1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_capsicum.h" 43 #include "opt_ddb.h" 44 #include "opt_ktrace.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 49 #include <sys/capsicum.h> 50 #include <sys/conf.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/filio.h> 55 #include <sys/jail.h> 56 #include <sys/kernel.h> 57 #include <sys/limits.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mount.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/selinfo.h> 64 #include <sys/poll.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/protosw.h> 68 #include <sys/racct.h> 69 #include <sys/resourcevar.h> 70 #include <sys/sbuf.h> 71 #include <sys/signalvar.h> 72 #include <sys/kdb.h> 73 #include <sys/smr.h> 74 #include <sys/stat.h> 75 #include <sys/sx.h> 76 #include <sys/syscallsubr.h> 77 #include <sys/sysctl.h> 78 #include <sys/sysproto.h> 79 #include <sys/unistd.h> 80 #include <sys/user.h> 81 #include <sys/vnode.h> 82 #include <sys/ktrace.h> 83 84 #include <net/vnet.h> 85 86 #include <security/audit/audit.h> 87 88 #include <vm/uma.h> 89 #include <vm/vm.h> 90 91 #include <ddb/ddb.h> 92 93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes"); 95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors"); 96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 97 "file desc to leader structures"); 98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 100 101 MALLOC_DECLARE(M_FADVISE); 102 103 static __read_mostly uma_zone_t file_zone; 104 static __read_mostly uma_zone_t filedesc0_zone; 105 __read_mostly uma_zone_t pwd_zone; 106 VFS_SMR_DECLARE; 107 108 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 109 struct thread *td, bool holdleaders, bool audit); 110 static int fd_first_free(struct filedesc *fdp, int low, int size); 111 static void fdgrowtable(struct filedesc *fdp, int nfd); 112 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 113 static void fdunused(struct filedesc *fdp, int fd); 114 static void fdused(struct filedesc *fdp, int fd); 115 static int getmaxfd(struct thread *td); 116 static u_long *filecaps_copy_prep(const struct filecaps *src); 117 static void filecaps_copy_finish(const struct filecaps *src, 118 struct filecaps *dst, u_long *ioctls); 119 static u_long *filecaps_free_prep(struct filecaps *fcaps); 120 static void filecaps_free_finish(u_long *ioctls); 121 122 static struct pwd *pwd_alloc(void); 123 124 /* 125 * Each process has: 126 * 127 * - An array of open file descriptors (fd_ofiles) 128 * - An array of file flags (fd_ofileflags) 129 * - A bitmap recording which descriptors are in use (fd_map) 130 * 131 * A process starts out with NDFILE descriptors. The value of NDFILE has 132 * been selected based the historical limit of 20 open files, and an 133 * assumption that the majority of processes, especially short-lived 134 * processes like shells, will never need more. 135 * 136 * If this initial allocation is exhausted, a larger descriptor table and 137 * map are allocated dynamically, and the pointers in the process's struct 138 * filedesc are updated to point to those. This is repeated every time 139 * the process runs out of file descriptors (provided it hasn't hit its 140 * resource limit). 141 * 142 * Since threads may hold references to individual descriptor table 143 * entries, the tables are never freed. Instead, they are placed on a 144 * linked list and freed only when the struct filedesc is released. 145 */ 146 #define NDFILE 20 147 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 148 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 149 #define NDSLOT(x) ((x) / NDENTRIES) 150 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 151 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 152 153 /* 154 * SLIST entry used to keep track of ofiles which must be reclaimed when 155 * the process exits. 156 */ 157 struct freetable { 158 struct fdescenttbl *ft_table; 159 SLIST_ENTRY(freetable) ft_next; 160 }; 161 162 /* 163 * Initial allocation: a filedesc structure + the head of SLIST used to 164 * keep track of old ofiles + enough space for NDFILE descriptors. 165 */ 166 167 struct fdescenttbl0 { 168 int fdt_nfiles; 169 struct filedescent fdt_ofiles[NDFILE]; 170 }; 171 172 struct filedesc0 { 173 struct filedesc fd_fd; 174 SLIST_HEAD(, freetable) fd_free; 175 struct fdescenttbl0 fd_dfiles; 176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 177 }; 178 179 /* 180 * Descriptor management. 181 */ 182 static int __exclusive_cache_line openfiles; /* actual number of open files */ 183 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 185 186 /* 187 * If low >= size, just return low. Otherwise find the first zero bit in the 188 * given bitmap, starting at low and not exceeding size - 1. Return size if 189 * not found. 190 */ 191 static int 192 fd_first_free(struct filedesc *fdp, int low, int size) 193 { 194 NDSLOTTYPE *map = fdp->fd_map; 195 NDSLOTTYPE mask; 196 int off, maxoff; 197 198 if (low >= size) 199 return (low); 200 201 off = NDSLOT(low); 202 if (low % NDENTRIES) { 203 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 204 if ((mask &= ~map[off]) != 0UL) 205 return (off * NDENTRIES + ffsl(mask) - 1); 206 ++off; 207 } 208 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 209 if (map[off] != ~0UL) 210 return (off * NDENTRIES + ffsl(~map[off]) - 1); 211 return (size); 212 } 213 214 /* 215 * Find the last used fd. 216 * 217 * Call this variant if fdp can't be modified by anyone else (e.g, during exec). 218 * Otherwise use fdlastfile. 219 */ 220 int 221 fdlastfile_single(struct filedesc *fdp) 222 { 223 NDSLOTTYPE *map = fdp->fd_map; 224 int off, minoff; 225 226 off = NDSLOT(fdp->fd_nfiles - 1); 227 for (minoff = NDSLOT(0); off >= minoff; --off) 228 if (map[off] != 0) 229 return (off * NDENTRIES + flsl(map[off]) - 1); 230 return (-1); 231 } 232 233 int 234 fdlastfile(struct filedesc *fdp) 235 { 236 237 FILEDESC_LOCK_ASSERT(fdp); 238 return (fdlastfile_single(fdp)); 239 } 240 241 static int 242 fdisused(struct filedesc *fdp, int fd) 243 { 244 245 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 246 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 247 248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 249 } 250 251 /* 252 * Mark a file descriptor as used. 253 */ 254 static void 255 fdused_init(struct filedesc *fdp, int fd) 256 { 257 258 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 259 260 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 261 } 262 263 static void 264 fdused(struct filedesc *fdp, int fd) 265 { 266 267 FILEDESC_XLOCK_ASSERT(fdp); 268 269 fdused_init(fdp, fd); 270 if (fd == fdp->fd_freefile) 271 fdp->fd_freefile++; 272 } 273 274 /* 275 * Mark a file descriptor as unused. 276 */ 277 static void 278 fdunused(struct filedesc *fdp, int fd) 279 { 280 281 FILEDESC_XLOCK_ASSERT(fdp); 282 283 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 284 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 285 ("fd=%d is still in use", fd)); 286 287 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 288 if (fd < fdp->fd_freefile) 289 fdp->fd_freefile = fd; 290 } 291 292 /* 293 * Free a file descriptor. 294 * 295 * Avoid some work if fdp is about to be destroyed. 296 */ 297 static inline void 298 fdefree_last(struct filedescent *fde) 299 { 300 301 filecaps_free(&fde->fde_caps); 302 } 303 304 static inline void 305 fdfree(struct filedesc *fdp, int fd) 306 { 307 struct filedescent *fde; 308 309 FILEDESC_XLOCK_ASSERT(fdp); 310 fde = &fdp->fd_ofiles[fd]; 311 #ifdef CAPABILITIES 312 seqc_write_begin(&fde->fde_seqc); 313 #endif 314 fde->fde_file = NULL; 315 #ifdef CAPABILITIES 316 seqc_write_end(&fde->fde_seqc); 317 #endif 318 fdefree_last(fde); 319 fdunused(fdp, fd); 320 } 321 322 /* 323 * System calls on descriptors. 324 */ 325 #ifndef _SYS_SYSPROTO_H_ 326 struct getdtablesize_args { 327 int dummy; 328 }; 329 #endif 330 /* ARGSUSED */ 331 int 332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 333 { 334 #ifdef RACCT 335 uint64_t lim; 336 #endif 337 338 td->td_retval[0] = getmaxfd(td); 339 #ifdef RACCT 340 PROC_LOCK(td->td_proc); 341 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 342 PROC_UNLOCK(td->td_proc); 343 if (lim < td->td_retval[0]) 344 td->td_retval[0] = lim; 345 #endif 346 return (0); 347 } 348 349 /* 350 * Duplicate a file descriptor to a particular value. 351 * 352 * Note: keep in mind that a potential race condition exists when closing 353 * descriptors from a shared descriptor table (via rfork). 354 */ 355 #ifndef _SYS_SYSPROTO_H_ 356 struct dup2_args { 357 u_int from; 358 u_int to; 359 }; 360 #endif 361 /* ARGSUSED */ 362 int 363 sys_dup2(struct thread *td, struct dup2_args *uap) 364 { 365 366 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 367 } 368 369 /* 370 * Duplicate a file descriptor. 371 */ 372 #ifndef _SYS_SYSPROTO_H_ 373 struct dup_args { 374 u_int fd; 375 }; 376 #endif 377 /* ARGSUSED */ 378 int 379 sys_dup(struct thread *td, struct dup_args *uap) 380 { 381 382 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 383 } 384 385 /* 386 * The file control system call. 387 */ 388 #ifndef _SYS_SYSPROTO_H_ 389 struct fcntl_args { 390 int fd; 391 int cmd; 392 long arg; 393 }; 394 #endif 395 /* ARGSUSED */ 396 int 397 sys_fcntl(struct thread *td, struct fcntl_args *uap) 398 { 399 400 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 401 } 402 403 int 404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 405 { 406 struct flock fl; 407 struct __oflock ofl; 408 intptr_t arg1; 409 int error, newcmd; 410 411 error = 0; 412 newcmd = cmd; 413 switch (cmd) { 414 case F_OGETLK: 415 case F_OSETLK: 416 case F_OSETLKW: 417 /* 418 * Convert old flock structure to new. 419 */ 420 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 421 fl.l_start = ofl.l_start; 422 fl.l_len = ofl.l_len; 423 fl.l_pid = ofl.l_pid; 424 fl.l_type = ofl.l_type; 425 fl.l_whence = ofl.l_whence; 426 fl.l_sysid = 0; 427 428 switch (cmd) { 429 case F_OGETLK: 430 newcmd = F_GETLK; 431 break; 432 case F_OSETLK: 433 newcmd = F_SETLK; 434 break; 435 case F_OSETLKW: 436 newcmd = F_SETLKW; 437 break; 438 } 439 arg1 = (intptr_t)&fl; 440 break; 441 case F_GETLK: 442 case F_SETLK: 443 case F_SETLKW: 444 case F_SETLK_REMOTE: 445 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 446 arg1 = (intptr_t)&fl; 447 break; 448 default: 449 arg1 = arg; 450 break; 451 } 452 if (error) 453 return (error); 454 error = kern_fcntl(td, fd, newcmd, arg1); 455 if (error) 456 return (error); 457 if (cmd == F_OGETLK) { 458 ofl.l_start = fl.l_start; 459 ofl.l_len = fl.l_len; 460 ofl.l_pid = fl.l_pid; 461 ofl.l_type = fl.l_type; 462 ofl.l_whence = fl.l_whence; 463 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 464 } else if (cmd == F_GETLK) { 465 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 466 } 467 return (error); 468 } 469 470 int 471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 472 { 473 struct filedesc *fdp; 474 struct flock *flp; 475 struct file *fp, *fp2; 476 struct filedescent *fde; 477 struct proc *p; 478 struct vnode *vp; 479 struct mount *mp; 480 int error, flg, seals, tmp; 481 uint64_t bsize; 482 off_t foffset; 483 484 error = 0; 485 flg = F_POSIX; 486 p = td->td_proc; 487 fdp = p->p_fd; 488 489 AUDIT_ARG_FD(cmd); 490 AUDIT_ARG_CMD(cmd); 491 switch (cmd) { 492 case F_DUPFD: 493 tmp = arg; 494 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 495 break; 496 497 case F_DUPFD_CLOEXEC: 498 tmp = arg; 499 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 500 break; 501 502 case F_DUP2FD: 503 tmp = arg; 504 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 505 break; 506 507 case F_DUP2FD_CLOEXEC: 508 tmp = arg; 509 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 510 break; 511 512 case F_GETFD: 513 error = EBADF; 514 FILEDESC_SLOCK(fdp); 515 fde = fdeget_locked(fdp, fd); 516 if (fde != NULL) { 517 td->td_retval[0] = 518 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 519 error = 0; 520 } 521 FILEDESC_SUNLOCK(fdp); 522 break; 523 524 case F_SETFD: 525 error = EBADF; 526 FILEDESC_XLOCK(fdp); 527 fde = fdeget_locked(fdp, fd); 528 if (fde != NULL) { 529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 531 error = 0; 532 } 533 FILEDESC_XUNLOCK(fdp); 534 break; 535 536 case F_GETFL: 537 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); 538 if (error != 0) 539 break; 540 td->td_retval[0] = OFLAGS(fp->f_flag); 541 fdrop(fp, td); 542 break; 543 544 case F_SETFL: 545 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); 546 if (error != 0) 547 break; 548 if (fp->f_ops == &path_fileops) { 549 fdrop(fp, td); 550 error = EBADF; 551 break; 552 } 553 do { 554 tmp = flg = fp->f_flag; 555 tmp &= ~FCNTLFLAGS; 556 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 557 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 558 tmp = fp->f_flag & FNONBLOCK; 559 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 560 if (error != 0) { 561 fdrop(fp, td); 562 break; 563 } 564 tmp = fp->f_flag & FASYNC; 565 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 566 if (error == 0) { 567 fdrop(fp, td); 568 break; 569 } 570 atomic_clear_int(&fp->f_flag, FNONBLOCK); 571 tmp = 0; 572 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 573 fdrop(fp, td); 574 break; 575 576 case F_GETOWN: 577 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 578 if (error != 0) 579 break; 580 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 581 if (error == 0) 582 td->td_retval[0] = tmp; 583 fdrop(fp, td); 584 break; 585 586 case F_SETOWN: 587 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 588 if (error != 0) 589 break; 590 tmp = arg; 591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 592 fdrop(fp, td); 593 break; 594 595 case F_SETLK_REMOTE: 596 error = priv_check(td, PRIV_NFS_LOCKD); 597 if (error != 0) 598 return (error); 599 flg = F_REMOTE; 600 goto do_setlk; 601 602 case F_SETLKW: 603 flg |= F_WAIT; 604 /* FALLTHROUGH F_SETLK */ 605 606 case F_SETLK: 607 do_setlk: 608 flp = (struct flock *)arg; 609 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { 610 error = EINVAL; 611 break; 612 } 613 614 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp); 615 if (error != 0) 616 break; 617 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 618 error = EBADF; 619 fdrop(fp, td); 620 break; 621 } 622 623 if (flp->l_whence == SEEK_CUR) { 624 foffset = foffset_get(fp); 625 if (foffset < 0 || 626 (flp->l_start > 0 && 627 foffset > OFF_MAX - flp->l_start)) { 628 error = EOVERFLOW; 629 fdrop(fp, td); 630 break; 631 } 632 flp->l_start += foffset; 633 } 634 635 vp = fp->f_vnode; 636 switch (flp->l_type) { 637 case F_RDLCK: 638 if ((fp->f_flag & FREAD) == 0) { 639 error = EBADF; 640 break; 641 } 642 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 643 PROC_LOCK(p->p_leader); 644 p->p_leader->p_flag |= P_ADVLOCK; 645 PROC_UNLOCK(p->p_leader); 646 } 647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 648 flp, flg); 649 break; 650 case F_WRLCK: 651 if ((fp->f_flag & FWRITE) == 0) { 652 error = EBADF; 653 break; 654 } 655 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 656 PROC_LOCK(p->p_leader); 657 p->p_leader->p_flag |= P_ADVLOCK; 658 PROC_UNLOCK(p->p_leader); 659 } 660 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 661 flp, flg); 662 break; 663 case F_UNLCK: 664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 665 flp, flg); 666 break; 667 case F_UNLCKSYS: 668 if (flg != F_REMOTE) { 669 error = EINVAL; 670 break; 671 } 672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 673 F_UNLCKSYS, flp, flg); 674 break; 675 default: 676 error = EINVAL; 677 break; 678 } 679 if (error != 0 || flp->l_type == F_UNLCK || 680 flp->l_type == F_UNLCKSYS) { 681 fdrop(fp, td); 682 break; 683 } 684 685 /* 686 * Check for a race with close. 687 * 688 * The vnode is now advisory locked (or unlocked, but this case 689 * is not really important) as the caller requested. 690 * We had to drop the filedesc lock, so we need to recheck if 691 * the descriptor is still valid, because if it was closed 692 * in the meantime we need to remove advisory lock from the 693 * vnode - close on any descriptor leading to an advisory 694 * locked vnode, removes that lock. 695 * We will return 0 on purpose in that case, as the result of 696 * successful advisory lock might have been externally visible 697 * already. This is fine - effectively we pretend to the caller 698 * that the closing thread was a bit slower and that the 699 * advisory lock succeeded before the close. 700 */ 701 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2); 702 if (error != 0) { 703 fdrop(fp, td); 704 break; 705 } 706 if (fp != fp2) { 707 flp->l_whence = SEEK_SET; 708 flp->l_start = 0; 709 flp->l_len = 0; 710 flp->l_type = F_UNLCK; 711 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 712 F_UNLCK, flp, F_POSIX); 713 } 714 fdrop(fp, td); 715 fdrop(fp2, td); 716 break; 717 718 case F_GETLK: 719 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp); 720 if (error != 0) 721 break; 722 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 723 error = EBADF; 724 fdrop(fp, td); 725 break; 726 } 727 flp = (struct flock *)arg; 728 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 729 flp->l_type != F_UNLCK) { 730 error = EINVAL; 731 fdrop(fp, td); 732 break; 733 } 734 if (flp->l_whence == SEEK_CUR) { 735 foffset = foffset_get(fp); 736 if ((flp->l_start > 0 && 737 foffset > OFF_MAX - flp->l_start) || 738 (flp->l_start < 0 && 739 foffset < OFF_MIN - flp->l_start)) { 740 error = EOVERFLOW; 741 fdrop(fp, td); 742 break; 743 } 744 flp->l_start += foffset; 745 } 746 vp = fp->f_vnode; 747 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 748 F_POSIX); 749 fdrop(fp, td); 750 break; 751 752 case F_ADD_SEALS: 753 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); 754 if (error != 0) 755 break; 756 error = fo_add_seals(fp, arg); 757 fdrop(fp, td); 758 break; 759 760 case F_GET_SEALS: 761 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); 762 if (error != 0) 763 break; 764 if (fo_get_seals(fp, &seals) == 0) 765 td->td_retval[0] = seals; 766 else 767 error = EINVAL; 768 fdrop(fp, td); 769 break; 770 771 case F_RDAHEAD: 772 arg = arg ? 128 * 1024: 0; 773 /* FALLTHROUGH */ 774 case F_READAHEAD: 775 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); 776 if (error != 0) 777 break; 778 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 779 fdrop(fp, td); 780 error = EBADF; 781 break; 782 } 783 vp = fp->f_vnode; 784 if (vp->v_type != VREG) { 785 fdrop(fp, td); 786 error = ENOTTY; 787 break; 788 } 789 790 /* 791 * Exclusive lock synchronizes against f_seqcount reads and 792 * writes in sequential_heuristic(). 793 */ 794 error = vn_lock(vp, LK_EXCLUSIVE); 795 if (error != 0) { 796 fdrop(fp, td); 797 break; 798 } 799 if (arg >= 0) { 800 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 801 arg = MIN(arg, INT_MAX - bsize + 1); 802 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, 803 (arg + bsize - 1) / bsize); 804 atomic_set_int(&fp->f_flag, FRDAHEAD); 805 } else { 806 atomic_clear_int(&fp->f_flag, FRDAHEAD); 807 } 808 VOP_UNLOCK(vp); 809 fdrop(fp, td); 810 break; 811 812 case F_ISUNIONSTACK: 813 /* 814 * Check if the vnode is part of a union stack (either the 815 * "union" flag from mount(2) or unionfs). 816 * 817 * Prior to introduction of this op libc's readdir would call 818 * fstatfs(2), in effect unnecessarily copying kilobytes of 819 * data just to check fs name and a mount flag. 820 * 821 * Fixing the code to handle everything in the kernel instead 822 * is a non-trivial endeavor and has low priority, thus this 823 * horrible kludge facilitates the current behavior in a much 824 * cheaper manner until someone(tm) sorts this out. 825 */ 826 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); 827 if (error != 0) 828 break; 829 if (fp->f_type != DTYPE_VNODE) { 830 fdrop(fp, td); 831 error = EBADF; 832 break; 833 } 834 vp = fp->f_vnode; 835 /* 836 * Since we don't prevent dooming the vnode even non-null mp 837 * found can become immediately stale. This is tolerable since 838 * mount points are type-stable (providing safe memory access) 839 * and any vfs op on this vnode going forward will return an 840 * error (meaning return value in this case is meaningless). 841 */ 842 mp = atomic_load_ptr(&vp->v_mount); 843 if (__predict_false(mp == NULL)) { 844 fdrop(fp, td); 845 error = EBADF; 846 break; 847 } 848 td->td_retval[0] = 0; 849 if (mp->mnt_kern_flag & MNTK_UNIONFS || 850 mp->mnt_flag & MNT_UNION) 851 td->td_retval[0] = 1; 852 fdrop(fp, td); 853 break; 854 855 default: 856 error = EINVAL; 857 break; 858 } 859 return (error); 860 } 861 862 static int 863 getmaxfd(struct thread *td) 864 { 865 866 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 867 } 868 869 /* 870 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 871 */ 872 int 873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 874 { 875 struct filedesc *fdp; 876 struct filedescent *oldfde, *newfde; 877 struct proc *p; 878 struct file *delfp, *oldfp; 879 u_long *oioctls, *nioctls; 880 int error, maxfd; 881 882 p = td->td_proc; 883 fdp = p->p_fd; 884 oioctls = NULL; 885 886 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 887 MPASS(mode < FDDUP_LASTMODE); 888 889 AUDIT_ARG_FD(old); 890 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 891 892 /* 893 * Verify we have a valid descriptor to dup from and possibly to 894 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 895 * return EINVAL when the new descriptor is out of bounds. 896 */ 897 if (old < 0) 898 return (EBADF); 899 if (new < 0) 900 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 901 maxfd = getmaxfd(td); 902 if (new >= maxfd) 903 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 904 905 error = EBADF; 906 FILEDESC_XLOCK(fdp); 907 if (fget_locked(fdp, old) == NULL) 908 goto unlock; 909 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 910 td->td_retval[0] = new; 911 if (flags & FDDUP_FLAG_CLOEXEC) 912 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 913 error = 0; 914 goto unlock; 915 } 916 917 oldfde = &fdp->fd_ofiles[old]; 918 oldfp = oldfde->fde_file; 919 if (!fhold(oldfp)) 920 goto unlock; 921 922 /* 923 * If the caller specified a file descriptor, make sure the file 924 * table is large enough to hold it, and grab it. Otherwise, just 925 * allocate a new descriptor the usual way. 926 */ 927 switch (mode) { 928 case FDDUP_NORMAL: 929 case FDDUP_FCNTL: 930 if ((error = fdalloc(td, new, &new)) != 0) { 931 fdrop(oldfp, td); 932 goto unlock; 933 } 934 break; 935 case FDDUP_MUSTREPLACE: 936 /* Target file descriptor must exist. */ 937 if (fget_locked(fdp, new) == NULL) { 938 fdrop(oldfp, td); 939 goto unlock; 940 } 941 break; 942 case FDDUP_FIXED: 943 if (new >= fdp->fd_nfiles) { 944 /* 945 * The resource limits are here instead of e.g. 946 * fdalloc(), because the file descriptor table may be 947 * shared between processes, so we can't really use 948 * racct_add()/racct_sub(). Instead of counting the 949 * number of actually allocated descriptors, just put 950 * the limit on the size of the file descriptor table. 951 */ 952 #ifdef RACCT 953 if (RACCT_ENABLED()) { 954 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); 955 if (error != 0) { 956 error = EMFILE; 957 fdrop(oldfp, td); 958 goto unlock; 959 } 960 } 961 #endif 962 fdgrowtable_exp(fdp, new + 1); 963 } 964 if (!fdisused(fdp, new)) 965 fdused(fdp, new); 966 break; 967 default: 968 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 969 } 970 971 KASSERT(old != new, ("new fd is same as old")); 972 973 /* Refetch oldfde because the table may have grown and old one freed. */ 974 oldfde = &fdp->fd_ofiles[old]; 975 KASSERT(oldfp == oldfde->fde_file, 976 ("fdt_ofiles shift from growth observed at fd %d", 977 old)); 978 979 newfde = &fdp->fd_ofiles[new]; 980 delfp = newfde->fde_file; 981 982 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 983 984 /* 985 * Duplicate the source descriptor. 986 */ 987 #ifdef CAPABILITIES 988 seqc_write_begin(&newfde->fde_seqc); 989 #endif 990 oioctls = filecaps_free_prep(&newfde->fde_caps); 991 memcpy(newfde, oldfde, fde_change_size); 992 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 993 nioctls); 994 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 995 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 996 else 997 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 998 #ifdef CAPABILITIES 999 seqc_write_end(&newfde->fde_seqc); 1000 #endif 1001 td->td_retval[0] = new; 1002 1003 error = 0; 1004 1005 if (delfp != NULL) { 1006 (void) closefp(fdp, new, delfp, td, true, false); 1007 FILEDESC_UNLOCK_ASSERT(fdp); 1008 } else { 1009 unlock: 1010 FILEDESC_XUNLOCK(fdp); 1011 } 1012 1013 filecaps_free_finish(oioctls); 1014 return (error); 1015 } 1016 1017 static void 1018 sigiofree(struct sigio *sigio) 1019 { 1020 crfree(sigio->sio_ucred); 1021 free(sigio, M_SIGIO); 1022 } 1023 1024 static struct sigio * 1025 funsetown_locked(struct sigio *sigio) 1026 { 1027 struct proc *p; 1028 struct pgrp *pg; 1029 1030 SIGIO_ASSERT_LOCKED(); 1031 1032 if (sigio == NULL) 1033 return (NULL); 1034 *(sigio->sio_myref) = NULL; 1035 if (sigio->sio_pgid < 0) { 1036 pg = sigio->sio_pgrp; 1037 PGRP_LOCK(pg); 1038 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 1039 sigio, sio_pgsigio); 1040 PGRP_UNLOCK(pg); 1041 } else { 1042 p = sigio->sio_proc; 1043 PROC_LOCK(p); 1044 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 1045 sigio, sio_pgsigio); 1046 PROC_UNLOCK(p); 1047 } 1048 return (sigio); 1049 } 1050 1051 /* 1052 * If sigio is on the list associated with a process or process group, 1053 * disable signalling from the device, remove sigio from the list and 1054 * free sigio. 1055 */ 1056 void 1057 funsetown(struct sigio **sigiop) 1058 { 1059 struct sigio *sigio; 1060 1061 /* Racy check, consumers must provide synchronization. */ 1062 if (*sigiop == NULL) 1063 return; 1064 1065 SIGIO_LOCK(); 1066 sigio = funsetown_locked(*sigiop); 1067 SIGIO_UNLOCK(); 1068 if (sigio != NULL) 1069 sigiofree(sigio); 1070 } 1071 1072 /* 1073 * Free a list of sigio structures. The caller must ensure that new sigio 1074 * structures cannot be added after this point. For process groups this is 1075 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves 1076 * as an interlock. 1077 */ 1078 void 1079 funsetownlst(struct sigiolst *sigiolst) 1080 { 1081 struct proc *p; 1082 struct pgrp *pg; 1083 struct sigio *sigio, *tmp; 1084 1085 /* Racy check. */ 1086 sigio = SLIST_FIRST(sigiolst); 1087 if (sigio == NULL) 1088 return; 1089 1090 p = NULL; 1091 pg = NULL; 1092 1093 SIGIO_LOCK(); 1094 sigio = SLIST_FIRST(sigiolst); 1095 if (sigio == NULL) { 1096 SIGIO_UNLOCK(); 1097 return; 1098 } 1099 1100 /* 1101 * Every entry of the list should belong to a single proc or pgrp. 1102 */ 1103 if (sigio->sio_pgid < 0) { 1104 pg = sigio->sio_pgrp; 1105 sx_assert(&proctree_lock, SX_XLOCKED); 1106 PGRP_LOCK(pg); 1107 } else /* if (sigio->sio_pgid > 0) */ { 1108 p = sigio->sio_proc; 1109 PROC_LOCK(p); 1110 KASSERT((p->p_flag & P_WEXIT) != 0, 1111 ("%s: process %p is not exiting", __func__, p)); 1112 } 1113 1114 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) { 1115 *sigio->sio_myref = NULL; 1116 if (pg != NULL) { 1117 KASSERT(sigio->sio_pgid < 0, 1118 ("Proc sigio in pgrp sigio list")); 1119 KASSERT(sigio->sio_pgrp == pg, 1120 ("Bogus pgrp in sigio list")); 1121 } else /* if (p != NULL) */ { 1122 KASSERT(sigio->sio_pgid > 0, 1123 ("Pgrp sigio in proc sigio list")); 1124 KASSERT(sigio->sio_proc == p, 1125 ("Bogus proc in sigio list")); 1126 } 1127 } 1128 1129 if (pg != NULL) 1130 PGRP_UNLOCK(pg); 1131 else 1132 PROC_UNLOCK(p); 1133 SIGIO_UNLOCK(); 1134 1135 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp) 1136 sigiofree(sigio); 1137 } 1138 1139 /* 1140 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1141 * 1142 * After permission checking, add a sigio structure to the sigio list for 1143 * the process or process group. 1144 */ 1145 int 1146 fsetown(pid_t pgid, struct sigio **sigiop) 1147 { 1148 struct proc *proc; 1149 struct pgrp *pgrp; 1150 struct sigio *osigio, *sigio; 1151 int ret; 1152 1153 if (pgid == 0) { 1154 funsetown(sigiop); 1155 return (0); 1156 } 1157 1158 ret = 0; 1159 1160 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1161 sigio->sio_pgid = pgid; 1162 sigio->sio_ucred = crhold(curthread->td_ucred); 1163 sigio->sio_myref = sigiop; 1164 1165 sx_slock(&proctree_lock); 1166 SIGIO_LOCK(); 1167 osigio = funsetown_locked(*sigiop); 1168 if (pgid > 0) { 1169 proc = pfind(pgid); 1170 if (proc == NULL) { 1171 ret = ESRCH; 1172 goto fail; 1173 } 1174 1175 /* 1176 * Policy - Don't allow a process to FSETOWN a process 1177 * in another session. 1178 * 1179 * Remove this test to allow maximum flexibility or 1180 * restrict FSETOWN to the current process or process 1181 * group for maximum safety. 1182 */ 1183 if (proc->p_session != curthread->td_proc->p_session) { 1184 PROC_UNLOCK(proc); 1185 ret = EPERM; 1186 goto fail; 1187 } 1188 1189 sigio->sio_proc = proc; 1190 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1191 PROC_UNLOCK(proc); 1192 } else /* if (pgid < 0) */ { 1193 pgrp = pgfind(-pgid); 1194 if (pgrp == NULL) { 1195 ret = ESRCH; 1196 goto fail; 1197 } 1198 1199 /* 1200 * Policy - Don't allow a process to FSETOWN a process 1201 * in another session. 1202 * 1203 * Remove this test to allow maximum flexibility or 1204 * restrict FSETOWN to the current process or process 1205 * group for maximum safety. 1206 */ 1207 if (pgrp->pg_session != curthread->td_proc->p_session) { 1208 PGRP_UNLOCK(pgrp); 1209 ret = EPERM; 1210 goto fail; 1211 } 1212 1213 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1214 sigio->sio_pgrp = pgrp; 1215 PGRP_UNLOCK(pgrp); 1216 } 1217 sx_sunlock(&proctree_lock); 1218 *sigiop = sigio; 1219 SIGIO_UNLOCK(); 1220 if (osigio != NULL) 1221 sigiofree(osigio); 1222 return (0); 1223 1224 fail: 1225 SIGIO_UNLOCK(); 1226 sx_sunlock(&proctree_lock); 1227 sigiofree(sigio); 1228 if (osigio != NULL) 1229 sigiofree(osigio); 1230 return (ret); 1231 } 1232 1233 /* 1234 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1235 */ 1236 pid_t 1237 fgetown(struct sigio **sigiop) 1238 { 1239 pid_t pgid; 1240 1241 SIGIO_LOCK(); 1242 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1243 SIGIO_UNLOCK(); 1244 return (pgid); 1245 } 1246 1247 static int 1248 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1249 bool audit) 1250 { 1251 int error; 1252 1253 FILEDESC_XLOCK_ASSERT(fdp); 1254 1255 /* 1256 * We now hold the fp reference that used to be owned by the 1257 * descriptor array. We have to unlock the FILEDESC *AFTER* 1258 * knote_fdclose to prevent a race of the fd getting opened, a knote 1259 * added, and deleteing a knote for the new fd. 1260 */ 1261 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1262 knote_fdclose(td, fd); 1263 1264 /* 1265 * We need to notify mqueue if the object is of type mqueue. 1266 */ 1267 if (__predict_false(fp->f_type == DTYPE_MQUEUE)) 1268 mq_fdclose(td, fd, fp); 1269 FILEDESC_XUNLOCK(fdp); 1270 1271 #ifdef AUDIT 1272 if (AUDITING_TD(td) && audit) 1273 audit_sysclose(td, fd, fp); 1274 #endif 1275 error = closef(fp, td); 1276 1277 /* 1278 * All paths leading up to closefp() will have already removed or 1279 * replaced the fd in the filedesc table, so a restart would not 1280 * operate on the same file. 1281 */ 1282 if (error == ERESTART) 1283 error = EINTR; 1284 1285 return (error); 1286 } 1287 1288 static int 1289 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1290 bool holdleaders, bool audit) 1291 { 1292 int error; 1293 1294 FILEDESC_XLOCK_ASSERT(fdp); 1295 1296 if (holdleaders) { 1297 if (td->td_proc->p_fdtol != NULL) { 1298 /* 1299 * Ask fdfree() to sleep to ensure that all relevant 1300 * process leaders can be traversed in closef(). 1301 */ 1302 fdp->fd_holdleaderscount++; 1303 } else { 1304 holdleaders = false; 1305 } 1306 } 1307 1308 error = closefp_impl(fdp, fd, fp, td, audit); 1309 if (holdleaders) { 1310 FILEDESC_XLOCK(fdp); 1311 fdp->fd_holdleaderscount--; 1312 if (fdp->fd_holdleaderscount == 0 && 1313 fdp->fd_holdleaderswakeup != 0) { 1314 fdp->fd_holdleaderswakeup = 0; 1315 wakeup(&fdp->fd_holdleaderscount); 1316 } 1317 FILEDESC_XUNLOCK(fdp); 1318 } 1319 return (error); 1320 } 1321 1322 static int 1323 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1324 bool holdleaders, bool audit) 1325 { 1326 1327 FILEDESC_XLOCK_ASSERT(fdp); 1328 1329 if (__predict_false(td->td_proc->p_fdtol != NULL)) { 1330 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit)); 1331 } else { 1332 return (closefp_impl(fdp, fd, fp, td, audit)); 1333 } 1334 } 1335 1336 /* 1337 * Close a file descriptor. 1338 */ 1339 #ifndef _SYS_SYSPROTO_H_ 1340 struct close_args { 1341 int fd; 1342 }; 1343 #endif 1344 /* ARGSUSED */ 1345 int 1346 sys_close(struct thread *td, struct close_args *uap) 1347 { 1348 1349 return (kern_close(td, uap->fd)); 1350 } 1351 1352 int 1353 kern_close(struct thread *td, int fd) 1354 { 1355 struct filedesc *fdp; 1356 struct file *fp; 1357 1358 fdp = td->td_proc->p_fd; 1359 1360 FILEDESC_XLOCK(fdp); 1361 if ((fp = fget_locked(fdp, fd)) == NULL) { 1362 FILEDESC_XUNLOCK(fdp); 1363 return (EBADF); 1364 } 1365 fdfree(fdp, fd); 1366 1367 /* closefp() drops the FILEDESC lock for us. */ 1368 return (closefp(fdp, fd, fp, td, true, true)); 1369 } 1370 1371 int 1372 kern_close_range(struct thread *td, u_int lowfd, u_int highfd) 1373 { 1374 struct filedesc *fdp; 1375 const struct fdescenttbl *fdt; 1376 struct file *fp; 1377 int fd; 1378 1379 /* 1380 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 1381 * open should not be a usage error. From a close_range() perspective, 1382 * close_range(3, ~0U, 0) in the same scenario should also likely not 1383 * be a usage error as all fd above 3 are in-fact already closed. 1384 */ 1385 if (highfd < lowfd) { 1386 return (EINVAL); 1387 } 1388 1389 fdp = td->td_proc->p_fd; 1390 FILEDESC_XLOCK(fdp); 1391 fdt = atomic_load_ptr(&fdp->fd_files); 1392 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1393 fd = lowfd; 1394 if (__predict_false(fd > highfd)) { 1395 goto out_locked; 1396 } 1397 for (;;) { 1398 fp = fdt->fdt_ofiles[fd].fde_file; 1399 if (fp == NULL) { 1400 if (fd == highfd) 1401 goto out_locked; 1402 } else { 1403 fdfree(fdp, fd); 1404 (void) closefp(fdp, fd, fp, td, true, true); 1405 if (fd == highfd) 1406 goto out_unlocked; 1407 FILEDESC_XLOCK(fdp); 1408 fdt = atomic_load_ptr(&fdp->fd_files); 1409 } 1410 fd++; 1411 } 1412 out_locked: 1413 FILEDESC_XUNLOCK(fdp); 1414 out_unlocked: 1415 return (0); 1416 } 1417 1418 #ifndef _SYS_SYSPROTO_H_ 1419 struct close_range_args { 1420 u_int lowfd; 1421 u_int highfd; 1422 int flags; 1423 }; 1424 #endif 1425 int 1426 sys_close_range(struct thread *td, struct close_range_args *uap) 1427 { 1428 1429 AUDIT_ARG_FD(uap->lowfd); 1430 AUDIT_ARG_CMD(uap->highfd); 1431 AUDIT_ARG_FFLAGS(uap->flags); 1432 1433 /* No flags currently defined */ 1434 if (uap->flags != 0) 1435 return (EINVAL); 1436 return (kern_close_range(td, uap->lowfd, uap->highfd)); 1437 } 1438 1439 #ifdef COMPAT_FREEBSD12 1440 /* 1441 * Close open file descriptors. 1442 */ 1443 #ifndef _SYS_SYSPROTO_H_ 1444 struct freebsd12_closefrom_args { 1445 int lowfd; 1446 }; 1447 #endif 1448 /* ARGSUSED */ 1449 int 1450 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) 1451 { 1452 u_int lowfd; 1453 1454 AUDIT_ARG_FD(uap->lowfd); 1455 1456 /* 1457 * Treat negative starting file descriptor values identical to 1458 * closefrom(0) which closes all files. 1459 */ 1460 lowfd = MAX(0, uap->lowfd); 1461 return (kern_close_range(td, lowfd, ~0U)); 1462 } 1463 #endif /* COMPAT_FREEBSD12 */ 1464 1465 #if defined(COMPAT_43) 1466 /* 1467 * Return status information about a file descriptor. 1468 */ 1469 #ifndef _SYS_SYSPROTO_H_ 1470 struct ofstat_args { 1471 int fd; 1472 struct ostat *sb; 1473 }; 1474 #endif 1475 /* ARGSUSED */ 1476 int 1477 ofstat(struct thread *td, struct ofstat_args *uap) 1478 { 1479 struct ostat oub; 1480 struct stat ub; 1481 int error; 1482 1483 error = kern_fstat(td, uap->fd, &ub); 1484 if (error == 0) { 1485 cvtstat(&ub, &oub); 1486 error = copyout(&oub, uap->sb, sizeof(oub)); 1487 } 1488 return (error); 1489 } 1490 #endif /* COMPAT_43 */ 1491 1492 #if defined(COMPAT_FREEBSD11) 1493 int 1494 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1495 { 1496 struct stat sb; 1497 struct freebsd11_stat osb; 1498 int error; 1499 1500 error = kern_fstat(td, uap->fd, &sb); 1501 if (error != 0) 1502 return (error); 1503 error = freebsd11_cvtstat(&sb, &osb); 1504 if (error == 0) 1505 error = copyout(&osb, uap->sb, sizeof(osb)); 1506 return (error); 1507 } 1508 #endif /* COMPAT_FREEBSD11 */ 1509 1510 /* 1511 * Return status information about a file descriptor. 1512 */ 1513 #ifndef _SYS_SYSPROTO_H_ 1514 struct fstat_args { 1515 int fd; 1516 struct stat *sb; 1517 }; 1518 #endif 1519 /* ARGSUSED */ 1520 int 1521 sys_fstat(struct thread *td, struct fstat_args *uap) 1522 { 1523 struct stat ub; 1524 int error; 1525 1526 error = kern_fstat(td, uap->fd, &ub); 1527 if (error == 0) 1528 error = copyout(&ub, uap->sb, sizeof(ub)); 1529 return (error); 1530 } 1531 1532 int 1533 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1534 { 1535 struct file *fp; 1536 int error; 1537 1538 AUDIT_ARG_FD(fd); 1539 1540 error = fget(td, fd, &cap_fstat_rights, &fp); 1541 if (__predict_false(error != 0)) 1542 return (error); 1543 1544 AUDIT_ARG_FILE(td->td_proc, fp); 1545 1546 error = fo_stat(fp, sbp, td->td_ucred, td); 1547 fdrop(fp, td); 1548 #ifdef __STAT_TIME_T_EXT 1549 sbp->st_atim_ext = 0; 1550 sbp->st_mtim_ext = 0; 1551 sbp->st_ctim_ext = 0; 1552 sbp->st_btim_ext = 0; 1553 #endif 1554 #ifdef KTRACE 1555 if (KTRPOINT(td, KTR_STRUCT)) 1556 ktrstat_error(sbp, error); 1557 #endif 1558 return (error); 1559 } 1560 1561 #if defined(COMPAT_FREEBSD11) 1562 /* 1563 * Return status information about a file descriptor. 1564 */ 1565 #ifndef _SYS_SYSPROTO_H_ 1566 struct freebsd11_nfstat_args { 1567 int fd; 1568 struct nstat *sb; 1569 }; 1570 #endif 1571 /* ARGSUSED */ 1572 int 1573 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1574 { 1575 struct nstat nub; 1576 struct stat ub; 1577 int error; 1578 1579 error = kern_fstat(td, uap->fd, &ub); 1580 if (error == 0) { 1581 freebsd11_cvtnstat(&ub, &nub); 1582 error = copyout(&nub, uap->sb, sizeof(nub)); 1583 } 1584 return (error); 1585 } 1586 #endif /* COMPAT_FREEBSD11 */ 1587 1588 /* 1589 * Return pathconf information about a file descriptor. 1590 */ 1591 #ifndef _SYS_SYSPROTO_H_ 1592 struct fpathconf_args { 1593 int fd; 1594 int name; 1595 }; 1596 #endif 1597 /* ARGSUSED */ 1598 int 1599 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1600 { 1601 long value; 1602 int error; 1603 1604 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1605 if (error == 0) 1606 td->td_retval[0] = value; 1607 return (error); 1608 } 1609 1610 int 1611 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1612 { 1613 struct file *fp; 1614 struct vnode *vp; 1615 int error; 1616 1617 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1618 if (error != 0) 1619 return (error); 1620 1621 if (name == _PC_ASYNC_IO) { 1622 *valuep = _POSIX_ASYNCHRONOUS_IO; 1623 goto out; 1624 } 1625 vp = fp->f_vnode; 1626 if (vp != NULL) { 1627 vn_lock(vp, LK_SHARED | LK_RETRY); 1628 error = VOP_PATHCONF(vp, name, valuep); 1629 VOP_UNLOCK(vp); 1630 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1631 if (name != _PC_PIPE_BUF) { 1632 error = EINVAL; 1633 } else { 1634 *valuep = PIPE_BUF; 1635 error = 0; 1636 } 1637 } else { 1638 error = EOPNOTSUPP; 1639 } 1640 out: 1641 fdrop(fp, td); 1642 return (error); 1643 } 1644 1645 /* 1646 * Copy filecaps structure allocating memory for ioctls array if needed. 1647 * 1648 * The last parameter indicates whether the fdtable is locked. If it is not and 1649 * ioctls are encountered, copying fails and the caller must lock the table. 1650 * 1651 * Note that if the table was not locked, the caller has to check the relevant 1652 * sequence counter to determine whether the operation was successful. 1653 */ 1654 bool 1655 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1656 { 1657 size_t size; 1658 1659 if (src->fc_ioctls != NULL && !locked) 1660 return (false); 1661 memcpy(dst, src, sizeof(*src)); 1662 if (src->fc_ioctls == NULL) 1663 return (true); 1664 1665 KASSERT(src->fc_nioctls > 0, 1666 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1667 1668 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1669 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1670 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1671 return (true); 1672 } 1673 1674 static u_long * 1675 filecaps_copy_prep(const struct filecaps *src) 1676 { 1677 u_long *ioctls; 1678 size_t size; 1679 1680 if (__predict_true(src->fc_ioctls == NULL)) 1681 return (NULL); 1682 1683 KASSERT(src->fc_nioctls > 0, 1684 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1685 1686 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1687 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1688 return (ioctls); 1689 } 1690 1691 static void 1692 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1693 u_long *ioctls) 1694 { 1695 size_t size; 1696 1697 *dst = *src; 1698 if (__predict_true(src->fc_ioctls == NULL)) { 1699 MPASS(ioctls == NULL); 1700 return; 1701 } 1702 1703 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1704 dst->fc_ioctls = ioctls; 1705 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1706 } 1707 1708 /* 1709 * Move filecaps structure to the new place and clear the old place. 1710 */ 1711 void 1712 filecaps_move(struct filecaps *src, struct filecaps *dst) 1713 { 1714 1715 *dst = *src; 1716 bzero(src, sizeof(*src)); 1717 } 1718 1719 /* 1720 * Fill the given filecaps structure with full rights. 1721 */ 1722 static void 1723 filecaps_fill(struct filecaps *fcaps) 1724 { 1725 1726 CAP_ALL(&fcaps->fc_rights); 1727 fcaps->fc_ioctls = NULL; 1728 fcaps->fc_nioctls = -1; 1729 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1730 } 1731 1732 /* 1733 * Free memory allocated within filecaps structure. 1734 */ 1735 void 1736 filecaps_free(struct filecaps *fcaps) 1737 { 1738 1739 free(fcaps->fc_ioctls, M_FILECAPS); 1740 bzero(fcaps, sizeof(*fcaps)); 1741 } 1742 1743 static u_long * 1744 filecaps_free_prep(struct filecaps *fcaps) 1745 { 1746 u_long *ioctls; 1747 1748 ioctls = fcaps->fc_ioctls; 1749 bzero(fcaps, sizeof(*fcaps)); 1750 return (ioctls); 1751 } 1752 1753 static void 1754 filecaps_free_finish(u_long *ioctls) 1755 { 1756 1757 free(ioctls, M_FILECAPS); 1758 } 1759 1760 /* 1761 * Validate the given filecaps structure. 1762 */ 1763 static void 1764 filecaps_validate(const struct filecaps *fcaps, const char *func) 1765 { 1766 1767 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1768 ("%s: invalid rights", func)); 1769 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1770 ("%s: invalid fcntls", func)); 1771 KASSERT(fcaps->fc_fcntls == 0 || 1772 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1773 ("%s: fcntls without CAP_FCNTL", func)); 1774 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1775 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1776 ("%s: invalid ioctls", func)); 1777 KASSERT(fcaps->fc_nioctls == 0 || 1778 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1779 ("%s: ioctls without CAP_IOCTL", func)); 1780 } 1781 1782 static void 1783 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1784 { 1785 int nfd1; 1786 1787 FILEDESC_XLOCK_ASSERT(fdp); 1788 1789 nfd1 = fdp->fd_nfiles * 2; 1790 if (nfd1 < nfd) 1791 nfd1 = nfd; 1792 fdgrowtable(fdp, nfd1); 1793 } 1794 1795 /* 1796 * Grow the file table to accommodate (at least) nfd descriptors. 1797 */ 1798 static void 1799 fdgrowtable(struct filedesc *fdp, int nfd) 1800 { 1801 struct filedesc0 *fdp0; 1802 struct freetable *ft; 1803 struct fdescenttbl *ntable; 1804 struct fdescenttbl *otable; 1805 int nnfiles, onfiles; 1806 NDSLOTTYPE *nmap, *omap; 1807 1808 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1809 1810 /* save old values */ 1811 onfiles = fdp->fd_nfiles; 1812 otable = fdp->fd_files; 1813 omap = fdp->fd_map; 1814 1815 /* compute the size of the new table */ 1816 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1817 if (nnfiles <= onfiles) 1818 /* the table is already large enough */ 1819 return; 1820 1821 /* 1822 * Allocate a new table. We need enough space for the number of 1823 * entries, file entries themselves and the struct freetable we will use 1824 * when we decommission the table and place it on the freelist. 1825 * We place the struct freetable in the middle so we don't have 1826 * to worry about padding. 1827 */ 1828 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1829 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1830 sizeof(struct freetable), 1831 M_FILEDESC, M_ZERO | M_WAITOK); 1832 /* copy the old data */ 1833 ntable->fdt_nfiles = nnfiles; 1834 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1835 onfiles * sizeof(ntable->fdt_ofiles[0])); 1836 1837 /* 1838 * Allocate a new map only if the old is not large enough. It will 1839 * grow at a slower rate than the table as it can map more 1840 * entries than the table can hold. 1841 */ 1842 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1843 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1844 M_ZERO | M_WAITOK); 1845 /* copy over the old data and update the pointer */ 1846 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1847 fdp->fd_map = nmap; 1848 } 1849 1850 /* 1851 * Make sure that ntable is correctly initialized before we replace 1852 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1853 * data. 1854 */ 1855 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1856 1857 /* 1858 * Free the old file table when not shared by other threads or processes. 1859 * The old file table is considered to be shared when either are true: 1860 * - The process has more than one thread. 1861 * - The file descriptor table has been shared via fdshare(). 1862 * 1863 * When shared, the old file table will be placed on a freelist 1864 * which will be processed when the struct filedesc is released. 1865 * 1866 * Note that if onfiles == NDFILE, we're dealing with the original 1867 * static allocation contained within (struct filedesc0 *)fdp, 1868 * which must not be freed. 1869 */ 1870 if (onfiles > NDFILE) { 1871 /* 1872 * Note we may be called here from fdinit while allocating a 1873 * table for a new process in which case ->p_fd points 1874 * elsewhere. 1875 */ 1876 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 1877 free(otable, M_FILEDESC); 1878 } else { 1879 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1880 fdp0 = (struct filedesc0 *)fdp; 1881 ft->ft_table = otable; 1882 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1883 } 1884 } 1885 /* 1886 * The map does not have the same possibility of threads still 1887 * holding references to it. So always free it as long as it 1888 * does not reference the original static allocation. 1889 */ 1890 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1891 free(omap, M_FILEDESC); 1892 } 1893 1894 /* 1895 * Allocate a file descriptor for the process. 1896 */ 1897 int 1898 fdalloc(struct thread *td, int minfd, int *result) 1899 { 1900 struct proc *p = td->td_proc; 1901 struct filedesc *fdp = p->p_fd; 1902 int fd, maxfd, allocfd; 1903 #ifdef RACCT 1904 int error; 1905 #endif 1906 1907 FILEDESC_XLOCK_ASSERT(fdp); 1908 1909 if (fdp->fd_freefile > minfd) 1910 minfd = fdp->fd_freefile; 1911 1912 maxfd = getmaxfd(td); 1913 1914 /* 1915 * Search the bitmap for a free descriptor starting at minfd. 1916 * If none is found, grow the file table. 1917 */ 1918 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1919 if (__predict_false(fd >= maxfd)) 1920 return (EMFILE); 1921 if (__predict_false(fd >= fdp->fd_nfiles)) { 1922 allocfd = min(fd * 2, maxfd); 1923 #ifdef RACCT 1924 if (RACCT_ENABLED()) { 1925 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 1926 if (error != 0) 1927 return (EMFILE); 1928 } 1929 #endif 1930 /* 1931 * fd is already equal to first free descriptor >= minfd, so 1932 * we only need to grow the table and we are done. 1933 */ 1934 fdgrowtable_exp(fdp, allocfd); 1935 } 1936 1937 /* 1938 * Perform some sanity checks, then mark the file descriptor as 1939 * used and return it to the caller. 1940 */ 1941 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1942 ("invalid descriptor %d", fd)); 1943 KASSERT(!fdisused(fdp, fd), 1944 ("fd_first_free() returned non-free descriptor")); 1945 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1946 ("file descriptor isn't free")); 1947 fdused(fdp, fd); 1948 *result = fd; 1949 return (0); 1950 } 1951 1952 /* 1953 * Allocate n file descriptors for the process. 1954 */ 1955 int 1956 fdallocn(struct thread *td, int minfd, int *fds, int n) 1957 { 1958 struct proc *p = td->td_proc; 1959 struct filedesc *fdp = p->p_fd; 1960 int i; 1961 1962 FILEDESC_XLOCK_ASSERT(fdp); 1963 1964 for (i = 0; i < n; i++) 1965 if (fdalloc(td, 0, &fds[i]) != 0) 1966 break; 1967 1968 if (i < n) { 1969 for (i--; i >= 0; i--) 1970 fdunused(fdp, fds[i]); 1971 return (EMFILE); 1972 } 1973 1974 return (0); 1975 } 1976 1977 /* 1978 * Create a new open file structure and allocate a file descriptor for the 1979 * process that refers to it. We add one reference to the file for the 1980 * descriptor table and one reference for resultfp. This is to prevent us 1981 * being preempted and the entry in the descriptor table closed after we 1982 * release the FILEDESC lock. 1983 */ 1984 int 1985 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1986 struct filecaps *fcaps) 1987 { 1988 struct file *fp; 1989 int error, fd; 1990 1991 MPASS(resultfp != NULL); 1992 MPASS(resultfd != NULL); 1993 1994 error = _falloc_noinstall(td, &fp, 2); 1995 if (__predict_false(error != 0)) { 1996 return (error); 1997 } 1998 1999 error = finstall_refed(td, fp, &fd, flags, fcaps); 2000 if (__predict_false(error != 0)) { 2001 falloc_abort(td, fp); 2002 return (error); 2003 } 2004 2005 *resultfp = fp; 2006 *resultfd = fd; 2007 2008 return (0); 2009 } 2010 2011 /* 2012 * Create a new open file structure without allocating a file descriptor. 2013 */ 2014 int 2015 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2016 { 2017 struct file *fp; 2018 int maxuserfiles = maxfiles - (maxfiles / 20); 2019 int openfiles_new; 2020 static struct timeval lastfail; 2021 static int curfail; 2022 2023 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2024 MPASS(n > 0); 2025 2026 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2027 if ((openfiles_new >= maxuserfiles && 2028 priv_check(td, PRIV_MAXFILES) != 0) || 2029 openfiles_new >= maxfiles) { 2030 atomic_subtract_int(&openfiles, 1); 2031 if (ppsratecheck(&lastfail, &curfail, 1)) { 2032 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2033 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2034 } 2035 return (ENFILE); 2036 } 2037 fp = uma_zalloc(file_zone, M_WAITOK); 2038 bzero(fp, sizeof(*fp)); 2039 refcount_init(&fp->f_count, n); 2040 fp->f_cred = crhold(td->td_ucred); 2041 fp->f_ops = &badfileops; 2042 *resultfp = fp; 2043 return (0); 2044 } 2045 2046 void 2047 falloc_abort(struct thread *td, struct file *fp) 2048 { 2049 2050 /* 2051 * For assertion purposes. 2052 */ 2053 refcount_init(&fp->f_count, 0); 2054 _fdrop(fp, td); 2055 } 2056 2057 /* 2058 * Install a file in a file descriptor table. 2059 */ 2060 void 2061 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2062 struct filecaps *fcaps) 2063 { 2064 struct filedescent *fde; 2065 2066 MPASS(fp != NULL); 2067 if (fcaps != NULL) 2068 filecaps_validate(fcaps, __func__); 2069 FILEDESC_XLOCK_ASSERT(fdp); 2070 2071 fde = &fdp->fd_ofiles[fd]; 2072 #ifdef CAPABILITIES 2073 seqc_write_begin(&fde->fde_seqc); 2074 #endif 2075 fde->fde_file = fp; 2076 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 2077 if (fcaps != NULL) 2078 filecaps_move(fcaps, &fde->fde_caps); 2079 else 2080 filecaps_fill(&fde->fde_caps); 2081 #ifdef CAPABILITIES 2082 seqc_write_end(&fde->fde_seqc); 2083 #endif 2084 } 2085 2086 int 2087 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2088 struct filecaps *fcaps) 2089 { 2090 struct filedesc *fdp = td->td_proc->p_fd; 2091 int error; 2092 2093 MPASS(fd != NULL); 2094 2095 FILEDESC_XLOCK(fdp); 2096 error = fdalloc(td, 0, fd); 2097 if (__predict_true(error == 0)) { 2098 _finstall(fdp, fp, *fd, flags, fcaps); 2099 } 2100 FILEDESC_XUNLOCK(fdp); 2101 return (error); 2102 } 2103 2104 int 2105 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2106 struct filecaps *fcaps) 2107 { 2108 int error; 2109 2110 MPASS(fd != NULL); 2111 2112 if (!fhold(fp)) 2113 return (EBADF); 2114 error = finstall_refed(td, fp, fd, flags, fcaps); 2115 if (__predict_false(error != 0)) { 2116 fdrop(fp, td); 2117 } 2118 return (error); 2119 } 2120 2121 /* 2122 * Build a new filedesc structure from another. 2123 * 2124 * If fdp is not NULL, return with it shared locked. 2125 */ 2126 struct filedesc * 2127 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile) 2128 { 2129 struct filedesc0 *newfdp0; 2130 struct filedesc *newfdp; 2131 2132 if (prepfiles) 2133 MPASS(lastfile != NULL); 2134 else 2135 MPASS(lastfile == NULL); 2136 2137 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2138 newfdp = &newfdp0->fd_fd; 2139 2140 /* Create the file descriptor table. */ 2141 FILEDESC_LOCK_INIT(newfdp); 2142 refcount_init(&newfdp->fd_refcnt, 1); 2143 refcount_init(&newfdp->fd_holdcnt, 1); 2144 newfdp->fd_map = newfdp0->fd_dmap; 2145 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2146 newfdp->fd_files->fdt_nfiles = NDFILE; 2147 2148 if (fdp == NULL) 2149 return (newfdp); 2150 2151 FILEDESC_SLOCK(fdp); 2152 if (!prepfiles) { 2153 FILEDESC_SUNLOCK(fdp); 2154 return (newfdp); 2155 } 2156 2157 for (;;) { 2158 *lastfile = fdlastfile(fdp); 2159 if (*lastfile < newfdp->fd_nfiles) 2160 break; 2161 FILEDESC_SUNLOCK(fdp); 2162 fdgrowtable(newfdp, *lastfile + 1); 2163 FILEDESC_SLOCK(fdp); 2164 } 2165 2166 return (newfdp); 2167 } 2168 2169 /* 2170 * Build a pwddesc structure from another. 2171 * Copy the current, root, and jail root vnode references. 2172 * 2173 * If pdp is not NULL, return with it shared locked. 2174 */ 2175 struct pwddesc * 2176 pdinit(struct pwddesc *pdp, bool keeplock) 2177 { 2178 struct pwddesc *newpdp; 2179 struct pwd *newpwd; 2180 2181 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2182 2183 PWDDESC_LOCK_INIT(newpdp); 2184 refcount_init(&newpdp->pd_refcount, 1); 2185 newpdp->pd_cmask = CMASK; 2186 2187 if (pdp == NULL) { 2188 newpwd = pwd_alloc(); 2189 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2190 return (newpdp); 2191 } 2192 2193 PWDDESC_XLOCK(pdp); 2194 newpwd = pwd_hold_pwddesc(pdp); 2195 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2196 if (!keeplock) 2197 PWDDESC_XUNLOCK(pdp); 2198 return (newpdp); 2199 } 2200 2201 /* 2202 * Hold either filedesc or pwddesc of the passed process. 2203 * 2204 * The process lock is used to synchronize against the target exiting and 2205 * freeing the data. 2206 * 2207 * Clearing can be ilustrated in 3 steps: 2208 * 1. set the pointer to NULL. Either routine can race against it, hence 2209 * atomic_load_ptr. 2210 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2211 * race to either still see the pointer or find NULL. It is still safe to 2212 * grab a reference as clearing is stalled. 2213 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2214 * guaranteed to see NULL, making it safe to finish clearing 2215 */ 2216 static struct filedesc * 2217 fdhold(struct proc *p) 2218 { 2219 struct filedesc *fdp; 2220 2221 PROC_LOCK_ASSERT(p, MA_OWNED); 2222 fdp = atomic_load_ptr(&p->p_fd); 2223 if (fdp != NULL) 2224 refcount_acquire(&fdp->fd_holdcnt); 2225 return (fdp); 2226 } 2227 2228 static struct pwddesc * 2229 pdhold(struct proc *p) 2230 { 2231 struct pwddesc *pdp; 2232 2233 PROC_LOCK_ASSERT(p, MA_OWNED); 2234 pdp = atomic_load_ptr(&p->p_pd); 2235 if (pdp != NULL) 2236 refcount_acquire(&pdp->pd_refcount); 2237 return (pdp); 2238 } 2239 2240 static void 2241 fddrop(struct filedesc *fdp) 2242 { 2243 2244 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2245 if (refcount_release(&fdp->fd_holdcnt) == 0) 2246 return; 2247 } 2248 2249 FILEDESC_LOCK_DESTROY(fdp); 2250 uma_zfree(filedesc0_zone, fdp); 2251 } 2252 2253 static void 2254 pddrop(struct pwddesc *pdp) 2255 { 2256 struct pwd *pwd; 2257 2258 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2259 return; 2260 2261 PWDDESC_XLOCK(pdp); 2262 if (refcount_release(&pdp->pd_refcount) == 0) { 2263 PWDDESC_XUNLOCK(pdp); 2264 return; 2265 } 2266 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2267 pwd_set(pdp, NULL); 2268 PWDDESC_XUNLOCK(pdp); 2269 pwd_drop(pwd); 2270 2271 PWDDESC_LOCK_DESTROY(pdp); 2272 free(pdp, M_PWDDESC); 2273 } 2274 2275 /* 2276 * Share a filedesc structure. 2277 */ 2278 struct filedesc * 2279 fdshare(struct filedesc *fdp) 2280 { 2281 2282 refcount_acquire(&fdp->fd_refcnt); 2283 return (fdp); 2284 } 2285 2286 /* 2287 * Share a pwddesc structure. 2288 */ 2289 struct pwddesc * 2290 pdshare(struct pwddesc *pdp) 2291 { 2292 refcount_acquire(&pdp->pd_refcount); 2293 return (pdp); 2294 } 2295 2296 /* 2297 * Unshare a filedesc structure, if necessary by making a copy 2298 */ 2299 void 2300 fdunshare(struct thread *td) 2301 { 2302 struct filedesc *tmp; 2303 struct proc *p = td->td_proc; 2304 2305 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2306 return; 2307 2308 tmp = fdcopy(p->p_fd); 2309 fdescfree(td); 2310 p->p_fd = tmp; 2311 } 2312 2313 /* 2314 * Unshare a pwddesc structure. 2315 */ 2316 void 2317 pdunshare(struct thread *td) 2318 { 2319 struct pwddesc *pdp; 2320 struct proc *p; 2321 2322 p = td->td_proc; 2323 /* Not shared. */ 2324 if (p->p_pd->pd_refcount == 1) 2325 return; 2326 2327 pdp = pdcopy(p->p_pd); 2328 pdescfree(td); 2329 p->p_pd = pdp; 2330 } 2331 2332 void 2333 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 2334 { 2335 2336 fdescfree(td); 2337 td->td_proc->p_fd = fdp; 2338 } 2339 2340 /* 2341 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2342 * this is to ease callers, not catch errors. 2343 */ 2344 struct filedesc * 2345 fdcopy(struct filedesc *fdp) 2346 { 2347 struct filedesc *newfdp; 2348 struct filedescent *nfde, *ofde; 2349 int i, lastfile; 2350 2351 MPASS(fdp != NULL); 2352 2353 newfdp = fdinit(fdp, true, &lastfile); 2354 /* copy all passable descriptors (i.e. not kqueue) */ 2355 newfdp->fd_freefile = -1; 2356 for (i = 0; i <= lastfile; ++i) { 2357 ofde = &fdp->fd_ofiles[i]; 2358 if (ofde->fde_file == NULL || 2359 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 || 2360 !fhold(ofde->fde_file)) { 2361 if (newfdp->fd_freefile == -1) 2362 newfdp->fd_freefile = i; 2363 continue; 2364 } 2365 nfde = &newfdp->fd_ofiles[i]; 2366 *nfde = *ofde; 2367 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2368 fdused_init(newfdp, i); 2369 } 2370 if (newfdp->fd_freefile == -1) 2371 newfdp->fd_freefile = i; 2372 FILEDESC_SUNLOCK(fdp); 2373 return (newfdp); 2374 } 2375 2376 /* 2377 * Copy a pwddesc structure. 2378 */ 2379 struct pwddesc * 2380 pdcopy(struct pwddesc *pdp) 2381 { 2382 struct pwddesc *newpdp; 2383 2384 MPASS(pdp != NULL); 2385 2386 newpdp = pdinit(pdp, true); 2387 newpdp->pd_cmask = pdp->pd_cmask; 2388 PWDDESC_XUNLOCK(pdp); 2389 return (newpdp); 2390 } 2391 2392 /* 2393 * Copies a filedesc structure, while remapping all file descriptors 2394 * stored inside using a translation table. 2395 * 2396 * File descriptors are copied over to the new file descriptor table, 2397 * regardless of whether the close-on-exec flag is set. 2398 */ 2399 int 2400 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 2401 struct filedesc **ret) 2402 { 2403 struct filedesc *newfdp; 2404 struct filedescent *nfde, *ofde; 2405 int error, i, lastfile; 2406 2407 MPASS(fdp != NULL); 2408 2409 newfdp = fdinit(fdp, true, &lastfile); 2410 if (nfds > lastfile + 1) { 2411 /* New table cannot be larger than the old one. */ 2412 error = E2BIG; 2413 goto bad; 2414 } 2415 /* Copy all passable descriptors (i.e. not kqueue). */ 2416 newfdp->fd_freefile = nfds; 2417 for (i = 0; i < nfds; ++i) { 2418 if (fds[i] < 0 || fds[i] > lastfile) { 2419 /* File descriptor out of bounds. */ 2420 error = EBADF; 2421 goto bad; 2422 } 2423 ofde = &fdp->fd_ofiles[fds[i]]; 2424 if (ofde->fde_file == NULL) { 2425 /* Unused file descriptor. */ 2426 error = EBADF; 2427 goto bad; 2428 } 2429 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2430 /* File descriptor cannot be passed. */ 2431 error = EINVAL; 2432 goto bad; 2433 } 2434 if (!fhold(ofde->fde_file)) { 2435 error = EBADF; 2436 goto bad; 2437 } 2438 nfde = &newfdp->fd_ofiles[i]; 2439 *nfde = *ofde; 2440 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2441 fdused_init(newfdp, i); 2442 } 2443 FILEDESC_SUNLOCK(fdp); 2444 *ret = newfdp; 2445 return (0); 2446 bad: 2447 FILEDESC_SUNLOCK(fdp); 2448 fdescfree_remapped(newfdp); 2449 return (error); 2450 } 2451 2452 /* 2453 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2454 * one of processes using it exits) and the table used to be shared. 2455 */ 2456 static void 2457 fdclearlocks(struct thread *td) 2458 { 2459 struct filedesc *fdp; 2460 struct filedesc_to_leader *fdtol; 2461 struct flock lf; 2462 struct file *fp; 2463 struct proc *p; 2464 struct vnode *vp; 2465 int i, lastfile; 2466 2467 p = td->td_proc; 2468 fdp = p->p_fd; 2469 fdtol = p->p_fdtol; 2470 MPASS(fdtol != NULL); 2471 2472 FILEDESC_XLOCK(fdp); 2473 KASSERT(fdtol->fdl_refcount > 0, 2474 ("filedesc_to_refcount botch: fdl_refcount=%d", 2475 fdtol->fdl_refcount)); 2476 if (fdtol->fdl_refcount == 1 && 2477 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2478 lastfile = fdlastfile(fdp); 2479 for (i = 0; i <= lastfile; i++) { 2480 fp = fdp->fd_ofiles[i].fde_file; 2481 if (fp == NULL || fp->f_type != DTYPE_VNODE || 2482 !fhold(fp)) 2483 continue; 2484 FILEDESC_XUNLOCK(fdp); 2485 lf.l_whence = SEEK_SET; 2486 lf.l_start = 0; 2487 lf.l_len = 0; 2488 lf.l_type = F_UNLCK; 2489 vp = fp->f_vnode; 2490 (void) VOP_ADVLOCK(vp, 2491 (caddr_t)p->p_leader, F_UNLCK, 2492 &lf, F_POSIX); 2493 FILEDESC_XLOCK(fdp); 2494 fdrop(fp, td); 2495 } 2496 } 2497 retry: 2498 if (fdtol->fdl_refcount == 1) { 2499 if (fdp->fd_holdleaderscount > 0 && 2500 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2501 /* 2502 * close() or kern_dup() has cleared a reference 2503 * in a shared file descriptor table. 2504 */ 2505 fdp->fd_holdleaderswakeup = 1; 2506 sx_sleep(&fdp->fd_holdleaderscount, 2507 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2508 goto retry; 2509 } 2510 if (fdtol->fdl_holdcount > 0) { 2511 /* 2512 * Ensure that fdtol->fdl_leader remains 2513 * valid in closef(). 2514 */ 2515 fdtol->fdl_wakeup = 1; 2516 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2517 "fdlhold", 0); 2518 goto retry; 2519 } 2520 } 2521 fdtol->fdl_refcount--; 2522 if (fdtol->fdl_refcount == 0 && 2523 fdtol->fdl_holdcount == 0) { 2524 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2525 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2526 } else 2527 fdtol = NULL; 2528 p->p_fdtol = NULL; 2529 FILEDESC_XUNLOCK(fdp); 2530 if (fdtol != NULL) 2531 free(fdtol, M_FILEDESC_TO_LEADER); 2532 } 2533 2534 /* 2535 * Release a filedesc structure. 2536 */ 2537 static void 2538 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2539 { 2540 struct filedesc0 *fdp0; 2541 struct freetable *ft, *tft; 2542 struct filedescent *fde; 2543 struct file *fp; 2544 int i, lastfile; 2545 2546 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2547 ("%s: fd table %p carries references", __func__, fdp)); 2548 2549 /* 2550 * Serialize with threads iterating over the table, if any. 2551 */ 2552 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2553 FILEDESC_XLOCK(fdp); 2554 FILEDESC_XUNLOCK(fdp); 2555 } 2556 2557 lastfile = fdlastfile_single(fdp); 2558 for (i = 0; i <= lastfile; i++) { 2559 fde = &fdp->fd_ofiles[i]; 2560 fp = fde->fde_file; 2561 if (fp != NULL) { 2562 fdefree_last(fde); 2563 if (needclose) 2564 (void) closef(fp, td); 2565 else 2566 fdrop(fp, td); 2567 } 2568 } 2569 2570 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2571 free(fdp->fd_map, M_FILEDESC); 2572 if (fdp->fd_nfiles > NDFILE) 2573 free(fdp->fd_files, M_FILEDESC); 2574 2575 fdp0 = (struct filedesc0 *)fdp; 2576 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2577 free(ft->ft_table, M_FILEDESC); 2578 2579 fddrop(fdp); 2580 } 2581 2582 void 2583 fdescfree(struct thread *td) 2584 { 2585 struct proc *p; 2586 struct filedesc *fdp; 2587 2588 p = td->td_proc; 2589 fdp = p->p_fd; 2590 MPASS(fdp != NULL); 2591 2592 #ifdef RACCT 2593 if (RACCT_ENABLED()) 2594 racct_set_unlocked(p, RACCT_NOFILE, 0); 2595 #endif 2596 2597 if (p->p_fdtol != NULL) 2598 fdclearlocks(td); 2599 2600 /* 2601 * Check fdhold for an explanation. 2602 */ 2603 atomic_store_ptr(&p->p_fd, NULL); 2604 atomic_thread_fence_seq_cst(); 2605 PROC_WAIT_UNLOCKED(p); 2606 2607 if (refcount_release(&fdp->fd_refcnt) == 0) 2608 return; 2609 2610 fdescfree_fds(td, fdp, 1); 2611 } 2612 2613 void 2614 pdescfree(struct thread *td) 2615 { 2616 struct proc *p; 2617 struct pwddesc *pdp; 2618 2619 p = td->td_proc; 2620 pdp = p->p_pd; 2621 MPASS(pdp != NULL); 2622 2623 /* 2624 * Check pdhold for an explanation. 2625 */ 2626 atomic_store_ptr(&p->p_pd, NULL); 2627 atomic_thread_fence_seq_cst(); 2628 PROC_WAIT_UNLOCKED(p); 2629 2630 pddrop(pdp); 2631 } 2632 2633 void 2634 fdescfree_remapped(struct filedesc *fdp) 2635 { 2636 #ifdef INVARIANTS 2637 /* fdescfree_fds() asserts that fd_refcnt == 0. */ 2638 if (!refcount_release(&fdp->fd_refcnt)) 2639 panic("%s: fd table %p has extra references", __func__, fdp); 2640 #endif 2641 fdescfree_fds(curthread, fdp, 0); 2642 } 2643 2644 /* 2645 * For setugid programs, we don't want to people to use that setugidness 2646 * to generate error messages which write to a file which otherwise would 2647 * otherwise be off-limits to the process. We check for filesystems where 2648 * the vnode can change out from under us after execve (like [lin]procfs). 2649 * 2650 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2651 * sufficient. We also don't check for setugidness since we know we are. 2652 */ 2653 static bool 2654 is_unsafe(struct file *fp) 2655 { 2656 struct vnode *vp; 2657 2658 if (fp->f_type != DTYPE_VNODE) 2659 return (false); 2660 2661 vp = fp->f_vnode; 2662 return ((vp->v_vflag & VV_PROCDEP) != 0); 2663 } 2664 2665 /* 2666 * Make this setguid thing safe, if at all possible. 2667 */ 2668 void 2669 fdsetugidsafety(struct thread *td) 2670 { 2671 struct filedesc *fdp; 2672 struct file *fp; 2673 int i; 2674 2675 fdp = td->td_proc->p_fd; 2676 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2677 ("the fdtable should not be shared")); 2678 MPASS(fdp->fd_nfiles >= 3); 2679 for (i = 0; i <= 2; i++) { 2680 fp = fdp->fd_ofiles[i].fde_file; 2681 if (fp != NULL && is_unsafe(fp)) { 2682 FILEDESC_XLOCK(fdp); 2683 knote_fdclose(td, i); 2684 /* 2685 * NULL-out descriptor prior to close to avoid 2686 * a race while close blocks. 2687 */ 2688 fdfree(fdp, i); 2689 FILEDESC_XUNLOCK(fdp); 2690 (void) closef(fp, td); 2691 } 2692 } 2693 } 2694 2695 /* 2696 * If a specific file object occupies a specific file descriptor, close the 2697 * file descriptor entry and drop a reference on the file object. This is a 2698 * convenience function to handle a subsequent error in a function that calls 2699 * falloc() that handles the race that another thread might have closed the 2700 * file descriptor out from under the thread creating the file object. 2701 */ 2702 void 2703 fdclose(struct thread *td, struct file *fp, int idx) 2704 { 2705 struct filedesc *fdp = td->td_proc->p_fd; 2706 2707 FILEDESC_XLOCK(fdp); 2708 if (fdp->fd_ofiles[idx].fde_file == fp) { 2709 fdfree(fdp, idx); 2710 FILEDESC_XUNLOCK(fdp); 2711 fdrop(fp, td); 2712 } else 2713 FILEDESC_XUNLOCK(fdp); 2714 } 2715 2716 /* 2717 * Close any files on exec? 2718 */ 2719 void 2720 fdcloseexec(struct thread *td) 2721 { 2722 struct filedesc *fdp; 2723 struct filedescent *fde; 2724 struct file *fp; 2725 int i, lastfile; 2726 2727 fdp = td->td_proc->p_fd; 2728 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2729 ("the fdtable should not be shared")); 2730 lastfile = fdlastfile_single(fdp); 2731 for (i = 0; i <= lastfile; i++) { 2732 fde = &fdp->fd_ofiles[i]; 2733 fp = fde->fde_file; 2734 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2735 (fde->fde_flags & UF_EXCLOSE))) { 2736 FILEDESC_XLOCK(fdp); 2737 fdfree(fdp, i); 2738 (void) closefp(fdp, i, fp, td, false, false); 2739 FILEDESC_UNLOCK_ASSERT(fdp); 2740 } 2741 } 2742 } 2743 2744 /* 2745 * It is unsafe for set[ug]id processes to be started with file 2746 * descriptors 0..2 closed, as these descriptors are given implicit 2747 * significance in the Standard C library. fdcheckstd() will create a 2748 * descriptor referencing /dev/null for each of stdin, stdout, and 2749 * stderr that is not already open. 2750 */ 2751 int 2752 fdcheckstd(struct thread *td) 2753 { 2754 struct filedesc *fdp; 2755 register_t save; 2756 int i, error, devnull; 2757 2758 fdp = td->td_proc->p_fd; 2759 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2760 ("the fdtable should not be shared")); 2761 MPASS(fdp->fd_nfiles >= 3); 2762 devnull = -1; 2763 for (i = 0; i <= 2; i++) { 2764 if (fdp->fd_ofiles[i].fde_file != NULL) 2765 continue; 2766 2767 save = td->td_retval[0]; 2768 if (devnull != -1) { 2769 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2770 } else { 2771 error = kern_openat(td, AT_FDCWD, "/dev/null", 2772 UIO_SYSSPACE, O_RDWR, 0); 2773 if (error == 0) { 2774 devnull = td->td_retval[0]; 2775 KASSERT(devnull == i, ("we didn't get our fd")); 2776 } 2777 } 2778 td->td_retval[0] = save; 2779 if (error != 0) 2780 return (error); 2781 } 2782 return (0); 2783 } 2784 2785 /* 2786 * Internal form of close. Decrement reference count on file structure. 2787 * Note: td may be NULL when closing a file that was being passed in a 2788 * message. 2789 */ 2790 int 2791 closef(struct file *fp, struct thread *td) 2792 { 2793 struct vnode *vp; 2794 struct flock lf; 2795 struct filedesc_to_leader *fdtol; 2796 struct filedesc *fdp; 2797 2798 MPASS(td != NULL); 2799 2800 /* 2801 * POSIX record locking dictates that any close releases ALL 2802 * locks owned by this process. This is handled by setting 2803 * a flag in the unlock to free ONLY locks obeying POSIX 2804 * semantics, and not to free BSD-style file locks. 2805 * If the descriptor was in a message, POSIX-style locks 2806 * aren't passed with the descriptor, and the thread pointer 2807 * will be NULL. Callers should be careful only to pass a 2808 * NULL thread pointer when there really is no owning 2809 * context that might have locks, or the locks will be 2810 * leaked. 2811 */ 2812 if (fp->f_type == DTYPE_VNODE) { 2813 vp = fp->f_vnode; 2814 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2815 lf.l_whence = SEEK_SET; 2816 lf.l_start = 0; 2817 lf.l_len = 0; 2818 lf.l_type = F_UNLCK; 2819 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2820 F_UNLCK, &lf, F_POSIX); 2821 } 2822 fdtol = td->td_proc->p_fdtol; 2823 if (fdtol != NULL) { 2824 /* 2825 * Handle special case where file descriptor table is 2826 * shared between multiple process leaders. 2827 */ 2828 fdp = td->td_proc->p_fd; 2829 FILEDESC_XLOCK(fdp); 2830 for (fdtol = fdtol->fdl_next; 2831 fdtol != td->td_proc->p_fdtol; 2832 fdtol = fdtol->fdl_next) { 2833 if ((fdtol->fdl_leader->p_flag & 2834 P_ADVLOCK) == 0) 2835 continue; 2836 fdtol->fdl_holdcount++; 2837 FILEDESC_XUNLOCK(fdp); 2838 lf.l_whence = SEEK_SET; 2839 lf.l_start = 0; 2840 lf.l_len = 0; 2841 lf.l_type = F_UNLCK; 2842 vp = fp->f_vnode; 2843 (void) VOP_ADVLOCK(vp, 2844 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2845 F_POSIX); 2846 FILEDESC_XLOCK(fdp); 2847 fdtol->fdl_holdcount--; 2848 if (fdtol->fdl_holdcount == 0 && 2849 fdtol->fdl_wakeup != 0) { 2850 fdtol->fdl_wakeup = 0; 2851 wakeup(fdtol); 2852 } 2853 } 2854 FILEDESC_XUNLOCK(fdp); 2855 } 2856 } 2857 return (fdrop_close(fp, td)); 2858 } 2859 2860 /* 2861 * Hack for file descriptor passing code. 2862 */ 2863 void 2864 closef_nothread(struct file *fp) 2865 { 2866 2867 fdrop(fp, NULL); 2868 } 2869 2870 /* 2871 * Initialize the file pointer with the specified properties. 2872 * 2873 * The ops are set with release semantics to be certain that the flags, type, 2874 * and data are visible when ops is. This is to prevent ops methods from being 2875 * called with bad data. 2876 */ 2877 void 2878 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2879 { 2880 fp->f_data = data; 2881 fp->f_flag = flag; 2882 fp->f_type = type; 2883 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2884 } 2885 2886 void 2887 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops) 2888 { 2889 fp->f_seqcount[UIO_READ] = 1; 2890 fp->f_seqcount[UIO_WRITE] = 1; 2891 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 2892 data, ops); 2893 } 2894 2895 int 2896 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2897 struct file **fpp, struct filecaps *havecapsp) 2898 { 2899 struct filedescent *fde; 2900 int error; 2901 2902 FILEDESC_LOCK_ASSERT(fdp); 2903 2904 fde = fdeget_locked(fdp, fd); 2905 if (fde == NULL) { 2906 error = EBADF; 2907 goto out; 2908 } 2909 2910 #ifdef CAPABILITIES 2911 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 2912 if (error != 0) 2913 goto out; 2914 #endif 2915 2916 if (havecapsp != NULL) 2917 filecaps_copy(&fde->fde_caps, havecapsp, true); 2918 2919 *fpp = fde->fde_file; 2920 2921 error = 0; 2922 out: 2923 return (error); 2924 } 2925 2926 int 2927 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2928 struct file **fpp, struct filecaps *havecapsp) 2929 { 2930 struct filedesc *fdp = td->td_proc->p_fd; 2931 int error; 2932 #ifndef CAPABILITIES 2933 error = fget_unlocked(fdp, fd, needrightsp, fpp); 2934 if (havecapsp != NULL && error == 0) 2935 filecaps_fill(havecapsp); 2936 #else 2937 struct file *fp; 2938 seqc_t seq; 2939 2940 *fpp = NULL; 2941 for (;;) { 2942 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq); 2943 if (error != 0) 2944 return (error); 2945 2946 if (havecapsp != NULL) { 2947 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2948 havecapsp, false)) { 2949 fdrop(fp, td); 2950 goto get_locked; 2951 } 2952 } 2953 2954 if (!fd_modified(fdp, fd, seq)) 2955 break; 2956 fdrop(fp, td); 2957 } 2958 2959 *fpp = fp; 2960 return (0); 2961 2962 get_locked: 2963 FILEDESC_SLOCK(fdp); 2964 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); 2965 if (error == 0 && !fhold(*fpp)) 2966 error = EBADF; 2967 FILEDESC_SUNLOCK(fdp); 2968 #endif 2969 return (error); 2970 } 2971 2972 #ifdef CAPABILITIES 2973 int 2974 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 2975 { 2976 const struct filedescent *fde; 2977 const struct fdescenttbl *fdt; 2978 struct filedesc *fdp; 2979 struct file *fp; 2980 struct vnode *vp; 2981 const cap_rights_t *haverights; 2982 cap_rights_t rights; 2983 seqc_t seq; 2984 2985 VFS_SMR_ASSERT_ENTERED(); 2986 2987 rights = *ndp->ni_rightsneeded; 2988 cap_rights_set_one(&rights, CAP_LOOKUP); 2989 2990 fdp = curproc->p_fd; 2991 fdt = fdp->fd_files; 2992 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 2993 return (EBADF); 2994 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 2995 fde = &fdt->fdt_ofiles[fd]; 2996 haverights = cap_rights_fde_inline(fde); 2997 fp = fde->fde_file; 2998 if (__predict_false(fp == NULL)) 2999 return (EAGAIN); 3000 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 3001 return (EAGAIN); 3002 *fsearch = ((fp->f_flag & FSEARCH) != 0); 3003 vp = fp->f_vnode; 3004 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3005 return (EAGAIN); 3006 } 3007 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 3008 return (EAGAIN); 3009 } 3010 /* 3011 * Use an acquire barrier to force re-reading of fdt so it is 3012 * refreshed for verification. 3013 */ 3014 atomic_thread_fence_acq(); 3015 fdt = fdp->fd_files; 3016 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq))) 3017 return (EAGAIN); 3018 /* 3019 * If file descriptor doesn't have all rights, 3020 * all lookups relative to it must also be 3021 * strictly relative. 3022 * 3023 * Not yet supported by fast path. 3024 */ 3025 CAP_ALL(&rights); 3026 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3027 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3028 ndp->ni_filecaps.fc_nioctls != -1) { 3029 #ifdef notyet 3030 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 3031 #else 3032 return (EAGAIN); 3033 #endif 3034 } 3035 *vpp = vp; 3036 return (0); 3037 } 3038 #else 3039 int 3040 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 3041 { 3042 const struct fdescenttbl *fdt; 3043 struct filedesc *fdp; 3044 struct file *fp; 3045 struct vnode *vp; 3046 3047 VFS_SMR_ASSERT_ENTERED(); 3048 3049 fdp = curproc->p_fd; 3050 fdt = fdp->fd_files; 3051 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3052 return (EBADF); 3053 fp = fdt->fdt_ofiles[fd].fde_file; 3054 if (__predict_false(fp == NULL)) 3055 return (EAGAIN); 3056 *fsearch = ((fp->f_flag & FSEARCH) != 0); 3057 vp = fp->f_vnode; 3058 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3059 return (EAGAIN); 3060 } 3061 /* 3062 * Use an acquire barrier to force re-reading of fdt so it is 3063 * refreshed for verification. 3064 */ 3065 atomic_thread_fence_acq(); 3066 fdt = fdp->fd_files; 3067 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3068 return (EAGAIN); 3069 filecaps_fill(&ndp->ni_filecaps); 3070 *vpp = vp; 3071 return (0); 3072 } 3073 #endif 3074 3075 int 3076 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3077 struct file **fpp, seqc_t *seqp) 3078 { 3079 #ifdef CAPABILITIES 3080 const struct filedescent *fde; 3081 #endif 3082 const struct fdescenttbl *fdt; 3083 struct file *fp; 3084 #ifdef CAPABILITIES 3085 seqc_t seq; 3086 cap_rights_t haverights; 3087 int error; 3088 #endif 3089 3090 fdt = fdp->fd_files; 3091 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3092 return (EBADF); 3093 /* 3094 * Fetch the descriptor locklessly. We avoid fdrop() races by 3095 * never raising a refcount above 0. To accomplish this we have 3096 * to use a cmpset loop rather than an atomic_add. The descriptor 3097 * must be re-verified once we acquire a reference to be certain 3098 * that the identity is still correct and we did not lose a race 3099 * due to preemption. 3100 */ 3101 for (;;) { 3102 #ifdef CAPABILITIES 3103 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3104 fde = &fdt->fdt_ofiles[fd]; 3105 haverights = *cap_rights_fde_inline(fde); 3106 fp = fde->fde_file; 3107 if (!seqc_consistent(fd_seqc(fdt, fd), seq)) 3108 continue; 3109 #else 3110 fp = fdt->fdt_ofiles[fd].fde_file; 3111 #endif 3112 if (fp == NULL) 3113 return (EBADF); 3114 #ifdef CAPABILITIES 3115 error = cap_check_inline(&haverights, needrightsp); 3116 if (error != 0) 3117 return (error); 3118 #endif 3119 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3120 /* 3121 * Force a reload. Other thread could reallocate the 3122 * table before this fd was closed, so it is possible 3123 * that there is a stale fp pointer in cached version. 3124 */ 3125 fdt = atomic_load_ptr(&fdp->fd_files); 3126 continue; 3127 } 3128 /* 3129 * Use an acquire barrier to force re-reading of fdt so it is 3130 * refreshed for verification. 3131 */ 3132 atomic_thread_fence_acq(); 3133 fdt = fdp->fd_files; 3134 #ifdef CAPABILITIES 3135 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq)) 3136 #else 3137 if (fp == fdt->fdt_ofiles[fd].fde_file) 3138 #endif 3139 break; 3140 fdrop(fp, curthread); 3141 } 3142 *fpp = fp; 3143 if (seqp != NULL) { 3144 #ifdef CAPABILITIES 3145 *seqp = seq; 3146 #endif 3147 } 3148 return (0); 3149 } 3150 3151 /* 3152 * See the comments in fget_unlocked_seq for an explanation of how this works. 3153 * 3154 * This is a simplified variant which bails out to the aforementioned routine 3155 * if anything goes wrong. In practice this only happens when userspace is 3156 * racing with itself. 3157 */ 3158 int 3159 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3160 struct file **fpp) 3161 { 3162 #ifdef CAPABILITIES 3163 const struct filedescent *fde; 3164 #endif 3165 const struct fdescenttbl *fdt; 3166 struct file *fp; 3167 #ifdef CAPABILITIES 3168 seqc_t seq; 3169 const cap_rights_t *haverights; 3170 #endif 3171 3172 fdt = fdp->fd_files; 3173 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3174 return (EBADF); 3175 #ifdef CAPABILITIES 3176 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3177 fde = &fdt->fdt_ofiles[fd]; 3178 haverights = cap_rights_fde_inline(fde); 3179 fp = fde->fde_file; 3180 #else 3181 fp = fdt->fdt_ofiles[fd].fde_file; 3182 #endif 3183 if (__predict_false(fp == NULL)) 3184 goto out_fallback; 3185 #ifdef CAPABILITIES 3186 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3187 goto out_fallback; 3188 #endif 3189 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3190 goto out_fallback; 3191 3192 /* 3193 * Use an acquire barrier to force re-reading of fdt so it is 3194 * refreshed for verification. 3195 */ 3196 atomic_thread_fence_acq(); 3197 fdt = fdp->fd_files; 3198 #ifdef CAPABILITIES 3199 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq))) 3200 #else 3201 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3202 #endif 3203 goto out_fdrop; 3204 *fpp = fp; 3205 return (0); 3206 out_fdrop: 3207 fdrop(fp, curthread); 3208 out_fallback: 3209 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL)); 3210 } 3211 3212 /* 3213 * Translate fd -> file when the caller guarantees the file descriptor table 3214 * can't be changed by others. 3215 * 3216 * Note this does not mean the file object itself is only visible to the caller, 3217 * merely that it wont disappear without having to be referenced. 3218 * 3219 * Must be paired with fput_only_user. 3220 */ 3221 #ifdef CAPABILITIES 3222 int 3223 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3224 struct file **fpp) 3225 { 3226 const struct filedescent *fde; 3227 const struct fdescenttbl *fdt; 3228 const cap_rights_t *haverights; 3229 struct file *fp; 3230 int error; 3231 3232 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3233 3234 if (__predict_false(fd >= fdp->fd_nfiles)) 3235 return (EBADF); 3236 3237 fdt = fdp->fd_files; 3238 fde = &fdt->fdt_ofiles[fd]; 3239 fp = fde->fde_file; 3240 if (__predict_false(fp == NULL)) 3241 return (EBADF); 3242 MPASS(refcount_load(&fp->f_count) > 0); 3243 haverights = cap_rights_fde_inline(fde); 3244 error = cap_check_inline(haverights, needrightsp); 3245 if (__predict_false(error != 0)) 3246 return (error); 3247 *fpp = fp; 3248 return (0); 3249 } 3250 #else 3251 int 3252 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3253 struct file **fpp) 3254 { 3255 struct file *fp; 3256 3257 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3258 3259 if (__predict_false(fd >= fdp->fd_nfiles)) 3260 return (EBADF); 3261 3262 fp = fdp->fd_ofiles[fd].fde_file; 3263 if (__predict_false(fp == NULL)) 3264 return (EBADF); 3265 3266 MPASS(refcount_load(&fp->f_count) > 0); 3267 *fpp = fp; 3268 return (0); 3269 } 3270 #endif 3271 3272 /* 3273 * Extract the file pointer associated with the specified descriptor for the 3274 * current user process. 3275 * 3276 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3277 * returned. 3278 * 3279 * File's rights will be checked against the capability rights mask. 3280 * 3281 * If an error occurred the non-zero error is returned and *fpp is set to 3282 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3283 * responsible for fdrop(). 3284 */ 3285 static __inline int 3286 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3287 cap_rights_t *needrightsp) 3288 { 3289 struct filedesc *fdp; 3290 struct file *fp; 3291 int error; 3292 3293 *fpp = NULL; 3294 fdp = td->td_proc->p_fd; 3295 error = fget_unlocked(fdp, fd, needrightsp, &fp); 3296 if (__predict_false(error != 0)) 3297 return (error); 3298 if (__predict_false(fp->f_ops == &badfileops)) { 3299 fdrop(fp, td); 3300 return (EBADF); 3301 } 3302 3303 /* 3304 * FREAD and FWRITE failure return EBADF as per POSIX. 3305 */ 3306 error = 0; 3307 switch (flags) { 3308 case FREAD: 3309 case FWRITE: 3310 if ((fp->f_flag & flags) == 0) 3311 error = EBADF; 3312 break; 3313 case FEXEC: 3314 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3315 ((fp->f_flag & FWRITE) != 0)) 3316 error = EBADF; 3317 break; 3318 case 0: 3319 break; 3320 default: 3321 KASSERT(0, ("wrong flags")); 3322 } 3323 3324 if (error != 0) { 3325 fdrop(fp, td); 3326 return (error); 3327 } 3328 3329 *fpp = fp; 3330 return (0); 3331 } 3332 3333 int 3334 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3335 { 3336 3337 return (_fget(td, fd, fpp, 0, rightsp)); 3338 } 3339 3340 int 3341 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, 3342 struct file **fpp) 3343 { 3344 int error; 3345 #ifndef CAPABILITIES 3346 error = _fget(td, fd, fpp, 0, rightsp); 3347 if (maxprotp != NULL) 3348 *maxprotp = VM_PROT_ALL; 3349 return (error); 3350 #else 3351 cap_rights_t fdrights; 3352 struct filedesc *fdp; 3353 struct file *fp; 3354 seqc_t seq; 3355 3356 *fpp = NULL; 3357 fdp = td->td_proc->p_fd; 3358 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3359 for (;;) { 3360 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); 3361 if (__predict_false(error != 0)) 3362 return (error); 3363 if (__predict_false(fp->f_ops == &badfileops)) { 3364 fdrop(fp, td); 3365 return (EBADF); 3366 } 3367 if (maxprotp != NULL) 3368 fdrights = *cap_rights(fdp, fd); 3369 if (!fd_modified(fdp, fd, seq)) 3370 break; 3371 fdrop(fp, td); 3372 } 3373 3374 /* 3375 * If requested, convert capability rights to access flags. 3376 */ 3377 if (maxprotp != NULL) 3378 *maxprotp = cap_rights_to_vmprot(&fdrights); 3379 *fpp = fp; 3380 return (0); 3381 #endif 3382 } 3383 3384 int 3385 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3386 { 3387 3388 return (_fget(td, fd, fpp, FREAD, rightsp)); 3389 } 3390 3391 int 3392 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3393 { 3394 3395 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3396 } 3397 3398 int 3399 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 3400 struct file **fpp) 3401 { 3402 struct filedesc *fdp = td->td_proc->p_fd; 3403 #ifndef CAPABILITIES 3404 return (fget_unlocked(fdp, fd, rightsp, fpp)); 3405 #else 3406 struct file *fp; 3407 int error; 3408 seqc_t seq; 3409 3410 *fpp = NULL; 3411 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3412 for (;;) { 3413 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); 3414 if (error != 0) 3415 return (error); 3416 error = cap_fcntl_check(fdp, fd, needfcntl); 3417 if (!fd_modified(fdp, fd, seq)) 3418 break; 3419 fdrop(fp, td); 3420 } 3421 if (error != 0) { 3422 fdrop(fp, td); 3423 return (error); 3424 } 3425 *fpp = fp; 3426 return (0); 3427 #endif 3428 } 3429 3430 /* 3431 * Like fget() but loads the underlying vnode, or returns an error if the 3432 * descriptor does not represent a vnode. Note that pipes use vnodes but 3433 * never have VM objects. The returned vnode will be vref()'d. 3434 * 3435 * XXX: what about the unused flags ? 3436 */ 3437 static __inline int 3438 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 3439 struct vnode **vpp) 3440 { 3441 struct file *fp; 3442 int error; 3443 3444 *vpp = NULL; 3445 error = _fget(td, fd, &fp, flags, needrightsp); 3446 if (error != 0) 3447 return (error); 3448 if (fp->f_vnode == NULL) { 3449 error = EINVAL; 3450 } else { 3451 *vpp = fp->f_vnode; 3452 vref(*vpp); 3453 } 3454 fdrop(fp, td); 3455 3456 return (error); 3457 } 3458 3459 int 3460 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3461 { 3462 3463 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3464 } 3465 3466 int 3467 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 3468 struct filecaps *havecaps, struct vnode **vpp) 3469 { 3470 struct filecaps caps; 3471 struct file *fp; 3472 int error; 3473 3474 error = fget_cap(td, fd, needrightsp, &fp, &caps); 3475 if (error != 0) 3476 return (error); 3477 if (fp->f_ops == &badfileops) { 3478 error = EBADF; 3479 goto out; 3480 } 3481 if (fp->f_vnode == NULL) { 3482 error = EINVAL; 3483 goto out; 3484 } 3485 3486 *havecaps = caps; 3487 *vpp = fp->f_vnode; 3488 vref(*vpp); 3489 fdrop(fp, td); 3490 3491 return (0); 3492 out: 3493 filecaps_free(&caps); 3494 fdrop(fp, td); 3495 return (error); 3496 } 3497 3498 int 3499 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3500 { 3501 3502 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3503 } 3504 3505 int 3506 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3507 { 3508 3509 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3510 } 3511 3512 #ifdef notyet 3513 int 3514 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 3515 struct vnode **vpp) 3516 { 3517 3518 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3519 } 3520 #endif 3521 3522 /* 3523 * Handle the last reference to a file being closed. 3524 * 3525 * Without the noinline attribute clang keeps inlining the func thorough this 3526 * file when fdrop is used. 3527 */ 3528 int __noinline 3529 _fdrop(struct file *fp, struct thread *td) 3530 { 3531 int error; 3532 #ifdef INVARIANTS 3533 int count; 3534 3535 count = refcount_load(&fp->f_count); 3536 if (count != 0) 3537 panic("fdrop: fp %p count %d", fp, count); 3538 #endif 3539 error = fo_close(fp, td); 3540 atomic_subtract_int(&openfiles, 1); 3541 crfree(fp->f_cred); 3542 free(fp->f_advice, M_FADVISE); 3543 uma_zfree(file_zone, fp); 3544 3545 return (error); 3546 } 3547 3548 /* 3549 * Apply an advisory lock on a file descriptor. 3550 * 3551 * Just attempt to get a record lock of the requested type on the entire file 3552 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 3553 */ 3554 #ifndef _SYS_SYSPROTO_H_ 3555 struct flock_args { 3556 int fd; 3557 int how; 3558 }; 3559 #endif 3560 /* ARGSUSED */ 3561 int 3562 sys_flock(struct thread *td, struct flock_args *uap) 3563 { 3564 struct file *fp; 3565 struct vnode *vp; 3566 struct flock lf; 3567 int error; 3568 3569 error = fget(td, uap->fd, &cap_flock_rights, &fp); 3570 if (error != 0) 3571 return (error); 3572 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 3573 fdrop(fp, td); 3574 return (EOPNOTSUPP); 3575 } 3576 3577 vp = fp->f_vnode; 3578 lf.l_whence = SEEK_SET; 3579 lf.l_start = 0; 3580 lf.l_len = 0; 3581 if (uap->how & LOCK_UN) { 3582 lf.l_type = F_UNLCK; 3583 atomic_clear_int(&fp->f_flag, FHASLOCK); 3584 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3585 goto done2; 3586 } 3587 if (uap->how & LOCK_EX) 3588 lf.l_type = F_WRLCK; 3589 else if (uap->how & LOCK_SH) 3590 lf.l_type = F_RDLCK; 3591 else { 3592 error = EBADF; 3593 goto done2; 3594 } 3595 atomic_set_int(&fp->f_flag, FHASLOCK); 3596 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3597 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3598 done2: 3599 fdrop(fp, td); 3600 return (error); 3601 } 3602 /* 3603 * Duplicate the specified descriptor to a free descriptor. 3604 */ 3605 int 3606 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3607 int openerror, int *indxp) 3608 { 3609 struct filedescent *newfde, *oldfde; 3610 struct file *fp; 3611 u_long *ioctls; 3612 int error, indx; 3613 3614 KASSERT(openerror == ENODEV || openerror == ENXIO, 3615 ("unexpected error %d in %s", openerror, __func__)); 3616 3617 /* 3618 * If the to-be-dup'd fd number is greater than the allowed number 3619 * of file descriptors, or the fd to be dup'd has already been 3620 * closed, then reject. 3621 */ 3622 FILEDESC_XLOCK(fdp); 3623 if ((fp = fget_locked(fdp, dfd)) == NULL) { 3624 FILEDESC_XUNLOCK(fdp); 3625 return (EBADF); 3626 } 3627 3628 error = fdalloc(td, 0, &indx); 3629 if (error != 0) { 3630 FILEDESC_XUNLOCK(fdp); 3631 return (error); 3632 } 3633 3634 /* 3635 * There are two cases of interest here. 3636 * 3637 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3638 * 3639 * For ENXIO steal away the file structure from (dfd) and store it in 3640 * (indx). (dfd) is effectively closed by this operation. 3641 */ 3642 switch (openerror) { 3643 case ENODEV: 3644 /* 3645 * Check that the mode the file is being opened for is a 3646 * subset of the mode of the existing descriptor. 3647 */ 3648 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3649 fdunused(fdp, indx); 3650 FILEDESC_XUNLOCK(fdp); 3651 return (EACCES); 3652 } 3653 if (!fhold(fp)) { 3654 fdunused(fdp, indx); 3655 FILEDESC_XUNLOCK(fdp); 3656 return (EBADF); 3657 } 3658 newfde = &fdp->fd_ofiles[indx]; 3659 oldfde = &fdp->fd_ofiles[dfd]; 3660 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3661 #ifdef CAPABILITIES 3662 seqc_write_begin(&newfde->fde_seqc); 3663 #endif 3664 memcpy(newfde, oldfde, fde_change_size); 3665 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3666 ioctls); 3667 #ifdef CAPABILITIES 3668 seqc_write_end(&newfde->fde_seqc); 3669 #endif 3670 break; 3671 case ENXIO: 3672 /* 3673 * Steal away the file pointer from dfd and stuff it into indx. 3674 */ 3675 newfde = &fdp->fd_ofiles[indx]; 3676 oldfde = &fdp->fd_ofiles[dfd]; 3677 #ifdef CAPABILITIES 3678 seqc_write_begin(&newfde->fde_seqc); 3679 #endif 3680 memcpy(newfde, oldfde, fde_change_size); 3681 oldfde->fde_file = NULL; 3682 fdunused(fdp, dfd); 3683 #ifdef CAPABILITIES 3684 seqc_write_end(&newfde->fde_seqc); 3685 #endif 3686 break; 3687 } 3688 FILEDESC_XUNLOCK(fdp); 3689 *indxp = indx; 3690 return (0); 3691 } 3692 3693 /* 3694 * This sysctl determines if we will allow a process to chroot(2) if it 3695 * has a directory open: 3696 * 0: disallowed for all processes. 3697 * 1: allowed for processes that were not already chroot(2)'ed. 3698 * 2: allowed for all processes. 3699 */ 3700 3701 static int chroot_allow_open_directories = 1; 3702 3703 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3704 &chroot_allow_open_directories, 0, 3705 "Allow a process to chroot(2) if it has a directory open"); 3706 3707 /* 3708 * Helper function for raised chroot(2) security function: Refuse if 3709 * any filedescriptors are open directories. 3710 */ 3711 static int 3712 chroot_refuse_vdir_fds(struct filedesc *fdp) 3713 { 3714 struct vnode *vp; 3715 struct file *fp; 3716 int fd, lastfile; 3717 3718 FILEDESC_LOCK_ASSERT(fdp); 3719 3720 lastfile = fdlastfile(fdp); 3721 for (fd = 0; fd <= lastfile; fd++) { 3722 fp = fget_locked(fdp, fd); 3723 if (fp == NULL) 3724 continue; 3725 if (fp->f_type == DTYPE_VNODE) { 3726 vp = fp->f_vnode; 3727 if (vp->v_type == VDIR) 3728 return (EPERM); 3729 } 3730 } 3731 return (0); 3732 } 3733 3734 static void 3735 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 3736 { 3737 3738 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 3739 vrefact(oldpwd->pwd_cdir); 3740 newpwd->pwd_cdir = oldpwd->pwd_cdir; 3741 } 3742 3743 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 3744 vrefact(oldpwd->pwd_rdir); 3745 newpwd->pwd_rdir = oldpwd->pwd_rdir; 3746 } 3747 3748 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 3749 vrefact(oldpwd->pwd_jdir); 3750 newpwd->pwd_jdir = oldpwd->pwd_jdir; 3751 } 3752 } 3753 3754 struct pwd * 3755 pwd_hold_pwddesc(struct pwddesc *pdp) 3756 { 3757 struct pwd *pwd; 3758 3759 PWDDESC_ASSERT_XLOCKED(pdp); 3760 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3761 if (pwd != NULL) 3762 refcount_acquire(&pwd->pwd_refcount); 3763 return (pwd); 3764 } 3765 3766 bool 3767 pwd_hold_smr(struct pwd *pwd) 3768 { 3769 3770 MPASS(pwd != NULL); 3771 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 3772 return (true); 3773 } 3774 return (false); 3775 } 3776 3777 struct pwd * 3778 pwd_hold(struct thread *td) 3779 { 3780 struct pwddesc *pdp; 3781 struct pwd *pwd; 3782 3783 pdp = td->td_proc->p_pd; 3784 3785 vfs_smr_enter(); 3786 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 3787 if (pwd_hold_smr(pwd)) { 3788 vfs_smr_exit(); 3789 return (pwd); 3790 } 3791 vfs_smr_exit(); 3792 PWDDESC_XLOCK(pdp); 3793 pwd = pwd_hold_pwddesc(pdp); 3794 MPASS(pwd != NULL); 3795 PWDDESC_XUNLOCK(pdp); 3796 return (pwd); 3797 } 3798 3799 static struct pwd * 3800 pwd_alloc(void) 3801 { 3802 struct pwd *pwd; 3803 3804 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 3805 bzero(pwd, sizeof(*pwd)); 3806 refcount_init(&pwd->pwd_refcount, 1); 3807 return (pwd); 3808 } 3809 3810 void 3811 pwd_drop(struct pwd *pwd) 3812 { 3813 3814 if (!refcount_release(&pwd->pwd_refcount)) 3815 return; 3816 3817 if (pwd->pwd_cdir != NULL) 3818 vrele(pwd->pwd_cdir); 3819 if (pwd->pwd_rdir != NULL) 3820 vrele(pwd->pwd_rdir); 3821 if (pwd->pwd_jdir != NULL) 3822 vrele(pwd->pwd_jdir); 3823 uma_zfree_smr(pwd_zone, pwd); 3824 } 3825 3826 /* 3827 * The caller is responsible for invoking priv_check() and 3828 * mac_vnode_check_chroot() to authorize this operation. 3829 */ 3830 int 3831 pwd_chroot(struct thread *td, struct vnode *vp) 3832 { 3833 struct pwddesc *pdp; 3834 struct filedesc *fdp; 3835 struct pwd *newpwd, *oldpwd; 3836 int error; 3837 3838 fdp = td->td_proc->p_fd; 3839 pdp = td->td_proc->p_pd; 3840 newpwd = pwd_alloc(); 3841 FILEDESC_SLOCK(fdp); 3842 PWDDESC_XLOCK(pdp); 3843 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3844 if (chroot_allow_open_directories == 0 || 3845 (chroot_allow_open_directories == 1 && 3846 oldpwd->pwd_rdir != rootvnode)) { 3847 error = chroot_refuse_vdir_fds(fdp); 3848 FILEDESC_SUNLOCK(fdp); 3849 if (error != 0) { 3850 PWDDESC_XUNLOCK(pdp); 3851 pwd_drop(newpwd); 3852 return (error); 3853 } 3854 } else { 3855 FILEDESC_SUNLOCK(fdp); 3856 } 3857 3858 vrefact(vp); 3859 newpwd->pwd_rdir = vp; 3860 if (oldpwd->pwd_jdir == NULL) { 3861 vrefact(vp); 3862 newpwd->pwd_jdir = vp; 3863 } 3864 pwd_fill(oldpwd, newpwd); 3865 pwd_set(pdp, newpwd); 3866 PWDDESC_XUNLOCK(pdp); 3867 pwd_drop(oldpwd); 3868 return (0); 3869 } 3870 3871 void 3872 pwd_chdir(struct thread *td, struct vnode *vp) 3873 { 3874 struct pwddesc *pdp; 3875 struct pwd *newpwd, *oldpwd; 3876 3877 VNPASS(vp->v_usecount > 0, vp); 3878 3879 newpwd = pwd_alloc(); 3880 pdp = td->td_proc->p_pd; 3881 PWDDESC_XLOCK(pdp); 3882 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3883 newpwd->pwd_cdir = vp; 3884 pwd_fill(oldpwd, newpwd); 3885 pwd_set(pdp, newpwd); 3886 PWDDESC_XUNLOCK(pdp); 3887 pwd_drop(oldpwd); 3888 } 3889 3890 /* 3891 * jail_attach(2) changes both root and working directories. 3892 */ 3893 int 3894 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 3895 { 3896 struct pwddesc *pdp; 3897 struct filedesc *fdp; 3898 struct pwd *newpwd, *oldpwd; 3899 int error; 3900 3901 fdp = td->td_proc->p_fd; 3902 pdp = td->td_proc->p_pd; 3903 newpwd = pwd_alloc(); 3904 FILEDESC_SLOCK(fdp); 3905 PWDDESC_XLOCK(pdp); 3906 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3907 error = chroot_refuse_vdir_fds(fdp); 3908 FILEDESC_SUNLOCK(fdp); 3909 if (error != 0) { 3910 PWDDESC_XUNLOCK(pdp); 3911 pwd_drop(newpwd); 3912 return (error); 3913 } 3914 3915 vrefact(vp); 3916 newpwd->pwd_rdir = vp; 3917 vrefact(vp); 3918 newpwd->pwd_cdir = vp; 3919 if (oldpwd->pwd_jdir == NULL) { 3920 vrefact(vp); 3921 newpwd->pwd_jdir = vp; 3922 } 3923 pwd_fill(oldpwd, newpwd); 3924 pwd_set(pdp, newpwd); 3925 PWDDESC_XUNLOCK(pdp); 3926 pwd_drop(oldpwd); 3927 return (0); 3928 } 3929 3930 void 3931 pwd_ensure_dirs(void) 3932 { 3933 struct pwddesc *pdp; 3934 struct pwd *oldpwd, *newpwd; 3935 3936 pdp = curproc->p_pd; 3937 PWDDESC_XLOCK(pdp); 3938 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3939 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) { 3940 PWDDESC_XUNLOCK(pdp); 3941 return; 3942 } 3943 PWDDESC_XUNLOCK(pdp); 3944 3945 newpwd = pwd_alloc(); 3946 PWDDESC_XLOCK(pdp); 3947 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3948 pwd_fill(oldpwd, newpwd); 3949 if (newpwd->pwd_cdir == NULL) { 3950 vrefact(rootvnode); 3951 newpwd->pwd_cdir = rootvnode; 3952 } 3953 if (newpwd->pwd_rdir == NULL) { 3954 vrefact(rootvnode); 3955 newpwd->pwd_rdir = rootvnode; 3956 } 3957 pwd_set(pdp, newpwd); 3958 PWDDESC_XUNLOCK(pdp); 3959 pwd_drop(oldpwd); 3960 } 3961 3962 void 3963 pwd_set_rootvnode(void) 3964 { 3965 struct pwddesc *pdp; 3966 struct pwd *oldpwd, *newpwd; 3967 3968 pdp = curproc->p_pd; 3969 3970 newpwd = pwd_alloc(); 3971 PWDDESC_XLOCK(pdp); 3972 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3973 vrefact(rootvnode); 3974 newpwd->pwd_cdir = rootvnode; 3975 vrefact(rootvnode); 3976 newpwd->pwd_rdir = rootvnode; 3977 pwd_fill(oldpwd, newpwd); 3978 pwd_set(pdp, newpwd); 3979 PWDDESC_XUNLOCK(pdp); 3980 pwd_drop(oldpwd); 3981 } 3982 3983 /* 3984 * Scan all active processes and prisons to see if any of them have a current 3985 * or root directory of `olddp'. If so, replace them with the new mount point. 3986 */ 3987 void 3988 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3989 { 3990 struct pwddesc *pdp; 3991 struct pwd *newpwd, *oldpwd; 3992 struct prison *pr; 3993 struct proc *p; 3994 int nrele; 3995 3996 if (vrefcnt(olddp) == 1) 3997 return; 3998 nrele = 0; 3999 newpwd = pwd_alloc(); 4000 sx_slock(&allproc_lock); 4001 FOREACH_PROC_IN_SYSTEM(p) { 4002 PROC_LOCK(p); 4003 pdp = pdhold(p); 4004 PROC_UNLOCK(p); 4005 if (pdp == NULL) 4006 continue; 4007 PWDDESC_XLOCK(pdp); 4008 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4009 if (oldpwd == NULL || 4010 (oldpwd->pwd_cdir != olddp && 4011 oldpwd->pwd_rdir != olddp && 4012 oldpwd->pwd_jdir != olddp)) { 4013 PWDDESC_XUNLOCK(pdp); 4014 pddrop(pdp); 4015 continue; 4016 } 4017 if (oldpwd->pwd_cdir == olddp) { 4018 vrefact(newdp); 4019 newpwd->pwd_cdir = newdp; 4020 } 4021 if (oldpwd->pwd_rdir == olddp) { 4022 vrefact(newdp); 4023 newpwd->pwd_rdir = newdp; 4024 } 4025 if (oldpwd->pwd_jdir == olddp) { 4026 vrefact(newdp); 4027 newpwd->pwd_jdir = newdp; 4028 } 4029 pwd_fill(oldpwd, newpwd); 4030 pwd_set(pdp, newpwd); 4031 PWDDESC_XUNLOCK(pdp); 4032 pwd_drop(oldpwd); 4033 pddrop(pdp); 4034 newpwd = pwd_alloc(); 4035 } 4036 sx_sunlock(&allproc_lock); 4037 pwd_drop(newpwd); 4038 if (rootvnode == olddp) { 4039 vrefact(newdp); 4040 rootvnode = newdp; 4041 nrele++; 4042 } 4043 mtx_lock(&prison0.pr_mtx); 4044 if (prison0.pr_root == olddp) { 4045 vrefact(newdp); 4046 prison0.pr_root = newdp; 4047 nrele++; 4048 } 4049 mtx_unlock(&prison0.pr_mtx); 4050 sx_slock(&allprison_lock); 4051 TAILQ_FOREACH(pr, &allprison, pr_list) { 4052 mtx_lock(&pr->pr_mtx); 4053 if (pr->pr_root == olddp) { 4054 vrefact(newdp); 4055 pr->pr_root = newdp; 4056 nrele++; 4057 } 4058 mtx_unlock(&pr->pr_mtx); 4059 } 4060 sx_sunlock(&allprison_lock); 4061 while (nrele--) 4062 vrele(olddp); 4063 } 4064 4065 struct filedesc_to_leader * 4066 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 4067 { 4068 struct filedesc_to_leader *fdtol; 4069 4070 fdtol = malloc(sizeof(struct filedesc_to_leader), 4071 M_FILEDESC_TO_LEADER, M_WAITOK); 4072 fdtol->fdl_refcount = 1; 4073 fdtol->fdl_holdcount = 0; 4074 fdtol->fdl_wakeup = 0; 4075 fdtol->fdl_leader = leader; 4076 if (old != NULL) { 4077 FILEDESC_XLOCK(fdp); 4078 fdtol->fdl_next = old->fdl_next; 4079 fdtol->fdl_prev = old; 4080 old->fdl_next = fdtol; 4081 fdtol->fdl_next->fdl_prev = fdtol; 4082 FILEDESC_XUNLOCK(fdp); 4083 } else { 4084 fdtol->fdl_next = fdtol; 4085 fdtol->fdl_prev = fdtol; 4086 } 4087 return (fdtol); 4088 } 4089 4090 static int 4091 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4092 { 4093 NDSLOTTYPE *map; 4094 struct filedesc *fdp; 4095 u_int namelen; 4096 int count, off, minoff; 4097 4098 namelen = arg2; 4099 if (namelen != 1) 4100 return (EINVAL); 4101 4102 if (*(int *)arg1 != 0) 4103 return (EINVAL); 4104 4105 fdp = curproc->p_fd; 4106 count = 0; 4107 FILEDESC_SLOCK(fdp); 4108 map = fdp->fd_map; 4109 off = NDSLOT(fdp->fd_nfiles - 1); 4110 for (minoff = NDSLOT(0); off >= minoff; --off) 4111 count += bitcountl(map[off]); 4112 FILEDESC_SUNLOCK(fdp); 4113 4114 return (SYSCTL_OUT(req, &count, sizeof(count))); 4115 } 4116 4117 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4118 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4119 "Number of open file descriptors"); 4120 4121 /* 4122 * Get file structures globally. 4123 */ 4124 static int 4125 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4126 { 4127 struct xfile xf; 4128 struct filedesc *fdp; 4129 struct file *fp; 4130 struct proc *p; 4131 int error, n, lastfile; 4132 4133 error = sysctl_wire_old_buffer(req, 0); 4134 if (error != 0) 4135 return (error); 4136 if (req->oldptr == NULL) { 4137 n = 0; 4138 sx_slock(&allproc_lock); 4139 FOREACH_PROC_IN_SYSTEM(p) { 4140 PROC_LOCK(p); 4141 if (p->p_state == PRS_NEW) { 4142 PROC_UNLOCK(p); 4143 continue; 4144 } 4145 fdp = fdhold(p); 4146 PROC_UNLOCK(p); 4147 if (fdp == NULL) 4148 continue; 4149 /* overestimates sparse tables. */ 4150 n += fdp->fd_nfiles; 4151 fddrop(fdp); 4152 } 4153 sx_sunlock(&allproc_lock); 4154 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4155 } 4156 error = 0; 4157 bzero(&xf, sizeof(xf)); 4158 xf.xf_size = sizeof(xf); 4159 sx_slock(&allproc_lock); 4160 FOREACH_PROC_IN_SYSTEM(p) { 4161 PROC_LOCK(p); 4162 if (p->p_state == PRS_NEW) { 4163 PROC_UNLOCK(p); 4164 continue; 4165 } 4166 if (p_cansee(req->td, p) != 0) { 4167 PROC_UNLOCK(p); 4168 continue; 4169 } 4170 xf.xf_pid = p->p_pid; 4171 xf.xf_uid = p->p_ucred->cr_uid; 4172 fdp = fdhold(p); 4173 PROC_UNLOCK(p); 4174 if (fdp == NULL) 4175 continue; 4176 FILEDESC_SLOCK(fdp); 4177 lastfile = fdlastfile(fdp); 4178 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile; 4179 n++) { 4180 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 4181 continue; 4182 xf.xf_fd = n; 4183 xf.xf_file = (uintptr_t)fp; 4184 xf.xf_data = (uintptr_t)fp->f_data; 4185 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4186 xf.xf_type = (uintptr_t)fp->f_type; 4187 xf.xf_count = refcount_load(&fp->f_count); 4188 xf.xf_msgcount = 0; 4189 xf.xf_offset = foffset_get(fp); 4190 xf.xf_flag = fp->f_flag; 4191 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4192 if (error) 4193 break; 4194 } 4195 FILEDESC_SUNLOCK(fdp); 4196 fddrop(fdp); 4197 if (error) 4198 break; 4199 } 4200 sx_sunlock(&allproc_lock); 4201 return (error); 4202 } 4203 4204 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4205 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4206 4207 #ifdef KINFO_FILE_SIZE 4208 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4209 #endif 4210 4211 static int 4212 xlate_fflags(int fflags) 4213 { 4214 static const struct { 4215 int fflag; 4216 int kf_fflag; 4217 } fflags_table[] = { 4218 { FAPPEND, KF_FLAG_APPEND }, 4219 { FASYNC, KF_FLAG_ASYNC }, 4220 { FFSYNC, KF_FLAG_FSYNC }, 4221 { FHASLOCK, KF_FLAG_HASLOCK }, 4222 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4223 { FREAD, KF_FLAG_READ }, 4224 { FWRITE, KF_FLAG_WRITE }, 4225 { O_CREAT, KF_FLAG_CREAT }, 4226 { O_DIRECT, KF_FLAG_DIRECT }, 4227 { O_EXCL, KF_FLAG_EXCL }, 4228 { O_EXEC, KF_FLAG_EXEC }, 4229 { O_EXLOCK, KF_FLAG_EXLOCK }, 4230 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4231 { O_SHLOCK, KF_FLAG_SHLOCK }, 4232 { O_TRUNC, KF_FLAG_TRUNC } 4233 }; 4234 unsigned int i; 4235 int kflags; 4236 4237 kflags = 0; 4238 for (i = 0; i < nitems(fflags_table); i++) 4239 if (fflags & fflags_table[i].fflag) 4240 kflags |= fflags_table[i].kf_fflag; 4241 return (kflags); 4242 } 4243 4244 /* Trim unused data from kf_path by truncating the structure size. */ 4245 void 4246 pack_kinfo(struct kinfo_file *kif) 4247 { 4248 4249 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4250 strlen(kif->kf_path) + 1; 4251 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4252 } 4253 4254 static void 4255 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4256 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4257 { 4258 int error; 4259 4260 bzero(kif, sizeof(*kif)); 4261 4262 /* Set a default type to allow for empty fill_kinfo() methods. */ 4263 kif->kf_type = KF_TYPE_UNKNOWN; 4264 kif->kf_flags = xlate_fflags(fp->f_flag); 4265 if (rightsp != NULL) 4266 kif->kf_cap_rights = *rightsp; 4267 else 4268 cap_rights_init_zero(&kif->kf_cap_rights); 4269 kif->kf_fd = fd; 4270 kif->kf_ref_count = refcount_load(&fp->f_count); 4271 kif->kf_offset = foffset_get(fp); 4272 4273 /* 4274 * This may drop the filedesc lock, so the 'fp' cannot be 4275 * accessed after this call. 4276 */ 4277 error = fo_fill_kinfo(fp, kif, fdp); 4278 if (error == 0) 4279 kif->kf_status |= KF_ATTR_VALID; 4280 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4281 pack_kinfo(kif); 4282 else 4283 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4284 } 4285 4286 static void 4287 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4288 struct kinfo_file *kif, int flags) 4289 { 4290 int error; 4291 4292 bzero(kif, sizeof(*kif)); 4293 4294 kif->kf_type = KF_TYPE_VNODE; 4295 error = vn_fill_kinfo_vnode(vp, kif); 4296 if (error == 0) 4297 kif->kf_status |= KF_ATTR_VALID; 4298 kif->kf_flags = xlate_fflags(fflags); 4299 cap_rights_init_zero(&kif->kf_cap_rights); 4300 kif->kf_fd = fd; 4301 kif->kf_ref_count = -1; 4302 kif->kf_offset = -1; 4303 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4304 pack_kinfo(kif); 4305 else 4306 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4307 vrele(vp); 4308 } 4309 4310 struct export_fd_buf { 4311 struct filedesc *fdp; 4312 struct pwddesc *pdp; 4313 struct sbuf *sb; 4314 ssize_t remainder; 4315 struct kinfo_file kif; 4316 int flags; 4317 }; 4318 4319 static int 4320 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4321 { 4322 struct kinfo_file *kif; 4323 4324 kif = &efbuf->kif; 4325 if (efbuf->remainder != -1) { 4326 if (efbuf->remainder < kif->kf_structsize) { 4327 /* Terminate export. */ 4328 efbuf->remainder = 0; 4329 return (0); 4330 } 4331 efbuf->remainder -= kif->kf_structsize; 4332 } 4333 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 4334 } 4335 4336 static int 4337 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4338 struct export_fd_buf *efbuf) 4339 { 4340 int error; 4341 4342 if (efbuf->remainder == 0) 4343 return (0); 4344 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4345 efbuf->flags); 4346 FILEDESC_SUNLOCK(efbuf->fdp); 4347 error = export_kinfo_to_sb(efbuf); 4348 FILEDESC_SLOCK(efbuf->fdp); 4349 return (error); 4350 } 4351 4352 static int 4353 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4354 struct export_fd_buf *efbuf) 4355 { 4356 int error; 4357 4358 if (efbuf->remainder == 0) 4359 return (0); 4360 if (efbuf->pdp != NULL) 4361 PWDDESC_XUNLOCK(efbuf->pdp); 4362 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4363 error = export_kinfo_to_sb(efbuf); 4364 if (efbuf->pdp != NULL) 4365 PWDDESC_XLOCK(efbuf->pdp); 4366 return (error); 4367 } 4368 4369 /* 4370 * Store a process file descriptor information to sbuf. 4371 * 4372 * Takes a locked proc as argument, and returns with the proc unlocked. 4373 */ 4374 int 4375 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4376 int flags) 4377 { 4378 struct file *fp; 4379 struct filedesc *fdp; 4380 struct pwddesc *pdp; 4381 struct export_fd_buf *efbuf; 4382 struct vnode *cttyvp, *textvp, *tracevp; 4383 struct pwd *pwd; 4384 int error, i, lastfile; 4385 cap_rights_t rights; 4386 4387 PROC_LOCK_ASSERT(p, MA_OWNED); 4388 4389 /* ktrace vnode */ 4390 tracevp = ktr_get_tracevp(p, true); 4391 /* text vnode */ 4392 textvp = p->p_textvp; 4393 if (textvp != NULL) 4394 vrefact(textvp); 4395 /* Controlling tty. */ 4396 cttyvp = NULL; 4397 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 4398 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 4399 if (cttyvp != NULL) 4400 vrefact(cttyvp); 4401 } 4402 fdp = fdhold(p); 4403 pdp = pdhold(p); 4404 PROC_UNLOCK(p); 4405 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4406 efbuf->fdp = NULL; 4407 efbuf->pdp = NULL; 4408 efbuf->sb = sb; 4409 efbuf->remainder = maxlen; 4410 efbuf->flags = flags; 4411 if (tracevp != NULL) 4412 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 4413 efbuf); 4414 if (textvp != NULL) 4415 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 4416 if (cttyvp != NULL) 4417 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 4418 efbuf); 4419 error = 0; 4420 if (pdp == NULL || fdp == NULL) 4421 goto fail; 4422 efbuf->fdp = fdp; 4423 efbuf->pdp = pdp; 4424 PWDDESC_XLOCK(pdp); 4425 pwd = pwd_hold_pwddesc(pdp); 4426 if (pwd != NULL) { 4427 /* working directory */ 4428 if (pwd->pwd_cdir != NULL) { 4429 vrefact(pwd->pwd_cdir); 4430 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, 4431 FREAD, efbuf); 4432 } 4433 /* root directory */ 4434 if (pwd->pwd_rdir != NULL) { 4435 vrefact(pwd->pwd_rdir); 4436 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, 4437 FREAD, efbuf); 4438 } 4439 /* jail directory */ 4440 if (pwd->pwd_jdir != NULL) { 4441 vrefact(pwd->pwd_jdir); 4442 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, 4443 FREAD, efbuf); 4444 } 4445 } 4446 PWDDESC_XUNLOCK(pdp); 4447 if (pwd != NULL) 4448 pwd_drop(pwd); 4449 FILEDESC_SLOCK(fdp); 4450 lastfile = fdlastfile(fdp); 4451 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) { 4452 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 4453 continue; 4454 #ifdef CAPABILITIES 4455 rights = *cap_rights(fdp, i); 4456 #else /* !CAPABILITIES */ 4457 rights = cap_no_rights; 4458 #endif 4459 /* 4460 * Create sysctl entry. It is OK to drop the filedesc 4461 * lock inside of export_file_to_sb() as we will 4462 * re-validate and re-evaluate its properties when the 4463 * loop continues. 4464 */ 4465 error = export_file_to_sb(fp, i, &rights, efbuf); 4466 if (error != 0 || efbuf->remainder == 0) 4467 break; 4468 } 4469 FILEDESC_SUNLOCK(fdp); 4470 fail: 4471 if (fdp != NULL) 4472 fddrop(fdp); 4473 if (pdp != NULL) 4474 pddrop(pdp); 4475 free(efbuf, M_TEMP); 4476 return (error); 4477 } 4478 4479 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 4480 4481 /* 4482 * Get per-process file descriptors for use by procstat(1), et al. 4483 */ 4484 static int 4485 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 4486 { 4487 struct sbuf sb; 4488 struct proc *p; 4489 ssize_t maxlen; 4490 u_int namelen; 4491 int error, error2, *name; 4492 4493 namelen = arg2; 4494 if (namelen != 1) 4495 return (EINVAL); 4496 4497 name = (int *)arg1; 4498 4499 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 4500 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4501 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4502 if (error != 0) { 4503 sbuf_delete(&sb); 4504 return (error); 4505 } 4506 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4507 error = kern_proc_filedesc_out(p, &sb, maxlen, 4508 KERN_FILEDESC_PACK_KINFO); 4509 error2 = sbuf_finish(&sb); 4510 sbuf_delete(&sb); 4511 return (error != 0 ? error : error2); 4512 } 4513 4514 #ifdef COMPAT_FREEBSD7 4515 #ifdef KINFO_OFILE_SIZE 4516 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 4517 #endif 4518 4519 static void 4520 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 4521 { 4522 4523 okif->kf_structsize = sizeof(*okif); 4524 okif->kf_type = kif->kf_type; 4525 okif->kf_fd = kif->kf_fd; 4526 okif->kf_ref_count = kif->kf_ref_count; 4527 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 4528 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 4529 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 4530 okif->kf_offset = kif->kf_offset; 4531 if (kif->kf_type == KF_TYPE_VNODE) 4532 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 4533 else 4534 okif->kf_vnode_type = KF_VTYPE_VNON; 4535 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 4536 if (kif->kf_type == KF_TYPE_SOCKET) { 4537 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 4538 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 4539 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 4540 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 4541 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 4542 } else { 4543 okif->kf_sa_local.ss_family = AF_UNSPEC; 4544 okif->kf_sa_peer.ss_family = AF_UNSPEC; 4545 } 4546 } 4547 4548 static int 4549 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 4550 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 4551 { 4552 int error; 4553 4554 vrefact(vp); 4555 PWDDESC_XUNLOCK(pdp); 4556 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 4557 kinfo_to_okinfo(kif, okif); 4558 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4559 PWDDESC_XLOCK(pdp); 4560 return (error); 4561 } 4562 4563 /* 4564 * Get per-process file descriptors for use by procstat(1), et al. 4565 */ 4566 static int 4567 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 4568 { 4569 struct kinfo_ofile *okif; 4570 struct kinfo_file *kif; 4571 struct filedesc *fdp; 4572 struct pwddesc *pdp; 4573 struct pwd *pwd; 4574 u_int namelen; 4575 int error, i, lastfile, *name; 4576 struct file *fp; 4577 struct proc *p; 4578 4579 namelen = arg2; 4580 if (namelen != 1) 4581 return (EINVAL); 4582 4583 name = (int *)arg1; 4584 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4585 if (error != 0) 4586 return (error); 4587 fdp = fdhold(p); 4588 if (fdp != NULL) 4589 pdp = pdhold(p); 4590 PROC_UNLOCK(p); 4591 if (fdp == NULL || pdp == NULL) { 4592 if (fdp != NULL) 4593 fddrop(fdp); 4594 return (ENOENT); 4595 } 4596 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 4597 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 4598 PWDDESC_XLOCK(pdp); 4599 pwd = pwd_hold_pwddesc(pdp); 4600 if (pwd != NULL) { 4601 if (pwd->pwd_cdir != NULL) 4602 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 4603 okif, pdp, req); 4604 if (pwd->pwd_rdir != NULL) 4605 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 4606 okif, pdp, req); 4607 if (pwd->pwd_jdir != NULL) 4608 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 4609 okif, pdp, req); 4610 } 4611 PWDDESC_XUNLOCK(pdp); 4612 if (pwd != NULL) 4613 pwd_drop(pwd); 4614 FILEDESC_SLOCK(fdp); 4615 lastfile = fdlastfile(fdp); 4616 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) { 4617 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 4618 continue; 4619 export_file_to_kinfo(fp, i, NULL, kif, fdp, 4620 KERN_FILEDESC_PACK_KINFO); 4621 FILEDESC_SUNLOCK(fdp); 4622 kinfo_to_okinfo(kif, okif); 4623 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4624 FILEDESC_SLOCK(fdp); 4625 if (error) 4626 break; 4627 } 4628 FILEDESC_SUNLOCK(fdp); 4629 fddrop(fdp); 4630 pddrop(pdp); 4631 free(kif, M_TEMP); 4632 free(okif, M_TEMP); 4633 return (0); 4634 } 4635 4636 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 4637 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 4638 "Process ofiledesc entries"); 4639 #endif /* COMPAT_FREEBSD7 */ 4640 4641 int 4642 vntype_to_kinfo(int vtype) 4643 { 4644 struct { 4645 int vtype; 4646 int kf_vtype; 4647 } vtypes_table[] = { 4648 { VBAD, KF_VTYPE_VBAD }, 4649 { VBLK, KF_VTYPE_VBLK }, 4650 { VCHR, KF_VTYPE_VCHR }, 4651 { VDIR, KF_VTYPE_VDIR }, 4652 { VFIFO, KF_VTYPE_VFIFO }, 4653 { VLNK, KF_VTYPE_VLNK }, 4654 { VNON, KF_VTYPE_VNON }, 4655 { VREG, KF_VTYPE_VREG }, 4656 { VSOCK, KF_VTYPE_VSOCK } 4657 }; 4658 unsigned int i; 4659 4660 /* 4661 * Perform vtype translation. 4662 */ 4663 for (i = 0; i < nitems(vtypes_table); i++) 4664 if (vtypes_table[i].vtype == vtype) 4665 return (vtypes_table[i].kf_vtype); 4666 4667 return (KF_VTYPE_UNKNOWN); 4668 } 4669 4670 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 4671 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 4672 "Process filedesc entries"); 4673 4674 /* 4675 * Store a process current working directory information to sbuf. 4676 * 4677 * Takes a locked proc as argument, and returns with the proc unlocked. 4678 */ 4679 int 4680 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 4681 { 4682 struct pwddesc *pdp; 4683 struct pwd *pwd; 4684 struct export_fd_buf *efbuf; 4685 struct vnode *cdir; 4686 int error; 4687 4688 PROC_LOCK_ASSERT(p, MA_OWNED); 4689 4690 pdp = pdhold(p); 4691 PROC_UNLOCK(p); 4692 if (pdp == NULL) 4693 return (EINVAL); 4694 4695 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4696 efbuf->pdp = pdp; 4697 efbuf->sb = sb; 4698 efbuf->remainder = maxlen; 4699 4700 PWDDESC_XLOCK(pdp); 4701 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4702 cdir = pwd->pwd_cdir; 4703 if (cdir == NULL) { 4704 error = EINVAL; 4705 } else { 4706 vrefact(cdir); 4707 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 4708 } 4709 PWDDESC_XUNLOCK(pdp); 4710 pddrop(pdp); 4711 free(efbuf, M_TEMP); 4712 return (error); 4713 } 4714 4715 /* 4716 * Get per-process current working directory. 4717 */ 4718 static int 4719 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 4720 { 4721 struct sbuf sb; 4722 struct proc *p; 4723 ssize_t maxlen; 4724 u_int namelen; 4725 int error, error2, *name; 4726 4727 namelen = arg2; 4728 if (namelen != 1) 4729 return (EINVAL); 4730 4731 name = (int *)arg1; 4732 4733 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 4734 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4735 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4736 if (error != 0) { 4737 sbuf_delete(&sb); 4738 return (error); 4739 } 4740 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4741 error = kern_proc_cwd_out(p, &sb, maxlen); 4742 error2 = sbuf_finish(&sb); 4743 sbuf_delete(&sb); 4744 return (error != 0 ? error : error2); 4745 } 4746 4747 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 4748 sysctl_kern_proc_cwd, "Process current working directory"); 4749 4750 #ifdef DDB 4751 /* 4752 * For the purposes of debugging, generate a human-readable string for the 4753 * file type. 4754 */ 4755 static const char * 4756 file_type_to_name(short type) 4757 { 4758 4759 switch (type) { 4760 case 0: 4761 return ("zero"); 4762 case DTYPE_VNODE: 4763 return ("vnode"); 4764 case DTYPE_SOCKET: 4765 return ("socket"); 4766 case DTYPE_PIPE: 4767 return ("pipe"); 4768 case DTYPE_FIFO: 4769 return ("fifo"); 4770 case DTYPE_KQUEUE: 4771 return ("kqueue"); 4772 case DTYPE_CRYPTO: 4773 return ("crypto"); 4774 case DTYPE_MQUEUE: 4775 return ("mqueue"); 4776 case DTYPE_SHM: 4777 return ("shm"); 4778 case DTYPE_SEM: 4779 return ("ksem"); 4780 case DTYPE_PTS: 4781 return ("pts"); 4782 case DTYPE_DEV: 4783 return ("dev"); 4784 case DTYPE_PROCDESC: 4785 return ("proc"); 4786 case DTYPE_EVENTFD: 4787 return ("eventfd"); 4788 case DTYPE_LINUXTFD: 4789 return ("ltimer"); 4790 default: 4791 return ("unkn"); 4792 } 4793 } 4794 4795 /* 4796 * For the purposes of debugging, identify a process (if any, perhaps one of 4797 * many) that references the passed file in its file descriptor array. Return 4798 * NULL if none. 4799 */ 4800 static struct proc * 4801 file_to_first_proc(struct file *fp) 4802 { 4803 struct filedesc *fdp; 4804 struct proc *p; 4805 int n; 4806 4807 FOREACH_PROC_IN_SYSTEM(p) { 4808 if (p->p_state == PRS_NEW) 4809 continue; 4810 fdp = p->p_fd; 4811 if (fdp == NULL) 4812 continue; 4813 for (n = 0; n < fdp->fd_nfiles; n++) { 4814 if (fp == fdp->fd_ofiles[n].fde_file) 4815 return (p); 4816 } 4817 } 4818 return (NULL); 4819 } 4820 4821 static void 4822 db_print_file(struct file *fp, int header) 4823 { 4824 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 4825 struct proc *p; 4826 4827 if (header) 4828 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 4829 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 4830 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 4831 "FCmd"); 4832 p = file_to_first_proc(fp); 4833 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 4834 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 4835 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 4836 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 4837 4838 #undef XPTRWIDTH 4839 } 4840 4841 DB_SHOW_COMMAND(file, db_show_file) 4842 { 4843 struct file *fp; 4844 4845 if (!have_addr) { 4846 db_printf("usage: show file <addr>\n"); 4847 return; 4848 } 4849 fp = (struct file *)addr; 4850 db_print_file(fp, 1); 4851 } 4852 4853 DB_SHOW_COMMAND(files, db_show_files) 4854 { 4855 struct filedesc *fdp; 4856 struct file *fp; 4857 struct proc *p; 4858 int header; 4859 int n; 4860 4861 header = 1; 4862 FOREACH_PROC_IN_SYSTEM(p) { 4863 if (p->p_state == PRS_NEW) 4864 continue; 4865 if ((fdp = p->p_fd) == NULL) 4866 continue; 4867 for (n = 0; n < fdp->fd_nfiles; ++n) { 4868 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 4869 continue; 4870 db_print_file(fp, header); 4871 header = 0; 4872 } 4873 } 4874 } 4875 #endif 4876 4877 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 4878 &maxfilesperproc, 0, "Maximum files allowed open per process"); 4879 4880 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 4881 &maxfiles, 0, "Maximum number of files"); 4882 4883 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 4884 &openfiles, 0, "System-wide number of open files"); 4885 4886 /* ARGSUSED*/ 4887 static void 4888 filelistinit(void *dummy) 4889 { 4890 4891 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 4892 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 4893 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 4894 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 4895 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 4896 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 4897 /* 4898 * XXXMJG this is a temporary hack due to boot ordering issues against 4899 * the vnode zone. 4900 */ 4901 vfs_smr = uma_zone_get_smr(pwd_zone); 4902 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 4903 } 4904 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 4905 4906 /*-------------------------------------------------------------------*/ 4907 4908 static int 4909 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 4910 int flags, struct thread *td) 4911 { 4912 4913 return (EBADF); 4914 } 4915 4916 static int 4917 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4918 struct thread *td) 4919 { 4920 4921 return (EINVAL); 4922 } 4923 4924 static int 4925 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 4926 struct thread *td) 4927 { 4928 4929 return (EBADF); 4930 } 4931 4932 static int 4933 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 4934 struct thread *td) 4935 { 4936 4937 return (0); 4938 } 4939 4940 static int 4941 badfo_kqfilter(struct file *fp, struct knote *kn) 4942 { 4943 4944 return (EBADF); 4945 } 4946 4947 static int 4948 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 4949 struct thread *td) 4950 { 4951 4952 return (EBADF); 4953 } 4954 4955 static int 4956 badfo_close(struct file *fp, struct thread *td) 4957 { 4958 4959 return (0); 4960 } 4961 4962 static int 4963 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4964 struct thread *td) 4965 { 4966 4967 return (EBADF); 4968 } 4969 4970 static int 4971 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4972 struct thread *td) 4973 { 4974 4975 return (EBADF); 4976 } 4977 4978 static int 4979 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4980 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4981 struct thread *td) 4982 { 4983 4984 return (EBADF); 4985 } 4986 4987 static int 4988 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 4989 { 4990 4991 return (0); 4992 } 4993 4994 struct fileops badfileops = { 4995 .fo_read = badfo_readwrite, 4996 .fo_write = badfo_readwrite, 4997 .fo_truncate = badfo_truncate, 4998 .fo_ioctl = badfo_ioctl, 4999 .fo_poll = badfo_poll, 5000 .fo_kqfilter = badfo_kqfilter, 5001 .fo_stat = badfo_stat, 5002 .fo_close = badfo_close, 5003 .fo_chmod = badfo_chmod, 5004 .fo_chown = badfo_chown, 5005 .fo_sendfile = badfo_sendfile, 5006 .fo_fill_kinfo = badfo_fill_kinfo, 5007 }; 5008 5009 static int 5010 path_poll(struct file *fp, int events, struct ucred *active_cred, 5011 struct thread *td) 5012 { 5013 return (POLLNVAL); 5014 } 5015 5016 static int 5017 path_close(struct file *fp, struct thread *td) 5018 { 5019 MPASS(fp->f_type == DTYPE_VNODE); 5020 fp->f_ops = &badfileops; 5021 vdrop(fp->f_vnode); 5022 return (0); 5023 } 5024 5025 struct fileops path_fileops = { 5026 .fo_read = badfo_readwrite, 5027 .fo_write = badfo_readwrite, 5028 .fo_truncate = badfo_truncate, 5029 .fo_ioctl = badfo_ioctl, 5030 .fo_poll = path_poll, 5031 .fo_kqfilter = vn_kqfilter_opath, 5032 .fo_stat = vn_statfile, 5033 .fo_close = path_close, 5034 .fo_chmod = badfo_chmod, 5035 .fo_chown = badfo_chown, 5036 .fo_sendfile = badfo_sendfile, 5037 .fo_fill_kinfo = vn_fill_kinfo, 5038 .fo_flags = DFLAG_PASSABLE, 5039 }; 5040 5041 int 5042 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5043 int flags, struct thread *td) 5044 { 5045 5046 return (EOPNOTSUPP); 5047 } 5048 5049 int 5050 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5051 struct thread *td) 5052 { 5053 5054 return (EINVAL); 5055 } 5056 5057 int 5058 invfo_ioctl(struct file *fp, u_long com, void *data, 5059 struct ucred *active_cred, struct thread *td) 5060 { 5061 5062 return (ENOTTY); 5063 } 5064 5065 int 5066 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5067 struct thread *td) 5068 { 5069 5070 return (poll_no_poll(events)); 5071 } 5072 5073 int 5074 invfo_kqfilter(struct file *fp, struct knote *kn) 5075 { 5076 5077 return (EINVAL); 5078 } 5079 5080 int 5081 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5082 struct thread *td) 5083 { 5084 5085 return (EINVAL); 5086 } 5087 5088 int 5089 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5090 struct thread *td) 5091 { 5092 5093 return (EINVAL); 5094 } 5095 5096 int 5097 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5098 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5099 struct thread *td) 5100 { 5101 5102 return (EINVAL); 5103 } 5104 5105 /*-------------------------------------------------------------------*/ 5106 5107 /* 5108 * File Descriptor pseudo-device driver (/dev/fd/). 5109 * 5110 * Opening minor device N dup()s the file (if any) connected to file 5111 * descriptor N belonging to the calling process. Note that this driver 5112 * consists of only the ``open()'' routine, because all subsequent 5113 * references to this file will be direct to the other driver. 5114 * 5115 * XXX: we could give this one a cloning event handler if necessary. 5116 */ 5117 5118 /* ARGSUSED */ 5119 static int 5120 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5121 { 5122 5123 /* 5124 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5125 * the file descriptor being sought for duplication. The error 5126 * return ensures that the vnode for this device will be released 5127 * by vn_open. Open will detect this special error and take the 5128 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5129 * will simply report the error. 5130 */ 5131 td->td_dupfd = dev2unit(dev); 5132 return (ENODEV); 5133 } 5134 5135 static struct cdevsw fildesc_cdevsw = { 5136 .d_version = D_VERSION, 5137 .d_open = fdopen, 5138 .d_name = "FD", 5139 }; 5140 5141 static void 5142 fildesc_drvinit(void *unused) 5143 { 5144 struct cdev *dev; 5145 5146 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5147 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5148 make_dev_alias(dev, "stdin"); 5149 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5150 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5151 make_dev_alias(dev, "stdout"); 5152 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5153 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5154 make_dev_alias(dev, "stderr"); 5155 } 5156 5157 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5158