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 37 #include "opt_capsicum.h" 38 #include "opt_ddb.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/systm.h> 42 #include <sys/capsicum.h> 43 #include <sys/conf.h> 44 #include <sys/fcntl.h> 45 #include <sys/file.h> 46 #include <sys/filedesc.h> 47 #include <sys/filio.h> 48 #include <sys/jail.h> 49 #include <sys/kernel.h> 50 #include <sys/limits.h> 51 #include <sys/lock.h> 52 #include <sys/malloc.h> 53 #include <sys/mount.h> 54 #include <sys/mutex.h> 55 #include <sys/namei.h> 56 #include <sys/selinfo.h> 57 #include <sys/poll.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/protosw.h> 61 #include <sys/racct.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sbuf.h> 64 #include <sys/signalvar.h> 65 #include <sys/kdb.h> 66 #include <sys/smr.h> 67 #include <sys/stat.h> 68 #include <sys/sx.h> 69 #include <sys/syscallsubr.h> 70 #include <sys/sysctl.h> 71 #include <sys/sysproto.h> 72 #include <sys/unistd.h> 73 #include <sys/user.h> 74 #include <sys/vnode.h> 75 #include <sys/ktrace.h> 76 77 #include <net/vnet.h> 78 79 #include <security/audit/audit.h> 80 81 #include <vm/uma.h> 82 #include <vm/vm.h> 83 84 #include <ddb/ddb.h> 85 86 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 87 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes"); 88 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors"); 89 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 90 "file desc to leader structures"); 91 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 92 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 93 94 MALLOC_DECLARE(M_FADVISE); 95 96 static __read_mostly uma_zone_t file_zone; 97 static __read_mostly uma_zone_t filedesc0_zone; 98 __read_mostly uma_zone_t pwd_zone; 99 VFS_SMR_DECLARE; 100 101 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 102 struct thread *td, bool holdleaders, bool audit); 103 static void export_file_to_kinfo(struct file *fp, int fd, 104 cap_rights_t *rightsp, struct kinfo_file *kif, 105 struct filedesc *fdp, int flags); 106 static int fd_first_free(struct filedesc *fdp, int low, int size); 107 static void fdgrowtable(struct filedesc *fdp, int nfd); 108 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 109 static void fdunused(struct filedesc *fdp, int fd); 110 static void fdused(struct filedesc *fdp, int fd); 111 static int fget_unlocked_seq(struct thread *td, int fd, 112 cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp); 113 static int getmaxfd(struct thread *td); 114 static u_long *filecaps_copy_prep(const struct filecaps *src); 115 static void filecaps_copy_finish(const struct filecaps *src, 116 struct filecaps *dst, u_long *ioctls); 117 static u_long *filecaps_free_prep(struct filecaps *fcaps); 118 static void filecaps_free_finish(u_long *ioctls); 119 120 static struct pwd *pwd_alloc(void); 121 122 /* 123 * Each process has: 124 * 125 * - An array of open file descriptors (fd_ofiles) 126 * - An array of file flags (fd_ofileflags) 127 * - A bitmap recording which descriptors are in use (fd_map) 128 * 129 * A process starts out with NDFILE descriptors. The value of NDFILE has 130 * been selected based the historical limit of 20 open files, and an 131 * assumption that the majority of processes, especially short-lived 132 * processes like shells, will never need more. 133 * 134 * If this initial allocation is exhausted, a larger descriptor table and 135 * map are allocated dynamically, and the pointers in the process's struct 136 * filedesc are updated to point to those. This is repeated every time 137 * the process runs out of file descriptors (provided it hasn't hit its 138 * resource limit). 139 * 140 * Since threads may hold references to individual descriptor table 141 * entries, the tables are never freed. Instead, they are placed on a 142 * linked list and freed only when the struct filedesc is released. 143 */ 144 #define NDFILE 20 145 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 146 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 147 #define NDSLOT(x) ((x) / NDENTRIES) 148 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 149 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 150 151 #define FILEDESC_FOREACH_FDE(fdp, _iterator, _fde) \ 152 struct filedesc *_fdp = (fdp); \ 153 int _lastfile = fdlastfile_single(_fdp); \ 154 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 155 if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL) 156 157 #define FILEDESC_FOREACH_FP(fdp, _iterator, _fp) \ 158 struct filedesc *_fdp = (fdp); \ 159 int _lastfile = fdlastfile_single(_fdp); \ 160 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 161 if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL) 162 163 /* 164 * SLIST entry used to keep track of ofiles which must be reclaimed when 165 * the process exits. 166 */ 167 struct freetable { 168 struct fdescenttbl *ft_table; 169 SLIST_ENTRY(freetable) ft_next; 170 }; 171 172 /* 173 * Initial allocation: a filedesc structure + the head of SLIST used to 174 * keep track of old ofiles + enough space for NDFILE descriptors. 175 */ 176 177 struct fdescenttbl0 { 178 int fdt_nfiles; 179 struct filedescent fdt_ofiles[NDFILE]; 180 }; 181 182 struct filedesc0 { 183 struct filedesc fd_fd; 184 SLIST_HEAD(, freetable) fd_free; 185 struct fdescenttbl0 fd_dfiles; 186 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 187 }; 188 189 /* 190 * Descriptor management. 191 */ 192 static int __exclusive_cache_line openfiles; /* actual number of open files */ 193 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 194 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 195 196 /* 197 * If low >= size, just return low. Otherwise find the first zero bit in the 198 * given bitmap, starting at low and not exceeding size - 1. Return size if 199 * not found. 200 */ 201 static int 202 fd_first_free(struct filedesc *fdp, int low, int size) 203 { 204 NDSLOTTYPE *map = fdp->fd_map; 205 NDSLOTTYPE mask; 206 int off, maxoff; 207 208 if (low >= size) 209 return (low); 210 211 off = NDSLOT(low); 212 if (low % NDENTRIES) { 213 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 214 if ((mask &= ~map[off]) != 0UL) 215 return (off * NDENTRIES + ffsl(mask) - 1); 216 ++off; 217 } 218 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 219 if (map[off] != ~0UL) 220 return (off * NDENTRIES + ffsl(~map[off]) - 1); 221 return (size); 222 } 223 224 /* 225 * Find the last used fd. 226 * 227 * Call this variant if fdp can't be modified by anyone else (e.g, during exec). 228 * Otherwise use fdlastfile. 229 */ 230 int 231 fdlastfile_single(struct filedesc *fdp) 232 { 233 NDSLOTTYPE *map = fdp->fd_map; 234 int off, minoff; 235 236 off = NDSLOT(fdp->fd_nfiles - 1); 237 for (minoff = NDSLOT(0); off >= minoff; --off) 238 if (map[off] != 0) 239 return (off * NDENTRIES + flsl(map[off]) - 1); 240 return (-1); 241 } 242 243 int 244 fdlastfile(struct filedesc *fdp) 245 { 246 247 FILEDESC_LOCK_ASSERT(fdp); 248 return (fdlastfile_single(fdp)); 249 } 250 251 static int 252 fdisused(struct filedesc *fdp, int fd) 253 { 254 255 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 256 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 257 258 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 259 } 260 261 /* 262 * Mark a file descriptor as used. 263 */ 264 static void 265 fdused_init(struct filedesc *fdp, int fd) 266 { 267 268 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 269 270 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 271 } 272 273 static void 274 fdused(struct filedesc *fdp, int fd) 275 { 276 277 FILEDESC_XLOCK_ASSERT(fdp); 278 279 fdused_init(fdp, fd); 280 if (fd == fdp->fd_freefile) 281 fdp->fd_freefile++; 282 } 283 284 /* 285 * Mark a file descriptor as unused. 286 */ 287 static void 288 fdunused(struct filedesc *fdp, int fd) 289 { 290 291 FILEDESC_XLOCK_ASSERT(fdp); 292 293 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 294 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 295 ("fd=%d is still in use", fd)); 296 297 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 298 if (fd < fdp->fd_freefile) 299 fdp->fd_freefile = fd; 300 } 301 302 /* 303 * Free a file descriptor. 304 * 305 * Avoid some work if fdp is about to be destroyed. 306 */ 307 static inline void 308 fdefree_last(struct filedescent *fde) 309 { 310 311 filecaps_free(&fde->fde_caps); 312 } 313 314 static inline void 315 fdfree(struct filedesc *fdp, int fd) 316 { 317 struct filedescent *fde; 318 319 FILEDESC_XLOCK_ASSERT(fdp); 320 fde = &fdp->fd_ofiles[fd]; 321 #ifdef CAPABILITIES 322 seqc_write_begin(&fde->fde_seqc); 323 #endif 324 fde->fde_file = NULL; 325 #ifdef CAPABILITIES 326 seqc_write_end(&fde->fde_seqc); 327 #endif 328 fdefree_last(fde); 329 fdunused(fdp, fd); 330 } 331 332 /* 333 * System calls on descriptors. 334 */ 335 #ifndef _SYS_SYSPROTO_H_ 336 struct getdtablesize_args { 337 int dummy; 338 }; 339 #endif 340 /* ARGSUSED */ 341 int 342 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 343 { 344 #ifdef RACCT 345 uint64_t lim; 346 #endif 347 348 td->td_retval[0] = getmaxfd(td); 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, intptr_t 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 *)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 *)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 *)arg, sizeof(ofl)); 474 } else if (cmd == F_GETLK) { 475 error = copyout(&fl, (void *)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 struct mount *mp; 490 struct kinfo_file *kif; 491 int error, flg, kif_sz, seals, tmp, got_set, got_cleared; 492 uint64_t bsize; 493 off_t foffset; 494 495 error = 0; 496 flg = F_POSIX; 497 p = td->td_proc; 498 fdp = p->p_fd; 499 500 AUDIT_ARG_FD(cmd); 501 AUDIT_ARG_CMD(cmd); 502 switch (cmd) { 503 case F_DUPFD: 504 tmp = arg; 505 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 506 break; 507 508 case F_DUPFD_CLOEXEC: 509 tmp = arg; 510 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 511 break; 512 513 case F_DUP2FD: 514 tmp = arg; 515 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 516 break; 517 518 case F_DUP2FD_CLOEXEC: 519 tmp = arg; 520 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 521 break; 522 523 case F_GETFD: 524 error = EBADF; 525 FILEDESC_SLOCK(fdp); 526 fde = fdeget_noref(fdp, fd); 527 if (fde != NULL) { 528 td->td_retval[0] = 529 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 530 error = 0; 531 } 532 FILEDESC_SUNLOCK(fdp); 533 break; 534 535 case F_SETFD: 536 error = EBADF; 537 FILEDESC_XLOCK(fdp); 538 fde = fdeget_noref(fdp, fd); 539 if (fde != NULL) { 540 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 541 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 542 error = 0; 543 } 544 FILEDESC_XUNLOCK(fdp); 545 break; 546 547 case F_GETFL: 548 error = fget_fcntl(td, fd, &cap_fcntl_rights, 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, &cap_fcntl_rights, F_SETFL, &fp); 557 if (error != 0) 558 break; 559 if (fp->f_ops == &path_fileops) { 560 fdrop(fp, td); 561 error = EBADF; 562 break; 563 } 564 do { 565 tmp = flg = fp->f_flag; 566 tmp &= ~FCNTLFLAGS; 567 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 568 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 569 got_set = tmp & ~flg; 570 got_cleared = flg & ~tmp; 571 tmp = fp->f_flag & FNONBLOCK; 572 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 573 if (error != 0) 574 goto revert_f_setfl; 575 tmp = fp->f_flag & FASYNC; 576 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 577 if (error == 0) { 578 fdrop(fp, td); 579 break; 580 } 581 atomic_clear_int(&fp->f_flag, FNONBLOCK); 582 tmp = 0; 583 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 584 revert_f_setfl: 585 do { 586 tmp = flg = fp->f_flag; 587 tmp &= ~FCNTLFLAGS; 588 tmp |= got_cleared; 589 tmp &= ~got_set; 590 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 591 fdrop(fp, td); 592 break; 593 594 case F_GETOWN: 595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 596 if (error != 0) 597 break; 598 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 599 if (error == 0) 600 td->td_retval[0] = tmp; 601 fdrop(fp, td); 602 break; 603 604 case F_SETOWN: 605 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 606 if (error != 0) 607 break; 608 tmp = arg; 609 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 610 fdrop(fp, td); 611 break; 612 613 case F_SETLK_REMOTE: 614 error = priv_check(td, PRIV_NFS_LOCKD); 615 if (error != 0) 616 return (error); 617 flg = F_REMOTE; 618 goto do_setlk; 619 620 case F_SETLKW: 621 flg |= F_WAIT; 622 /* FALLTHROUGH F_SETLK */ 623 624 case F_SETLK: 625 do_setlk: 626 flp = (struct flock *)arg; 627 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { 628 error = EINVAL; 629 break; 630 } 631 632 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 633 if (error != 0) 634 break; 635 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 636 error = EBADF; 637 fdrop(fp, td); 638 break; 639 } 640 641 if (flp->l_whence == SEEK_CUR) { 642 foffset = foffset_get(fp); 643 if (foffset < 0 || 644 (flp->l_start > 0 && 645 foffset > OFF_MAX - flp->l_start)) { 646 error = EOVERFLOW; 647 fdrop(fp, td); 648 break; 649 } 650 flp->l_start += foffset; 651 } 652 653 vp = fp->f_vnode; 654 switch (flp->l_type) { 655 case F_RDLCK: 656 if ((fp->f_flag & FREAD) == 0) { 657 error = EBADF; 658 break; 659 } 660 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 661 PROC_LOCK(p->p_leader); 662 p->p_leader->p_flag |= P_ADVLOCK; 663 PROC_UNLOCK(p->p_leader); 664 } 665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 666 flp, flg); 667 break; 668 case F_WRLCK: 669 if ((fp->f_flag & FWRITE) == 0) { 670 error = EBADF; 671 break; 672 } 673 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 674 PROC_LOCK(p->p_leader); 675 p->p_leader->p_flag |= P_ADVLOCK; 676 PROC_UNLOCK(p->p_leader); 677 } 678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 679 flp, flg); 680 break; 681 case F_UNLCK: 682 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 683 flp, flg); 684 break; 685 case F_UNLCKSYS: 686 if (flg != F_REMOTE) { 687 error = EINVAL; 688 break; 689 } 690 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 691 F_UNLCKSYS, flp, flg); 692 break; 693 default: 694 error = EINVAL; 695 break; 696 } 697 if (error != 0 || flp->l_type == F_UNLCK || 698 flp->l_type == F_UNLCKSYS) { 699 fdrop(fp, td); 700 break; 701 } 702 703 /* 704 * Check for a race with close. 705 * 706 * The vnode is now advisory locked (or unlocked, but this case 707 * is not really important) as the caller requested. 708 * We had to drop the filedesc lock, so we need to recheck if 709 * the descriptor is still valid, because if it was closed 710 * in the meantime we need to remove advisory lock from the 711 * vnode - close on any descriptor leading to an advisory 712 * locked vnode, removes that lock. 713 * We will return 0 on purpose in that case, as the result of 714 * successful advisory lock might have been externally visible 715 * already. This is fine - effectively we pretend to the caller 716 * that the closing thread was a bit slower and that the 717 * advisory lock succeeded before the close. 718 */ 719 error = fget_unlocked(td, fd, &cap_no_rights, &fp2); 720 if (error != 0) { 721 fdrop(fp, td); 722 break; 723 } 724 if (fp != fp2) { 725 flp->l_whence = SEEK_SET; 726 flp->l_start = 0; 727 flp->l_len = 0; 728 flp->l_type = F_UNLCK; 729 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 730 F_UNLCK, flp, F_POSIX); 731 } 732 fdrop(fp, td); 733 fdrop(fp2, td); 734 break; 735 736 case F_GETLK: 737 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 738 if (error != 0) 739 break; 740 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 741 error = EBADF; 742 fdrop(fp, td); 743 break; 744 } 745 flp = (struct flock *)arg; 746 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 747 flp->l_type != F_UNLCK) { 748 error = EINVAL; 749 fdrop(fp, td); 750 break; 751 } 752 if (flp->l_whence == SEEK_CUR) { 753 foffset = foffset_get(fp); 754 if ((flp->l_start > 0 && 755 foffset > OFF_MAX - flp->l_start) || 756 (flp->l_start < 0 && 757 foffset < OFF_MIN - flp->l_start)) { 758 error = EOVERFLOW; 759 fdrop(fp, td); 760 break; 761 } 762 flp->l_start += foffset; 763 } 764 vp = fp->f_vnode; 765 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 766 F_POSIX); 767 fdrop(fp, td); 768 break; 769 770 case F_ADD_SEALS: 771 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 772 if (error != 0) 773 break; 774 error = fo_add_seals(fp, arg); 775 fdrop(fp, td); 776 break; 777 778 case F_GET_SEALS: 779 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 780 if (error != 0) 781 break; 782 if (fo_get_seals(fp, &seals) == 0) 783 td->td_retval[0] = seals; 784 else 785 error = EINVAL; 786 fdrop(fp, td); 787 break; 788 789 case F_RDAHEAD: 790 arg = arg ? 128 * 1024: 0; 791 /* FALLTHROUGH */ 792 case F_READAHEAD: 793 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 794 if (error != 0) 795 break; 796 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 797 fdrop(fp, td); 798 error = EBADF; 799 break; 800 } 801 vp = fp->f_vnode; 802 if (vp->v_type != VREG) { 803 fdrop(fp, td); 804 error = ENOTTY; 805 break; 806 } 807 808 /* 809 * Exclusive lock synchronizes against f_seqcount reads and 810 * writes in sequential_heuristic(). 811 */ 812 error = vn_lock(vp, LK_EXCLUSIVE); 813 if (error != 0) { 814 fdrop(fp, td); 815 break; 816 } 817 if (arg >= 0) { 818 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 819 arg = MIN(arg, INT_MAX - bsize + 1); 820 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, 821 (arg + bsize - 1) / bsize); 822 atomic_set_int(&fp->f_flag, FRDAHEAD); 823 } else { 824 atomic_clear_int(&fp->f_flag, FRDAHEAD); 825 } 826 VOP_UNLOCK(vp); 827 fdrop(fp, td); 828 break; 829 830 case F_ISUNIONSTACK: 831 /* 832 * Check if the vnode is part of a union stack (either the 833 * "union" flag from mount(2) or unionfs). 834 * 835 * Prior to introduction of this op libc's readdir would call 836 * fstatfs(2), in effect unnecessarily copying kilobytes of 837 * data just to check fs name and a mount flag. 838 * 839 * Fixing the code to handle everything in the kernel instead 840 * is a non-trivial endeavor and has low priority, thus this 841 * horrible kludge facilitates the current behavior in a much 842 * cheaper manner until someone(tm) sorts this out. 843 */ 844 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 845 if (error != 0) 846 break; 847 if (fp->f_type != DTYPE_VNODE) { 848 fdrop(fp, td); 849 error = EBADF; 850 break; 851 } 852 vp = fp->f_vnode; 853 /* 854 * Since we don't prevent dooming the vnode even non-null mp 855 * found can become immediately stale. This is tolerable since 856 * mount points are type-stable (providing safe memory access) 857 * and any vfs op on this vnode going forward will return an 858 * error (meaning return value in this case is meaningless). 859 */ 860 mp = atomic_load_ptr(&vp->v_mount); 861 if (__predict_false(mp == NULL)) { 862 fdrop(fp, td); 863 error = EBADF; 864 break; 865 } 866 td->td_retval[0] = 0; 867 if (mp->mnt_kern_flag & MNTK_UNIONFS || 868 mp->mnt_flag & MNT_UNION) 869 td->td_retval[0] = 1; 870 fdrop(fp, td); 871 break; 872 873 case F_KINFO: 874 #ifdef CAPABILITY_MODE 875 if (CAP_TRACING(td)) 876 ktrcapfail(CAPFAIL_SYSCALL, &cmd); 877 if (IN_CAPABILITY_MODE(td)) { 878 error = ECAPMODE; 879 break; 880 } 881 #endif 882 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz)); 883 if (error != 0) 884 break; 885 if (kif_sz != sizeof(*kif)) { 886 error = EINVAL; 887 break; 888 } 889 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO); 890 FILEDESC_SLOCK(fdp); 891 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL); 892 if (error == 0 && fhold(fp)) { 893 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0); 894 FILEDESC_SUNLOCK(fdp); 895 fdrop(fp, td); 896 if ((kif->kf_status & KF_ATTR_VALID) != 0) { 897 kif->kf_structsize = sizeof(*kif); 898 error = copyout(kif, (void *)arg, sizeof(*kif)); 899 } else { 900 error = EBADF; 901 } 902 } else { 903 FILEDESC_SUNLOCK(fdp); 904 if (error == 0) 905 error = EBADF; 906 } 907 free(kif, M_TEMP); 908 break; 909 910 default: 911 error = EINVAL; 912 break; 913 } 914 return (error); 915 } 916 917 static int 918 getmaxfd(struct thread *td) 919 { 920 921 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 922 } 923 924 /* 925 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 926 */ 927 int 928 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 929 { 930 struct filedesc *fdp; 931 struct filedescent *oldfde, *newfde; 932 struct proc *p; 933 struct file *delfp, *oldfp; 934 u_long *oioctls, *nioctls; 935 int error, maxfd; 936 937 p = td->td_proc; 938 fdp = p->p_fd; 939 oioctls = NULL; 940 941 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 942 MPASS(mode < FDDUP_LASTMODE); 943 944 AUDIT_ARG_FD(old); 945 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 946 947 /* 948 * Verify we have a valid descriptor to dup from and possibly to 949 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 950 * return EINVAL when the new descriptor is out of bounds. 951 */ 952 if (old < 0) 953 return (EBADF); 954 if (new < 0) 955 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 956 maxfd = getmaxfd(td); 957 if (new >= maxfd) 958 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 959 960 error = EBADF; 961 FILEDESC_XLOCK(fdp); 962 if (fget_noref(fdp, old) == NULL) 963 goto unlock; 964 if (mode == FDDUP_FIXED && old == new) { 965 td->td_retval[0] = new; 966 if (flags & FDDUP_FLAG_CLOEXEC) 967 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 968 error = 0; 969 goto unlock; 970 } 971 972 oldfde = &fdp->fd_ofiles[old]; 973 oldfp = oldfde->fde_file; 974 if (!fhold(oldfp)) 975 goto unlock; 976 977 /* 978 * If the caller specified a file descriptor, make sure the file 979 * table is large enough to hold it, and grab it. Otherwise, just 980 * allocate a new descriptor the usual way. 981 */ 982 switch (mode) { 983 case FDDUP_NORMAL: 984 case FDDUP_FCNTL: 985 if ((error = fdalloc(td, new, &new)) != 0) { 986 fdrop(oldfp, td); 987 goto unlock; 988 } 989 break; 990 case FDDUP_FIXED: 991 if (new >= fdp->fd_nfiles) { 992 /* 993 * The resource limits are here instead of e.g. 994 * fdalloc(), because the file descriptor table may be 995 * shared between processes, so we can't really use 996 * racct_add()/racct_sub(). Instead of counting the 997 * number of actually allocated descriptors, just put 998 * the limit on the size of the file descriptor table. 999 */ 1000 #ifdef RACCT 1001 if (RACCT_ENABLED()) { 1002 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); 1003 if (error != 0) { 1004 error = EMFILE; 1005 fdrop(oldfp, td); 1006 goto unlock; 1007 } 1008 } 1009 #endif 1010 fdgrowtable_exp(fdp, new + 1); 1011 } 1012 if (!fdisused(fdp, new)) 1013 fdused(fdp, new); 1014 break; 1015 default: 1016 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 1017 } 1018 1019 KASSERT(old != new, ("new fd is same as old")); 1020 1021 /* Refetch oldfde because the table may have grown and old one freed. */ 1022 oldfde = &fdp->fd_ofiles[old]; 1023 KASSERT(oldfp == oldfde->fde_file, 1024 ("fdt_ofiles shift from growth observed at fd %d", 1025 old)); 1026 1027 newfde = &fdp->fd_ofiles[new]; 1028 delfp = newfde->fde_file; 1029 1030 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 1031 1032 /* 1033 * Duplicate the source descriptor. 1034 */ 1035 #ifdef CAPABILITIES 1036 seqc_write_begin(&newfde->fde_seqc); 1037 #endif 1038 oioctls = filecaps_free_prep(&newfde->fde_caps); 1039 fde_copy(oldfde, newfde); 1040 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 1041 nioctls); 1042 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 1043 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 1044 else 1045 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 1046 #ifdef CAPABILITIES 1047 seqc_write_end(&newfde->fde_seqc); 1048 #endif 1049 td->td_retval[0] = new; 1050 1051 error = 0; 1052 1053 if (delfp != NULL) { 1054 (void) closefp(fdp, new, delfp, td, true, false); 1055 FILEDESC_UNLOCK_ASSERT(fdp); 1056 } else { 1057 unlock: 1058 FILEDESC_XUNLOCK(fdp); 1059 } 1060 1061 filecaps_free_finish(oioctls); 1062 return (error); 1063 } 1064 1065 static void 1066 sigiofree(struct sigio *sigio) 1067 { 1068 crfree(sigio->sio_ucred); 1069 free(sigio, M_SIGIO); 1070 } 1071 1072 static struct sigio * 1073 funsetown_locked(struct sigio *sigio) 1074 { 1075 struct proc *p; 1076 struct pgrp *pg; 1077 1078 SIGIO_ASSERT_LOCKED(); 1079 1080 if (sigio == NULL) 1081 return (NULL); 1082 *sigio->sio_myref = NULL; 1083 if (sigio->sio_pgid < 0) { 1084 pg = sigio->sio_pgrp; 1085 PGRP_LOCK(pg); 1086 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); 1087 PGRP_UNLOCK(pg); 1088 } else { 1089 p = sigio->sio_proc; 1090 PROC_LOCK(p); 1091 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 1092 PROC_UNLOCK(p); 1093 } 1094 return (sigio); 1095 } 1096 1097 /* 1098 * If sigio is on the list associated with a process or process group, 1099 * disable signalling from the device, remove sigio from the list and 1100 * free sigio. 1101 */ 1102 void 1103 funsetown(struct sigio **sigiop) 1104 { 1105 struct sigio *sigio; 1106 1107 /* Racy check, consumers must provide synchronization. */ 1108 if (*sigiop == NULL) 1109 return; 1110 1111 SIGIO_LOCK(); 1112 sigio = funsetown_locked(*sigiop); 1113 SIGIO_UNLOCK(); 1114 if (sigio != NULL) 1115 sigiofree(sigio); 1116 } 1117 1118 /* 1119 * Free a list of sigio structures. The caller must ensure that new sigio 1120 * structures cannot be added after this point. For process groups this is 1121 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves 1122 * as an interlock. 1123 */ 1124 void 1125 funsetownlst(struct sigiolst *sigiolst) 1126 { 1127 struct proc *p; 1128 struct pgrp *pg; 1129 struct sigio *sigio, *tmp; 1130 1131 /* Racy check. */ 1132 sigio = SLIST_FIRST(sigiolst); 1133 if (sigio == NULL) 1134 return; 1135 1136 p = NULL; 1137 pg = NULL; 1138 1139 SIGIO_LOCK(); 1140 sigio = SLIST_FIRST(sigiolst); 1141 if (sigio == NULL) { 1142 SIGIO_UNLOCK(); 1143 return; 1144 } 1145 1146 /* 1147 * Every entry of the list should belong to a single proc or pgrp. 1148 */ 1149 if (sigio->sio_pgid < 0) { 1150 pg = sigio->sio_pgrp; 1151 sx_assert(&proctree_lock, SX_XLOCKED); 1152 PGRP_LOCK(pg); 1153 } else /* if (sigio->sio_pgid > 0) */ { 1154 p = sigio->sio_proc; 1155 PROC_LOCK(p); 1156 KASSERT((p->p_flag & P_WEXIT) != 0, 1157 ("%s: process %p is not exiting", __func__, p)); 1158 } 1159 1160 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) { 1161 *sigio->sio_myref = NULL; 1162 if (pg != NULL) { 1163 KASSERT(sigio->sio_pgid < 0, 1164 ("Proc sigio in pgrp sigio list")); 1165 KASSERT(sigio->sio_pgrp == pg, 1166 ("Bogus pgrp in sigio list")); 1167 } else /* if (p != NULL) */ { 1168 KASSERT(sigio->sio_pgid > 0, 1169 ("Pgrp sigio in proc sigio list")); 1170 KASSERT(sigio->sio_proc == p, 1171 ("Bogus proc in sigio list")); 1172 } 1173 } 1174 1175 if (pg != NULL) 1176 PGRP_UNLOCK(pg); 1177 else 1178 PROC_UNLOCK(p); 1179 SIGIO_UNLOCK(); 1180 1181 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp) 1182 sigiofree(sigio); 1183 } 1184 1185 /* 1186 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1187 * 1188 * After permission checking, add a sigio structure to the sigio list for 1189 * the process or process group. 1190 */ 1191 int 1192 fsetown(pid_t pgid, struct sigio **sigiop) 1193 { 1194 struct proc *proc; 1195 struct pgrp *pgrp; 1196 struct sigio *osigio, *sigio; 1197 int ret; 1198 1199 if (pgid == 0) { 1200 funsetown(sigiop); 1201 return (0); 1202 } 1203 1204 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1205 sigio->sio_pgid = pgid; 1206 sigio->sio_ucred = crhold(curthread->td_ucred); 1207 sigio->sio_myref = sigiop; 1208 1209 ret = 0; 1210 if (pgid > 0) { 1211 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc); 1212 SIGIO_LOCK(); 1213 osigio = funsetown_locked(*sigiop); 1214 if (ret == 0) { 1215 PROC_LOCK(proc); 1216 _PRELE(proc); 1217 if ((proc->p_flag & P_WEXIT) != 0) { 1218 ret = ESRCH; 1219 } else if (proc->p_session != 1220 curthread->td_proc->p_session) { 1221 /* 1222 * Policy - Don't allow a process to FSETOWN a 1223 * process in another session. 1224 * 1225 * Remove this test to allow maximum flexibility 1226 * or restrict FSETOWN to the current process or 1227 * process group for maximum safety. 1228 */ 1229 ret = EPERM; 1230 } else { 1231 sigio->sio_proc = proc; 1232 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, 1233 sio_pgsigio); 1234 } 1235 PROC_UNLOCK(proc); 1236 } 1237 } else /* if (pgid < 0) */ { 1238 sx_slock(&proctree_lock); 1239 SIGIO_LOCK(); 1240 osigio = funsetown_locked(*sigiop); 1241 pgrp = pgfind(-pgid); 1242 if (pgrp == NULL) { 1243 ret = ESRCH; 1244 } else { 1245 if (pgrp->pg_session != curthread->td_proc->p_session) { 1246 /* 1247 * Policy - Don't allow a process to FSETOWN a 1248 * process in another session. 1249 * 1250 * Remove this test to allow maximum flexibility 1251 * or restrict FSETOWN to the current process or 1252 * process group for maximum safety. 1253 */ 1254 ret = EPERM; 1255 } else { 1256 sigio->sio_pgrp = pgrp; 1257 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, 1258 sio_pgsigio); 1259 } 1260 PGRP_UNLOCK(pgrp); 1261 } 1262 sx_sunlock(&proctree_lock); 1263 } 1264 if (ret == 0) 1265 *sigiop = sigio; 1266 SIGIO_UNLOCK(); 1267 if (osigio != NULL) 1268 sigiofree(osigio); 1269 return (ret); 1270 } 1271 1272 /* 1273 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1274 */ 1275 pid_t 1276 fgetown(struct sigio **sigiop) 1277 { 1278 pid_t pgid; 1279 1280 SIGIO_LOCK(); 1281 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1282 SIGIO_UNLOCK(); 1283 return (pgid); 1284 } 1285 1286 static int 1287 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1288 bool audit) 1289 { 1290 int error; 1291 1292 FILEDESC_XLOCK_ASSERT(fdp); 1293 1294 /* 1295 * We now hold the fp reference that used to be owned by the 1296 * descriptor array. We have to unlock the FILEDESC *AFTER* 1297 * knote_fdclose to prevent a race of the fd getting opened, a knote 1298 * added, and deleteing a knote for the new fd. 1299 */ 1300 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1301 knote_fdclose(td, fd); 1302 1303 /* 1304 * We need to notify mqueue if the object is of type mqueue. 1305 */ 1306 if (__predict_false(fp->f_type == DTYPE_MQUEUE)) 1307 mq_fdclose(td, fd, fp); 1308 FILEDESC_XUNLOCK(fdp); 1309 1310 #ifdef AUDIT 1311 if (AUDITING_TD(td) && audit) 1312 audit_sysclose(td, fd, fp); 1313 #endif 1314 error = closef(fp, td); 1315 1316 /* 1317 * All paths leading up to closefp() will have already removed or 1318 * replaced the fd in the filedesc table, so a restart would not 1319 * operate on the same file. 1320 */ 1321 if (error == ERESTART) 1322 error = EINTR; 1323 1324 return (error); 1325 } 1326 1327 static int 1328 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1329 bool holdleaders, bool audit) 1330 { 1331 int error; 1332 1333 FILEDESC_XLOCK_ASSERT(fdp); 1334 1335 if (holdleaders) { 1336 if (td->td_proc->p_fdtol != NULL) { 1337 /* 1338 * Ask fdfree() to sleep to ensure that all relevant 1339 * process leaders can be traversed in closef(). 1340 */ 1341 fdp->fd_holdleaderscount++; 1342 } else { 1343 holdleaders = false; 1344 } 1345 } 1346 1347 error = closefp_impl(fdp, fd, fp, td, audit); 1348 if (holdleaders) { 1349 FILEDESC_XLOCK(fdp); 1350 fdp->fd_holdleaderscount--; 1351 if (fdp->fd_holdleaderscount == 0 && 1352 fdp->fd_holdleaderswakeup != 0) { 1353 fdp->fd_holdleaderswakeup = 0; 1354 wakeup(&fdp->fd_holdleaderscount); 1355 } 1356 FILEDESC_XUNLOCK(fdp); 1357 } 1358 return (error); 1359 } 1360 1361 static int 1362 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1363 bool holdleaders, bool audit) 1364 { 1365 1366 FILEDESC_XLOCK_ASSERT(fdp); 1367 1368 if (__predict_false(td->td_proc->p_fdtol != NULL)) { 1369 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit)); 1370 } else { 1371 return (closefp_impl(fdp, fd, fp, td, audit)); 1372 } 1373 } 1374 1375 /* 1376 * Close a file descriptor. 1377 */ 1378 #ifndef _SYS_SYSPROTO_H_ 1379 struct close_args { 1380 int fd; 1381 }; 1382 #endif 1383 /* ARGSUSED */ 1384 int 1385 sys_close(struct thread *td, struct close_args *uap) 1386 { 1387 1388 return (kern_close(td, uap->fd)); 1389 } 1390 1391 int 1392 kern_close(struct thread *td, int fd) 1393 { 1394 struct filedesc *fdp; 1395 struct file *fp; 1396 1397 fdp = td->td_proc->p_fd; 1398 1399 FILEDESC_XLOCK(fdp); 1400 if ((fp = fget_noref(fdp, fd)) == NULL) { 1401 FILEDESC_XUNLOCK(fdp); 1402 return (EBADF); 1403 } 1404 fdfree(fdp, fd); 1405 1406 /* closefp() drops the FILEDESC lock for us. */ 1407 return (closefp(fdp, fd, fp, td, true, true)); 1408 } 1409 1410 static int 1411 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd) 1412 { 1413 struct filedesc *fdp; 1414 struct fdescenttbl *fdt; 1415 struct filedescent *fde; 1416 int fd; 1417 1418 fdp = td->td_proc->p_fd; 1419 FILEDESC_XLOCK(fdp); 1420 fdt = atomic_load_ptr(&fdp->fd_files); 1421 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1422 fd = lowfd; 1423 if (__predict_false(fd > highfd)) { 1424 goto out_locked; 1425 } 1426 for (; fd <= highfd; fd++) { 1427 fde = &fdt->fdt_ofiles[fd]; 1428 if (fde->fde_file != NULL) 1429 fde->fde_flags |= UF_EXCLOSE; 1430 } 1431 out_locked: 1432 FILEDESC_XUNLOCK(fdp); 1433 return (0); 1434 } 1435 1436 static int 1437 close_range_impl(struct thread *td, u_int lowfd, u_int highfd) 1438 { 1439 struct filedesc *fdp; 1440 const struct fdescenttbl *fdt; 1441 struct file *fp; 1442 int fd; 1443 1444 fdp = td->td_proc->p_fd; 1445 FILEDESC_XLOCK(fdp); 1446 fdt = atomic_load_ptr(&fdp->fd_files); 1447 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1448 fd = lowfd; 1449 if (__predict_false(fd > highfd)) { 1450 goto out_locked; 1451 } 1452 for (;;) { 1453 fp = fdt->fdt_ofiles[fd].fde_file; 1454 if (fp == NULL) { 1455 if (fd == highfd) 1456 goto out_locked; 1457 } else { 1458 fdfree(fdp, fd); 1459 (void) closefp(fdp, fd, fp, td, true, true); 1460 if (fd == highfd) 1461 goto out_unlocked; 1462 FILEDESC_XLOCK(fdp); 1463 fdt = atomic_load_ptr(&fdp->fd_files); 1464 } 1465 fd++; 1466 } 1467 out_locked: 1468 FILEDESC_XUNLOCK(fdp); 1469 out_unlocked: 1470 return (0); 1471 } 1472 1473 int 1474 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd) 1475 { 1476 1477 /* 1478 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 1479 * open should not be a usage error. From a close_range() perspective, 1480 * close_range(3, ~0U, 0) in the same scenario should also likely not 1481 * be a usage error as all fd above 3 are in-fact already closed. 1482 */ 1483 if (highfd < lowfd) { 1484 return (EINVAL); 1485 } 1486 1487 if ((flags & CLOSE_RANGE_CLOEXEC) != 0) 1488 return (close_range_cloexec(td, lowfd, highfd)); 1489 1490 return (close_range_impl(td, lowfd, highfd)); 1491 } 1492 1493 #ifndef _SYS_SYSPROTO_H_ 1494 struct close_range_args { 1495 u_int lowfd; 1496 u_int highfd; 1497 int flags; 1498 }; 1499 #endif 1500 int 1501 sys_close_range(struct thread *td, struct close_range_args *uap) 1502 { 1503 1504 AUDIT_ARG_FD(uap->lowfd); 1505 AUDIT_ARG_CMD(uap->highfd); 1506 AUDIT_ARG_FFLAGS(uap->flags); 1507 1508 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0) 1509 return (EINVAL); 1510 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd)); 1511 } 1512 1513 #ifdef COMPAT_FREEBSD12 1514 /* 1515 * Close open file descriptors. 1516 */ 1517 #ifndef _SYS_SYSPROTO_H_ 1518 struct freebsd12_closefrom_args { 1519 int lowfd; 1520 }; 1521 #endif 1522 /* ARGSUSED */ 1523 int 1524 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) 1525 { 1526 u_int lowfd; 1527 1528 AUDIT_ARG_FD(uap->lowfd); 1529 1530 /* 1531 * Treat negative starting file descriptor values identical to 1532 * closefrom(0) which closes all files. 1533 */ 1534 lowfd = MAX(0, uap->lowfd); 1535 return (kern_close_range(td, 0, lowfd, ~0U)); 1536 } 1537 #endif /* COMPAT_FREEBSD12 */ 1538 1539 #if defined(COMPAT_43) 1540 /* 1541 * Return status information about a file descriptor. 1542 */ 1543 #ifndef _SYS_SYSPROTO_H_ 1544 struct ofstat_args { 1545 int fd; 1546 struct ostat *sb; 1547 }; 1548 #endif 1549 /* ARGSUSED */ 1550 int 1551 ofstat(struct thread *td, struct ofstat_args *uap) 1552 { 1553 struct ostat oub; 1554 struct stat ub; 1555 int error; 1556 1557 error = kern_fstat(td, uap->fd, &ub); 1558 if (error == 0) { 1559 cvtstat(&ub, &oub); 1560 error = copyout(&oub, uap->sb, sizeof(oub)); 1561 } 1562 return (error); 1563 } 1564 #endif /* COMPAT_43 */ 1565 1566 #if defined(COMPAT_FREEBSD11) 1567 int 1568 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1569 { 1570 struct stat sb; 1571 struct freebsd11_stat osb; 1572 int error; 1573 1574 error = kern_fstat(td, uap->fd, &sb); 1575 if (error != 0) 1576 return (error); 1577 error = freebsd11_cvtstat(&sb, &osb); 1578 if (error == 0) 1579 error = copyout(&osb, uap->sb, sizeof(osb)); 1580 return (error); 1581 } 1582 #endif /* COMPAT_FREEBSD11 */ 1583 1584 /* 1585 * Return status information about a file descriptor. 1586 */ 1587 #ifndef _SYS_SYSPROTO_H_ 1588 struct fstat_args { 1589 int fd; 1590 struct stat *sb; 1591 }; 1592 #endif 1593 /* ARGSUSED */ 1594 int 1595 sys_fstat(struct thread *td, struct fstat_args *uap) 1596 { 1597 struct stat ub; 1598 int error; 1599 1600 error = kern_fstat(td, uap->fd, &ub); 1601 if (error == 0) 1602 error = copyout(&ub, uap->sb, sizeof(ub)); 1603 return (error); 1604 } 1605 1606 int 1607 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1608 { 1609 struct file *fp; 1610 int error; 1611 1612 AUDIT_ARG_FD(fd); 1613 1614 error = fget(td, fd, &cap_fstat_rights, &fp); 1615 if (__predict_false(error != 0)) 1616 return (error); 1617 1618 AUDIT_ARG_FILE(td->td_proc, fp); 1619 1620 sbp->st_filerev = 0; 1621 sbp->st_bsdflags = 0; 1622 error = fo_stat(fp, sbp, td->td_ucred); 1623 fdrop(fp, td); 1624 #ifdef __STAT_TIME_T_EXT 1625 sbp->st_atim_ext = 0; 1626 sbp->st_mtim_ext = 0; 1627 sbp->st_ctim_ext = 0; 1628 sbp->st_btim_ext = 0; 1629 #endif 1630 #ifdef KTRACE 1631 if (KTRPOINT(td, KTR_STRUCT)) 1632 ktrstat_error(sbp, error); 1633 #endif 1634 return (error); 1635 } 1636 1637 #if defined(COMPAT_FREEBSD11) 1638 /* 1639 * Return status information about a file descriptor. 1640 */ 1641 #ifndef _SYS_SYSPROTO_H_ 1642 struct freebsd11_nfstat_args { 1643 int fd; 1644 struct nstat *sb; 1645 }; 1646 #endif 1647 /* ARGSUSED */ 1648 int 1649 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1650 { 1651 struct nstat nub; 1652 struct stat ub; 1653 int error; 1654 1655 error = kern_fstat(td, uap->fd, &ub); 1656 if (error != 0) 1657 return (error); 1658 error = freebsd11_cvtnstat(&ub, &nub); 1659 if (error != 0) 1660 error = copyout(&nub, uap->sb, sizeof(nub)); 1661 return (error); 1662 } 1663 #endif /* COMPAT_FREEBSD11 */ 1664 1665 /* 1666 * Return pathconf information about a file descriptor. 1667 */ 1668 #ifndef _SYS_SYSPROTO_H_ 1669 struct fpathconf_args { 1670 int fd; 1671 int name; 1672 }; 1673 #endif 1674 /* ARGSUSED */ 1675 int 1676 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1677 { 1678 long value; 1679 int error; 1680 1681 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1682 if (error == 0) 1683 td->td_retval[0] = value; 1684 return (error); 1685 } 1686 1687 int 1688 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1689 { 1690 struct file *fp; 1691 struct vnode *vp; 1692 int error; 1693 1694 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1695 if (error != 0) 1696 return (error); 1697 1698 if (name == _PC_ASYNC_IO) { 1699 *valuep = _POSIX_ASYNCHRONOUS_IO; 1700 goto out; 1701 } 1702 vp = fp->f_vnode; 1703 if (vp != NULL) { 1704 vn_lock(vp, LK_SHARED | LK_RETRY); 1705 error = VOP_PATHCONF(vp, name, valuep); 1706 VOP_UNLOCK(vp); 1707 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1708 if (name != _PC_PIPE_BUF) { 1709 error = EINVAL; 1710 } else { 1711 *valuep = PIPE_BUF; 1712 error = 0; 1713 } 1714 } else { 1715 error = EOPNOTSUPP; 1716 } 1717 out: 1718 fdrop(fp, td); 1719 return (error); 1720 } 1721 1722 /* 1723 * Copy filecaps structure allocating memory for ioctls array if needed. 1724 * 1725 * The last parameter indicates whether the fdtable is locked. If it is not and 1726 * ioctls are encountered, copying fails and the caller must lock the table. 1727 * 1728 * Note that if the table was not locked, the caller has to check the relevant 1729 * sequence counter to determine whether the operation was successful. 1730 */ 1731 bool 1732 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1733 { 1734 size_t size; 1735 1736 if (src->fc_ioctls != NULL && !locked) 1737 return (false); 1738 memcpy(dst, src, sizeof(*src)); 1739 if (src->fc_ioctls == NULL) 1740 return (true); 1741 1742 KASSERT(src->fc_nioctls > 0, 1743 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1744 1745 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1746 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1747 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1748 return (true); 1749 } 1750 1751 static u_long * 1752 filecaps_copy_prep(const struct filecaps *src) 1753 { 1754 u_long *ioctls; 1755 size_t size; 1756 1757 if (__predict_true(src->fc_ioctls == NULL)) 1758 return (NULL); 1759 1760 KASSERT(src->fc_nioctls > 0, 1761 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1762 1763 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1764 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1765 return (ioctls); 1766 } 1767 1768 static void 1769 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1770 u_long *ioctls) 1771 { 1772 size_t size; 1773 1774 *dst = *src; 1775 if (__predict_true(src->fc_ioctls == NULL)) { 1776 MPASS(ioctls == NULL); 1777 return; 1778 } 1779 1780 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1781 dst->fc_ioctls = ioctls; 1782 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1783 } 1784 1785 /* 1786 * Move filecaps structure to the new place and clear the old place. 1787 */ 1788 void 1789 filecaps_move(struct filecaps *src, struct filecaps *dst) 1790 { 1791 1792 *dst = *src; 1793 bzero(src, sizeof(*src)); 1794 } 1795 1796 /* 1797 * Fill the given filecaps structure with full rights. 1798 */ 1799 static void 1800 filecaps_fill(struct filecaps *fcaps) 1801 { 1802 1803 CAP_ALL(&fcaps->fc_rights); 1804 fcaps->fc_ioctls = NULL; 1805 fcaps->fc_nioctls = -1; 1806 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1807 } 1808 1809 /* 1810 * Free memory allocated within filecaps structure. 1811 */ 1812 static void 1813 filecaps_free_ioctl(struct filecaps *fcaps) 1814 { 1815 1816 free(fcaps->fc_ioctls, M_FILECAPS); 1817 fcaps->fc_ioctls = NULL; 1818 } 1819 1820 void 1821 filecaps_free(struct filecaps *fcaps) 1822 { 1823 1824 filecaps_free_ioctl(fcaps); 1825 bzero(fcaps, sizeof(*fcaps)); 1826 } 1827 1828 static u_long * 1829 filecaps_free_prep(struct filecaps *fcaps) 1830 { 1831 u_long *ioctls; 1832 1833 ioctls = fcaps->fc_ioctls; 1834 bzero(fcaps, sizeof(*fcaps)); 1835 return (ioctls); 1836 } 1837 1838 static void 1839 filecaps_free_finish(u_long *ioctls) 1840 { 1841 1842 free(ioctls, M_FILECAPS); 1843 } 1844 1845 /* 1846 * Validate the given filecaps structure. 1847 */ 1848 static void 1849 filecaps_validate(const struct filecaps *fcaps, const char *func) 1850 { 1851 1852 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1853 ("%s: invalid rights", func)); 1854 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1855 ("%s: invalid fcntls", func)); 1856 KASSERT(fcaps->fc_fcntls == 0 || 1857 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1858 ("%s: fcntls without CAP_FCNTL", func)); 1859 /* 1860 * open calls without WANTIOCTLCAPS free caps but leave the counter 1861 */ 1862 #if 0 1863 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1864 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1865 ("%s: invalid ioctls", func)); 1866 #endif 1867 KASSERT(fcaps->fc_nioctls == 0 || 1868 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1869 ("%s: ioctls without CAP_IOCTL", func)); 1870 } 1871 1872 static void 1873 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1874 { 1875 int nfd1; 1876 1877 FILEDESC_XLOCK_ASSERT(fdp); 1878 1879 nfd1 = fdp->fd_nfiles * 2; 1880 if (nfd1 < nfd) 1881 nfd1 = nfd; 1882 fdgrowtable(fdp, nfd1); 1883 } 1884 1885 /* 1886 * Grow the file table to accommodate (at least) nfd descriptors. 1887 */ 1888 static void 1889 fdgrowtable(struct filedesc *fdp, int nfd) 1890 { 1891 struct filedesc0 *fdp0; 1892 struct freetable *ft; 1893 struct fdescenttbl *ntable; 1894 struct fdescenttbl *otable; 1895 int nnfiles, onfiles; 1896 NDSLOTTYPE *nmap, *omap; 1897 1898 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1899 1900 /* save old values */ 1901 onfiles = fdp->fd_nfiles; 1902 otable = fdp->fd_files; 1903 omap = fdp->fd_map; 1904 1905 /* compute the size of the new table */ 1906 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1907 if (nnfiles <= onfiles) 1908 /* the table is already large enough */ 1909 return; 1910 1911 /* 1912 * Allocate a new table. We need enough space for the number of 1913 * entries, file entries themselves and the struct freetable we will use 1914 * when we decommission the table and place it on the freelist. 1915 * We place the struct freetable in the middle so we don't have 1916 * to worry about padding. 1917 */ 1918 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1919 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1920 sizeof(struct freetable), 1921 M_FILEDESC, M_ZERO | M_WAITOK); 1922 /* copy the old data */ 1923 ntable->fdt_nfiles = nnfiles; 1924 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1925 onfiles * sizeof(ntable->fdt_ofiles[0])); 1926 1927 /* 1928 * Allocate a new map only if the old is not large enough. It will 1929 * grow at a slower rate than the table as it can map more 1930 * entries than the table can hold. 1931 */ 1932 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1933 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1934 M_ZERO | M_WAITOK); 1935 /* copy over the old data and update the pointer */ 1936 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1937 fdp->fd_map = nmap; 1938 } 1939 1940 /* 1941 * Make sure that ntable is correctly initialized before we replace 1942 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1943 * data. 1944 */ 1945 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1946 1947 /* 1948 * Free the old file table when not shared by other threads or processes. 1949 * The old file table is considered to be shared when either are true: 1950 * - The process has more than one thread. 1951 * - The file descriptor table has been shared via fdshare(). 1952 * 1953 * When shared, the old file table will be placed on a freelist 1954 * which will be processed when the struct filedesc is released. 1955 * 1956 * Note that if onfiles == NDFILE, we're dealing with the original 1957 * static allocation contained within (struct filedesc0 *)fdp, 1958 * which must not be freed. 1959 */ 1960 if (onfiles > NDFILE) { 1961 /* 1962 * Note we may be called here from fdinit while allocating a 1963 * table for a new process in which case ->p_fd points 1964 * elsewhere. 1965 */ 1966 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 1967 free(otable, M_FILEDESC); 1968 } else { 1969 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1970 fdp0 = (struct filedesc0 *)fdp; 1971 ft->ft_table = otable; 1972 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1973 } 1974 } 1975 /* 1976 * The map does not have the same possibility of threads still 1977 * holding references to it. So always free it as long as it 1978 * does not reference the original static allocation. 1979 */ 1980 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1981 free(omap, M_FILEDESC); 1982 } 1983 1984 /* 1985 * Allocate a file descriptor for the process. 1986 */ 1987 int 1988 fdalloc(struct thread *td, int minfd, int *result) 1989 { 1990 struct proc *p = td->td_proc; 1991 struct filedesc *fdp = p->p_fd; 1992 int fd, maxfd, allocfd; 1993 #ifdef RACCT 1994 int error; 1995 #endif 1996 1997 FILEDESC_XLOCK_ASSERT(fdp); 1998 1999 if (fdp->fd_freefile > minfd) 2000 minfd = fdp->fd_freefile; 2001 2002 maxfd = getmaxfd(td); 2003 2004 /* 2005 * Search the bitmap for a free descriptor starting at minfd. 2006 * If none is found, grow the file table. 2007 */ 2008 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 2009 if (__predict_false(fd >= maxfd)) 2010 return (EMFILE); 2011 if (__predict_false(fd >= fdp->fd_nfiles)) { 2012 allocfd = min(fd * 2, maxfd); 2013 #ifdef RACCT 2014 if (RACCT_ENABLED()) { 2015 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 2016 if (error != 0) 2017 return (EMFILE); 2018 } 2019 #endif 2020 /* 2021 * fd is already equal to first free descriptor >= minfd, so 2022 * we only need to grow the table and we are done. 2023 */ 2024 fdgrowtable_exp(fdp, allocfd); 2025 } 2026 2027 /* 2028 * Perform some sanity checks, then mark the file descriptor as 2029 * used and return it to the caller. 2030 */ 2031 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 2032 ("invalid descriptor %d", fd)); 2033 KASSERT(!fdisused(fdp, fd), 2034 ("fd_first_free() returned non-free descriptor")); 2035 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 2036 ("file descriptor isn't free")); 2037 fdused(fdp, fd); 2038 *result = fd; 2039 return (0); 2040 } 2041 2042 /* 2043 * Allocate n file descriptors for the process. 2044 */ 2045 int 2046 fdallocn(struct thread *td, int minfd, int *fds, int n) 2047 { 2048 struct proc *p = td->td_proc; 2049 struct filedesc *fdp = p->p_fd; 2050 int i; 2051 2052 FILEDESC_XLOCK_ASSERT(fdp); 2053 2054 for (i = 0; i < n; i++) 2055 if (fdalloc(td, 0, &fds[i]) != 0) 2056 break; 2057 2058 if (i < n) { 2059 for (i--; i >= 0; i--) 2060 fdunused(fdp, fds[i]); 2061 return (EMFILE); 2062 } 2063 2064 return (0); 2065 } 2066 2067 /* 2068 * Create a new open file structure and allocate a file descriptor for the 2069 * process that refers to it. We add one reference to the file for the 2070 * descriptor table and one reference for resultfp. This is to prevent us 2071 * being preempted and the entry in the descriptor table closed after we 2072 * release the FILEDESC lock. 2073 */ 2074 int 2075 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 2076 struct filecaps *fcaps) 2077 { 2078 struct file *fp; 2079 int error, fd; 2080 2081 MPASS(resultfp != NULL); 2082 MPASS(resultfd != NULL); 2083 2084 error = _falloc_noinstall(td, &fp, 2); 2085 if (__predict_false(error != 0)) { 2086 return (error); 2087 } 2088 2089 error = finstall_refed(td, fp, &fd, flags, fcaps); 2090 if (__predict_false(error != 0)) { 2091 falloc_abort(td, fp); 2092 return (error); 2093 } 2094 2095 *resultfp = fp; 2096 *resultfd = fd; 2097 2098 return (0); 2099 } 2100 2101 /* 2102 * Create a new open file structure without allocating a file descriptor. 2103 */ 2104 int 2105 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2106 { 2107 struct file *fp; 2108 int maxuserfiles = maxfiles - (maxfiles / 20); 2109 int openfiles_new; 2110 static struct timeval lastfail; 2111 static int curfail; 2112 2113 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2114 MPASS(n > 0); 2115 2116 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2117 if ((openfiles_new >= maxuserfiles && 2118 priv_check(td, PRIV_MAXFILES) != 0) || 2119 openfiles_new >= maxfiles) { 2120 atomic_subtract_int(&openfiles, 1); 2121 if (ppsratecheck(&lastfail, &curfail, 1)) { 2122 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2123 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2124 } 2125 return (ENFILE); 2126 } 2127 fp = uma_zalloc(file_zone, M_WAITOK); 2128 bzero(fp, sizeof(*fp)); 2129 refcount_init(&fp->f_count, n); 2130 fp->f_cred = crhold(td->td_ucred); 2131 fp->f_ops = &badfileops; 2132 *resultfp = fp; 2133 return (0); 2134 } 2135 2136 void 2137 falloc_abort(struct thread *td, struct file *fp) 2138 { 2139 2140 /* 2141 * For assertion purposes. 2142 */ 2143 refcount_init(&fp->f_count, 0); 2144 _fdrop(fp, td); 2145 } 2146 2147 /* 2148 * Install a file in a file descriptor table. 2149 */ 2150 void 2151 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2152 struct filecaps *fcaps) 2153 { 2154 struct filedescent *fde; 2155 2156 MPASS(fp != NULL); 2157 if (fcaps != NULL) 2158 filecaps_validate(fcaps, __func__); 2159 FILEDESC_XLOCK_ASSERT(fdp); 2160 2161 fde = &fdp->fd_ofiles[fd]; 2162 #ifdef CAPABILITIES 2163 seqc_write_begin(&fde->fde_seqc); 2164 #endif 2165 fde->fde_file = fp; 2166 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 2167 if (fcaps != NULL) 2168 filecaps_move(fcaps, &fde->fde_caps); 2169 else 2170 filecaps_fill(&fde->fde_caps); 2171 #ifdef CAPABILITIES 2172 seqc_write_end(&fde->fde_seqc); 2173 #endif 2174 } 2175 2176 int 2177 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2178 struct filecaps *fcaps) 2179 { 2180 struct filedesc *fdp = td->td_proc->p_fd; 2181 int error; 2182 2183 MPASS(fd != NULL); 2184 2185 FILEDESC_XLOCK(fdp); 2186 error = fdalloc(td, 0, fd); 2187 if (__predict_true(error == 0)) { 2188 _finstall(fdp, fp, *fd, flags, fcaps); 2189 } 2190 FILEDESC_XUNLOCK(fdp); 2191 return (error); 2192 } 2193 2194 int 2195 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2196 struct filecaps *fcaps) 2197 { 2198 int error; 2199 2200 MPASS(fd != NULL); 2201 2202 if (!fhold(fp)) 2203 return (EBADF); 2204 error = finstall_refed(td, fp, fd, flags, fcaps); 2205 if (__predict_false(error != 0)) { 2206 fdrop(fp, td); 2207 } 2208 return (error); 2209 } 2210 2211 /* 2212 * Build a new filedesc structure from another. 2213 * 2214 * If fdp is not NULL, return with it shared locked. 2215 */ 2216 struct filedesc * 2217 fdinit(void) 2218 { 2219 struct filedesc0 *newfdp0; 2220 struct filedesc *newfdp; 2221 2222 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2223 newfdp = &newfdp0->fd_fd; 2224 2225 /* Create the file descriptor table. */ 2226 FILEDESC_LOCK_INIT(newfdp); 2227 refcount_init(&newfdp->fd_refcnt, 1); 2228 refcount_init(&newfdp->fd_holdcnt, 1); 2229 newfdp->fd_map = newfdp0->fd_dmap; 2230 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2231 newfdp->fd_files->fdt_nfiles = NDFILE; 2232 2233 return (newfdp); 2234 } 2235 2236 /* 2237 * Build a pwddesc structure from another. 2238 * Copy the current, root, and jail root vnode references. 2239 * 2240 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked. 2241 */ 2242 struct pwddesc * 2243 pdinit(struct pwddesc *pdp, bool keeplock) 2244 { 2245 struct pwddesc *newpdp; 2246 struct pwd *newpwd; 2247 2248 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2249 2250 PWDDESC_LOCK_INIT(newpdp); 2251 refcount_init(&newpdp->pd_refcount, 1); 2252 newpdp->pd_cmask = CMASK; 2253 2254 if (pdp == NULL) { 2255 newpwd = pwd_alloc(); 2256 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2257 return (newpdp); 2258 } 2259 2260 PWDDESC_XLOCK(pdp); 2261 newpwd = pwd_hold_pwddesc(pdp); 2262 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2263 if (!keeplock) 2264 PWDDESC_XUNLOCK(pdp); 2265 return (newpdp); 2266 } 2267 2268 /* 2269 * Hold either filedesc or pwddesc of the passed process. 2270 * 2271 * The process lock is used to synchronize against the target exiting and 2272 * freeing the data. 2273 * 2274 * Clearing can be ilustrated in 3 steps: 2275 * 1. set the pointer to NULL. Either routine can race against it, hence 2276 * atomic_load_ptr. 2277 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2278 * race to either still see the pointer or find NULL. It is still safe to 2279 * grab a reference as clearing is stalled. 2280 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2281 * guaranteed to see NULL, making it safe to finish clearing 2282 */ 2283 static struct filedesc * 2284 fdhold(struct proc *p) 2285 { 2286 struct filedesc *fdp; 2287 2288 PROC_LOCK_ASSERT(p, MA_OWNED); 2289 fdp = atomic_load_ptr(&p->p_fd); 2290 if (fdp != NULL) 2291 refcount_acquire(&fdp->fd_holdcnt); 2292 return (fdp); 2293 } 2294 2295 static struct pwddesc * 2296 pdhold(struct proc *p) 2297 { 2298 struct pwddesc *pdp; 2299 2300 PROC_LOCK_ASSERT(p, MA_OWNED); 2301 pdp = atomic_load_ptr(&p->p_pd); 2302 if (pdp != NULL) 2303 refcount_acquire(&pdp->pd_refcount); 2304 return (pdp); 2305 } 2306 2307 static void 2308 fddrop(struct filedesc *fdp) 2309 { 2310 2311 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2312 if (refcount_release(&fdp->fd_holdcnt) == 0) 2313 return; 2314 } 2315 2316 FILEDESC_LOCK_DESTROY(fdp); 2317 uma_zfree(filedesc0_zone, fdp); 2318 } 2319 2320 static void 2321 pddrop(struct pwddesc *pdp) 2322 { 2323 struct pwd *pwd; 2324 2325 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2326 return; 2327 2328 PWDDESC_XLOCK(pdp); 2329 if (refcount_release(&pdp->pd_refcount) == 0) { 2330 PWDDESC_XUNLOCK(pdp); 2331 return; 2332 } 2333 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2334 pwd_set(pdp, NULL); 2335 PWDDESC_XUNLOCK(pdp); 2336 pwd_drop(pwd); 2337 2338 PWDDESC_LOCK_DESTROY(pdp); 2339 free(pdp, M_PWDDESC); 2340 } 2341 2342 /* 2343 * Share a filedesc structure. 2344 */ 2345 struct filedesc * 2346 fdshare(struct filedesc *fdp) 2347 { 2348 2349 refcount_acquire(&fdp->fd_refcnt); 2350 return (fdp); 2351 } 2352 2353 /* 2354 * Share a pwddesc structure. 2355 */ 2356 struct pwddesc * 2357 pdshare(struct pwddesc *pdp) 2358 { 2359 refcount_acquire(&pdp->pd_refcount); 2360 return (pdp); 2361 } 2362 2363 /* 2364 * Unshare a filedesc structure, if necessary by making a copy 2365 */ 2366 void 2367 fdunshare(struct thread *td) 2368 { 2369 struct filedesc *tmp; 2370 struct proc *p = td->td_proc; 2371 2372 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2373 return; 2374 2375 tmp = fdcopy(p->p_fd); 2376 fdescfree(td); 2377 p->p_fd = tmp; 2378 } 2379 2380 /* 2381 * Unshare a pwddesc structure. 2382 */ 2383 void 2384 pdunshare(struct thread *td) 2385 { 2386 struct pwddesc *pdp; 2387 struct proc *p; 2388 2389 p = td->td_proc; 2390 /* Not shared. */ 2391 if (refcount_load(&p->p_pd->pd_refcount) == 1) 2392 return; 2393 2394 pdp = pdcopy(p->p_pd); 2395 pdescfree(td); 2396 p->p_pd = pdp; 2397 } 2398 2399 /* 2400 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2401 * this is to ease callers, not catch errors. 2402 */ 2403 struct filedesc * 2404 fdcopy(struct filedesc *fdp) 2405 { 2406 struct filedesc *newfdp; 2407 struct filedescent *nfde, *ofde; 2408 int i, lastfile; 2409 2410 MPASS(fdp != NULL); 2411 2412 newfdp = fdinit(); 2413 FILEDESC_SLOCK(fdp); 2414 for (;;) { 2415 lastfile = fdlastfile(fdp); 2416 if (lastfile < newfdp->fd_nfiles) 2417 break; 2418 FILEDESC_SUNLOCK(fdp); 2419 fdgrowtable(newfdp, lastfile + 1); 2420 FILEDESC_SLOCK(fdp); 2421 } 2422 /* copy all passable descriptors (i.e. not kqueue) */ 2423 newfdp->fd_freefile = fdp->fd_freefile; 2424 FILEDESC_FOREACH_FDE(fdp, i, ofde) { 2425 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 || 2426 !fhold(ofde->fde_file)) { 2427 if (newfdp->fd_freefile == fdp->fd_freefile) 2428 newfdp->fd_freefile = i; 2429 continue; 2430 } 2431 nfde = &newfdp->fd_ofiles[i]; 2432 *nfde = *ofde; 2433 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2434 fdused_init(newfdp, i); 2435 } 2436 MPASS(newfdp->fd_freefile != -1); 2437 FILEDESC_SUNLOCK(fdp); 2438 return (newfdp); 2439 } 2440 2441 /* 2442 * Copy a pwddesc structure. 2443 */ 2444 struct pwddesc * 2445 pdcopy(struct pwddesc *pdp) 2446 { 2447 struct pwddesc *newpdp; 2448 2449 MPASS(pdp != NULL); 2450 2451 newpdp = pdinit(pdp, true); 2452 newpdp->pd_cmask = pdp->pd_cmask; 2453 PWDDESC_XUNLOCK(pdp); 2454 return (newpdp); 2455 } 2456 2457 /* 2458 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2459 * one of processes using it exits) and the table used to be shared. 2460 */ 2461 static void 2462 fdclearlocks(struct thread *td) 2463 { 2464 struct filedesc *fdp; 2465 struct filedesc_to_leader *fdtol; 2466 struct flock lf; 2467 struct file *fp; 2468 struct proc *p; 2469 struct vnode *vp; 2470 int i; 2471 2472 p = td->td_proc; 2473 fdp = p->p_fd; 2474 fdtol = p->p_fdtol; 2475 MPASS(fdtol != NULL); 2476 2477 FILEDESC_XLOCK(fdp); 2478 KASSERT(fdtol->fdl_refcount > 0, 2479 ("filedesc_to_refcount botch: fdl_refcount=%d", 2480 fdtol->fdl_refcount)); 2481 if (fdtol->fdl_refcount == 1 && 2482 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2483 FILEDESC_FOREACH_FP(fdp, i, fp) { 2484 if (fp->f_type != DTYPE_VNODE || 2485 !fhold(fp)) 2486 continue; 2487 FILEDESC_XUNLOCK(fdp); 2488 lf.l_whence = SEEK_SET; 2489 lf.l_start = 0; 2490 lf.l_len = 0; 2491 lf.l_type = F_UNLCK; 2492 vp = fp->f_vnode; 2493 (void) VOP_ADVLOCK(vp, 2494 (caddr_t)p->p_leader, F_UNLCK, 2495 &lf, F_POSIX); 2496 FILEDESC_XLOCK(fdp); 2497 fdrop(fp, td); 2498 } 2499 } 2500 retry: 2501 if (fdtol->fdl_refcount == 1) { 2502 if (fdp->fd_holdleaderscount > 0 && 2503 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2504 /* 2505 * close() or kern_dup() has cleared a reference 2506 * in a shared file descriptor table. 2507 */ 2508 fdp->fd_holdleaderswakeup = 1; 2509 sx_sleep(&fdp->fd_holdleaderscount, 2510 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2511 goto retry; 2512 } 2513 if (fdtol->fdl_holdcount > 0) { 2514 /* 2515 * Ensure that fdtol->fdl_leader remains 2516 * valid in closef(). 2517 */ 2518 fdtol->fdl_wakeup = 1; 2519 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2520 "fdlhold", 0); 2521 goto retry; 2522 } 2523 } 2524 fdtol->fdl_refcount--; 2525 if (fdtol->fdl_refcount == 0 && 2526 fdtol->fdl_holdcount == 0) { 2527 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2528 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2529 } else 2530 fdtol = NULL; 2531 p->p_fdtol = NULL; 2532 FILEDESC_XUNLOCK(fdp); 2533 if (fdtol != NULL) 2534 free(fdtol, M_FILEDESC_TO_LEADER); 2535 } 2536 2537 /* 2538 * Release a filedesc structure. 2539 */ 2540 static void 2541 fdescfree_fds(struct thread *td, struct filedesc *fdp) 2542 { 2543 struct filedesc0 *fdp0; 2544 struct freetable *ft, *tft; 2545 struct filedescent *fde; 2546 struct file *fp; 2547 int i; 2548 2549 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2550 ("%s: fd table %p carries references", __func__, fdp)); 2551 2552 /* 2553 * Serialize with threads iterating over the table, if any. 2554 */ 2555 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2556 FILEDESC_XLOCK(fdp); 2557 FILEDESC_XUNLOCK(fdp); 2558 } 2559 2560 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2561 fp = fde->fde_file; 2562 fdefree_last(fde); 2563 (void) closef(fp, td); 2564 } 2565 2566 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2567 free(fdp->fd_map, M_FILEDESC); 2568 if (fdp->fd_nfiles > NDFILE) 2569 free(fdp->fd_files, M_FILEDESC); 2570 2571 fdp0 = (struct filedesc0 *)fdp; 2572 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2573 free(ft->ft_table, M_FILEDESC); 2574 2575 fddrop(fdp); 2576 } 2577 2578 void 2579 fdescfree(struct thread *td) 2580 { 2581 struct proc *p; 2582 struct filedesc *fdp; 2583 2584 p = td->td_proc; 2585 fdp = p->p_fd; 2586 MPASS(fdp != NULL); 2587 2588 #ifdef RACCT 2589 if (RACCT_ENABLED()) 2590 racct_set_unlocked(p, RACCT_NOFILE, 0); 2591 #endif 2592 2593 if (p->p_fdtol != NULL) 2594 fdclearlocks(td); 2595 2596 /* 2597 * Check fdhold for an explanation. 2598 */ 2599 atomic_store_ptr(&p->p_fd, NULL); 2600 atomic_thread_fence_seq_cst(); 2601 PROC_WAIT_UNLOCKED(p); 2602 2603 if (refcount_release(&fdp->fd_refcnt) == 0) 2604 return; 2605 2606 fdescfree_fds(td, fdp); 2607 } 2608 2609 void 2610 pdescfree(struct thread *td) 2611 { 2612 struct proc *p; 2613 struct pwddesc *pdp; 2614 2615 p = td->td_proc; 2616 pdp = p->p_pd; 2617 MPASS(pdp != NULL); 2618 2619 /* 2620 * Check pdhold for an explanation. 2621 */ 2622 atomic_store_ptr(&p->p_pd, NULL); 2623 atomic_thread_fence_seq_cst(); 2624 PROC_WAIT_UNLOCKED(p); 2625 2626 pddrop(pdp); 2627 } 2628 2629 /* 2630 * For setugid programs, we don't want to people to use that setugidness 2631 * to generate error messages which write to a file which otherwise would 2632 * otherwise be off-limits to the process. We check for filesystems where 2633 * the vnode can change out from under us after execve (like [lin]procfs). 2634 * 2635 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2636 * sufficient. We also don't check for setugidness since we know we are. 2637 */ 2638 static bool 2639 is_unsafe(struct file *fp) 2640 { 2641 struct vnode *vp; 2642 2643 if (fp->f_type != DTYPE_VNODE) 2644 return (false); 2645 2646 vp = fp->f_vnode; 2647 return ((vp->v_vflag & VV_PROCDEP) != 0); 2648 } 2649 2650 /* 2651 * Make this setguid thing safe, if at all possible. 2652 */ 2653 void 2654 fdsetugidsafety(struct thread *td) 2655 { 2656 struct filedesc *fdp; 2657 struct file *fp; 2658 int i; 2659 2660 fdp = td->td_proc->p_fd; 2661 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2662 ("the fdtable should not be shared")); 2663 MPASS(fdp->fd_nfiles >= 3); 2664 for (i = 0; i <= 2; i++) { 2665 fp = fdp->fd_ofiles[i].fde_file; 2666 if (fp != NULL && is_unsafe(fp)) { 2667 FILEDESC_XLOCK(fdp); 2668 knote_fdclose(td, i); 2669 /* 2670 * NULL-out descriptor prior to close to avoid 2671 * a race while close blocks. 2672 */ 2673 fdfree(fdp, i); 2674 FILEDESC_XUNLOCK(fdp); 2675 (void) closef(fp, td); 2676 } 2677 } 2678 } 2679 2680 /* 2681 * If a specific file object occupies a specific file descriptor, close the 2682 * file descriptor entry and drop a reference on the file object. This is a 2683 * convenience function to handle a subsequent error in a function that calls 2684 * falloc() that handles the race that another thread might have closed the 2685 * file descriptor out from under the thread creating the file object. 2686 */ 2687 void 2688 fdclose(struct thread *td, struct file *fp, int idx) 2689 { 2690 struct filedesc *fdp = td->td_proc->p_fd; 2691 2692 FILEDESC_XLOCK(fdp); 2693 if (fdp->fd_ofiles[idx].fde_file == fp) { 2694 fdfree(fdp, idx); 2695 FILEDESC_XUNLOCK(fdp); 2696 fdrop(fp, td); 2697 } else 2698 FILEDESC_XUNLOCK(fdp); 2699 } 2700 2701 /* 2702 * Close any files on exec? 2703 */ 2704 void 2705 fdcloseexec(struct thread *td) 2706 { 2707 struct filedesc *fdp; 2708 struct filedescent *fde; 2709 struct file *fp; 2710 int i; 2711 2712 fdp = td->td_proc->p_fd; 2713 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2714 ("the fdtable should not be shared")); 2715 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2716 fp = fde->fde_file; 2717 if (fp->f_type == DTYPE_MQUEUE || 2718 (fde->fde_flags & UF_EXCLOSE)) { 2719 FILEDESC_XLOCK(fdp); 2720 fdfree(fdp, i); 2721 (void) closefp(fdp, i, fp, td, false, false); 2722 FILEDESC_UNLOCK_ASSERT(fdp); 2723 } 2724 } 2725 } 2726 2727 /* 2728 * It is unsafe for set[ug]id processes to be started with file 2729 * descriptors 0..2 closed, as these descriptors are given implicit 2730 * significance in the Standard C library. fdcheckstd() will create a 2731 * descriptor referencing /dev/null for each of stdin, stdout, and 2732 * stderr that is not already open. 2733 */ 2734 int 2735 fdcheckstd(struct thread *td) 2736 { 2737 struct filedesc *fdp; 2738 register_t save; 2739 int i, error, devnull; 2740 2741 fdp = td->td_proc->p_fd; 2742 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2743 ("the fdtable should not be shared")); 2744 MPASS(fdp->fd_nfiles >= 3); 2745 devnull = -1; 2746 for (i = 0; i <= 2; i++) { 2747 if (fdp->fd_ofiles[i].fde_file != NULL) 2748 continue; 2749 2750 save = td->td_retval[0]; 2751 if (devnull != -1) { 2752 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2753 } else { 2754 error = kern_openat(td, AT_FDCWD, "/dev/null", 2755 UIO_SYSSPACE, O_RDWR, 0); 2756 if (error == 0) { 2757 devnull = td->td_retval[0]; 2758 KASSERT(devnull == i, ("we didn't get our fd")); 2759 } 2760 } 2761 td->td_retval[0] = save; 2762 if (error != 0) 2763 return (error); 2764 } 2765 return (0); 2766 } 2767 2768 /* 2769 * Internal form of close. Decrement reference count on file structure. 2770 * Note: td may be NULL when closing a file that was being passed in a 2771 * message. 2772 */ 2773 int 2774 closef(struct file *fp, struct thread *td) 2775 { 2776 struct vnode *vp; 2777 struct flock lf; 2778 struct filedesc_to_leader *fdtol; 2779 struct filedesc *fdp; 2780 2781 MPASS(td != NULL); 2782 2783 /* 2784 * POSIX record locking dictates that any close releases ALL 2785 * locks owned by this process. This is handled by setting 2786 * a flag in the unlock to free ONLY locks obeying POSIX 2787 * semantics, and not to free BSD-style file locks. 2788 * If the descriptor was in a message, POSIX-style locks 2789 * aren't passed with the descriptor, and the thread pointer 2790 * will be NULL. Callers should be careful only to pass a 2791 * NULL thread pointer when there really is no owning 2792 * context that might have locks, or the locks will be 2793 * leaked. 2794 */ 2795 if (fp->f_type == DTYPE_VNODE) { 2796 vp = fp->f_vnode; 2797 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2798 lf.l_whence = SEEK_SET; 2799 lf.l_start = 0; 2800 lf.l_len = 0; 2801 lf.l_type = F_UNLCK; 2802 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2803 F_UNLCK, &lf, F_POSIX); 2804 } 2805 fdtol = td->td_proc->p_fdtol; 2806 if (fdtol != NULL) { 2807 /* 2808 * Handle special case where file descriptor table is 2809 * shared between multiple process leaders. 2810 */ 2811 fdp = td->td_proc->p_fd; 2812 FILEDESC_XLOCK(fdp); 2813 for (fdtol = fdtol->fdl_next; 2814 fdtol != td->td_proc->p_fdtol; 2815 fdtol = fdtol->fdl_next) { 2816 if ((fdtol->fdl_leader->p_flag & 2817 P_ADVLOCK) == 0) 2818 continue; 2819 fdtol->fdl_holdcount++; 2820 FILEDESC_XUNLOCK(fdp); 2821 lf.l_whence = SEEK_SET; 2822 lf.l_start = 0; 2823 lf.l_len = 0; 2824 lf.l_type = F_UNLCK; 2825 vp = fp->f_vnode; 2826 (void) VOP_ADVLOCK(vp, 2827 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2828 F_POSIX); 2829 FILEDESC_XLOCK(fdp); 2830 fdtol->fdl_holdcount--; 2831 if (fdtol->fdl_holdcount == 0 && 2832 fdtol->fdl_wakeup != 0) { 2833 fdtol->fdl_wakeup = 0; 2834 wakeup(fdtol); 2835 } 2836 } 2837 FILEDESC_XUNLOCK(fdp); 2838 } 2839 } 2840 return (fdrop_close(fp, td)); 2841 } 2842 2843 /* 2844 * Hack for file descriptor passing code. 2845 */ 2846 void 2847 closef_nothread(struct file *fp) 2848 { 2849 2850 fdrop(fp, NULL); 2851 } 2852 2853 /* 2854 * Initialize the file pointer with the specified properties. 2855 * 2856 * The ops are set with release semantics to be certain that the flags, type, 2857 * and data are visible when ops is. This is to prevent ops methods from being 2858 * called with bad data. 2859 */ 2860 void 2861 finit(struct file *fp, u_int flag, short type, void *data, 2862 const struct fileops *ops) 2863 { 2864 fp->f_data = data; 2865 fp->f_flag = flag; 2866 fp->f_type = type; 2867 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2868 } 2869 2870 void 2871 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops) 2872 { 2873 fp->f_seqcount[UIO_READ] = 1; 2874 fp->f_seqcount[UIO_WRITE] = 1; 2875 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 2876 data, ops); 2877 } 2878 2879 int 2880 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2881 struct file **fpp, struct filecaps *havecapsp) 2882 { 2883 struct filedescent *fde; 2884 int error; 2885 2886 FILEDESC_LOCK_ASSERT(fdp); 2887 2888 *fpp = NULL; 2889 fde = fdeget_noref(fdp, fd); 2890 if (fde == NULL) { 2891 error = EBADF; 2892 goto out; 2893 } 2894 2895 #ifdef CAPABILITIES 2896 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 2897 if (error != 0) 2898 goto out; 2899 #endif 2900 2901 if (havecapsp != NULL) 2902 filecaps_copy(&fde->fde_caps, havecapsp, true); 2903 2904 *fpp = fde->fde_file; 2905 2906 error = 0; 2907 out: 2908 return (error); 2909 } 2910 2911 #ifdef CAPABILITIES 2912 int 2913 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2914 struct file **fpp, struct filecaps *havecapsp) 2915 { 2916 struct filedesc *fdp = td->td_proc->p_fd; 2917 int error; 2918 struct file *fp; 2919 seqc_t seq; 2920 2921 *fpp = NULL; 2922 for (;;) { 2923 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq); 2924 if (error != 0) 2925 return (error); 2926 2927 if (havecapsp != NULL) { 2928 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2929 havecapsp, false)) { 2930 fdrop(fp, td); 2931 goto get_locked; 2932 } 2933 } 2934 2935 if (!fd_modified(fdp, fd, seq)) 2936 break; 2937 fdrop(fp, td); 2938 } 2939 2940 *fpp = fp; 2941 return (0); 2942 2943 get_locked: 2944 FILEDESC_SLOCK(fdp); 2945 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp); 2946 if (error == 0 && !fhold(*fpp)) 2947 error = EBADF; 2948 FILEDESC_SUNLOCK(fdp); 2949 return (error); 2950 } 2951 #else 2952 int 2953 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2954 struct file **fpp, struct filecaps *havecapsp) 2955 { 2956 int error; 2957 error = fget_unlocked(td, fd, needrightsp, fpp); 2958 if (havecapsp != NULL && error == 0) 2959 filecaps_fill(havecapsp); 2960 2961 return (error); 2962 } 2963 #endif 2964 2965 int 2966 fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp) 2967 { 2968 struct filedesc *fdp; 2969 struct file *fp; 2970 int error; 2971 2972 if (p == td->td_proc) /* curproc */ 2973 return (fget_unlocked(td, fd, &cap_no_rights, fpp)); 2974 2975 PROC_LOCK(p); 2976 fdp = fdhold(p); 2977 PROC_UNLOCK(p); 2978 if (fdp == NULL) 2979 return (ENOENT); 2980 FILEDESC_SLOCK(fdp); 2981 if (refcount_load(&fdp->fd_refcnt) != 0) { 2982 fp = fget_noref(fdp, fd); 2983 if (fp != NULL && fhold(fp)) { 2984 *fpp = fp; 2985 error = 0; 2986 } else { 2987 error = EBADF; 2988 } 2989 } else { 2990 error = ENOENT; 2991 } 2992 FILEDESC_SUNLOCK(fdp); 2993 fddrop(fdp); 2994 return (error); 2995 } 2996 2997 int 2998 fget_remote_foreach(struct thread *td, struct proc *p, 2999 int (*fn)(struct proc *, int, struct file *, void *), void *arg) 3000 { 3001 struct filedesc *fdp; 3002 struct fdescenttbl *fdt; 3003 struct file *fp; 3004 int error, error1, fd, highfd; 3005 3006 error = 0; 3007 PROC_LOCK(p); 3008 fdp = fdhold(p); 3009 PROC_UNLOCK(p); 3010 if (fdp == NULL) 3011 return (ENOENT); 3012 3013 FILEDESC_SLOCK(fdp); 3014 if (refcount_load(&fdp->fd_refcnt) != 0) { 3015 fdt = atomic_load_ptr(&fdp->fd_files); 3016 highfd = fdt->fdt_nfiles - 1; 3017 FILEDESC_SUNLOCK(fdp); 3018 } else { 3019 error = ENOENT; 3020 FILEDESC_SUNLOCK(fdp); 3021 goto out; 3022 } 3023 3024 for (fd = 0; fd <= highfd; fd++) { 3025 error1 = fget_remote(td, p, fd, &fp); 3026 if (error1 != 0) 3027 continue; 3028 error = fn(p, fd, fp, arg); 3029 fdrop(fp, td); 3030 if (error != 0) 3031 break; 3032 } 3033 out: 3034 fddrop(fdp); 3035 return (error); 3036 } 3037 3038 #ifdef CAPABILITIES 3039 int 3040 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 3041 { 3042 const struct filedescent *fde; 3043 const struct fdescenttbl *fdt; 3044 struct filedesc *fdp; 3045 struct file *fp; 3046 struct vnode *vp; 3047 const cap_rights_t *haverights; 3048 cap_rights_t rights; 3049 seqc_t seq; 3050 int fd; 3051 3052 VFS_SMR_ASSERT_ENTERED(); 3053 3054 fd = ndp->ni_dirfd; 3055 rights = *ndp->ni_rightsneeded; 3056 cap_rights_set_one(&rights, CAP_LOOKUP); 3057 3058 fdp = curproc->p_fd; 3059 fdt = fdp->fd_files; 3060 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3061 return (EBADF); 3062 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3063 fde = &fdt->fdt_ofiles[fd]; 3064 haverights = cap_rights_fde_inline(fde); 3065 fp = fde->fde_file; 3066 if (__predict_false(fp == NULL)) 3067 return (EAGAIN); 3068 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 3069 return (EAGAIN); 3070 *fsearch = ((fp->f_flag & FSEARCH) != 0); 3071 vp = fp->f_vnode; 3072 if (__predict_false(vp == NULL)) { 3073 return (EAGAIN); 3074 } 3075 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 3076 return (EAGAIN); 3077 } 3078 /* 3079 * Use an acquire barrier to force re-reading of fdt so it is 3080 * refreshed for verification. 3081 */ 3082 atomic_thread_fence_acq(); 3083 fdt = fdp->fd_files; 3084 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3085 return (EAGAIN); 3086 /* 3087 * If file descriptor doesn't have all rights, 3088 * all lookups relative to it must also be 3089 * strictly relative. 3090 * 3091 * Not yet supported by fast path. 3092 */ 3093 CAP_ALL(&rights); 3094 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3095 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3096 ndp->ni_filecaps.fc_nioctls != -1) { 3097 #ifdef notyet 3098 ndp->ni_lcf |= NI_LCF_STRICTREL; 3099 #else 3100 return (EAGAIN); 3101 #endif 3102 } 3103 *vpp = vp; 3104 return (0); 3105 } 3106 #else 3107 int 3108 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 3109 { 3110 const struct fdescenttbl *fdt; 3111 struct filedesc *fdp; 3112 struct file *fp; 3113 struct vnode *vp; 3114 int fd; 3115 3116 VFS_SMR_ASSERT_ENTERED(); 3117 3118 fd = ndp->ni_dirfd; 3119 fdp = curproc->p_fd; 3120 fdt = fdp->fd_files; 3121 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3122 return (EBADF); 3123 fp = fdt->fdt_ofiles[fd].fde_file; 3124 if (__predict_false(fp == NULL)) 3125 return (EAGAIN); 3126 *fsearch = ((fp->f_flag & FSEARCH) != 0); 3127 vp = fp->f_vnode; 3128 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3129 return (EAGAIN); 3130 } 3131 /* 3132 * Use an acquire barrier to force re-reading of fdt so it is 3133 * refreshed for verification. 3134 */ 3135 atomic_thread_fence_acq(); 3136 fdt = fdp->fd_files; 3137 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3138 return (EAGAIN); 3139 filecaps_fill(&ndp->ni_filecaps); 3140 *vpp = vp; 3141 return (0); 3142 } 3143 #endif 3144 3145 int 3146 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp) 3147 { 3148 struct thread *td; 3149 struct file *fp; 3150 struct vnode *vp; 3151 struct componentname *cnp; 3152 cap_rights_t rights; 3153 int error; 3154 3155 td = curthread; 3156 rights = *ndp->ni_rightsneeded; 3157 cap_rights_set_one(&rights, CAP_LOOKUP); 3158 cnp = &ndp->ni_cnd; 3159 3160 error = fget_cap(td, ndp->ni_dirfd, &rights, &fp, &ndp->ni_filecaps); 3161 if (__predict_false(error != 0)) 3162 return (error); 3163 if (__predict_false(fp->f_ops == &badfileops)) { 3164 error = EBADF; 3165 goto out_free; 3166 } 3167 vp = fp->f_vnode; 3168 if (__predict_false(vp == NULL)) { 3169 error = ENOTDIR; 3170 goto out_free; 3171 } 3172 vrefact(vp); 3173 /* 3174 * XXX does not check for VDIR, handled by namei_setup 3175 */ 3176 if ((fp->f_flag & FSEARCH) != 0) 3177 cnp->cn_flags |= NOEXECCHECK; 3178 fdrop(fp, td); 3179 3180 #ifdef CAPABILITIES 3181 /* 3182 * If file descriptor doesn't have all rights, 3183 * all lookups relative to it must also be 3184 * strictly relative. 3185 */ 3186 CAP_ALL(&rights); 3187 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3188 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3189 ndp->ni_filecaps.fc_nioctls != -1) { 3190 ndp->ni_lcf |= NI_LCF_STRICTREL; 3191 ndp->ni_resflags |= NIRES_STRICTREL; 3192 } 3193 #endif 3194 3195 /* 3196 * TODO: avoid copying ioctl caps if it can be helped to begin with 3197 */ 3198 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0) 3199 filecaps_free_ioctl(&ndp->ni_filecaps); 3200 3201 *vpp = vp; 3202 return (0); 3203 3204 out_free: 3205 filecaps_free(&ndp->ni_filecaps); 3206 fdrop(fp, td); 3207 return (error); 3208 } 3209 3210 /* 3211 * Fetch the descriptor locklessly. 3212 * 3213 * We avoid fdrop() races by never raising a refcount above 0. To accomplish 3214 * this we have to use a cmpset loop rather than an atomic_add. The descriptor 3215 * must be re-verified once we acquire a reference to be certain that the 3216 * identity is still correct and we did not lose a race due to preemption. 3217 * 3218 * Force a reload of fdt when looping. Another thread could reallocate 3219 * the table before this fd was closed, so it is possible that there is 3220 * a stale fp pointer in cached version. 3221 */ 3222 #ifdef CAPABILITIES 3223 static int 3224 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp, 3225 struct file **fpp, seqc_t *seqp) 3226 { 3227 struct filedesc *fdp; 3228 const struct filedescent *fde; 3229 const struct fdescenttbl *fdt; 3230 struct file *fp; 3231 seqc_t seq; 3232 cap_rights_t haverights; 3233 int error; 3234 3235 fdp = td->td_proc->p_fd; 3236 fdt = fdp->fd_files; 3237 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3238 return (EBADF); 3239 3240 for (;;) { 3241 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3242 fde = &fdt->fdt_ofiles[fd]; 3243 haverights = *cap_rights_fde_inline(fde); 3244 fp = fde->fde_file; 3245 if (__predict_false(fp == NULL)) { 3246 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3247 return (EBADF); 3248 fdt = atomic_load_ptr(&fdp->fd_files); 3249 continue; 3250 } 3251 error = cap_check_inline(&haverights, needrightsp); 3252 if (__predict_false(error != 0)) { 3253 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3254 return (error); 3255 fdt = atomic_load_ptr(&fdp->fd_files); 3256 continue; 3257 } 3258 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3259 fdt = atomic_load_ptr(&fdp->fd_files); 3260 continue; 3261 } 3262 /* 3263 * Use an acquire barrier to force re-reading of fdt so it is 3264 * refreshed for verification. 3265 */ 3266 atomic_thread_fence_acq(); 3267 fdt = fdp->fd_files; 3268 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)) 3269 break; 3270 fdrop(fp, td); 3271 } 3272 *fpp = fp; 3273 if (seqp != NULL) { 3274 *seqp = seq; 3275 } 3276 return (0); 3277 } 3278 #else 3279 static int 3280 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp, 3281 struct file **fpp, seqc_t *seqp __unused) 3282 { 3283 struct filedesc *fdp; 3284 const struct fdescenttbl *fdt; 3285 struct file *fp; 3286 3287 fdp = td->td_proc->p_fd; 3288 fdt = fdp->fd_files; 3289 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3290 return (EBADF); 3291 3292 for (;;) { 3293 fp = fdt->fdt_ofiles[fd].fde_file; 3294 if (__predict_false(fp == NULL)) 3295 return (EBADF); 3296 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3297 fdt = atomic_load_ptr(&fdp->fd_files); 3298 continue; 3299 } 3300 /* 3301 * Use an acquire barrier to force re-reading of fdt so it is 3302 * refreshed for verification. 3303 */ 3304 atomic_thread_fence_acq(); 3305 fdt = fdp->fd_files; 3306 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file)) 3307 break; 3308 fdrop(fp, td); 3309 } 3310 *fpp = fp; 3311 return (0); 3312 } 3313 #endif 3314 3315 /* 3316 * See the comments in fget_unlocked_seq for an explanation of how this works. 3317 * 3318 * This is a simplified variant which bails out to the aforementioned routine 3319 * if anything goes wrong. In practice this only happens when userspace is 3320 * racing with itself. 3321 */ 3322 int 3323 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp, 3324 struct file **fpp) 3325 { 3326 struct filedesc *fdp; 3327 #ifdef CAPABILITIES 3328 const struct filedescent *fde; 3329 #endif 3330 const struct fdescenttbl *fdt; 3331 struct file *fp; 3332 #ifdef CAPABILITIES 3333 seqc_t seq; 3334 const cap_rights_t *haverights; 3335 #endif 3336 3337 fdp = td->td_proc->p_fd; 3338 fdt = fdp->fd_files; 3339 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { 3340 *fpp = NULL; 3341 return (EBADF); 3342 } 3343 #ifdef CAPABILITIES 3344 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3345 fde = &fdt->fdt_ofiles[fd]; 3346 haverights = cap_rights_fde_inline(fde); 3347 fp = fde->fde_file; 3348 #else 3349 fp = fdt->fdt_ofiles[fd].fde_file; 3350 #endif 3351 if (__predict_false(fp == NULL)) 3352 goto out_fallback; 3353 #ifdef CAPABILITIES 3354 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3355 goto out_fallback; 3356 #endif 3357 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3358 goto out_fallback; 3359 3360 /* 3361 * Use an acquire barrier to force re-reading of fdt so it is 3362 * refreshed for verification. 3363 */ 3364 atomic_thread_fence_acq(); 3365 fdt = fdp->fd_files; 3366 #ifdef CAPABILITIES 3367 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3368 #else 3369 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3370 #endif 3371 goto out_fdrop; 3372 *fpp = fp; 3373 return (0); 3374 out_fdrop: 3375 fdrop(fp, td); 3376 out_fallback: 3377 *fpp = NULL; 3378 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL)); 3379 } 3380 3381 /* 3382 * Translate fd -> file when the caller guarantees the file descriptor table 3383 * can't be changed by others. 3384 * 3385 * Note this does not mean the file object itself is only visible to the caller, 3386 * merely that it wont disappear without having to be referenced. 3387 * 3388 * Must be paired with fput_only_user. 3389 */ 3390 #ifdef CAPABILITIES 3391 int 3392 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3393 struct file **fpp) 3394 { 3395 const struct filedescent *fde; 3396 const struct fdescenttbl *fdt; 3397 const cap_rights_t *haverights; 3398 struct file *fp; 3399 int error; 3400 3401 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3402 3403 *fpp = NULL; 3404 if (__predict_false(fd >= fdp->fd_nfiles)) 3405 return (EBADF); 3406 3407 fdt = fdp->fd_files; 3408 fde = &fdt->fdt_ofiles[fd]; 3409 fp = fde->fde_file; 3410 if (__predict_false(fp == NULL)) 3411 return (EBADF); 3412 MPASS(refcount_load(&fp->f_count) > 0); 3413 haverights = cap_rights_fde_inline(fde); 3414 error = cap_check_inline(haverights, needrightsp); 3415 if (__predict_false(error != 0)) 3416 return (error); 3417 *fpp = fp; 3418 return (0); 3419 } 3420 #else 3421 int 3422 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3423 struct file **fpp) 3424 { 3425 struct file *fp; 3426 3427 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3428 3429 *fpp = NULL; 3430 if (__predict_false(fd >= fdp->fd_nfiles)) 3431 return (EBADF); 3432 3433 fp = fdp->fd_ofiles[fd].fde_file; 3434 if (__predict_false(fp == NULL)) 3435 return (EBADF); 3436 3437 MPASS(refcount_load(&fp->f_count) > 0); 3438 *fpp = fp; 3439 return (0); 3440 } 3441 #endif 3442 3443 /* 3444 * Extract the file pointer associated with the specified descriptor for the 3445 * current user process. 3446 * 3447 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3448 * returned. 3449 * 3450 * File's rights will be checked against the capability rights mask. 3451 * 3452 * If an error occurred the non-zero error is returned and *fpp is set to 3453 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3454 * responsible for fdrop(). 3455 */ 3456 static __inline int 3457 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3458 cap_rights_t *needrightsp) 3459 { 3460 struct file *fp; 3461 int error; 3462 3463 *fpp = NULL; 3464 error = fget_unlocked(td, fd, needrightsp, &fp); 3465 if (__predict_false(error != 0)) 3466 return (error); 3467 if (__predict_false(fp->f_ops == &badfileops)) { 3468 fdrop(fp, td); 3469 return (EBADF); 3470 } 3471 3472 /* 3473 * FREAD and FWRITE failure return EBADF as per POSIX. 3474 */ 3475 error = 0; 3476 switch (flags) { 3477 case FREAD: 3478 case FWRITE: 3479 if ((fp->f_flag & flags) == 0) 3480 error = EBADF; 3481 break; 3482 case FEXEC: 3483 if (fp->f_ops != &path_fileops && 3484 ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3485 (fp->f_flag & FWRITE) != 0)) 3486 error = EBADF; 3487 break; 3488 case 0: 3489 break; 3490 default: 3491 KASSERT(0, ("wrong flags")); 3492 } 3493 3494 if (error != 0) { 3495 fdrop(fp, td); 3496 return (error); 3497 } 3498 3499 *fpp = fp; 3500 return (0); 3501 } 3502 3503 int 3504 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3505 { 3506 3507 return (_fget(td, fd, fpp, 0, rightsp)); 3508 } 3509 3510 int 3511 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, 3512 struct file **fpp) 3513 { 3514 int error; 3515 #ifndef CAPABILITIES 3516 error = _fget(td, fd, fpp, 0, rightsp); 3517 if (maxprotp != NULL) 3518 *maxprotp = VM_PROT_ALL; 3519 return (error); 3520 #else 3521 cap_rights_t fdrights; 3522 struct filedesc *fdp; 3523 struct file *fp; 3524 seqc_t seq; 3525 3526 *fpp = NULL; 3527 fdp = td->td_proc->p_fd; 3528 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3529 for (;;) { 3530 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq); 3531 if (__predict_false(error != 0)) 3532 return (error); 3533 if (__predict_false(fp->f_ops == &badfileops)) { 3534 fdrop(fp, td); 3535 return (EBADF); 3536 } 3537 if (maxprotp != NULL) 3538 fdrights = *cap_rights(fdp, fd); 3539 if (!fd_modified(fdp, fd, seq)) 3540 break; 3541 fdrop(fp, td); 3542 } 3543 3544 /* 3545 * If requested, convert capability rights to access flags. 3546 */ 3547 if (maxprotp != NULL) 3548 *maxprotp = cap_rights_to_vmprot(&fdrights); 3549 *fpp = fp; 3550 return (0); 3551 #endif 3552 } 3553 3554 int 3555 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3556 { 3557 3558 return (_fget(td, fd, fpp, FREAD, rightsp)); 3559 } 3560 3561 int 3562 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3563 { 3564 3565 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3566 } 3567 3568 int 3569 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 3570 struct file **fpp) 3571 { 3572 #ifndef CAPABILITIES 3573 return (fget_unlocked(td, fd, rightsp, fpp)); 3574 #else 3575 struct filedesc *fdp = td->td_proc->p_fd; 3576 struct file *fp; 3577 int error; 3578 seqc_t seq; 3579 3580 *fpp = NULL; 3581 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3582 for (;;) { 3583 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq); 3584 if (error != 0) 3585 return (error); 3586 error = cap_fcntl_check(fdp, fd, needfcntl); 3587 if (!fd_modified(fdp, fd, seq)) 3588 break; 3589 fdrop(fp, td); 3590 } 3591 if (error != 0) { 3592 fdrop(fp, td); 3593 return (error); 3594 } 3595 *fpp = fp; 3596 return (0); 3597 #endif 3598 } 3599 3600 /* 3601 * Like fget() but loads the underlying vnode, or returns an error if the 3602 * descriptor does not represent a vnode. Note that pipes use vnodes but 3603 * never have VM objects. The returned vnode will be vref()'d. 3604 * 3605 * XXX: what about the unused flags ? 3606 */ 3607 static __inline int 3608 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 3609 struct vnode **vpp) 3610 { 3611 struct file *fp; 3612 int error; 3613 3614 *vpp = NULL; 3615 error = _fget(td, fd, &fp, flags, needrightsp); 3616 if (error != 0) 3617 return (error); 3618 if (fp->f_vnode == NULL) { 3619 error = EINVAL; 3620 } else { 3621 *vpp = fp->f_vnode; 3622 vrefact(*vpp); 3623 } 3624 fdrop(fp, td); 3625 3626 return (error); 3627 } 3628 3629 int 3630 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3631 { 3632 3633 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3634 } 3635 3636 int 3637 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 3638 struct filecaps *havecaps, struct vnode **vpp) 3639 { 3640 struct filecaps caps; 3641 struct file *fp; 3642 int error; 3643 3644 error = fget_cap(td, fd, needrightsp, &fp, &caps); 3645 if (error != 0) 3646 return (error); 3647 if (fp->f_ops == &badfileops) { 3648 error = EBADF; 3649 goto out; 3650 } 3651 if (fp->f_vnode == NULL) { 3652 error = EINVAL; 3653 goto out; 3654 } 3655 3656 *havecaps = caps; 3657 *vpp = fp->f_vnode; 3658 vrefact(*vpp); 3659 fdrop(fp, td); 3660 3661 return (0); 3662 out: 3663 filecaps_free(&caps); 3664 fdrop(fp, td); 3665 return (error); 3666 } 3667 3668 int 3669 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3670 { 3671 3672 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3673 } 3674 3675 int 3676 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3677 { 3678 3679 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3680 } 3681 3682 #ifdef notyet 3683 int 3684 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 3685 struct vnode **vpp) 3686 { 3687 3688 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3689 } 3690 #endif 3691 3692 /* 3693 * Handle the last reference to a file being closed. 3694 * 3695 * Without the noinline attribute clang keeps inlining the func thorough this 3696 * file when fdrop is used. 3697 */ 3698 int __noinline 3699 _fdrop(struct file *fp, struct thread *td) 3700 { 3701 int error; 3702 3703 KASSERT(refcount_load(&fp->f_count) == 0, 3704 ("fdrop: fp %p count %d", fp, refcount_load(&fp->f_count))); 3705 3706 error = fo_close(fp, td); 3707 atomic_subtract_int(&openfiles, 1); 3708 crfree(fp->f_cred); 3709 free(fp->f_advice, M_FADVISE); 3710 uma_zfree(file_zone, fp); 3711 3712 return (error); 3713 } 3714 3715 /* 3716 * Apply an advisory lock on a file descriptor. 3717 * 3718 * Just attempt to get a record lock of the requested type on the entire file 3719 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 3720 */ 3721 #ifndef _SYS_SYSPROTO_H_ 3722 struct flock_args { 3723 int fd; 3724 int how; 3725 }; 3726 #endif 3727 /* ARGSUSED */ 3728 int 3729 sys_flock(struct thread *td, struct flock_args *uap) 3730 { 3731 struct file *fp; 3732 struct vnode *vp; 3733 struct flock lf; 3734 int error; 3735 3736 error = fget(td, uap->fd, &cap_flock_rights, &fp); 3737 if (error != 0) 3738 return (error); 3739 error = EOPNOTSUPP; 3740 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 3741 goto done; 3742 } 3743 if (fp->f_ops == &path_fileops) { 3744 goto done; 3745 } 3746 3747 error = 0; 3748 vp = fp->f_vnode; 3749 lf.l_whence = SEEK_SET; 3750 lf.l_start = 0; 3751 lf.l_len = 0; 3752 if (uap->how & LOCK_UN) { 3753 lf.l_type = F_UNLCK; 3754 atomic_clear_int(&fp->f_flag, FHASLOCK); 3755 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3756 goto done; 3757 } 3758 if (uap->how & LOCK_EX) 3759 lf.l_type = F_WRLCK; 3760 else if (uap->how & LOCK_SH) 3761 lf.l_type = F_RDLCK; 3762 else { 3763 error = EBADF; 3764 goto done; 3765 } 3766 atomic_set_int(&fp->f_flag, FHASLOCK); 3767 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3768 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3769 done: 3770 fdrop(fp, td); 3771 return (error); 3772 } 3773 /* 3774 * Duplicate the specified descriptor to a free descriptor. 3775 */ 3776 int 3777 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3778 int openerror, int *indxp) 3779 { 3780 struct filedescent *newfde, *oldfde; 3781 struct file *fp; 3782 u_long *ioctls; 3783 int error, indx; 3784 3785 KASSERT(openerror == ENODEV || openerror == ENXIO, 3786 ("unexpected error %d in %s", openerror, __func__)); 3787 3788 /* 3789 * If the to-be-dup'd fd number is greater than the allowed number 3790 * of file descriptors, or the fd to be dup'd has already been 3791 * closed, then reject. 3792 */ 3793 FILEDESC_XLOCK(fdp); 3794 if ((fp = fget_noref(fdp, dfd)) == NULL) { 3795 FILEDESC_XUNLOCK(fdp); 3796 return (EBADF); 3797 } 3798 3799 error = fdalloc(td, 0, &indx); 3800 if (error != 0) { 3801 FILEDESC_XUNLOCK(fdp); 3802 return (error); 3803 } 3804 3805 /* 3806 * There are two cases of interest here. 3807 * 3808 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3809 * 3810 * For ENXIO steal away the file structure from (dfd) and store it in 3811 * (indx). (dfd) is effectively closed by this operation. 3812 */ 3813 switch (openerror) { 3814 case ENODEV: 3815 /* 3816 * Check that the mode the file is being opened for is a 3817 * subset of the mode of the existing descriptor. 3818 */ 3819 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3820 fdunused(fdp, indx); 3821 FILEDESC_XUNLOCK(fdp); 3822 return (EACCES); 3823 } 3824 if (!fhold(fp)) { 3825 fdunused(fdp, indx); 3826 FILEDESC_XUNLOCK(fdp); 3827 return (EBADF); 3828 } 3829 newfde = &fdp->fd_ofiles[indx]; 3830 oldfde = &fdp->fd_ofiles[dfd]; 3831 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3832 #ifdef CAPABILITIES 3833 seqc_write_begin(&newfde->fde_seqc); 3834 #endif 3835 fde_copy(oldfde, newfde); 3836 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3837 ioctls); 3838 #ifdef CAPABILITIES 3839 seqc_write_end(&newfde->fde_seqc); 3840 #endif 3841 break; 3842 case ENXIO: 3843 /* 3844 * Steal away the file pointer from dfd and stuff it into indx. 3845 */ 3846 newfde = &fdp->fd_ofiles[indx]; 3847 oldfde = &fdp->fd_ofiles[dfd]; 3848 #ifdef CAPABILITIES 3849 seqc_write_begin(&oldfde->fde_seqc); 3850 seqc_write_begin(&newfde->fde_seqc); 3851 #endif 3852 fde_copy(oldfde, newfde); 3853 oldfde->fde_file = NULL; 3854 fdunused(fdp, dfd); 3855 #ifdef CAPABILITIES 3856 seqc_write_end(&newfde->fde_seqc); 3857 seqc_write_end(&oldfde->fde_seqc); 3858 #endif 3859 break; 3860 } 3861 FILEDESC_XUNLOCK(fdp); 3862 *indxp = indx; 3863 return (0); 3864 } 3865 3866 /* 3867 * This sysctl determines if we will allow a process to chroot(2) if it 3868 * has a directory open: 3869 * 0: disallowed for all processes. 3870 * 1: allowed for processes that were not already chroot(2)'ed. 3871 * 2: allowed for all processes. 3872 */ 3873 3874 static int chroot_allow_open_directories = 1; 3875 3876 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3877 &chroot_allow_open_directories, 0, 3878 "Allow a process to chroot(2) if it has a directory open"); 3879 3880 /* 3881 * Helper function for raised chroot(2) security function: Refuse if 3882 * any filedescriptors are open directories. 3883 */ 3884 static int 3885 chroot_refuse_vdir_fds(struct filedesc *fdp) 3886 { 3887 struct vnode *vp; 3888 struct file *fp; 3889 int i; 3890 3891 FILEDESC_LOCK_ASSERT(fdp); 3892 3893 FILEDESC_FOREACH_FP(fdp, i, fp) { 3894 if (fp->f_type == DTYPE_VNODE) { 3895 vp = fp->f_vnode; 3896 if (vp->v_type == VDIR) 3897 return (EPERM); 3898 } 3899 } 3900 return (0); 3901 } 3902 3903 static void 3904 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 3905 { 3906 3907 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 3908 vrefact(oldpwd->pwd_cdir); 3909 newpwd->pwd_cdir = oldpwd->pwd_cdir; 3910 } 3911 3912 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 3913 vrefact(oldpwd->pwd_rdir); 3914 newpwd->pwd_rdir = oldpwd->pwd_rdir; 3915 } 3916 3917 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 3918 vrefact(oldpwd->pwd_jdir); 3919 newpwd->pwd_jdir = oldpwd->pwd_jdir; 3920 } 3921 3922 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) { 3923 vrefact(oldpwd->pwd_adir); 3924 newpwd->pwd_adir = oldpwd->pwd_adir; 3925 } 3926 } 3927 3928 struct pwd * 3929 pwd_hold_pwddesc(struct pwddesc *pdp) 3930 { 3931 struct pwd *pwd; 3932 3933 PWDDESC_ASSERT_XLOCKED(pdp); 3934 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3935 if (pwd != NULL) 3936 refcount_acquire(&pwd->pwd_refcount); 3937 return (pwd); 3938 } 3939 3940 bool 3941 pwd_hold_smr(struct pwd *pwd) 3942 { 3943 3944 MPASS(pwd != NULL); 3945 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 3946 return (true); 3947 } 3948 return (false); 3949 } 3950 3951 struct pwd * 3952 pwd_hold(struct thread *td) 3953 { 3954 struct pwddesc *pdp; 3955 struct pwd *pwd; 3956 3957 pdp = td->td_proc->p_pd; 3958 3959 vfs_smr_enter(); 3960 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 3961 if (pwd_hold_smr(pwd)) { 3962 vfs_smr_exit(); 3963 return (pwd); 3964 } 3965 vfs_smr_exit(); 3966 PWDDESC_XLOCK(pdp); 3967 pwd = pwd_hold_pwddesc(pdp); 3968 MPASS(pwd != NULL); 3969 PWDDESC_XUNLOCK(pdp); 3970 return (pwd); 3971 } 3972 3973 struct pwd * 3974 pwd_hold_proc(struct proc *p) 3975 { 3976 struct pwddesc *pdp; 3977 struct pwd *pwd; 3978 3979 PROC_ASSERT_HELD(p); 3980 PROC_LOCK(p); 3981 pdp = pdhold(p); 3982 MPASS(pdp != NULL); 3983 PROC_UNLOCK(p); 3984 3985 PWDDESC_XLOCK(pdp); 3986 pwd = pwd_hold_pwddesc(pdp); 3987 MPASS(pwd != NULL); 3988 PWDDESC_XUNLOCK(pdp); 3989 pddrop(pdp); 3990 return (pwd); 3991 } 3992 3993 static struct pwd * 3994 pwd_alloc(void) 3995 { 3996 struct pwd *pwd; 3997 3998 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 3999 bzero(pwd, sizeof(*pwd)); 4000 refcount_init(&pwd->pwd_refcount, 1); 4001 return (pwd); 4002 } 4003 4004 void 4005 pwd_drop(struct pwd *pwd) 4006 { 4007 4008 if (!refcount_release(&pwd->pwd_refcount)) 4009 return; 4010 4011 if (pwd->pwd_cdir != NULL) 4012 vrele(pwd->pwd_cdir); 4013 if (pwd->pwd_rdir != NULL) 4014 vrele(pwd->pwd_rdir); 4015 if (pwd->pwd_jdir != NULL) 4016 vrele(pwd->pwd_jdir); 4017 if (pwd->pwd_adir != NULL) 4018 vrele(pwd->pwd_adir); 4019 uma_zfree_smr(pwd_zone, pwd); 4020 } 4021 4022 /* 4023 * The caller is responsible for invoking priv_check() and 4024 * mac_vnode_check_chroot() to authorize this operation. 4025 */ 4026 int 4027 pwd_chroot(struct thread *td, struct vnode *vp) 4028 { 4029 struct pwddesc *pdp; 4030 struct filedesc *fdp; 4031 struct pwd *newpwd, *oldpwd; 4032 int error; 4033 4034 fdp = td->td_proc->p_fd; 4035 pdp = td->td_proc->p_pd; 4036 newpwd = pwd_alloc(); 4037 FILEDESC_SLOCK(fdp); 4038 PWDDESC_XLOCK(pdp); 4039 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4040 if (chroot_allow_open_directories == 0 || 4041 (chroot_allow_open_directories == 1 && 4042 oldpwd->pwd_rdir != rootvnode)) { 4043 error = chroot_refuse_vdir_fds(fdp); 4044 FILEDESC_SUNLOCK(fdp); 4045 if (error != 0) { 4046 PWDDESC_XUNLOCK(pdp); 4047 pwd_drop(newpwd); 4048 return (error); 4049 } 4050 } else { 4051 FILEDESC_SUNLOCK(fdp); 4052 } 4053 4054 vrefact(vp); 4055 newpwd->pwd_rdir = vp; 4056 vrefact(vp); 4057 newpwd->pwd_adir = vp; 4058 if (oldpwd->pwd_jdir == NULL) { 4059 vrefact(vp); 4060 newpwd->pwd_jdir = vp; 4061 } 4062 pwd_fill(oldpwd, newpwd); 4063 pwd_set(pdp, newpwd); 4064 PWDDESC_XUNLOCK(pdp); 4065 pwd_drop(oldpwd); 4066 return (0); 4067 } 4068 4069 void 4070 pwd_chdir(struct thread *td, struct vnode *vp) 4071 { 4072 struct pwddesc *pdp; 4073 struct pwd *newpwd, *oldpwd; 4074 4075 VNPASS(vp->v_usecount > 0, vp); 4076 4077 newpwd = pwd_alloc(); 4078 pdp = td->td_proc->p_pd; 4079 PWDDESC_XLOCK(pdp); 4080 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4081 newpwd->pwd_cdir = vp; 4082 pwd_fill(oldpwd, newpwd); 4083 pwd_set(pdp, newpwd); 4084 PWDDESC_XUNLOCK(pdp); 4085 pwd_drop(oldpwd); 4086 } 4087 4088 /* 4089 * Process is transitioning to/from a non-native ABI. 4090 */ 4091 void 4092 pwd_altroot(struct thread *td, struct vnode *altroot_vp) 4093 { 4094 struct pwddesc *pdp; 4095 struct pwd *newpwd, *oldpwd; 4096 4097 newpwd = pwd_alloc(); 4098 pdp = td->td_proc->p_pd; 4099 PWDDESC_XLOCK(pdp); 4100 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4101 if (altroot_vp != NULL) { 4102 /* 4103 * Native process to a non-native ABI. 4104 */ 4105 4106 vrefact(altroot_vp); 4107 newpwd->pwd_adir = altroot_vp; 4108 } else { 4109 /* 4110 * Non-native process to the native ABI. 4111 */ 4112 4113 vrefact(oldpwd->pwd_rdir); 4114 newpwd->pwd_adir = oldpwd->pwd_rdir; 4115 } 4116 pwd_fill(oldpwd, newpwd); 4117 pwd_set(pdp, newpwd); 4118 PWDDESC_XUNLOCK(pdp); 4119 pwd_drop(oldpwd); 4120 } 4121 4122 /* 4123 * jail_attach(2) changes both root and working directories. 4124 */ 4125 int 4126 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 4127 { 4128 struct pwddesc *pdp; 4129 struct filedesc *fdp; 4130 struct pwd *newpwd, *oldpwd; 4131 int error; 4132 4133 fdp = td->td_proc->p_fd; 4134 pdp = td->td_proc->p_pd; 4135 newpwd = pwd_alloc(); 4136 FILEDESC_SLOCK(fdp); 4137 PWDDESC_XLOCK(pdp); 4138 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4139 error = chroot_refuse_vdir_fds(fdp); 4140 FILEDESC_SUNLOCK(fdp); 4141 if (error != 0) { 4142 PWDDESC_XUNLOCK(pdp); 4143 pwd_drop(newpwd); 4144 return (error); 4145 } 4146 4147 vrefact(vp); 4148 newpwd->pwd_rdir = vp; 4149 vrefact(vp); 4150 newpwd->pwd_cdir = vp; 4151 if (oldpwd->pwd_jdir == NULL) { 4152 vrefact(vp); 4153 newpwd->pwd_jdir = vp; 4154 } 4155 vrefact(vp); 4156 newpwd->pwd_adir = vp; 4157 pwd_fill(oldpwd, newpwd); 4158 pwd_set(pdp, newpwd); 4159 PWDDESC_XUNLOCK(pdp); 4160 pwd_drop(oldpwd); 4161 return (0); 4162 } 4163 4164 void 4165 pwd_ensure_dirs(void) 4166 { 4167 struct pwddesc *pdp; 4168 struct pwd *oldpwd, *newpwd; 4169 4170 pdp = curproc->p_pd; 4171 PWDDESC_XLOCK(pdp); 4172 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4173 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL && 4174 oldpwd->pwd_adir != NULL) { 4175 PWDDESC_XUNLOCK(pdp); 4176 return; 4177 } 4178 PWDDESC_XUNLOCK(pdp); 4179 4180 newpwd = pwd_alloc(); 4181 PWDDESC_XLOCK(pdp); 4182 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4183 pwd_fill(oldpwd, newpwd); 4184 if (newpwd->pwd_cdir == NULL) { 4185 vrefact(rootvnode); 4186 newpwd->pwd_cdir = rootvnode; 4187 } 4188 if (newpwd->pwd_rdir == NULL) { 4189 vrefact(rootvnode); 4190 newpwd->pwd_rdir = rootvnode; 4191 } 4192 if (newpwd->pwd_adir == NULL) { 4193 vrefact(rootvnode); 4194 newpwd->pwd_adir = rootvnode; 4195 } 4196 pwd_set(pdp, newpwd); 4197 PWDDESC_XUNLOCK(pdp); 4198 pwd_drop(oldpwd); 4199 } 4200 4201 void 4202 pwd_set_rootvnode(void) 4203 { 4204 struct pwddesc *pdp; 4205 struct pwd *oldpwd, *newpwd; 4206 4207 pdp = curproc->p_pd; 4208 4209 newpwd = pwd_alloc(); 4210 PWDDESC_XLOCK(pdp); 4211 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4212 vrefact(rootvnode); 4213 newpwd->pwd_cdir = rootvnode; 4214 vrefact(rootvnode); 4215 newpwd->pwd_rdir = rootvnode; 4216 vrefact(rootvnode); 4217 newpwd->pwd_adir = rootvnode; 4218 pwd_fill(oldpwd, newpwd); 4219 pwd_set(pdp, newpwd); 4220 PWDDESC_XUNLOCK(pdp); 4221 pwd_drop(oldpwd); 4222 } 4223 4224 /* 4225 * Scan all active processes and prisons to see if any of them have a current 4226 * or root directory of `olddp'. If so, replace them with the new mount point. 4227 */ 4228 void 4229 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 4230 { 4231 struct pwddesc *pdp; 4232 struct pwd *newpwd, *oldpwd; 4233 struct prison *pr; 4234 struct proc *p; 4235 int nrele; 4236 4237 if (vrefcnt(olddp) == 1) 4238 return; 4239 nrele = 0; 4240 newpwd = pwd_alloc(); 4241 sx_slock(&allproc_lock); 4242 FOREACH_PROC_IN_SYSTEM(p) { 4243 PROC_LOCK(p); 4244 pdp = pdhold(p); 4245 PROC_UNLOCK(p); 4246 if (pdp == NULL) 4247 continue; 4248 PWDDESC_XLOCK(pdp); 4249 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4250 if (oldpwd == NULL || 4251 (oldpwd->pwd_cdir != olddp && 4252 oldpwd->pwd_rdir != olddp && 4253 oldpwd->pwd_jdir != olddp && 4254 oldpwd->pwd_adir != olddp)) { 4255 PWDDESC_XUNLOCK(pdp); 4256 pddrop(pdp); 4257 continue; 4258 } 4259 if (oldpwd->pwd_cdir == olddp) { 4260 vrefact(newdp); 4261 newpwd->pwd_cdir = newdp; 4262 } 4263 if (oldpwd->pwd_rdir == olddp) { 4264 vrefact(newdp); 4265 newpwd->pwd_rdir = newdp; 4266 } 4267 if (oldpwd->pwd_jdir == olddp) { 4268 vrefact(newdp); 4269 newpwd->pwd_jdir = newdp; 4270 } 4271 if (oldpwd->pwd_adir == olddp) { 4272 vrefact(newdp); 4273 newpwd->pwd_adir = newdp; 4274 } 4275 pwd_fill(oldpwd, newpwd); 4276 pwd_set(pdp, newpwd); 4277 PWDDESC_XUNLOCK(pdp); 4278 pwd_drop(oldpwd); 4279 pddrop(pdp); 4280 newpwd = pwd_alloc(); 4281 } 4282 sx_sunlock(&allproc_lock); 4283 pwd_drop(newpwd); 4284 if (rootvnode == olddp) { 4285 vrefact(newdp); 4286 rootvnode = newdp; 4287 nrele++; 4288 } 4289 mtx_lock(&prison0.pr_mtx); 4290 if (prison0.pr_root == olddp) { 4291 vrefact(newdp); 4292 prison0.pr_root = newdp; 4293 nrele++; 4294 } 4295 mtx_unlock(&prison0.pr_mtx); 4296 sx_slock(&allprison_lock); 4297 TAILQ_FOREACH(pr, &allprison, pr_list) { 4298 mtx_lock(&pr->pr_mtx); 4299 if (pr->pr_root == olddp) { 4300 vrefact(newdp); 4301 pr->pr_root = newdp; 4302 nrele++; 4303 } 4304 mtx_unlock(&pr->pr_mtx); 4305 } 4306 sx_sunlock(&allprison_lock); 4307 while (nrele--) 4308 vrele(olddp); 4309 } 4310 4311 int 4312 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp) 4313 { 4314 struct file *fp; 4315 struct vnode *vp; 4316 int error, i; 4317 4318 error = 0; 4319 FILEDESC_SLOCK(fdp); 4320 FILEDESC_FOREACH_FP(fdp, i, fp) { 4321 if (fp->f_type != DTYPE_VNODE || 4322 (atomic_load_int(&fp->f_flag) & FWRITE) == 0) 4323 continue; 4324 vp = fp->f_vnode; 4325 if (vp->v_mount == mp) { 4326 error = EDEADLK; 4327 break; 4328 } 4329 } 4330 FILEDESC_SUNLOCK(fdp); 4331 return (error); 4332 } 4333 4334 struct filedesc_to_leader * 4335 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, 4336 struct proc *leader) 4337 { 4338 struct filedesc_to_leader *fdtol; 4339 4340 fdtol = malloc(sizeof(struct filedesc_to_leader), 4341 M_FILEDESC_TO_LEADER, M_WAITOK); 4342 fdtol->fdl_refcount = 1; 4343 fdtol->fdl_holdcount = 0; 4344 fdtol->fdl_wakeup = 0; 4345 fdtol->fdl_leader = leader; 4346 if (old != NULL) { 4347 FILEDESC_XLOCK(fdp); 4348 fdtol->fdl_next = old->fdl_next; 4349 fdtol->fdl_prev = old; 4350 old->fdl_next = fdtol; 4351 fdtol->fdl_next->fdl_prev = fdtol; 4352 FILEDESC_XUNLOCK(fdp); 4353 } else { 4354 fdtol->fdl_next = fdtol; 4355 fdtol->fdl_prev = fdtol; 4356 } 4357 return (fdtol); 4358 } 4359 4360 struct filedesc_to_leader * 4361 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp) 4362 { 4363 FILEDESC_XLOCK(fdp); 4364 fdtol->fdl_refcount++; 4365 FILEDESC_XUNLOCK(fdp); 4366 return (fdtol); 4367 } 4368 4369 static int 4370 filedesc_nfiles(struct filedesc *fdp) 4371 { 4372 NDSLOTTYPE *map; 4373 int count, off, minoff; 4374 4375 if (fdp == NULL) 4376 return (0); 4377 count = 0; 4378 FILEDESC_SLOCK(fdp); 4379 map = fdp->fd_map; 4380 off = NDSLOT(fdp->fd_nfiles - 1); 4381 for (minoff = NDSLOT(0); off >= minoff; --off) 4382 count += bitcountl(map[off]); 4383 FILEDESC_SUNLOCK(fdp); 4384 return (count); 4385 } 4386 4387 int 4388 proc_nfiles(struct proc *p) 4389 { 4390 struct filedesc *fdp; 4391 int res; 4392 4393 PROC_LOCK(p); 4394 fdp = fdhold(p); 4395 PROC_UNLOCK(p); 4396 res = filedesc_nfiles(fdp); 4397 fddrop(fdp); 4398 return (res); 4399 } 4400 4401 static int 4402 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4403 { 4404 u_int namelen; 4405 int count; 4406 4407 namelen = arg2; 4408 if (namelen != 1) 4409 return (EINVAL); 4410 4411 if (*(int *)arg1 != 0) 4412 return (EINVAL); 4413 4414 count = filedesc_nfiles(curproc->p_fd); 4415 return (SYSCTL_OUT(req, &count, sizeof(count))); 4416 } 4417 4418 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4419 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4420 "Number of open file descriptors"); 4421 4422 /* 4423 * Get file structures globally. 4424 */ 4425 static int 4426 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4427 { 4428 struct xfile xf; 4429 struct filedesc *fdp; 4430 struct file *fp; 4431 struct proc *p; 4432 int error, n; 4433 4434 error = sysctl_wire_old_buffer(req, 0); 4435 if (error != 0) 4436 return (error); 4437 if (req->oldptr == NULL) { 4438 n = 0; 4439 sx_slock(&allproc_lock); 4440 FOREACH_PROC_IN_SYSTEM(p) { 4441 PROC_LOCK(p); 4442 if (p->p_state == PRS_NEW) { 4443 PROC_UNLOCK(p); 4444 continue; 4445 } 4446 fdp = fdhold(p); 4447 PROC_UNLOCK(p); 4448 if (fdp == NULL) 4449 continue; 4450 /* overestimates sparse tables. */ 4451 n += fdp->fd_nfiles; 4452 fddrop(fdp); 4453 } 4454 sx_sunlock(&allproc_lock); 4455 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4456 } 4457 error = 0; 4458 bzero(&xf, sizeof(xf)); 4459 xf.xf_size = sizeof(xf); 4460 sx_slock(&allproc_lock); 4461 FOREACH_PROC_IN_SYSTEM(p) { 4462 PROC_LOCK(p); 4463 if (p->p_state == PRS_NEW) { 4464 PROC_UNLOCK(p); 4465 continue; 4466 } 4467 if (p_cansee(req->td, p) != 0) { 4468 PROC_UNLOCK(p); 4469 continue; 4470 } 4471 xf.xf_pid = p->p_pid; 4472 xf.xf_uid = p->p_ucred->cr_uid; 4473 fdp = fdhold(p); 4474 PROC_UNLOCK(p); 4475 if (fdp == NULL) 4476 continue; 4477 FILEDESC_SLOCK(fdp); 4478 if (refcount_load(&fdp->fd_refcnt) == 0) 4479 goto nextproc; 4480 FILEDESC_FOREACH_FP(fdp, n, fp) { 4481 xf.xf_fd = n; 4482 xf.xf_file = (uintptr_t)fp; 4483 xf.xf_data = (uintptr_t)fp->f_data; 4484 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4485 xf.xf_type = (uintptr_t)fp->f_type; 4486 xf.xf_count = refcount_load(&fp->f_count); 4487 xf.xf_msgcount = 0; 4488 xf.xf_offset = foffset_get(fp); 4489 xf.xf_flag = fp->f_flag; 4490 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4491 4492 /* 4493 * There is no need to re-check the fdtable refcount 4494 * here since the filedesc lock is not dropped in the 4495 * loop body. 4496 */ 4497 if (error != 0) 4498 break; 4499 } 4500 nextproc: 4501 FILEDESC_SUNLOCK(fdp); 4502 fddrop(fdp); 4503 if (error) 4504 break; 4505 } 4506 sx_sunlock(&allproc_lock); 4507 return (error); 4508 } 4509 4510 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4511 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4512 4513 #ifdef KINFO_FILE_SIZE 4514 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4515 #endif 4516 4517 static int 4518 xlate_fflags(int fflags) 4519 { 4520 static const struct { 4521 int fflag; 4522 int kf_fflag; 4523 } fflags_table[] = { 4524 { FAPPEND, KF_FLAG_APPEND }, 4525 { FASYNC, KF_FLAG_ASYNC }, 4526 { FFSYNC, KF_FLAG_FSYNC }, 4527 { FHASLOCK, KF_FLAG_HASLOCK }, 4528 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4529 { FREAD, KF_FLAG_READ }, 4530 { FWRITE, KF_FLAG_WRITE }, 4531 { O_CREAT, KF_FLAG_CREAT }, 4532 { O_DIRECT, KF_FLAG_DIRECT }, 4533 { O_EXCL, KF_FLAG_EXCL }, 4534 { O_EXEC, KF_FLAG_EXEC }, 4535 { O_EXLOCK, KF_FLAG_EXLOCK }, 4536 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4537 { O_SHLOCK, KF_FLAG_SHLOCK }, 4538 { O_TRUNC, KF_FLAG_TRUNC } 4539 }; 4540 unsigned int i; 4541 int kflags; 4542 4543 kflags = 0; 4544 for (i = 0; i < nitems(fflags_table); i++) 4545 if (fflags & fflags_table[i].fflag) 4546 kflags |= fflags_table[i].kf_fflag; 4547 return (kflags); 4548 } 4549 4550 /* Trim unused data from kf_path by truncating the structure size. */ 4551 void 4552 pack_kinfo(struct kinfo_file *kif) 4553 { 4554 4555 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4556 strlen(kif->kf_path) + 1; 4557 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4558 } 4559 4560 static void 4561 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4562 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4563 { 4564 int error; 4565 4566 bzero(kif, sizeof(*kif)); 4567 4568 /* Set a default type to allow for empty fill_kinfo() methods. */ 4569 kif->kf_type = KF_TYPE_UNKNOWN; 4570 kif->kf_flags = xlate_fflags(fp->f_flag); 4571 if (rightsp != NULL) 4572 kif->kf_cap_rights = *rightsp; 4573 else 4574 cap_rights_init_zero(&kif->kf_cap_rights); 4575 kif->kf_fd = fd; 4576 kif->kf_ref_count = refcount_load(&fp->f_count); 4577 kif->kf_offset = foffset_get(fp); 4578 4579 /* 4580 * This may drop the filedesc lock, so the 'fp' cannot be 4581 * accessed after this call. 4582 */ 4583 error = fo_fill_kinfo(fp, kif, fdp); 4584 if (error == 0) 4585 kif->kf_status |= KF_ATTR_VALID; 4586 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4587 pack_kinfo(kif); 4588 else 4589 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4590 } 4591 4592 static void 4593 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4594 struct kinfo_file *kif, int flags) 4595 { 4596 int error; 4597 4598 bzero(kif, sizeof(*kif)); 4599 4600 kif->kf_type = KF_TYPE_VNODE; 4601 error = vn_fill_kinfo_vnode(vp, kif); 4602 if (error == 0) 4603 kif->kf_status |= KF_ATTR_VALID; 4604 kif->kf_flags = xlate_fflags(fflags); 4605 cap_rights_init_zero(&kif->kf_cap_rights); 4606 kif->kf_fd = fd; 4607 kif->kf_ref_count = -1; 4608 kif->kf_offset = -1; 4609 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4610 pack_kinfo(kif); 4611 else 4612 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4613 vrele(vp); 4614 } 4615 4616 struct export_fd_buf { 4617 struct filedesc *fdp; 4618 struct pwddesc *pdp; 4619 struct sbuf *sb; 4620 ssize_t remainder; 4621 struct kinfo_file kif; 4622 int flags; 4623 }; 4624 4625 static int 4626 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4627 { 4628 struct kinfo_file *kif; 4629 4630 kif = &efbuf->kif; 4631 if (efbuf->remainder != -1) { 4632 if (efbuf->remainder < kif->kf_structsize) 4633 return (ENOMEM); 4634 efbuf->remainder -= kif->kf_structsize; 4635 } 4636 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0) 4637 return (sbuf_error(efbuf->sb)); 4638 return (0); 4639 } 4640 4641 static int 4642 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4643 struct export_fd_buf *efbuf) 4644 { 4645 int error; 4646 4647 if (efbuf->remainder == 0) 4648 return (ENOMEM); 4649 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4650 efbuf->flags); 4651 FILEDESC_SUNLOCK(efbuf->fdp); 4652 error = export_kinfo_to_sb(efbuf); 4653 FILEDESC_SLOCK(efbuf->fdp); 4654 return (error); 4655 } 4656 4657 static int 4658 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4659 struct export_fd_buf *efbuf) 4660 { 4661 int error; 4662 4663 if (efbuf->remainder == 0) 4664 return (ENOMEM); 4665 if (efbuf->pdp != NULL) 4666 PWDDESC_XUNLOCK(efbuf->pdp); 4667 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4668 error = export_kinfo_to_sb(efbuf); 4669 if (efbuf->pdp != NULL) 4670 PWDDESC_XLOCK(efbuf->pdp); 4671 return (error); 4672 } 4673 4674 /* 4675 * Store a process file descriptor information to sbuf. 4676 * 4677 * Takes a locked proc as argument, and returns with the proc unlocked. 4678 */ 4679 int 4680 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4681 int flags) 4682 { 4683 struct file *fp; 4684 struct filedesc *fdp; 4685 struct pwddesc *pdp; 4686 struct export_fd_buf *efbuf; 4687 struct vnode *cttyvp, *textvp, *tracevp; 4688 struct pwd *pwd; 4689 int error, i; 4690 cap_rights_t rights; 4691 4692 PROC_LOCK_ASSERT(p, MA_OWNED); 4693 4694 /* ktrace vnode */ 4695 tracevp = ktr_get_tracevp(p, true); 4696 /* text vnode */ 4697 textvp = p->p_textvp; 4698 if (textvp != NULL) 4699 vrefact(textvp); 4700 /* Controlling tty. */ 4701 cttyvp = NULL; 4702 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 4703 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 4704 if (cttyvp != NULL) 4705 vrefact(cttyvp); 4706 } 4707 fdp = fdhold(p); 4708 pdp = pdhold(p); 4709 PROC_UNLOCK(p); 4710 4711 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4712 efbuf->fdp = NULL; 4713 efbuf->pdp = NULL; 4714 efbuf->sb = sb; 4715 efbuf->remainder = maxlen; 4716 efbuf->flags = flags; 4717 4718 error = 0; 4719 if (tracevp != NULL) 4720 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, 4721 FREAD | FWRITE, efbuf); 4722 if (error == 0 && textvp != NULL) 4723 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, 4724 efbuf); 4725 if (error == 0 && cttyvp != NULL) 4726 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, 4727 FREAD | FWRITE, efbuf); 4728 if (error != 0 || pdp == NULL || fdp == NULL) 4729 goto fail; 4730 efbuf->fdp = fdp; 4731 efbuf->pdp = pdp; 4732 PWDDESC_XLOCK(pdp); 4733 pwd = pwd_hold_pwddesc(pdp); 4734 if (pwd != NULL) { 4735 /* working directory */ 4736 if (pwd->pwd_cdir != NULL) { 4737 vrefact(pwd->pwd_cdir); 4738 error = export_vnode_to_sb(pwd->pwd_cdir, 4739 KF_FD_TYPE_CWD, FREAD, efbuf); 4740 } 4741 /* root directory */ 4742 if (error == 0 && pwd->pwd_rdir != NULL) { 4743 vrefact(pwd->pwd_rdir); 4744 error = export_vnode_to_sb(pwd->pwd_rdir, 4745 KF_FD_TYPE_ROOT, FREAD, efbuf); 4746 } 4747 /* jail directory */ 4748 if (error == 0 && pwd->pwd_jdir != NULL) { 4749 vrefact(pwd->pwd_jdir); 4750 error = export_vnode_to_sb(pwd->pwd_jdir, 4751 KF_FD_TYPE_JAIL, FREAD, efbuf); 4752 } 4753 } 4754 PWDDESC_XUNLOCK(pdp); 4755 if (error != 0) 4756 goto fail; 4757 if (pwd != NULL) 4758 pwd_drop(pwd); 4759 FILEDESC_SLOCK(fdp); 4760 if (refcount_load(&fdp->fd_refcnt) == 0) 4761 goto skip; 4762 FILEDESC_FOREACH_FP(fdp, i, fp) { 4763 #ifdef CAPABILITIES 4764 rights = *cap_rights(fdp, i); 4765 #else /* !CAPABILITIES */ 4766 rights = cap_no_rights; 4767 #endif 4768 /* 4769 * Create sysctl entry. It is OK to drop the filedesc 4770 * lock inside of export_file_to_sb() as we will 4771 * re-validate and re-evaluate its properties when the 4772 * loop continues. 4773 */ 4774 error = export_file_to_sb(fp, i, &rights, efbuf); 4775 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4776 break; 4777 } 4778 skip: 4779 FILEDESC_SUNLOCK(fdp); 4780 fail: 4781 if (fdp != NULL) 4782 fddrop(fdp); 4783 if (pdp != NULL) 4784 pddrop(pdp); 4785 free(efbuf, M_TEMP); 4786 return (error); 4787 } 4788 4789 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 4790 4791 /* 4792 * Get per-process file descriptors for use by procstat(1), et al. 4793 */ 4794 static int 4795 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 4796 { 4797 struct sbuf sb; 4798 struct proc *p; 4799 ssize_t maxlen; 4800 u_int namelen; 4801 int error, error2, *name; 4802 4803 namelen = arg2; 4804 if (namelen != 1) 4805 return (EINVAL); 4806 4807 name = (int *)arg1; 4808 4809 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 4810 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4811 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4812 if (error != 0) { 4813 sbuf_delete(&sb); 4814 return (error); 4815 } 4816 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4817 error = kern_proc_filedesc_out(p, &sb, maxlen, 4818 KERN_FILEDESC_PACK_KINFO); 4819 error2 = sbuf_finish(&sb); 4820 sbuf_delete(&sb); 4821 return (error != 0 ? error : error2); 4822 } 4823 4824 #ifdef COMPAT_FREEBSD7 4825 #ifdef KINFO_OFILE_SIZE 4826 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 4827 #endif 4828 4829 static void 4830 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 4831 { 4832 4833 okif->kf_structsize = sizeof(*okif); 4834 okif->kf_type = kif->kf_type; 4835 okif->kf_fd = kif->kf_fd; 4836 okif->kf_ref_count = kif->kf_ref_count; 4837 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 4838 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 4839 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 4840 okif->kf_offset = kif->kf_offset; 4841 if (kif->kf_type == KF_TYPE_VNODE) 4842 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 4843 else 4844 okif->kf_vnode_type = KF_VTYPE_VNON; 4845 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 4846 if (kif->kf_type == KF_TYPE_SOCKET) { 4847 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 4848 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 4849 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 4850 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 4851 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 4852 } else { 4853 okif->kf_sa_local.ss_family = AF_UNSPEC; 4854 okif->kf_sa_peer.ss_family = AF_UNSPEC; 4855 } 4856 } 4857 4858 static int 4859 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 4860 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 4861 { 4862 int error; 4863 4864 vrefact(vp); 4865 PWDDESC_XUNLOCK(pdp); 4866 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 4867 kinfo_to_okinfo(kif, okif); 4868 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4869 PWDDESC_XLOCK(pdp); 4870 return (error); 4871 } 4872 4873 /* 4874 * Get per-process file descriptors for use by procstat(1), et al. 4875 */ 4876 static int 4877 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 4878 { 4879 struct kinfo_ofile *okif; 4880 struct kinfo_file *kif; 4881 struct filedesc *fdp; 4882 struct pwddesc *pdp; 4883 struct pwd *pwd; 4884 u_int namelen; 4885 int error, i, *name; 4886 struct file *fp; 4887 struct proc *p; 4888 4889 namelen = arg2; 4890 if (namelen != 1) 4891 return (EINVAL); 4892 4893 name = (int *)arg1; 4894 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4895 if (error != 0) 4896 return (error); 4897 fdp = fdhold(p); 4898 if (fdp != NULL) 4899 pdp = pdhold(p); 4900 PROC_UNLOCK(p); 4901 if (fdp == NULL || pdp == NULL) { 4902 if (fdp != NULL) 4903 fddrop(fdp); 4904 return (ENOENT); 4905 } 4906 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 4907 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 4908 PWDDESC_XLOCK(pdp); 4909 pwd = pwd_hold_pwddesc(pdp); 4910 if (pwd != NULL) { 4911 if (pwd->pwd_cdir != NULL) 4912 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 4913 okif, pdp, req); 4914 if (pwd->pwd_rdir != NULL) 4915 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 4916 okif, pdp, req); 4917 if (pwd->pwd_jdir != NULL) 4918 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 4919 okif, pdp, req); 4920 } 4921 PWDDESC_XUNLOCK(pdp); 4922 if (pwd != NULL) 4923 pwd_drop(pwd); 4924 FILEDESC_SLOCK(fdp); 4925 if (refcount_load(&fdp->fd_refcnt) == 0) 4926 goto skip; 4927 FILEDESC_FOREACH_FP(fdp, i, fp) { 4928 export_file_to_kinfo(fp, i, NULL, kif, fdp, 4929 KERN_FILEDESC_PACK_KINFO); 4930 FILEDESC_SUNLOCK(fdp); 4931 kinfo_to_okinfo(kif, okif); 4932 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4933 FILEDESC_SLOCK(fdp); 4934 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4935 break; 4936 } 4937 skip: 4938 FILEDESC_SUNLOCK(fdp); 4939 fddrop(fdp); 4940 pddrop(pdp); 4941 free(kif, M_TEMP); 4942 free(okif, M_TEMP); 4943 return (0); 4944 } 4945 4946 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 4947 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 4948 "Process ofiledesc entries"); 4949 #endif /* COMPAT_FREEBSD7 */ 4950 4951 int 4952 vntype_to_kinfo(int vtype) 4953 { 4954 struct { 4955 int vtype; 4956 int kf_vtype; 4957 } vtypes_table[] = { 4958 { VBAD, KF_VTYPE_VBAD }, 4959 { VBLK, KF_VTYPE_VBLK }, 4960 { VCHR, KF_VTYPE_VCHR }, 4961 { VDIR, KF_VTYPE_VDIR }, 4962 { VFIFO, KF_VTYPE_VFIFO }, 4963 { VLNK, KF_VTYPE_VLNK }, 4964 { VNON, KF_VTYPE_VNON }, 4965 { VREG, KF_VTYPE_VREG }, 4966 { VSOCK, KF_VTYPE_VSOCK } 4967 }; 4968 unsigned int i; 4969 4970 /* 4971 * Perform vtype translation. 4972 */ 4973 for (i = 0; i < nitems(vtypes_table); i++) 4974 if (vtypes_table[i].vtype == vtype) 4975 return (vtypes_table[i].kf_vtype); 4976 4977 return (KF_VTYPE_UNKNOWN); 4978 } 4979 4980 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 4981 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 4982 "Process filedesc entries"); 4983 4984 /* 4985 * Store a process current working directory information to sbuf. 4986 * 4987 * Takes a locked proc as argument, and returns with the proc unlocked. 4988 */ 4989 int 4990 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 4991 { 4992 struct pwddesc *pdp; 4993 struct pwd *pwd; 4994 struct export_fd_buf *efbuf; 4995 struct vnode *cdir; 4996 int error; 4997 4998 PROC_LOCK_ASSERT(p, MA_OWNED); 4999 5000 pdp = pdhold(p); 5001 PROC_UNLOCK(p); 5002 if (pdp == NULL) 5003 return (EINVAL); 5004 5005 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 5006 efbuf->fdp = NULL; 5007 efbuf->pdp = pdp; 5008 efbuf->sb = sb; 5009 efbuf->remainder = maxlen; 5010 efbuf->flags = 0; 5011 5012 PWDDESC_XLOCK(pdp); 5013 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 5014 cdir = pwd->pwd_cdir; 5015 if (cdir == NULL) { 5016 error = EINVAL; 5017 } else { 5018 vrefact(cdir); 5019 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 5020 } 5021 PWDDESC_XUNLOCK(pdp); 5022 pddrop(pdp); 5023 free(efbuf, M_TEMP); 5024 return (error); 5025 } 5026 5027 /* 5028 * Get per-process current working directory. 5029 */ 5030 static int 5031 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 5032 { 5033 struct sbuf sb; 5034 struct proc *p; 5035 ssize_t maxlen; 5036 u_int namelen; 5037 int error, error2, *name; 5038 5039 namelen = arg2; 5040 if (namelen != 1) 5041 return (EINVAL); 5042 5043 name = (int *)arg1; 5044 5045 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 5046 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5047 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5048 if (error != 0) { 5049 sbuf_delete(&sb); 5050 return (error); 5051 } 5052 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5053 error = kern_proc_cwd_out(p, &sb, maxlen); 5054 error2 = sbuf_finish(&sb); 5055 sbuf_delete(&sb); 5056 return (error != 0 ? error : error2); 5057 } 5058 5059 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 5060 sysctl_kern_proc_cwd, "Process current working directory"); 5061 5062 #ifdef DDB 5063 /* 5064 * For the purposes of debugging, generate a human-readable string for the 5065 * file type. 5066 */ 5067 static const char * 5068 file_type_to_name(short type) 5069 { 5070 5071 switch (type) { 5072 case 0: 5073 return ("zero"); 5074 case DTYPE_VNODE: 5075 return ("vnode"); 5076 case DTYPE_SOCKET: 5077 return ("socket"); 5078 case DTYPE_PIPE: 5079 return ("pipe"); 5080 case DTYPE_FIFO: 5081 return ("fifo"); 5082 case DTYPE_KQUEUE: 5083 return ("kqueue"); 5084 case DTYPE_CRYPTO: 5085 return ("crypto"); 5086 case DTYPE_MQUEUE: 5087 return ("mqueue"); 5088 case DTYPE_SHM: 5089 return ("shm"); 5090 case DTYPE_SEM: 5091 return ("ksem"); 5092 case DTYPE_PTS: 5093 return ("pts"); 5094 case DTYPE_DEV: 5095 return ("dev"); 5096 case DTYPE_PROCDESC: 5097 return ("proc"); 5098 case DTYPE_EVENTFD: 5099 return ("eventfd"); 5100 case DTYPE_TIMERFD: 5101 return ("timerfd"); 5102 default: 5103 return ("unkn"); 5104 } 5105 } 5106 5107 /* 5108 * For the purposes of debugging, identify a process (if any, perhaps one of 5109 * many) that references the passed file in its file descriptor array. Return 5110 * NULL if none. 5111 */ 5112 static struct proc * 5113 file_to_first_proc(struct file *fp) 5114 { 5115 struct filedesc *fdp; 5116 struct proc *p; 5117 int n; 5118 5119 FOREACH_PROC_IN_SYSTEM(p) { 5120 if (p->p_state == PRS_NEW) 5121 continue; 5122 fdp = p->p_fd; 5123 if (fdp == NULL) 5124 continue; 5125 for (n = 0; n < fdp->fd_nfiles; n++) { 5126 if (fp == fdp->fd_ofiles[n].fde_file) 5127 return (p); 5128 } 5129 } 5130 return (NULL); 5131 } 5132 5133 static void 5134 db_print_file(struct file *fp, int header) 5135 { 5136 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 5137 struct proc *p; 5138 5139 if (header) 5140 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 5141 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 5142 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 5143 "FCmd"); 5144 p = file_to_first_proc(fp); 5145 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 5146 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 5147 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 5148 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 5149 5150 #undef XPTRWIDTH 5151 } 5152 5153 DB_SHOW_COMMAND(file, db_show_file) 5154 { 5155 struct file *fp; 5156 5157 if (!have_addr) { 5158 db_printf("usage: show file <addr>\n"); 5159 return; 5160 } 5161 fp = (struct file *)addr; 5162 db_print_file(fp, 1); 5163 } 5164 5165 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE) 5166 { 5167 struct filedesc *fdp; 5168 struct file *fp; 5169 struct proc *p; 5170 int header; 5171 int n; 5172 5173 header = 1; 5174 FOREACH_PROC_IN_SYSTEM(p) { 5175 if (p->p_state == PRS_NEW) 5176 continue; 5177 if ((fdp = p->p_fd) == NULL) 5178 continue; 5179 for (n = 0; n < fdp->fd_nfiles; ++n) { 5180 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 5181 continue; 5182 db_print_file(fp, header); 5183 header = 0; 5184 } 5185 } 5186 } 5187 #endif 5188 5189 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, 5190 CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5191 &maxfilesperproc, 0, "Maximum files allowed open per process"); 5192 5193 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5194 &maxfiles, 0, "Maximum number of files"); 5195 5196 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 5197 &openfiles, 0, "System-wide number of open files"); 5198 5199 /* ARGSUSED*/ 5200 static void 5201 filelistinit(void *dummy) 5202 { 5203 5204 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 5205 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 5206 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 5207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 5208 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 5209 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 5210 /* 5211 * XXXMJG this is a temporary hack due to boot ordering issues against 5212 * the vnode zone. 5213 */ 5214 vfs_smr = uma_zone_get_smr(pwd_zone); 5215 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 5216 } 5217 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 5218 5219 /*-------------------------------------------------------------------*/ 5220 5221 static int 5222 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 5223 int flags, struct thread *td) 5224 { 5225 5226 return (EBADF); 5227 } 5228 5229 static int 5230 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5231 struct thread *td) 5232 { 5233 5234 return (EINVAL); 5235 } 5236 5237 static int 5238 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 5239 struct thread *td) 5240 { 5241 5242 return (EBADF); 5243 } 5244 5245 static int 5246 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 5247 struct thread *td) 5248 { 5249 5250 return (0); 5251 } 5252 5253 static int 5254 badfo_kqfilter(struct file *fp, struct knote *kn) 5255 { 5256 5257 return (EBADF); 5258 } 5259 5260 static int 5261 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 5262 { 5263 5264 return (EBADF); 5265 } 5266 5267 static int 5268 badfo_close(struct file *fp, struct thread *td) 5269 { 5270 5271 return (0); 5272 } 5273 5274 static int 5275 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5276 struct thread *td) 5277 { 5278 5279 return (EBADF); 5280 } 5281 5282 static int 5283 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5284 struct thread *td) 5285 { 5286 5287 return (EBADF); 5288 } 5289 5290 static int 5291 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5292 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5293 struct thread *td) 5294 { 5295 5296 return (EBADF); 5297 } 5298 5299 static int 5300 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 5301 { 5302 5303 return (0); 5304 } 5305 5306 const struct fileops badfileops = { 5307 .fo_read = badfo_readwrite, 5308 .fo_write = badfo_readwrite, 5309 .fo_truncate = badfo_truncate, 5310 .fo_ioctl = badfo_ioctl, 5311 .fo_poll = badfo_poll, 5312 .fo_kqfilter = badfo_kqfilter, 5313 .fo_stat = badfo_stat, 5314 .fo_close = badfo_close, 5315 .fo_chmod = badfo_chmod, 5316 .fo_chown = badfo_chown, 5317 .fo_sendfile = badfo_sendfile, 5318 .fo_fill_kinfo = badfo_fill_kinfo, 5319 }; 5320 5321 static int 5322 path_poll(struct file *fp, int events, struct ucred *active_cred, 5323 struct thread *td) 5324 { 5325 return (POLLNVAL); 5326 } 5327 5328 static int 5329 path_close(struct file *fp, struct thread *td) 5330 { 5331 MPASS(fp->f_type == DTYPE_VNODE); 5332 fp->f_ops = &badfileops; 5333 vrele(fp->f_vnode); 5334 return (0); 5335 } 5336 5337 const struct fileops path_fileops = { 5338 .fo_read = badfo_readwrite, 5339 .fo_write = badfo_readwrite, 5340 .fo_truncate = badfo_truncate, 5341 .fo_ioctl = badfo_ioctl, 5342 .fo_poll = path_poll, 5343 .fo_kqfilter = vn_kqfilter_opath, 5344 .fo_stat = vn_statfile, 5345 .fo_close = path_close, 5346 .fo_chmod = badfo_chmod, 5347 .fo_chown = badfo_chown, 5348 .fo_sendfile = badfo_sendfile, 5349 .fo_fill_kinfo = vn_fill_kinfo, 5350 .fo_cmp = vn_cmp, 5351 .fo_flags = DFLAG_PASSABLE, 5352 }; 5353 5354 int 5355 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5356 int flags, struct thread *td) 5357 { 5358 5359 return (EOPNOTSUPP); 5360 } 5361 5362 int 5363 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5364 struct thread *td) 5365 { 5366 5367 return (EINVAL); 5368 } 5369 5370 int 5371 invfo_ioctl(struct file *fp, u_long com, void *data, 5372 struct ucred *active_cred, struct thread *td) 5373 { 5374 5375 return (ENOTTY); 5376 } 5377 5378 int 5379 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5380 struct thread *td) 5381 { 5382 5383 return (poll_no_poll(events)); 5384 } 5385 5386 int 5387 invfo_kqfilter(struct file *fp, struct knote *kn) 5388 { 5389 5390 return (EINVAL); 5391 } 5392 5393 int 5394 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5395 struct thread *td) 5396 { 5397 5398 return (EINVAL); 5399 } 5400 5401 int 5402 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5403 struct thread *td) 5404 { 5405 5406 return (EINVAL); 5407 } 5408 5409 int 5410 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5411 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5412 struct thread *td) 5413 { 5414 5415 return (EINVAL); 5416 } 5417 5418 /*-------------------------------------------------------------------*/ 5419 5420 /* 5421 * File Descriptor pseudo-device driver (/dev/fd/). 5422 * 5423 * Opening minor device N dup()s the file (if any) connected to file 5424 * descriptor N belonging to the calling process. Note that this driver 5425 * consists of only the ``open()'' routine, because all subsequent 5426 * references to this file will be direct to the other driver. 5427 * 5428 * XXX: we could give this one a cloning event handler if necessary. 5429 */ 5430 5431 /* ARGSUSED */ 5432 static int 5433 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5434 { 5435 5436 /* 5437 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5438 * the file descriptor being sought for duplication. The error 5439 * return ensures that the vnode for this device will be released 5440 * by vn_open. Open will detect this special error and take the 5441 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5442 * will simply report the error. 5443 */ 5444 td->td_dupfd = dev2unit(dev); 5445 return (ENODEV); 5446 } 5447 5448 static struct cdevsw fildesc_cdevsw = { 5449 .d_version = D_VERSION, 5450 .d_open = fdopen, 5451 .d_name = "FD", 5452 }; 5453 5454 static void 5455 fildesc_drvinit(void *unused) 5456 { 5457 struct cdev *dev; 5458 5459 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5460 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5461 make_dev_alias(dev, "stdin"); 5462 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5463 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5464 make_dev_alias(dev, "stdout"); 5465 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5466 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5467 make_dev_alias(dev, "stderr"); 5468 } 5469 5470 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5471