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