHARED | LK_RETRY); 1462 error = VOP_PATHCONF(vp, name, valuep); 1463 VOP_UNLOCK(vp, 0); 1464 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1465 if (name != _PC_PIPE_BUF) { 1466 error = EINVAL; 1467 } else { 1468 *valuep = PIPE_BUF; 1469 error = 0; 1470 } 1471 } else { 1472 error = EOPNOTSUPP; 1473 } 1474 out: 1475 fdrop(fp, td); 1476 return (error); 1477 } 1478 1479 /* 1480 * Initialize filecaps structure. 1481 */ 1482 void 1483 filecaps_init(struct filecaps *fcaps) 1484 { 1485 1486 bzero(fcaps, sizeof(*fcaps)); 1487 fcaps->fc_nioctls = -1; 1488 } 1489 1490 /* 1491 * Copy filecaps structure allocating memory for ioctls array if needed. 1492 * 1493 * The last parameter indicates whether the fdtable is locked. If it is not and 1494 * ioctls are encountered, copying fails and the caller must lock the table. 1495 * 1496 * Note that if the table was not locked, the caller has to check the relevant 1497 * sequence counter to determine whether the operation was successful. 1498 */ 1499 bool 1500 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1501 { 1502 size_t size; 1503 1504 if (src->fc_ioctls != NULL && !locked) 1505 return (false); 1506 *dst = *src; 1507 if (src->fc_ioctls == NULL) 1508 return (true); 1509 1510 KASSERT(src->fc_nioctls > 0, 1511 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1512 1513 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1514 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1515 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1516 return (true); 1517 } 1518 1519 static u_long * 1520 filecaps_copy_prep(const struct filecaps *src) 1521 { 1522 u_long *ioctls; 1523 size_t size; 1524 1525 if (src->fc_ioctls == NULL) 1526 return (NULL); 1527 1528 KASSERT(src->fc_nioctls > 0, 1529 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1530 1531 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1532 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1533 return (ioctls); 1534 } 1535 1536 static void 1537 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1538 u_long *ioctls) 1539 { 1540 size_t size; 1541 1542 *dst = *src; 1543 if (src->fc_ioctls == NULL) { 1544 MPASS(ioctls == NULL); 1545 return; 1546 } 1547 1548 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1549 dst->fc_ioctls = ioctls; 1550 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1551 } 1552 1553 /* 1554 * Move filecaps structure to the new place and clear the old place. 1555 */ 1556 void 1557 filecaps_move(struct filecaps *src, struct filecaps *dst) 1558 { 1559 1560 *dst = *src; 1561 bzero(src, sizeof(*src)); 1562 } 1563 1564 /* 1565 * Fill the given filecaps structure with full rights. 1566 */ 1567 static void 1568 filecaps_fill(struct filecaps *fcaps) 1569 { 1570 1571 CAP_ALL(&fcaps->fc_rights); 1572 fcaps->fc_ioctls = NULL; 1573 fcaps->fc_nioctls = -1; 1574 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1575 } 1576 1577 /* 1578 * Free memory allocated within filecaps structure. 1579 */ 1580 void 1581 filecaps_free(struct filecaps *fcaps) 1582 { 1583 1584 free(fcaps->fc_ioctls, M_FILECAPS); 1585 bzero(fcaps, sizeof(*fcaps)); 1586 } 1587 1588 static u_long * 1589 filecaps_free_prep(struct filecaps *fcaps) 1590 { 1591 u_long *ioctls; 1592 1593 ioctls = fcaps->fc_ioctls; 1594 bzero(fcaps, sizeof(*fcaps)); 1595 return (ioctls); 1596 } 1597 1598 static void 1599 filecaps_free_finish(u_long *ioctls) 1600 { 1601 1602 free(ioctls, M_FILECAPS); 1603 } 1604 1605 /* 1606 * Validate the given filecaps structure. 1607 */ 1608 static void 1609 filecaps_validate(const struct filecaps *fcaps, const char *func) 1610 { 1611 1612 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1613 ("%s: invalid rights", func)); 1614 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1615 ("%s: invalid fcntls", func)); 1616 KASSERT(fcaps->fc_fcntls == 0 || 1617 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1618 ("%s: fcntls without CAP_FCNTL", func)); 1619 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1620 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1621 ("%s: invalid ioctls", func)); 1622 KASSERT(fcaps->fc_nioctls == 0 || 1623 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1624 ("%s: ioctls without CAP_IOCTL", func)); 1625 } 1626 1627 static void 1628 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1629 { 1630 int nfd1; 1631 1632 FILEDESC_XLOCK_ASSERT(fdp); 1633 1634 nfd1 = fdp->fd_nfiles * 2; 1635 if (nfd1 < nfd) 1636 nfd1 = nfd; 1637 fdgrowtable(fdp, nfd1); 1638 } 1639 1640 /* 1641 * Grow the file table to accommodate (at least) nfd descriptors. 1642 */ 1643 static void 1644 fdgrowtable(struct filedesc *fdp, int nfd) 1645 { 1646 struct filedesc0 *fdp0; 1647 struct freetable *ft; 1648 struct fdescenttbl *ntable; 1649 struct fdescenttbl *otable; 1650 int nnfiles, onfiles; 1651 NDSLOTTYPE *nmap, *omap; 1652 1653 /* 1654 * If lastfile is -1 this struct filedesc was just allocated and we are 1655 * growing it to accommodate for the one we are going to copy from. There 1656 * is no need to have a lock on this one as it's not visible to anyone. 1657 */ 1658 if (fdp->fd_lastfile != -1) 1659 FILEDESC_XLOCK_ASSERT(fdp); 1660 1661 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1662 1663 /* save old values */ 1664 onfiles = fdp->fd_nfiles; 1665 otable = fdp->fd_files; 1666 omap = fdp->fd_map; 1667 1668 /* compute the size of the new table */ 1669 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1670 if (nnfiles <= onfiles) 1671 /* the table is already large enough */ 1672 return; 1673 1674 /* 1675 * Allocate a new table. We need enough space for the number of 1676 * entries, file entries themselves and the struct freetable we will use 1677 * when we decommission the table and place it on the freelist. 1678 * We place the struct freetable in the middle so we don't have 1679 * to worry about padding. 1680 */ 1681 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1682 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1683 sizeof(struct freetable), 1684 M_FILEDESC, M_ZERO | M_WAITOK); 1685 /* copy the old data */ 1686 ntable->fdt_nfiles = nnfiles; 1687 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1688 onfiles * sizeof(ntable->fdt_ofiles[0])); 1689 1690 /* 1691 * Allocate a new map only if the old is not large enough. It will 1692 * grow at a slower rate than the table as it can map more 1693 * entries than the table can hold. 1694 */ 1695 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1696 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1697 M_ZERO | M_WAITOK); 1698 /* copy over the old data and update the pointer */ 1699 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1700 fdp->fd_map = nmap; 1701 } 1702 1703 /* 1704 * Make sure that ntable is correctly initialized before we replace 1705 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1706 * data. 1707 */ 1708 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1709 1710 /* 1711 * Do not free the old file table, as some threads may still 1712 * reference entries within it. Instead, place it on a freelist 1713 * which will be processed when the struct filedesc is released. 1714 * 1715 * Note that if onfiles == NDFILE, we're dealing with the original 1716 * static allocation contained within (struct filedesc0 *)fdp, 1717 * which must not be freed. 1718 */ 1719 if (onfiles > NDFILE) { 1720 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1721 fdp0 = (struct filedesc0 *)fdp; 1722 ft->ft_table = otable; 1723 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1724 } 1725 /* 1726 * The map does not have the same possibility of threads still 1727 * holding references to it. So always free it as long as it 1728 * does not reference the original static allocation. 1729 */ 1730 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1731 free(omap, M_FILEDESC); 1732 } 1733 1734 /* 1735 * Allocate a file descriptor for the process. 1736 */ 1737 int 1738 fdalloc(struct thread *td, int minfd, int *result) 1739 { 1740 struct proc *p = td->td_proc; 1741 struct filedesc *fdp = p->p_fd; 1742 int fd, maxfd, allocfd; 1743 #ifdef RACCT 1744 int error; 1745 #endif 1746 1747 FILEDESC_XLOCK_ASSERT(fdp); 1748 1749 if (fdp->fd_freefile > minfd) 1750 minfd = fdp->fd_freefile; 1751 1752 maxfd = getmaxfd(td); 1753 1754 /* 1755 * Search the bitmap for a free descriptor starting at minfd. 1756 * If none is found, grow the file table. 1757 */ 1758 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1759 if (fd >= maxfd) 1760 return (EMFILE); 1761 if (fd >= fdp->fd_nfiles) { 1762 allocfd = min(fd * 2, maxfd); 1763 #ifdef RACCT 1764 if (racct_enable) { 1765 PROC_LOCK(p); 1766 error = racct_set(p, RACCT_NOFILE, allocfd); 1767 PROC_UNLOCK(p); 1768 if (error != 0) 1769 return (EMFILE); 1770 } 1771 #endif 1772 /* 1773 * fd is already equal to first free descriptor >= minfd, so 1774 * we only need to grow the table and we are done. 1775 */ 1776 fdgrowtable_exp(fdp, allocfd); 1777 } 1778 1779 /* 1780 * Perform some sanity checks, then mark the file descriptor as 1781 * used and return it to the caller. 1782 */ 1783 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1784 ("invalid descriptor %d", fd)); 1785 KASSERT(!fdisused(fdp, fd), 1786 ("fd_first_free() returned non-free descriptor")); 1787 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1788 ("file descriptor isn't free")); 1789 fdused(fdp, fd); 1790 *result = fd; 1791 return (0); 1792 } 1793 1794 /* 1795 * Allocate n file descriptors for the process. 1796 */ 1797 int 1798 fdallocn(struct thread *td, int minfd, int *fds, int n) 1799 { 1800 struct proc *p = td->td_proc; 1801 struct filedesc *fdp = p->p_fd; 1802 int i; 1803 1804 FILEDESC_XLOCK_ASSERT(fdp); 1805 1806 for (i = 0; i < n; i++) 1807 if (fdalloc(td, 0, &fds[i]) != 0) 1808 break; 1809 1810 if (i < n) { 1811 for (i--; i >= 0; i--) 1812 fdunused(fdp, fds[i]); 1813 return (EMFILE); 1814 } 1815 1816 return (0); 1817 } 1818 1819 /* 1820 * Create a new open file structure and allocate a file descriptor for the 1821 * process that refers to it. We add one reference to the file for the 1822 * descriptor table and one reference for resultfp. This is to prevent us 1823 * being preempted and the entry in the descriptor table closed after we 1824 * release the FILEDESC lock. 1825 */ 1826 int 1827 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1828 struct filecaps *fcaps) 1829 { 1830 struct file *fp; 1831 int error, fd; 1832 1833 error = falloc_noinstall(td, &fp); 1834 if (error) 1835 return (error); /* no reference held on error */ 1836 1837 error = finstall(td, fp, &fd, flags, fcaps); 1838 if (error) { 1839 fdrop(fp, td); /* one reference (fp only) */ 1840 return (error); 1841 } 1842 1843 if (resultfp != NULL) 1844 *resultfp = fp; /* copy out result */ 1845 else 1846 fdrop(fp, td); /* release local reference */ 1847 1848 if (resultfd != NULL) 1849 *resultfd = fd; 1850 1851 return (0); 1852 } 1853 1854 /* 1855 * Create a new open file structure without allocating a file descriptor. 1856 */ 1857 int 1858 falloc_noinstall(struct thread *td, struct file **resultfp) 1859 { 1860 struct file *fp; 1861 int maxuserfiles = maxfiles - (maxfiles / 20); 1862 int openfiles_new; 1863 static struct timeval lastfail; 1864 static int curfail; 1865 1866 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1867 1868 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 1869 if ((openfiles_new >= maxuserfiles && 1870 priv_check(td, PRIV_MAXFILES) != 0) || 1871 openfiles_new >= maxfiles) { 1872 atomic_subtract_int(&openfiles, 1); 1873 if (ppsratecheck(&lastfail, &curfail, 1)) { 1874 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 1875 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 1876 } 1877 return (ENFILE); 1878 } 1879 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1880 refcount_init(&fp->f_count, 1); 1881 fp->f_cred = crhold(td->td_ucred); 1882 fp->f_ops = &badfileops; 1883 *resultfp = fp; 1884 return (0); 1885 } 1886 1887 /* 1888 * Install a file in a file descriptor table. 1889 */ 1890 void 1891 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1892 struct filecaps *fcaps) 1893 { 1894 struct filedescent *fde; 1895 1896 MPASS(fp != NULL); 1897 if (fcaps != NULL) 1898 filecaps_validate(fcaps, __func__); 1899 FILEDESC_XLOCK_ASSERT(fdp); 1900 1901 fde = &fdp->fd_ofiles[fd]; 1902 #ifdef CAPABILITIES 1903 seq_write_begin(&fde->fde_seq); 1904 #endif 1905 fde->fde_file = fp; 1906 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1907 if (fcaps != NULL) 1908 filecaps_move(fcaps, &fde->fde_caps); 1909 else 1910 filecaps_fill(&fde->fde_caps); 1911 #ifdef CAPABILITIES 1912 seq_write_end(&fde->fde_seq); 1913 #endif 1914 } 1915 1916 int 1917 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1918 struct filecaps *fcaps) 1919 { 1920 struct filedesc *fdp = td->td_proc->p_fd; 1921 int error; 1922 1923 MPASS(fd != NULL); 1924 1925 FILEDESC_XLOCK(fdp); 1926 if ((error = fdalloc(td, 0, fd))) { 1927 FILEDESC_XUNLOCK(fdp); 1928 return (error); 1929 } 1930 fhold(fp); 1931 _finstall(fdp, fp, *fd, flags, fcaps); 1932 FILEDESC_XUNLOCK(fdp); 1933 return (0); 1934 } 1935 1936 /* 1937 * Build a new filedesc structure from another. 1938 * Copy the current, root, and jail root vnode references. 1939 * 1940 * If fdp is not NULL, return with it shared locked. 1941 */ 1942 struct filedesc * 1943 fdinit(struct filedesc *fdp, bool prepfiles) 1944 { 1945 struct filedesc0 *newfdp0; 1946 struct filedesc *newfdp; 1947 1948 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1949 newfdp = &newfdp0->fd_fd; 1950 1951 /* Create the file descriptor table. */ 1952 FILEDESC_LOCK_INIT(newfdp); 1953 refcount_init(&newfdp->fd_refcnt, 1); 1954 refcount_init(&newfdp->fd_holdcnt, 1); 1955 newfdp->fd_cmask = CMASK; 1956 newfdp->fd_map = newfdp0->fd_dmap; 1957 newfdp->fd_lastfile = -1; 1958 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1959 newfdp->fd_files->fdt_nfiles = NDFILE; 1960 1961 if (fdp == NULL) 1962 return (newfdp); 1963 1964 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1965 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1966 1967 FILEDESC_SLOCK(fdp); 1968 newfdp->fd_cdir = fdp->fd_cdir; 1969 if (newfdp->fd_cdir) 1970 vrefact(newfdp->fd_cdir); 1971 newfdp->fd_rdir = fdp->fd_rdir; 1972 if (newfdp->fd_rdir) 1973 vrefact(newfdp->fd_rdir); 1974 newfdp->fd_jdir = fdp->fd_jdir; 1975 if (newfdp->fd_jdir) 1976 vrefact(newfdp->fd_jdir); 1977 1978 if (!prepfiles) { 1979 FILEDESC_SUNLOCK(fdp); 1980 } else { 1981 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1982 FILEDESC_SUNLOCK(fdp); 1983 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1984 FILEDESC_SLOCK(fdp); 1985 } 1986 } 1987 1988 return (newfdp); 1989 } 1990 1991 static struct filedesc * 1992 fdhold(struct proc *p) 1993 { 1994 struct filedesc *fdp; 1995 1996 PROC_LOCK_ASSERT(p, MA_OWNED); 1997 fdp = p->p_fd; 1998 if (fdp != NULL) 1999 refcount_acquire(&fdp->fd_holdcnt); 2000 return (fdp); 2001 } 2002 2003 static void 2004 fddrop(struct filedesc *fdp) 2005 { 2006 2007 if (fdp->fd_holdcnt > 1) { 2008 if (refcount_release(&fdp->fd_holdcnt) == 0) 2009 return; 2010 } 2011 2012 FILEDESC_LOCK_DESTROY(fdp); 2013 uma_zfree(filedesc0_zone, fdp); 2014 } 2015 2016 /* 2017 * Share a filedesc structure. 2018 */ 2019 struct filedesc * 2020 fdshare(struct filedesc *fdp) 2021 { 2022 2023 refcount_acquire(&fdp->fd_refcnt); 2024 return (fdp); 2025 } 2026 2027 /* 2028 * Unshare a filedesc structure, if necessary by making a copy 2029 */ 2030 void 2031 fdunshare(struct thread *td) 2032 { 2033 struct filedesc *tmp; 2034 struct proc *p = td->td_proc; 2035 2036 if (p->p_fd->fd_refcnt == 1) 2037 return; 2038 2039 tmp = fdcopy(p->p_fd); 2040 fdescfree(td); 2041 p->p_fd = tmp; 2042 } 2043 2044 void 2045 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 2046 { 2047 2048 fdescfree(td); 2049 td->td_proc->p_fd = fdp; 2050 } 2051 2052 /* 2053 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2054 * this is to ease callers, not catch errors. 2055 */ 2056 struct filedesc * 2057 fdcopy(struct filedesc *fdp) 2058 { 2059 struct filedesc *newfdp; 2060 struct filedescent *nfde, *ofde; 2061 int i; 2062 2063 MPASS(fdp != NULL); 2064 2065 newfdp = fdinit(fdp, true); 2066 /* copy all passable descriptors (i.e. not kqueue) */ 2067 newfdp->fd_freefile = -1; 2068 for (i = 0; i <= fdp->fd_lastfile; ++i) { 2069 ofde = &fdp->fd_ofiles[i]; 2070 if (ofde->fde_file == NULL || 2071 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2072 if (newfdp->fd_freefile == -1) 2073 newfdp->fd_freefile = i; 2074 continue; 2075 } 2076 nfde = &newfdp->fd_ofiles[i]; 2077 *nfde = *ofde; 2078 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2079 fhold(nfde->fde_file); 2080 fdused_init(newfdp, i); 2081 newfdp->fd_lastfile = i; 2082 } 2083 if (newfdp->fd_freefile == -1) 2084 newfdp->fd_freefile = i; 2085 newfdp->fd_cmask = fdp->fd_cmask; 2086 FILEDESC_SUNLOCK(fdp); 2087 return (newfdp); 2088 } 2089 2090 /* 2091 * Copies a filedesc structure, while remapping all file descriptors 2092 * stored inside using a translation table. 2093 * 2094 * File descriptors are copied over to the new file descriptor table, 2095 * regardless of whether the close-on-exec flag is set. 2096 */ 2097 int 2098 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 2099 struct filedesc **ret) 2100 { 2101 struct filedesc *newfdp; 2102 struct filedescent *nfde, *ofde; 2103 int error, i; 2104 2105 MPASS(fdp != NULL); 2106 2107 newfdp = fdinit(fdp, true); 2108 if (nfds > fdp->fd_lastfile + 1) { 2109 /* New table cannot be larger than the old one. */ 2110 error = E2BIG; 2111 goto bad; 2112 } 2113 /* Copy all passable descriptors (i.e. not kqueue). */ 2114 newfdp->fd_freefile = nfds; 2115 for (i = 0; i < nfds; ++i) { 2116 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { 2117 /* File descriptor out of bounds. */ 2118 error = EBADF; 2119 goto bad; 2120 } 2121 ofde = &fdp->fd_ofiles[fds[i]]; 2122 if (ofde->fde_file == NULL) { 2123 /* Unused file descriptor. */ 2124 error = EBADF; 2125 goto bad; 2126 } 2127 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2128 /* File descriptor cannot be passed. */ 2129 error = EINVAL; 2130 goto bad; 2131 } 2132 nfde = &newfdp->fd_ofiles[i]; 2133 *nfde = *ofde; 2134 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2135 fhold(nfde->fde_file); 2136 fdused_init(newfdp, i); 2137 newfdp->fd_lastfile = i; 2138 } 2139 newfdp->fd_cmask = fdp->fd_cmask; 2140 FILEDESC_SUNLOCK(fdp); 2141 *ret = newfdp; 2142 return (0); 2143 bad: 2144 FILEDESC_SUNLOCK(fdp); 2145 fdescfree_remapped(newfdp); 2146 return (error); 2147 } 2148 2149 /* 2150 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2151 * one of processes using it exits) and the table used to be shared. 2152 */ 2153 static void 2154 fdclearlocks(struct thread *td) 2155 { 2156 struct filedesc *fdp; 2157 struct filedesc_to_leader *fdtol; 2158 struct flock lf; 2159 struct file *fp; 2160 struct proc *p; 2161 struct vnode *vp; 2162 int i; 2163 2164 p = td->td_proc; 2165 fdp = p->p_fd; 2166 fdtol = p->p_fdtol; 2167 MPASS(fdtol != NULL); 2168 2169 FILEDESC_XLOCK(fdp); 2170 KASSERT(fdtol->fdl_refcount > 0, 2171 ("filedesc_to_refcount botch: fdl_refcount=%d", 2172 fdtol->fdl_refcount)); 2173 if (fdtol->fdl_refcount == 1 && 2174 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2175 for (i = 0; i <= fdp->fd_lastfile; i++) { 2176 fp = fdp->fd_ofiles[i].fde_file; 2177 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2178 continue; 2179 fhold(fp); 2180 FILEDESC_XUNLOCK(fdp); 2181 lf.l_whence = SEEK_SET; 2182 lf.l_start = 0; 2183 lf.l_len = 0; 2184 lf.l_type = F_UNLCK; 2185 vp = fp->f_vnode; 2186 (void) VOP_ADVLOCK(vp, 2187 (caddr_t)p->p_leader, F_UNLCK, 2188 &lf, F_POSIX); 2189 FILEDESC_XLOCK(fdp); 2190 fdrop(fp, td); 2191 } 2192 } 2193 retry: 2194 if (fdtol->fdl_refcount == 1) { 2195 if (fdp->fd_holdleaderscount > 0 && 2196 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2197 /* 2198 * close() or kern_dup() has cleared a reference 2199 * in a shared file descriptor table. 2200 */ 2201 fdp->fd_holdleaderswakeup = 1; 2202 sx_sleep(&fdp->fd_holdleaderscount, 2203 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2204 goto retry; 2205 } 2206 if (fdtol->fdl_holdcount > 0) { 2207 /* 2208 * Ensure that fdtol->fdl_leader remains 2209 * valid in closef(). 2210 */ 2211 fdtol->fdl_wakeup = 1; 2212 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2213 "fdlhold", 0); 2214 goto retry; 2215 } 2216 } 2217 fdtol->fdl_refcount--; 2218 if (fdtol->fdl_refcount == 0 && 2219 fdtol->fdl_holdcount == 0) { 2220 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2221 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2222 } else 2223 fdtol = NULL; 2224 p->p_fdtol = NULL; 2225 FILEDESC_XUNLOCK(fdp); 2226 if (fdtol != NULL) 2227 free(fdtol, M_FILEDESC_TO_LEADER); 2228 } 2229 2230 /* 2231 * Release a filedesc structure. 2232 */ 2233 static void 2234 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2235 { 2236 struct filedesc0 *fdp0; 2237 struct freetable *ft, *tft; 2238 struct filedescent *fde; 2239 struct file *fp; 2240 int i; 2241 2242 for (i = 0; i <= fdp->fd_lastfile; i++) { 2243 fde = &fdp->fd_ofiles[i]; 2244 fp = fde->fde_file; 2245 if (fp != NULL) { 2246 fdefree_last(fde); 2247 if (needclose) 2248 (void) closef(fp, td); 2249 else 2250 fdrop(fp, td); 2251 } 2252 } 2253 2254 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2255 free(fdp->fd_map, M_FILEDESC); 2256 if (fdp->fd_nfiles > NDFILE) 2257 free(fdp->fd_files, M_FILEDESC); 2258 2259 fdp0 = (struct filedesc0 *)fdp; 2260 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2261 free(ft->ft_table, M_FILEDESC); 2262 2263 fddrop(fdp); 2264 } 2265 2266 void 2267 fdescfree(struct thread *td) 2268 { 2269 struct proc *p; 2270 struct filedesc *fdp; 2271 struct vnode *cdir, *jdir, *rdir; 2272 2273 p = td->td_proc; 2274 fdp = p->p_fd; 2275 MPASS(fdp != NULL); 2276 2277 #ifdef RACCT 2278 if (racct_enable) { 2279 PROC_LOCK(p); 2280 racct_set(p, RACCT_NOFILE, 0); 2281 PROC_UNLOCK(p); 2282 } 2283 #endif 2284 2285 if (p->p_fdtol != NULL) 2286 fdclearlocks(td); 2287 2288 PROC_LOCK(p); 2289 p->p_fd = NULL; 2290 PROC_UNLOCK(p); 2291 2292 if (refcount_release(&fdp->fd_refcnt) == 0) 2293 return; 2294 2295 FILEDESC_XLOCK(fdp); 2296 cdir = fdp->fd_cdir; 2297 fdp->fd_cdir = NULL; 2298 rdir = fdp->fd_rdir; 2299 fdp->fd_rdir = NULL; 2300 jdir = fdp->fd_jdir; 2301 fdp->fd_jdir = NULL; 2302 FILEDESC_XUNLOCK(fdp); 2303 2304 if (cdir != NULL) 2305 vrele(cdir); 2306 if (rdir != NULL) 2307 vrele(rdir); 2308 if (jdir != NULL) 2309 vrele(jdir); 2310 2311 fdescfree_fds(td, fdp, 1); 2312 } 2313 2314 void 2315 fdescfree_remapped(struct filedesc *fdp) 2316 { 2317 2318 if (fdp->fd_cdir != NULL) 2319 vrele(fdp->fd_cdir); 2320 if (fdp->fd_rdir != NULL) 2321 vrele(fdp->fd_rdir); 2322 if (fdp->fd_jdir != NULL) 2323 vrele(fdp->fd_jdir); 2324 2325 fdescfree_fds(curthread, fdp, 0); 2326 } 2327 2328 /* 2329 * For setugid programs, we don't want to people to use that setugidness 2330 * to generate error messages which write to a file which otherwise would 2331 * otherwise be off-limits to the process. We check for filesystems where 2332 * the vnode can change out from under us after execve (like [lin]procfs). 2333 * 2334 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2335 * sufficient. We also don't check for setugidness since we know we are. 2336 */ 2337 static bool 2338 is_unsafe(struct file *fp) 2339 { 2340 struct vnode *vp; 2341 2342 if (fp->f_type != DTYPE_VNODE) 2343 return (false); 2344 2345 vp = fp->f_vnode; 2346 return ((vp->v_vflag & VV_PROCDEP) != 0); 2347 } 2348 2349 /* 2350 * Make this setguid thing safe, if at all possible. 2351 */ 2352 void 2353 fdsetugidsafety(struct thread *td) 2354 { 2355 struct filedesc *fdp; 2356 struct file *fp; 2357 int i; 2358 2359 fdp = td->td_proc->p_fd; 2360 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2361 MPASS(fdp->fd_nfiles >= 3); 2362 for (i = 0; i <= 2; i++) { 2363 fp = fdp->fd_ofiles[i].fde_file; 2364 if (fp != NULL && is_unsafe(fp)) { 2365 FILEDESC_XLOCK(fdp); 2366 knote_fdclose(td, i); 2367 /* 2368 * NULL-out descriptor prior to close to avoid 2369 * a race while close blocks. 2370 */ 2371 fdfree(fdp, i); 2372 FILEDESC_XUNLOCK(fdp); 2373 (void) closef(fp, td); 2374 } 2375 } 2376 } 2377 2378 /* 2379 * If a specific file object occupies a specific file descriptor, close the 2380 * file descriptor entry and drop a reference on the file object. This is a 2381 * convenience function to handle a subsequent error in a function that calls 2382 * falloc() that handles the race that another thread might have closed the 2383 * file descriptor out from under the thread creating the file object. 2384 */ 2385 void 2386 fdclose(struct thread *td, struct file *fp, int idx) 2387 { 2388 struct filedesc *fdp = td->td_proc->p_fd; 2389 2390 FILEDESC_XLOCK(fdp); 2391 if (fdp->fd_ofiles[idx].fde_file == fp) { 2392 fdfree(fdp, idx); 2393 FILEDESC_XUNLOCK(fdp); 2394 fdrop(fp, td); 2395 } else 2396 FILEDESC_XUNLOCK(fdp); 2397 } 2398 2399 /* 2400 * Close any files on exec? 2401 */ 2402 void 2403 fdcloseexec(struct thread *td) 2404 { 2405 struct filedesc *fdp; 2406 struct filedescent *fde; 2407 struct file *fp; 2408 int i; 2409 2410 fdp = td->td_proc->p_fd; 2411 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2412 for (i = 0; i <= fdp->fd_lastfile; i++) { 2413 fde = &fdp->fd_ofiles[i]; 2414 fp = fde->fde_file; 2415 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2416 (fde->fde_flags & UF_EXCLOSE))) { 2417 FILEDESC_XLOCK(fdp); 2418 fdfree(fdp, i); 2419 (void) closefp(fdp, i, fp, td, 0); 2420 FILEDESC_UNLOCK_ASSERT(fdp); 2421 } 2422 } 2423 } 2424 2425 /* 2426 * It is unsafe for set[ug]id processes to be started with file 2427 * descriptors 0..2 closed, as these descriptors are given implicit 2428 * significance in the Standard C library. fdcheckstd() will create a 2429 * descriptor referencing /dev/null for each of stdin, stdout, and 2430 * stderr that is not already open. 2431 */ 2432 int 2433 fdcheckstd(struct thread *td) 2434 { 2435 struct filedesc *fdp; 2436 register_t save; 2437 int i, error, devnull; 2438 2439 fdp = td->td_proc->p_fd; 2440 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2441 MPASS(fdp->fd_nfiles >= 3); 2442 devnull = -1; 2443 for (i = 0; i <= 2; i++) { 2444 if (fdp->fd_ofiles[i].fde_file != NULL) 2445 continue; 2446 2447 save = td->td_retval[0]; 2448 if (devnull != -1) { 2449 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2450 } else { 2451 error = kern_openat(td, AT_FDCWD, "/dev/null", 2452 UIO_SYSSPACE, O_RDWR, 0); 2453 if (error == 0) { 2454 devnull = td->td_retval[0]; 2455 KASSERT(devnull == i, ("we didn't get our fd")); 2456 } 2457 } 2458 td->td_retval[0] = save; 2459 if (error != 0) 2460 return (error); 2461 } 2462 return (0); 2463 } 2464 2465 /* 2466 * Internal form of close. Decrement reference count on file structure. 2467 * Note: td may be NULL when closing a file that was being passed in a 2468 * message. 2469 * 2470 * XXXRW: Giant is not required for the caller, but often will be held; this 2471 * makes it moderately likely the Giant will be recursed in the VFS case. 2472 */ 2473 int 2474 closef(struct file *fp, struct thread *td) 2475 { 2476 struct vnode *vp; 2477 struct flock lf; 2478 struct filedesc_to_leader *fdtol; 2479 struct filedesc *fdp; 2480 2481 /* 2482 * POSIX record locking dictates that any close releases ALL 2483 * locks owned by this process. This is handled by setting 2484 * a flag in the unlock to free ONLY locks obeying POSIX 2485 * semantics, and not to free BSD-style file locks. 2486 * If the descriptor was in a message, POSIX-style locks 2487 * aren't passed with the descriptor, and the thread pointer 2488 * will be NULL. Callers should be careful only to pass a 2489 * NULL thread pointer when there really is no owning 2490 * context that might have locks, or the locks will be 2491 * leaked. 2492 */ 2493 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2494 vp = fp->f_vnode; 2495 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2496 lf.l_whence = SEEK_SET; 2497 lf.l_start = 0; 2498 lf.l_len = 0; 2499 lf.l_type = F_UNLCK; 2500 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2501 F_UNLCK, &lf, F_POSIX); 2502 } 2503 fdtol = td->td_proc->p_fdtol; 2504 if (fdtol != NULL) { 2505 /* 2506 * Handle special case where file descriptor table is 2507 * shared between multiple process leaders. 2508 */ 2509 fdp = td->td_proc->p_fd; 2510 FILEDESC_XLOCK(fdp); 2511 for (fdtol = fdtol->fdl_next; 2512 fdtol != td->td_proc->p_fdtol; 2513 fdtol = fdtol->fdl_next) { 2514 if ((fdtol->fdl_leader->p_flag & 2515 P_ADVLOCK) == 0) 2516 continue; 2517 fdtol->fdl_holdcount++; 2518 FILEDESC_XUNLOCK(fdp); 2519 lf.l_whence = SEEK_SET; 2520 lf.l_start = 0; 2521 lf.l_len = 0; 2522 lf.l_type = F_UNLCK; 2523 vp = fp->f_vnode; 2524 (void) VOP_ADVLOCK(vp, 2525 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2526 F_POSIX); 2527 FILEDESC_XLOCK(fdp); 2528 fdtol->fdl_holdcount--; 2529 if (fdtol->fdl_holdcount == 0 && 2530 fdtol->fdl_wakeup != 0) { 2531 fdtol->fdl_wakeup = 0; 2532 wakeup(fdtol); 2533 } 2534 } 2535 FILEDESC_XUNLOCK(fdp); 2536 } 2537 } 2538 return (fdrop(fp, td)); 2539 } 2540 2541 /* 2542 * Initialize the file pointer with the specified properties. 2543 * 2544 * The ops are set with release semantics to be certain that the flags, type, 2545 * and data are visible when ops is. This is to prevent ops methods from being 2546 * called with bad data. 2547 */ 2548 void 2549 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2550 { 2551 fp->f_data = data; 2552 fp->f_flag = flag; 2553 fp->f_type = type; 2554 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2555 } 2556 2557 int 2558 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2559 struct file **fpp, struct filecaps *havecapsp) 2560 { 2561 struct filedescent *fde; 2562 int error; 2563 2564 FILEDESC_LOCK_ASSERT(fdp); 2565 2566 fde = fdeget_locked(fdp, fd); 2567 if (fde == NULL) { 2568 error = EBADF; 2569 goto out; 2570 } 2571 2572 #ifdef CAPABILITIES 2573 error = cap_check(cap_rights_fde(fde), needrightsp); 2574 if (error != 0) 2575 goto out; 2576 #endif 2577 2578 if (havecapsp != NULL) 2579 filecaps_copy(&fde->fde_caps, havecapsp, true); 2580 2581 *fpp = fde->fde_file; 2582 2583 error = 0; 2584 out: 2585 return (error); 2586 } 2587 2588 int 2589 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2590 struct file **fpp, struct filecaps *havecapsp) 2591 { 2592 struct filedesc *fdp = td->td_proc->p_fd; 2593 int error; 2594 #ifndef CAPABILITIES 2595 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL); 2596 if (error == 0 && havecapsp != NULL) 2597 filecaps_fill(havecapsp); 2598 #else 2599 struct file *fp; 2600 seq_t seq; 2601 2602 for (;;) { 2603 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq); 2604 if (error != 0) 2605 return (error); 2606 2607 if (havecapsp != NULL) { 2608 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2609 havecapsp, false)) { 2610 fdrop(fp, td); 2611 goto get_locked; 2612 } 2613 } 2614 2615 if (!fd_modified(fdp, fd, seq)) 2616 break; 2617 fdrop(fp, td); 2618 } 2619 2620 *fpp = fp; 2621 return (0); 2622 2623 get_locked: 2624 FILEDESC_SLOCK(fdp); 2625 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); 2626 if (error == 0) 2627 fhold(*fpp); 2628 FILEDESC_SUNLOCK(fdp); 2629 #endif 2630 return (error); 2631 } 2632 2633 int 2634 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2635 struct file **fpp, seq_t *seqp) 2636 { 2637 #ifdef CAPABILITIES 2638 struct filedescent *fde; 2639 #endif 2640 struct fdescenttbl *fdt; 2641 struct file *fp; 2642 u_int count; 2643 #ifdef CAPABILITIES 2644 seq_t seq; 2645 cap_rights_t haverights; 2646 int error; 2647 #endif 2648 2649 fdt = fdp->fd_files; 2650 if ((u_int)fd >= fdt->fdt_nfiles) 2651 return (EBADF); 2652 /* 2653 * Fetch the descriptor locklessly. We avoid fdrop() races by 2654 * never raising a refcount above 0. To accomplish this we have 2655 * to use a cmpset loop rather than an atomic_add. The descriptor 2656 * must be re-verified once we acquire a reference to be certain 2657 * that the identity is still correct and we did not lose a race 2658 * due to preemption. 2659 */ 2660 for (;;) { 2661 #ifdef CAPABILITIES 2662 seq = seq_read(fd_seq(fdt, fd)); 2663 fde = &fdt->fdt_ofiles[fd]; 2664 haverights = *cap_rights_fde(fde); 2665 fp = fde->fde_file; 2666 if (!seq_consistent(fd_seq(fdt, fd), seq)) 2667 continue; 2668 #else 2669 fp = fdt->fdt_ofiles[fd].fde_file; 2670 #endif 2671 if (fp == NULL) 2672 return (EBADF); 2673 #ifdef CAPABILITIES 2674 error = cap_check(&haverights, needrightsp); 2675 if (error != 0) 2676 return (error); 2677 #endif 2678 count = fp->f_count; 2679 retry: 2680 if (count == 0) { 2681 /* 2682 * Force a reload. Other thread could reallocate the 2683 * table before this fd was closed, so it possible that 2684 * there is a stale fp pointer in cached version. 2685 */ 2686 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2687 continue; 2688 } 2689 /* 2690 * Use an acquire barrier to force re-reading of fdt so it is 2691 * refreshed for verification. 2692 */ 2693 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0) 2694 goto retry; 2695 fdt = fdp->fd_files; 2696 #ifdef CAPABILITIES 2697 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2698 #else 2699 if (fp == fdt->fdt_ofiles[fd].fde_file) 2700 #endif 2701 break; 2702 fdrop(fp, curthread); 2703 } 2704 *fpp = fp; 2705 if (seqp != NULL) { 2706 #ifdef CAPABILITIES 2707 *seqp = seq; 2708 #endif 2709 } 2710 return (0); 2711 } 2712 2713 /* 2714 * Extract the file pointer associated with the specified descriptor for the 2715 * current user process. 2716 * 2717 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2718 * returned. 2719 * 2720 * File's rights will be checked against the capability rights mask. 2721 * 2722 * If an error occurred the non-zero error is returned and *fpp is set to 2723 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2724 * responsible for fdrop(). 2725 */ 2726 static __inline int 2727 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2728 cap_rights_t *needrightsp, seq_t *seqp) 2729 { 2730 struct filedesc *fdp; 2731 struct file *fp; 2732 int error; 2733 2734 *fpp = NULL; 2735 fdp = td->td_proc->p_fd; 2736 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2737 if (error != 0) 2738 return (error); 2739 if (fp->f_ops == &badfileops) { 2740 fdrop(fp, td); 2741 return (EBADF); 2742 } 2743 2744 /* 2745 * FREAD and FWRITE failure return EBADF as per POSIX. 2746 */ 2747 error = 0; 2748 switch (flags) { 2749 case FREAD: 2750 case FWRITE: 2751 if ((fp->f_flag & flags) == 0) 2752 error = EBADF; 2753 break; 2754 case FEXEC: 2755 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2756 ((fp->f_flag & FWRITE) != 0)) 2757 error = EBADF; 2758 break; 2759 case 0: 2760 break; 2761 default: 2762 KASSERT(0, ("wrong flags")); 2763 } 2764 2765 if (error != 0) { 2766 fdrop(fp, td); 2767 return (error); 2768 } 2769 2770 *fpp = fp; 2771 return (0); 2772 } 2773 2774 int 2775 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2776 { 2777 2778 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2779 } 2780 2781 int 2782 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2783 struct file **fpp) 2784 { 2785 int error; 2786 #ifndef CAPABILITIES 2787 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2788 if (maxprotp != NULL) 2789 *maxprotp = VM_PROT_ALL; 2790 #else 2791 struct filedesc *fdp = td->td_proc->p_fd; 2792 seq_t seq; 2793 2794 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2795 for (;;) { 2796 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2797 if (error != 0) 2798 return (error); 2799 /* 2800 * If requested, convert capability rights to access flags. 2801 */ 2802 if (maxprotp != NULL) 2803 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2804 if (!fd_modified(fdp, fd, seq)) 2805 break; 2806 fdrop(*fpp, td); 2807 } 2808 #endif 2809 return (error); 2810 } 2811 2812 int 2813 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2814 { 2815 2816 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2817 } 2818 2819 int 2820 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2821 { 2822 2823 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2824 } 2825 2826 int 2827 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2828 struct file **fpp) 2829 { 2830 struct filedesc *fdp = td->td_proc->p_fd; 2831 #ifndef CAPABILITIES 2832 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2833 #else 2834 int error; 2835 seq_t seq; 2836 2837 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2838 for (;;) { 2839 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2840 if (error != 0) 2841 return (error); 2842 error = cap_fcntl_check(fdp, fd, needfcntl); 2843 if (!fd_modified(fdp, fd, seq)) 2844 break; 2845 fdrop(*fpp, td); 2846 } 2847 if (error != 0) { 2848 fdrop(*fpp, td); 2849 *fpp = NULL; 2850 } 2851 return (error); 2852 #endif 2853 } 2854 2855 /* 2856 * Like fget() but loads the underlying vnode, or returns an error if the 2857 * descriptor does not represent a vnode. Note that pipes use vnodes but 2858 * never have VM objects. The returned vnode will be vref()'d. 2859 * 2860 * XXX: what about the unused flags ? 2861 */ 2862 static __inline int 2863 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2864 struct vnode **vpp) 2865 { 2866 struct file *fp; 2867 int error; 2868 2869 *vpp = NULL; 2870 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2871 if (error != 0) 2872 return (error); 2873 if (fp->f_vnode == NULL) { 2874 error = EINVAL; 2875 } else { 2876 *vpp = fp->f_vnode; 2877 vrefact(*vpp); 2878 } 2879 fdrop(fp, td); 2880 2881 return (error); 2882 } 2883 2884 int 2885 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2886 { 2887 2888 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2889 } 2890 2891 int 2892 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2893 struct filecaps *havecaps, struct vnode **vpp) 2894 { 2895 struct filedesc *fdp; 2896 struct filecaps caps; 2897 struct file *fp; 2898 int error; 2899 2900 fdp = td->td_proc->p_fd; 2901 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps); 2902 if (error != 0) 2903 return (error); 2904 if (fp->f_ops == &badfileops) { 2905 error = EBADF; 2906 goto out; 2907 } 2908 if (fp->f_vnode == NULL) { 2909 error = EINVAL; 2910 goto out; 2911 } 2912 2913 *havecaps = caps; 2914 *vpp = fp->f_vnode; 2915 vrefact(*vpp); 2916 2917 return (0); 2918 out: 2919 filecaps_free(&caps); 2920 return (error); 2921 } 2922 2923 int 2924 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2925 { 2926 2927 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2928 } 2929 2930 int 2931 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2932 { 2933 2934 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2935 } 2936 2937 #ifdef notyet 2938 int 2939 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2940 struct vnode **vpp) 2941 { 2942 2943 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2944 } 2945 #endif 2946 2947 /* 2948 * Handle the last reference to a file being closed. 2949 */ 2950 int 2951 _fdrop(struct file *fp, struct thread *td) 2952 { 2953 int error; 2954 2955 if (fp->f_count != 0) 2956 panic("fdrop: count %d", fp->f_count); 2957 error = fo_close(fp, td); 2958 atomic_subtract_int(&openfiles, 1); 2959 crfree(fp->f_cred); 2960 free(fp->f_advice, M_FADVISE); 2961 uma_zfree(file_zone, fp); 2962 2963 return (error); 2964 } 2965 2966 /* 2967 * Apply an advisory lock on a file descriptor. 2968 * 2969 * Just attempt to get a record lock of the requested type on the entire file 2970 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2971 */ 2972 #ifndef _SYS_SYSPROTO_H_ 2973 struct flock_args { 2974 int fd; 2975 int how; 2976 }; 2977 #endif 2978 /* ARGSUSED */ 2979 int 2980 sys_flock(struct thread *td, struct flock_args *uap) 2981 { 2982 struct file *fp; 2983 struct vnode *vp; 2984 struct flock lf; 2985 cap_rights_t rights; 2986 int error; 2987 2988 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2989 if (error != 0) 2990 return (error); 2991 if (fp->f_type != DTYPE_VNODE) { 2992 fdrop(fp, td); 2993 return (EOPNOTSUPP); 2994 } 2995 2996 vp = fp->f_vnode; 2997 lf.l_whence = SEEK_SET; 2998 lf.l_start = 0; 2999 lf.l_len = 0; 3000 if (uap->how & LOCK_UN) { 3001 lf.l_type = F_UNLCK; 3002 atomic_clear_int(&fp->f_flag, FHASLOCK); 3003 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3004 goto done2; 3005 } 3006 if (uap->how & LOCK_EX) 3007 lf.l_type = F_WRLCK; 3008 else if (uap->how & LOCK_SH) 3009 lf.l_type = F_RDLCK; 3010 else { 3011 error = EBADF; 3012 goto done2; 3013 } 3014 atomic_set_int(&fp->f_flag, FHASLOCK); 3015 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3016 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3017 done2: 3018 fdrop(fp, td); 3019 return (error); 3020 } 3021 /* 3022 * Duplicate the specified descriptor to a free descriptor. 3023 */ 3024 int 3025 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3026 int openerror, int *indxp) 3027 { 3028 struct filedescent *newfde, *oldfde; 3029 struct file *fp; 3030 u_long *ioctls; 3031 int error, indx; 3032 3033 KASSERT(openerror == ENODEV || openerror == ENXIO, 3034 ("unexpected error %d in %s", openerror, __func__)); 3035 3036 /* 3037 * If the to-be-dup'd fd number is greater than the allowed number 3038 * of file descriptors, or the fd to be dup'd has already been 3039 * closed, then reject. 3040 */ 3041 FILEDESC_XLOCK(fdp); 3042 if ((fp = fget_locked(fdp, dfd)) == NULL) { 3043 FILEDESC_XUNLOCK(fdp); 3044 return (EBADF); 3045 } 3046 3047 error = fdalloc(td, 0, &indx); 3048 if (error != 0) { 3049 FILEDESC_XUNLOCK(fdp); 3050 return (error); 3051 } 3052 3053 /* 3054 * There are two cases of interest here. 3055 * 3056 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3057 * 3058 * For ENXIO steal away the file structure from (dfd) and store it in 3059 * (indx). (dfd) is effectively closed by this operation. 3060 */ 3061 switch (openerror) { 3062 case ENODEV: 3063 /* 3064 * Check that the mode the file is being opened for is a 3065 * subset of the mode of the existing descriptor. 3066 */ 3067 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3068 fdunused(fdp, indx); 3069 FILEDESC_XUNLOCK(fdp); 3070 return (EACCES); 3071 } 3072 fhold(fp); 3073 newfde = &fdp->fd_ofiles[indx]; 3074 oldfde = &fdp->fd_ofiles[dfd]; 3075 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3076 #ifdef CAPABILITIES 3077 seq_write_begin(&newfde->fde_seq); 3078 #endif 3079 memcpy(newfde, oldfde, fde_change_size); 3080 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3081 ioctls); 3082 #ifdef CAPABILITIES 3083 seq_write_end(&newfde->fde_seq); 3084 #endif 3085 break; 3086 case ENXIO: 3087 /* 3088 * Steal away the file pointer from dfd and stuff it into indx. 3089 */ 3090 newfde = &fdp->fd_ofiles[indx]; 3091 oldfde = &fdp->fd_ofiles[dfd]; 3092 #ifdef CAPABILITIES 3093 seq_write_begin(&newfde->fde_seq); 3094 #endif 3095 memcpy(newfde, oldfde, fde_change_size); 3096 oldfde->fde_file = NULL; 3097 fdunused(fdp, dfd); 3098 #ifdef CAPABILITIES 3099 seq_write_end(&newfde->fde_seq); 3100 #endif 3101 break; 3102 } 3103 FILEDESC_XUNLOCK(fdp); 3104 *indxp = indx; 3105 return (0); 3106 } 3107 3108 /* 3109 * This sysctl determines if we will allow a process to chroot(2) if it 3110 * has a directory open: 3111 * 0: disallowed for all processes. 3112 * 1: allowed for processes that were not already chroot(2)'ed. 3113 * 2: allowed for all processes. 3114 */ 3115 3116 static int chroot_allow_open_directories = 1; 3117 3118 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3119 &chroot_allow_open_directories, 0, 3120 "Allow a process to chroot(2) if it has a directory open"); 3121 3122 /* 3123 * Helper function for raised chroot(2) security function: Refuse if 3124 * any filedescriptors are open directories. 3125 */ 3126 static int 3127 chroot_refuse_vdir_fds(struct filedesc *fdp) 3128 { 3129 struct vnode *vp; 3130 struct file *fp; 3131 int fd; 3132 3133 FILEDESC_LOCK_ASSERT(fdp); 3134 3135 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 3136 fp = fget_locked(fdp, fd); 3137 if (fp == NULL) 3138 continue; 3139 if (fp->f_type == DTYPE_VNODE) { 3140 vp = fp->f_vnode; 3141 if (vp->v_type == VDIR) 3142 return (EPERM); 3143 } 3144 } 3145 return (0); 3146 } 3147 3148 /* 3149 * Common routine for kern_chroot() and jail_attach(). The caller is 3150 * responsible for invoking priv_check() and mac_vnode_check_chroot() to 3151 * authorize this operation. 3152 */ 3153 int 3154 pwd_chroot(struct thread *td, struct vnode *vp) 3155 { 3156 struct filedesc *fdp; 3157 struct vnode *oldvp; 3158 int error; 3159 3160 fdp = td->td_proc->p_fd; 3161 FILEDESC_XLOCK(fdp); 3162 if (chroot_allow_open_directories == 0 || 3163 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 3164 error = chroot_refuse_vdir_fds(fdp); 3165 if (error != 0) { 3166 FILEDESC_XUNLOCK(fdp); 3167 return (error); 3168 } 3169 } 3170 oldvp = fdp->fd_rdir; 3171 vrefact(vp); 3172 fdp->fd_rdir = vp; 3173 if (fdp->fd_jdir == NULL) { 3174 vrefact(vp); 3175 fdp->fd_jdir = vp; 3176 } 3177 FILEDESC_XUNLOCK(fdp); 3178 vrele(oldvp); 3179 return (0); 3180 } 3181 3182 void 3183 pwd_chdir(struct thread *td, struct vnode *vp) 3184 { 3185 struct filedesc *fdp; 3186 struct vnode *oldvp; 3187 3188 fdp = td->td_proc->p_fd; 3189 FILEDESC_XLOCK(fdp); 3190 VNASSERT(vp->v_usecount > 0, vp, 3191 ("chdir to a vnode with zero usecount")); 3192 oldvp = fdp->fd_cdir; 3193 fdp->fd_cdir = vp; 3194 FILEDESC_XUNLOCK(fdp); 3195 vrele(oldvp); 3196 } 3197 3198 /* 3199 * Scan all active processes and prisons to see if any of them have a current 3200 * or root directory of `olddp'. If so, replace them with the new mount point. 3201 */ 3202 void 3203 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3204 { 3205 struct filedesc *fdp; 3206 struct prison *pr; 3207 struct proc *p; 3208 int nrele; 3209 3210 if (vrefcnt(olddp) == 1) 3211 return; 3212 nrele = 0; 3213 sx_slock(&allproc_lock); 3214 FOREACH_PROC_IN_SYSTEM(p) { 3215 PROC_LOCK(p); 3216 fdp = fdhold(p); 3217 PROC_UNLOCK(p); 3218 if (fdp == NULL) 3219 continue; 3220 FILEDESC_XLOCK(fdp); 3221 if (fdp->fd_cdir == olddp) { 3222 vrefact(newdp); 3223 fdp->fd_cdir = newdp; 3224 nrele++; 3225 } 3226 if (fdp->fd_rdir == olddp) { 3227 vrefact(newdp); 3228 fdp->fd_rdir = newdp; 3229 nrele++; 3230 } 3231 if (fdp->fd_jdir == olddp) { 3232 vrefact(newdp); 3233 fdp->fd_jdir = newdp; 3234 nrele++; 3235 } 3236 FILEDESC_XUNLOCK(fdp); 3237 fddrop(fdp); 3238 } 3239 sx_sunlock(&allproc_lock); 3240 if (rootvnode == olddp) { 3241 vrefact(newdp); 3242 rootvnode = newdp; 3243 nrele++; 3244 } 3245 mtx_lock(&prison0.pr_mtx); 3246 if (prison0.pr_root == olddp) { 3247 vrefact(newdp); 3248 prison0.pr_root = newdp; 3249 nrele++; 3250 } 3251 mtx_unlock(&prison0.pr_mtx); 3252 sx_slock(&allprison_lock); 3253 TAILQ_FOREACH(pr, &allprison, pr_list) { 3254 mtx_lock(&pr->pr_mtx); 3255 if (pr->pr_root == olddp) { 3256 vrefact(newdp); 3257 pr->pr_root = newdp; 3258 nrele++; 3259 } 3260 mtx_unlock(&pr->pr_mtx); 3261 } 3262 sx_sunlock(&allprison_lock); 3263 while (nrele--) 3264 vrele(olddp); 3265 } 3266 3267 struct filedesc_to_leader * 3268 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 3269 { 3270 struct filedesc_to_leader *fdtol; 3271 3272 fdtol = malloc(sizeof(struct filedesc_to_leader), 3273 M_FILEDESC_TO_LEADER, M_WAITOK); 3274 fdtol->fdl_refcount = 1; 3275 fdtol->fdl_holdcount = 0; 3276 fdtol->fdl_wakeup = 0; 3277 fdtol->fdl_leader = leader; 3278 if (old != NULL) { 3279 FILEDESC_XLOCK(fdp); 3280 fdtol->fdl_next = old->fdl_next; 3281 fdtol->fdl_prev = old; 3282 old->fdl_next = fdtol; 3283 fdtol->fdl_next->fdl_prev = fdtol; 3284 FILEDESC_XUNLOCK(fdp); 3285 } else { 3286 fdtol->fdl_next = fdtol; 3287 fdtol->fdl_prev = fdtol; 3288 } 3289 return (fdtol); 3290 } 3291 3292 static int 3293 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 3294 { 3295 struct filedesc *fdp; 3296 int i, count, slots; 3297 3298 if (*(int *)arg1 != 0) 3299 return (EINVAL); 3300 3301 fdp = curproc->p_fd; 3302 count = 0; 3303 FILEDESC_SLOCK(fdp); 3304 slots = NDSLOTS(fdp->fd_lastfile + 1); 3305 for (i = 0; i < slots; i++) 3306 count += bitcountl(fdp->fd_map[i]); 3307 FILEDESC_SUNLOCK(fdp); 3308 3309 return (SYSCTL_OUT(req, &count, sizeof(count))); 3310 } 3311 3312 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 3313 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 3314 "Number of open file descriptors"); 3315 3316 /* 3317 * Get file structures globally. 3318 */ 3319 static int 3320 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 3321 { 3322 struct xfile xf; 3323 struct filedesc *fdp; 3324 struct file *fp; 3325 struct proc *p; 3326 int error, n; 3327 3328 error = sysctl_wire_old_buffer(req, 0); 3329 if (error != 0) 3330 return (error); 3331 if (req->oldptr == NULL) { 3332 n = 0; 3333 sx_slock(&allproc_lock); 3334 FOREACH_PROC_IN_SYSTEM(p) { 3335 PROC_LOCK(p); 3336 if (p->p_state == PRS_NEW) { 3337 PROC_UNLOCK(p); 3338 continue; 3339 } 3340 fdp = fdhold(p); 3341 PROC_UNLOCK(p); 3342 if (fdp == NULL) 3343 continue; 3344 /* overestimates sparse tables. */ 3345 if (fdp->fd_lastfile > 0) 3346 n += fdp->fd_lastfile; 3347 fddrop(fdp); 3348 } 3349 sx_sunlock(&allproc_lock); 3350 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3351 } 3352 error = 0; 3353 bzero(&xf, sizeof(xf)); 3354 xf.xf_size = sizeof(xf); 3355 sx_slock(&allproc_lock); 3356 FOREACH_PROC_IN_SYSTEM(p) { 3357 PROC_LOCK(p); 3358 if (p->p_state == PRS_NEW) { 3359 PROC_UNLOCK(p); 3360 continue; 3361 } 3362 if (p_cansee(req->td, p) != 0) { 3363 PROC_UNLOCK(p); 3364 continue; 3365 } 3366 xf.xf_pid = p->p_pid; 3367 xf.xf_uid = p->p_ucred->cr_uid; 3368 fdp = fdhold(p); 3369 PROC_UNLOCK(p); 3370 if (fdp == NULL) 3371 continue; 3372 FILEDESC_SLOCK(fdp); 3373 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3374 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3375 continue; 3376 xf.xf_fd = n; 3377 xf.xf_file = fp; 3378 xf.xf_data = fp->f_data; 3379 xf.xf_vnode = fp->f_vnode; 3380 xf.xf_type = fp->f_type; 3381 xf.xf_count = fp->f_count; 3382 xf.xf_msgcount = 0; 3383 xf.xf_offset = foffset_get(fp); 3384 xf.xf_flag = fp->f_flag; 3385 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3386 if (error) 3387 break; 3388 } 3389 FILEDESC_SUNLOCK(fdp); 3390 fddrop(fdp); 3391 if (error) 3392 break; 3393 } 3394 sx_sunlock(&allproc_lock); 3395 return (error); 3396 } 3397 3398 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3399 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3400 3401 #ifdef KINFO_FILE_SIZE 3402 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3403 #endif 3404 3405 static int 3406 xlate_fflags(int fflags) 3407 { 3408 static const struct { 3409 int fflag; 3410 int kf_fflag; 3411 } fflags_table[] = { 3412 { FAPPEND, KF_FLAG_APPEND }, 3413 { FASYNC, KF_FLAG_ASYNC }, 3414 { FFSYNC, KF_FLAG_FSYNC }, 3415 { FHASLOCK, KF_FLAG_HASLOCK }, 3416 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3417 { FREAD, KF_FLAG_READ }, 3418 { FWRITE, KF_FLAG_WRITE }, 3419 { O_CREAT, KF_FLAG_CREAT }, 3420 { O_DIRECT, KF_FLAG_DIRECT }, 3421 { O_EXCL, KF_FLAG_EXCL }, 3422 { O_EXEC, KF_FLAG_EXEC }, 3423 { O_EXLOCK, KF_FLAG_EXLOCK }, 3424 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3425 { O_SHLOCK, KF_FLAG_SHLOCK }, 3426 { O_TRUNC, KF_FLAG_TRUNC } 3427 }; 3428 unsigned int i; 3429 int kflags; 3430 3431 kflags = 0; 3432 for (i = 0; i < nitems(fflags_table); i++) 3433 if (fflags & fflags_table[i].fflag) 3434 kflags |= fflags_table[i].kf_fflag; 3435 return (kflags); 3436 } 3437 3438 /* Trim unused data from kf_path by truncating the structure size. */ 3439 static void 3440 pack_kinfo(struct kinfo_file *kif) 3441 { 3442 3443 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3444 strlen(kif->kf_path) + 1; 3445 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3446 } 3447 3448 static void 3449 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3450 struct kinfo_file *kif, struct filedesc *fdp, int flags) 3451 { 3452 int error; 3453 3454 bzero(kif, sizeof(*kif)); 3455 3456 /* Set a default type to allow for empty fill_kinfo() methods. */ 3457 kif->kf_type = KF_TYPE_UNKNOWN; 3458 kif->kf_flags = xlate_fflags(fp->f_flag); 3459 if (rightsp != NULL) 3460 kif->kf_cap_rights = *rightsp; 3461 else 3462 cap_rights_init(&kif->kf_cap_rights); 3463 kif->kf_fd = fd; 3464 kif->kf_ref_count = fp->f_count; 3465 kif->kf_offset = foffset_get(fp); 3466 3467 /* 3468 * This may drop the filedesc lock, so the 'fp' cannot be 3469 * accessed after this call. 3470 */ 3471 error = fo_fill_kinfo(fp, kif, fdp); 3472 if (error == 0) 3473 kif->kf_status |= KF_ATTR_VALID; 3474 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3475 pack_kinfo(kif); 3476 else 3477 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3478 } 3479 3480 static void 3481 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3482 struct kinfo_file *kif, int flags) 3483 { 3484 int error; 3485 3486 bzero(kif, sizeof(*kif)); 3487 3488 kif->kf_type = KF_TYPE_VNODE; 3489 error = vn_fill_kinfo_vnode(vp, kif); 3490 if (error == 0) 3491 kif->kf_status |= KF_ATTR_VALID; 3492 kif->kf_flags = xlate_fflags(fflags); 3493 cap_rights_init(&kif->kf_cap_rights); 3494 kif->kf_fd = fd; 3495 kif->kf_ref_count = -1; 3496 kif->kf_offset = -1; 3497 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3498 pack_kinfo(kif); 3499 else 3500 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3501 vrele(vp); 3502 } 3503 3504 struct export_fd_buf { 3505 struct filedesc *fdp; 3506 struct sbuf *sb; 3507 ssize_t remainder; 3508 struct kinfo_file kif; 3509 int flags; 3510 }; 3511 3512 static int 3513 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3514 { 3515 struct kinfo_file *kif; 3516 3517 kif = &efbuf->kif; 3518 if (efbuf->remainder != -1) { 3519 if (efbuf->remainder < kif->kf_structsize) { 3520 /* Terminate export. */ 3521 efbuf->remainder = 0; 3522 return (0); 3523 } 3524 efbuf->remainder -= kif->kf_structsize; 3525 } 3526 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3527 } 3528 3529 static int 3530 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3531 struct export_fd_buf *efbuf) 3532 { 3533 int error; 3534 3535 if (efbuf->remainder == 0) 3536 return (0); 3537 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 3538 efbuf->flags); 3539 FILEDESC_SUNLOCK(efbuf->fdp); 3540 error = export_kinfo_to_sb(efbuf); 3541 FILEDESC_SLOCK(efbuf->fdp); 3542 return (error); 3543 } 3544 3545 static int 3546 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3547 struct export_fd_buf *efbuf) 3548 { 3549 int error; 3550 3551 if (efbuf->remainder == 0) 3552 return (0); 3553 if (efbuf->fdp != NULL) 3554 FILEDESC_SUNLOCK(efbuf->fdp); 3555 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 3556 error = export_kinfo_to_sb(efbuf); 3557 if (efbuf->fdp != NULL) 3558 FILEDESC_SLOCK(efbuf->fdp); 3559 return (error); 3560 } 3561 3562 /* 3563 * Store a process file descriptor information to sbuf. 3564 * 3565 * Takes a locked proc as argument, and returns with the proc unlocked. 3566 */ 3567 int 3568 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 3569 int flags) 3570 { 3571 struct file *fp; 3572 struct filedesc *fdp; 3573 struct export_fd_buf *efbuf; 3574 struct vnode *cttyvp, *textvp, *tracevp; 3575 int error, i; 3576 cap_rights_t rights; 3577 3578 PROC_LOCK_ASSERT(p, MA_OWNED); 3579 3580 /* ktrace vnode */ 3581 tracevp = p->p_tracevp; 3582 if (tracevp != NULL) 3583 vrefact(tracevp); 3584 /* text vnode */ 3585 textvp = p->p_textvp; 3586 if (textvp != NULL) 3587 vrefact(textvp); 3588 /* Controlling tty. */ 3589 cttyvp = NULL; 3590 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3591 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3592 if (cttyvp != NULL) 3593 vrefact(cttyvp); 3594 } 3595 fdp = fdhold(p); 3596 PROC_UNLOCK(p); 3597 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3598 efbuf->fdp = NULL; 3599 efbuf->sb = sb; 3600 efbuf->remainder = maxlen; 3601 efbuf->flags = flags; 3602 if (tracevp != NULL) 3603 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3604 efbuf); 3605 if (textvp != NULL) 3606 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3607 if (cttyvp != NULL) 3608 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3609 efbuf); 3610 error = 0; 3611 if (fdp == NULL) 3612 goto fail; 3613 efbuf->fdp = fdp; 3614 FILEDESC_SLOCK(fdp); 3615 /* working directory */ 3616 if (fdp->fd_cdir != NULL) { 3617 vrefact(fdp->fd_cdir); 3618 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3619 } 3620 /* root directory */ 3621 if (fdp->fd_rdir != NULL) { 3622 vrefact(fdp->fd_rdir); 3623 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3624 } 3625 /* jail directory */ 3626 if (fdp->fd_jdir != NULL) { 3627 vrefact(fdp->fd_jdir); 3628 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3629 } 3630 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3631 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3632 continue; 3633 #ifdef CAPABILITIES 3634 rights = *cap_rights(fdp, i); 3635 #else /* !CAPABILITIES */ 3636 cap_rights_init(&rights); 3637 #endif 3638 /* 3639 * Create sysctl entry. It is OK to drop the filedesc 3640 * lock inside of export_file_to_sb() as we will 3641 * re-validate and re-evaluate its properties when the 3642 * loop continues. 3643 */ 3644 error = export_file_to_sb(fp, i, &rights, efbuf); 3645 if (error != 0 || efbuf->remainder == 0) 3646 break; 3647 } 3648 FILEDESC_SUNLOCK(fdp); 3649 fddrop(fdp); 3650 fail: 3651 free(efbuf, M_TEMP); 3652 return (error); 3653 } 3654 3655 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3656 3657 /* 3658 * Get per-process file descriptors for use by procstat(1), et al. 3659 */ 3660 static int 3661 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3662 { 3663 struct sbuf sb; 3664 struct proc *p; 3665 ssize_t maxlen; 3666 int error, error2, *name; 3667 3668 name = (int *)arg1; 3669 3670 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3671 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3672 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3673 if (error != 0) { 3674 sbuf_delete(&sb); 3675 return (error); 3676 } 3677 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3678 error = kern_proc_filedesc_out(p, &sb, maxlen, 3679 KERN_FILEDESC_PACK_KINFO); 3680 error2 = sbuf_finish(&sb); 3681 sbuf_delete(&sb); 3682 return (error != 0 ? error : error2); 3683 } 3684 3685 #ifdef COMPAT_FREEBSD7 3686 #ifdef KINFO_OFILE_SIZE 3687 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3688 #endif 3689 3690 static void 3691 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3692 { 3693 3694 okif->kf_structsize = sizeof(*okif); 3695 okif->kf_type = kif->kf_type; 3696 okif->kf_fd = kif->kf_fd; 3697 okif->kf_ref_count = kif->kf_ref_count; 3698 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3699 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3700 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3701 okif->kf_offset = kif->kf_offset; 3702 if (kif->kf_type == KF_TYPE_VNODE) 3703 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 3704 else 3705 okif->kf_vnode_type = KF_VTYPE_VNON; 3706 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3707 if (kif->kf_type == KF_TYPE_SOCKET) { 3708 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 3709 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 3710 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 3711 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 3712 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 3713 } else { 3714 okif->kf_sa_local.ss_family = AF_UNSPEC; 3715 okif->kf_sa_peer.ss_family = AF_UNSPEC; 3716 } 3717 } 3718 3719 static int 3720 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3721 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3722 { 3723 int error; 3724 3725 vrefact(vp); 3726 FILEDESC_SUNLOCK(fdp); 3727 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 3728 kinfo_to_okinfo(kif, okif); 3729 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3730 FILEDESC_SLOCK(fdp); 3731 return (error); 3732 } 3733 3734 /* 3735 * Get per-process file descriptors for use by procstat(1), et al. 3736 */ 3737 static int 3738 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3739 { 3740 struct kinfo_ofile *okif; 3741 struct kinfo_file *kif; 3742 struct filedesc *fdp; 3743 int error, i, *name; 3744 struct file *fp; 3745 struct proc *p; 3746 3747 name = (int *)arg1; 3748 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3749 if (error != 0) 3750 return (error); 3751 fdp = fdhold(p); 3752 PROC_UNLOCK(p); 3753 if (fdp == NULL) 3754 return (ENOENT); 3755 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3756 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3757 FILEDESC_SLOCK(fdp); 3758 if (fdp->fd_cdir != NULL) 3759 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3760 okif, fdp, req); 3761 if (fdp->fd_rdir != NULL) 3762 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3763 okif, fdp, req); 3764 if (fdp->fd_jdir != NULL) 3765 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3766 okif, fdp, req); 3767 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3768 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3769 continue; 3770 export_file_to_kinfo(fp, i, NULL, kif, fdp, 3771 KERN_FILEDESC_PACK_KINFO); 3772 FILEDESC_SUNLOCK(fdp); 3773 kinfo_to_okinfo(kif, okif); 3774 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3775 FILEDESC_SLOCK(fdp); 3776 if (error) 3777 break; 3778 } 3779 FILEDESC_SUNLOCK(fdp); 3780 fddrop(fdp); 3781 free(kif, M_TEMP); 3782 free(okif, M_TEMP); 3783 return (0); 3784 } 3785 3786 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3787 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3788 "Process ofiledesc entries"); 3789 #endif /* COMPAT_FREEBSD7 */ 3790 3791 int 3792 vntype_to_kinfo(int vtype) 3793 { 3794 struct { 3795 int vtype; 3796 int kf_vtype; 3797 } vtypes_table[] = { 3798 { VBAD, KF_VTYPE_VBAD }, 3799 { VBLK, KF_VTYPE_VBLK }, 3800 { VCHR, KF_VTYPE_VCHR }, 3801 { VDIR, KF_VTYPE_VDIR }, 3802 { VFIFO, KF_VTYPE_VFIFO }, 3803 { VLNK, KF_VTYPE_VLNK }, 3804 { VNON, KF_VTYPE_VNON }, 3805 { VREG, KF_VTYPE_VREG }, 3806 { VSOCK, KF_VTYPE_VSOCK } 3807 }; 3808 unsigned int i; 3809 3810 /* 3811 * Perform vtype translation. 3812 */ 3813 for (i = 0; i < nitems(vtypes_table); i++) 3814 if (vtypes_table[i].vtype == vtype) 3815 return (vtypes_table[i].kf_vtype); 3816 3817 return (KF_VTYPE_UNKNOWN); 3818 } 3819 3820 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3821 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3822 "Process filedesc entries"); 3823 3824 /* 3825 * Store a process current working directory information to sbuf. 3826 * 3827 * Takes a locked proc as argument, and returns with the proc unlocked. 3828 */ 3829 int 3830 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3831 { 3832 struct filedesc *fdp; 3833 struct export_fd_buf *efbuf; 3834 int error; 3835 3836 PROC_LOCK_ASSERT(p, MA_OWNED); 3837 3838 fdp = fdhold(p); 3839 PROC_UNLOCK(p); 3840 if (fdp == NULL) 3841 return (EINVAL); 3842 3843 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3844 efbuf->fdp = fdp; 3845 efbuf->sb = sb; 3846 efbuf->remainder = maxlen; 3847 3848 FILEDESC_SLOCK(fdp); 3849 if (fdp->fd_cdir == NULL) 3850 error = EINVAL; 3851 else { 3852 vrefact(fdp->fd_cdir); 3853 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3854 FREAD, efbuf); 3855 } 3856 FILEDESC_SUNLOCK(fdp); 3857 fddrop(fdp); 3858 free(efbuf, M_TEMP); 3859 return (error); 3860 } 3861 3862 /* 3863 * Get per-process current working directory. 3864 */ 3865 static int 3866 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3867 { 3868 struct sbuf sb; 3869 struct proc *p; 3870 ssize_t maxlen; 3871 int error, error2, *name; 3872 3873 name = (int *)arg1; 3874 3875 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3876 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3877 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3878 if (error != 0) { 3879 sbuf_delete(&sb); 3880 return (error); 3881 } 3882 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3883 error = kern_proc_cwd_out(p, &sb, maxlen); 3884 error2 = sbuf_finish(&sb); 3885 sbuf_delete(&sb); 3886 return (error != 0 ? error : error2); 3887 } 3888 3889 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3890 sysctl_kern_proc_cwd, "Process current working directory"); 3891 3892 #ifdef DDB 3893 /* 3894 * For the purposes of debugging, generate a human-readable string for the 3895 * file type. 3896 */ 3897 static const char * 3898 file_type_to_name(short type) 3899 { 3900 3901 switch (type) { 3902 case 0: 3903 return ("zero"); 3904 case DTYPE_VNODE: 3905 return ("vnode"); 3906 case DTYPE_SOCKET: 3907 return ("socket"); 3908 case DTYPE_PIPE: 3909 return ("pipe"); 3910 case DTYPE_FIFO: 3911 return ("fifo"); 3912 case DTYPE_KQUEUE: 3913 return ("kqueue"); 3914 case DTYPE_CRYPTO: 3915 return ("crypto"); 3916 case DTYPE_MQUEUE: 3917 return ("mqueue"); 3918 case DTYPE_SHM: 3919 return ("shm"); 3920 case DTYPE_SEM: 3921 return ("ksem"); 3922 case DTYPE_PTS: 3923 return ("pts"); 3924 case DTYPE_DEV: 3925 return ("dev"); 3926 case DTYPE_PROCDESC: 3927 return ("proc"); 3928 case DTYPE_LINUXEFD: 3929 return ("levent"); 3930 case DTYPE_LINUXTFD: 3931 return ("ltimer"); 3932 default: 3933 return ("unkn"); 3934 } 3935 } 3936 3937 /* 3938 * For the purposes of debugging, identify a process (if any, perhaps one of 3939 * many) that references the passed file in its file descriptor array. Return 3940 * NULL if none. 3941 */ 3942 static struct proc * 3943 file_to_first_proc(struct file *fp) 3944 { 3945 struct filedesc *fdp; 3946 struct proc *p; 3947 int n; 3948 3949 FOREACH_PROC_IN_SYSTEM(p) { 3950 if (p->p_state == PRS_NEW) 3951 continue; 3952 fdp = p->p_fd; 3953 if (fdp == NULL) 3954 continue; 3955 for (n = 0; n <= fdp->fd_lastfile; n++) { 3956 if (fp == fdp->fd_ofiles[n].fde_file) 3957 return (p); 3958 } 3959 } 3960 return (NULL); 3961 } 3962 3963 static void 3964 db_print_file(struct file *fp, int header) 3965 { 3966 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 3967 struct proc *p; 3968 3969 if (header) 3970 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 3971 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 3972 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 3973 "FCmd"); 3974 p = file_to_first_proc(fp); 3975 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 3976 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 3977 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode, 3978 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3979 3980 #undef XPTRWIDTH 3981 } 3982 3983 DB_SHOW_COMMAND(file, db_show_file) 3984 { 3985 struct file *fp; 3986 3987 if (!have_addr) { 3988 db_printf("usage: show file <addr>\n"); 3989 return; 3990 } 3991 fp = (struct file *)addr; 3992 db_print_file(fp, 1); 3993 } 3994 3995 DB_SHOW_COMMAND(files, db_show_files) 3996 { 3997 struct filedesc *fdp; 3998 struct file *fp; 3999 struct proc *p; 4000 int header; 4001 int n; 4002 4003 header = 1; 4004 FOREACH_PROC_IN_SYSTEM(p) { 4005 if (p->p_state == PRS_NEW) 4006 continue; 4007 if ((fdp = p->p_fd) == NULL) 4008 continue; 4009 for (n = 0; n <= fdp->fd_lastfile; ++n) { 4010 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 4011 continue; 4012 db_print_file(fp, header); 4013 header = 0; 4014 } 4015 } 4016 } 4017 #endif 4018 4019 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 4020 &maxfilesperproc, 0, "Maximum files allowed open per process"); 4021 4022 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 4023 &maxfiles, 0, "Maximum number of files"); 4024 4025 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 4026 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 4027 4028 /* ARGSUSED*/ 4029 static void 4030 filelistinit(void *dummy) 4031 { 4032 4033 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 4034 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 4035 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 4036 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 4037 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 4038 } 4039 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 4040 4041 /*-------------------------------------------------------------------*/ 4042 4043 static int 4044 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 4045 int flags, struct thread *td) 4046 { 4047 4048 return (EBADF); 4049 } 4050 4051 static int 4052 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4053 struct thread *td) 4054 { 4055 4056 return (EINVAL); 4057 } 4058 4059 static int 4060 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 4061 struct thread *td) 4062 { 4063 4064 return (EBADF); 4065 } 4066 4067 static int 4068 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 4069 struct thread *td) 4070 { 4071 4072 return (0); 4073 } 4074 4075 static int 4076 badfo_kqfilter(struct file *fp, struct knote *kn) 4077 { 4078 4079 return (EBADF); 4080 } 4081 4082 static int 4083 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 4084 struct thread *td) 4085 { 4086 4087 return (EBADF); 4088 } 4089 4090 static int 4091 badfo_close(struct file *fp, struct thread *td) 4092 { 4093 4094 return (0); 4095 } 4096 4097 static int 4098 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4099 struct thread *td) 4100 { 4101 4102 return (EBADF); 4103 } 4104 4105 static int 4106 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4107 struct thread *td) 4108 { 4109 4110 return (EBADF); 4111 } 4112 4113 static int 4114 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4115 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4116 struct thread *td) 4117 { 4118 4119 return (EBADF); 4120 } 4121 4122 static int 4123 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 4124 { 4125 4126 return (0); 4127 } 4128 4129 struct fileops badfileops = { 4130 .fo_read = badfo_readwrite, 4131 .fo_write = badfo_readwrite, 4132 .fo_truncate = badfo_truncate, 4133 .fo_ioctl = badfo_ioctl, 4134 .fo_poll = badfo_poll, 4135 .fo_kqfilter = badfo_kqfilter, 4136 .fo_stat = badfo_stat, 4137 .fo_close = badfo_close, 4138 .fo_chmod = badfo_chmod, 4139 .fo_chown = badfo_chown, 4140 .fo_sendfile = badfo_sendfile, 4141 .fo_fill_kinfo = badfo_fill_kinfo, 4142 }; 4143 4144 int 4145 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 4146 int flags, struct thread *td) 4147 { 4148 4149 return (EOPNOTSUPP); 4150 } 4151 4152 int 4153 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4154 struct thread *td) 4155 { 4156 4157 return (EINVAL); 4158 } 4159 4160 int 4161 invfo_ioctl(struct file *fp, u_long com, void *data, 4162 struct ucred *active_cred, struct thread *td) 4163 { 4164 4165 return (ENOTTY); 4166 } 4167 4168 int 4169 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 4170 struct thread *td) 4171 { 4172 4173 return (poll_no_poll(events)); 4174 } 4175 4176 int 4177 invfo_kqfilter(struct file *fp, struct knote *kn) 4178 { 4179 4180 return (EINVAL); 4181 } 4182 4183 int 4184 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4185 struct thread *td) 4186 { 4187 4188 return (EINVAL); 4189 } 4190 4191 int 4192 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4193 struct thread *td) 4194 { 4195 4196 return (EINVAL); 4197 } 4198 4199 int 4200 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4201 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4202 struct thread *td) 4203 { 4204 4205 return (EINVAL); 4206 } 4207 4208 /*-------------------------------------------------------------------*/ 4209 4210 /* 4211 * File Descriptor pseudo-device driver (/dev/fd/). 4212 * 4213 * Opening minor device N dup()s the file (if any) connected to file 4214 * descriptor N belonging to the calling process. Note that this driver 4215 * consists of only the ``open()'' routine, because all subsequent 4216 * references to this file will be direct to the other driver. 4217 * 4218 * XXX: we could give this one a cloning event handler if necessary. 4219 */ 4220 4221 /* ARGSUSED */ 4222 static int 4223 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4224 { 4225 4226 /* 4227 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4228 * the file descriptor being sought for duplication. The error 4229 * return ensures that the vnode for this device will be released 4230 * by vn_open. Open will detect this special error and take the 4231 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4232 * will simply report the error. 4233 */ 4234 td->td_dupfd = dev2unit(dev); 4235 return (ENODEV); 4236 } 4237 4238 static struct cdevsw fildesc_cdevsw = { 4239 .d_version = D_VERSION, 4240 .d_open = fdopen, 4241 .d_name = "FD", 4242 }; 4243 4244 static void 4245 fildesc_drvinit(void *unused) 4246 { 4247 struct cdev *dev; 4248 4249 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4250 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4251 make_dev_alias(dev, "stdin"); 4252 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4253 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4254 make_dev_alias(dev, "stdout"); 4255 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4256 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4257 make_dev_alias(dev, "stderr"); 4258 } 4259 4260 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4261