1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37 #include "opt_capsicum.h" 38 #include "opt_ddb.h" 39 #include "opt_ktrace.h" 40 41 #define EXTERR_CATEGORY EXTERR_CAT_FILEDESC 42 #include <sys/systm.h> 43 #include <sys/capsicum.h> 44 #include <sys/conf.h> 45 #include <sys/exterrvar.h> 46 #include <sys/fcntl.h> 47 #include <sys/file.h> 48 #include <sys/filedesc.h> 49 #include <sys/filio.h> 50 #include <sys/jail.h> 51 #include <sys/kernel.h> 52 #include <sys/limits.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mount.h> 56 #include <sys/mutex.h> 57 #include <sys/namei.h> 58 #include <sys/selinfo.h> 59 #include <sys/poll.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/protosw.h> 63 #include <sys/racct.h> 64 #include <sys/resourcevar.h> 65 #include <sys/sbuf.h> 66 #include <sys/signalvar.h> 67 #include <sys/kdb.h> 68 #include <sys/smr.h> 69 #include <sys/stat.h> 70 #include <sys/sx.h> 71 #include <sys/syscallsubr.h> 72 #include <sys/sysctl.h> 73 #include <sys/sysproto.h> 74 #include <sys/unistd.h> 75 #include <sys/user.h> 76 #include <sys/vnode.h> 77 #include <sys/ktrace.h> 78 79 #include <net/vnet.h> 80 81 #include <security/audit/audit.h> 82 83 #include <vm/uma.h> 84 #include <vm/vm.h> 85 86 #include <ddb/ddb.h> 87 88 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 89 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes"); 90 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors"); 91 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 92 "file desc to leader structures"); 93 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 94 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 95 96 MALLOC_DECLARE(M_FADVISE); 97 98 static __read_mostly uma_zone_t file_zone; 99 static __read_mostly uma_zone_t filedesc0_zone; 100 __read_mostly uma_zone_t pwd_zone; 101 VFS_SMR_DECLARE; 102 103 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 104 struct thread *td, bool holdleaders, bool audit); 105 static void export_file_to_kinfo(struct file *fp, int fd, 106 cap_rights_t *rightsp, struct kinfo_file *kif, 107 struct filedesc *fdp, int flags); 108 static int fd_first_free(struct filedesc *fdp, int low, int size); 109 static void fdgrowtable(struct filedesc *fdp, int nfd); 110 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 111 static void fdunused(struct filedesc *fdp, int fd); 112 static void fdused(struct filedesc *fdp, int fd); 113 static int fget_unlocked_seq(struct thread *td, int fd, 114 const cap_rights_t *needrightsp, uint8_t *flagsp, 115 struct file **fpp, seqc_t *seqp); 116 static int getmaxfd(struct thread *td); 117 static u_long *filecaps_copy_prep(const struct filecaps *src); 118 static void filecaps_copy_finish(const struct filecaps *src, 119 struct filecaps *dst, u_long *ioctls); 120 static u_long *filecaps_free_prep(struct filecaps *fcaps); 121 static void filecaps_free_finish(u_long *ioctls); 122 123 static struct pwd *pwd_alloc(void); 124 125 /* 126 * Each process has: 127 * 128 * - An array of open file descriptors (fd_ofiles) 129 * - An array of file flags (fd_ofileflags) 130 * - A bitmap recording which descriptors are in use (fd_map) 131 * 132 * A process starts out with NDFILE descriptors. The value of NDFILE has 133 * been selected based the historical limit of 20 open files, and an 134 * assumption that the majority of processes, especially short-lived 135 * processes like shells, will never need more. 136 * 137 * If this initial allocation is exhausted, a larger descriptor table and 138 * map are allocated dynamically, and the pointers in the process's struct 139 * filedesc are updated to point to those. This is repeated every time 140 * the process runs out of file descriptors (provided it hasn't hit its 141 * resource limit). 142 * 143 * Since threads may hold references to individual descriptor table 144 * entries, the tables are never freed. Instead, they are placed on a 145 * linked list and freed only when the struct filedesc is released. 146 */ 147 #define NDFILE 20 148 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 149 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 150 #define NDSLOT(x) ((x) / NDENTRIES) 151 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 152 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 153 154 #define FILEDESC_FOREACH_FDE(fdp, _iterator, _fde) \ 155 struct filedesc *_fdp = (fdp); \ 156 int _lastfile = fdlastfile_single(_fdp); \ 157 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 158 if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL) 159 160 #define FILEDESC_FOREACH_FP(fdp, _iterator, _fp) \ 161 struct filedesc *_fdp = (fdp); \ 162 int _lastfile = fdlastfile_single(_fdp); \ 163 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 164 if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL) 165 166 /* 167 * SLIST entry used to keep track of ofiles which must be reclaimed when 168 * the process exits. 169 */ 170 struct freetable { 171 struct fdescenttbl *ft_table; 172 SLIST_ENTRY(freetable) ft_next; 173 }; 174 175 /* 176 * Initial allocation: a filedesc structure + the head of SLIST used to 177 * keep track of old ofiles + enough space for NDFILE descriptors. 178 */ 179 180 struct fdescenttbl0 { 181 int fdt_nfiles; 182 struct filedescent fdt_ofiles[NDFILE]; 183 }; 184 185 struct filedesc0 { 186 struct filedesc fd_fd; 187 SLIST_HEAD(, freetable) fd_free; 188 struct fdescenttbl0 fd_dfiles; 189 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 190 }; 191 192 /* 193 * Descriptor management. 194 */ 195 static int __exclusive_cache_line openfiles; /* actual number of open files */ 196 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 197 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 198 199 /* 200 * If low >= size, just return low. Otherwise find the first zero bit in the 201 * given bitmap, starting at low and not exceeding size - 1. Return size if 202 * not found. 203 */ 204 static int 205 fd_first_free(struct filedesc *fdp, int low, int size) 206 { 207 NDSLOTTYPE *map = fdp->fd_map; 208 NDSLOTTYPE mask; 209 int off, maxoff; 210 211 if (low >= size) 212 return (low); 213 214 off = NDSLOT(low); 215 if (low % NDENTRIES) { 216 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 217 if ((mask &= ~map[off]) != 0UL) 218 return (off * NDENTRIES + ffsl(mask) - 1); 219 ++off; 220 } 221 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 222 if (map[off] != ~0UL) 223 return (off * NDENTRIES + ffsl(~map[off]) - 1); 224 return (size); 225 } 226 227 /* 228 * Find the last used fd. 229 * 230 * Call this variant if fdp can't be modified by anyone else (e.g, during exec). 231 * Otherwise use fdlastfile. 232 */ 233 int 234 fdlastfile_single(struct filedesc *fdp) 235 { 236 NDSLOTTYPE *map = fdp->fd_map; 237 int off, minoff; 238 239 off = NDSLOT(fdp->fd_nfiles - 1); 240 for (minoff = NDSLOT(0); off >= minoff; --off) 241 if (map[off] != 0) 242 return (off * NDENTRIES + flsl(map[off]) - 1); 243 return (-1); 244 } 245 246 int 247 fdlastfile(struct filedesc *fdp) 248 { 249 250 FILEDESC_LOCK_ASSERT(fdp); 251 return (fdlastfile_single(fdp)); 252 } 253 254 static int 255 fdisused(struct filedesc *fdp, int fd) 256 { 257 258 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 259 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 260 261 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 262 } 263 264 /* 265 * Mark a file descriptor as used. 266 */ 267 static void 268 fdused_init(struct filedesc *fdp, int fd) 269 { 270 271 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 272 273 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 274 } 275 276 static void 277 fdused(struct filedesc *fdp, int fd) 278 { 279 280 FILEDESC_XLOCK_ASSERT(fdp); 281 282 fdused_init(fdp, fd); 283 if (fd == fdp->fd_freefile) 284 fdp->fd_freefile++; 285 } 286 287 /* 288 * Mark a file descriptor as unused. 289 */ 290 static void 291 fdunused(struct filedesc *fdp, int fd) 292 { 293 294 FILEDESC_XLOCK_ASSERT(fdp); 295 296 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 297 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 298 ("fd=%d is still in use", fd)); 299 300 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 301 if (fd < fdp->fd_freefile) 302 fdp->fd_freefile = fd; 303 } 304 305 /* 306 * Free a file descriptor. 307 * 308 * Avoid some work if fdp is about to be destroyed. 309 */ 310 static inline void 311 fdefree_last(struct filedescent *fde) 312 { 313 314 filecaps_free(&fde->fde_caps); 315 } 316 317 static inline void 318 fdfree(struct filedesc *fdp, int fd) 319 { 320 struct filedescent *fde; 321 322 FILEDESC_XLOCK_ASSERT(fdp); 323 fde = &fdp->fd_ofiles[fd]; 324 #ifdef CAPABILITIES 325 seqc_write_begin(&fde->fde_seqc); 326 #endif 327 fde->fde_file = NULL; 328 #ifdef CAPABILITIES 329 seqc_write_end(&fde->fde_seqc); 330 #endif 331 fdefree_last(fde); 332 fdunused(fdp, fd); 333 } 334 335 /* 336 * System calls on descriptors. 337 */ 338 #ifndef _SYS_SYSPROTO_H_ 339 struct getdtablesize_args { 340 int dummy; 341 }; 342 #endif 343 /* ARGSUSED */ 344 int 345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 346 { 347 #ifdef RACCT 348 uint64_t lim; 349 #endif 350 351 td->td_retval[0] = getmaxfd(td); 352 #ifdef RACCT 353 PROC_LOCK(td->td_proc); 354 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 355 PROC_UNLOCK(td->td_proc); 356 if (lim < td->td_retval[0]) 357 td->td_retval[0] = lim; 358 #endif 359 return (0); 360 } 361 362 /* 363 * Duplicate a file descriptor to a particular value. 364 * 365 * Note: keep in mind that a potential race condition exists when closing 366 * descriptors from a shared descriptor table (via rfork). 367 */ 368 #ifndef _SYS_SYSPROTO_H_ 369 struct dup2_args { 370 u_int from; 371 u_int to; 372 }; 373 #endif 374 /* ARGSUSED */ 375 int 376 sys_dup2(struct thread *td, struct dup2_args *uap) 377 { 378 379 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 380 } 381 382 /* 383 * Duplicate a file descriptor. 384 */ 385 #ifndef _SYS_SYSPROTO_H_ 386 struct dup_args { 387 u_int fd; 388 }; 389 #endif 390 /* ARGSUSED */ 391 int 392 sys_dup(struct thread *td, struct dup_args *uap) 393 { 394 395 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 396 } 397 398 /* 399 * The file control system call. 400 */ 401 #ifndef _SYS_SYSPROTO_H_ 402 struct fcntl_args { 403 int fd; 404 int cmd; 405 long arg; 406 }; 407 #endif 408 /* ARGSUSED */ 409 int 410 sys_fcntl(struct thread *td, struct fcntl_args *uap) 411 { 412 413 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 414 } 415 416 int 417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, intptr_t arg) 418 { 419 struct flock fl; 420 struct __oflock ofl; 421 intptr_t arg1; 422 int error, newcmd; 423 424 error = 0; 425 newcmd = cmd; 426 switch (cmd) { 427 case F_OGETLK: 428 case F_OSETLK: 429 case F_OSETLKW: 430 /* 431 * Convert old flock structure to new. 432 */ 433 error = copyin((void *)arg, &ofl, sizeof(ofl)); 434 fl.l_start = ofl.l_start; 435 fl.l_len = ofl.l_len; 436 fl.l_pid = ofl.l_pid; 437 fl.l_type = ofl.l_type; 438 fl.l_whence = ofl.l_whence; 439 fl.l_sysid = 0; 440 441 switch (cmd) { 442 case F_OGETLK: 443 newcmd = F_GETLK; 444 break; 445 case F_OSETLK: 446 newcmd = F_SETLK; 447 break; 448 case F_OSETLKW: 449 newcmd = F_SETLKW; 450 break; 451 } 452 arg1 = (intptr_t)&fl; 453 break; 454 case F_GETLK: 455 case F_SETLK: 456 case F_SETLKW: 457 case F_SETLK_REMOTE: 458 error = copyin((void *)arg, &fl, sizeof(fl)); 459 arg1 = (intptr_t)&fl; 460 break; 461 default: 462 arg1 = arg; 463 break; 464 } 465 if (error) 466 return (error); 467 error = kern_fcntl(td, fd, newcmd, arg1); 468 if (error) 469 return (error); 470 if (cmd == F_OGETLK) { 471 ofl.l_start = fl.l_start; 472 ofl.l_len = fl.l_len; 473 ofl.l_pid = fl.l_pid; 474 ofl.l_type = fl.l_type; 475 ofl.l_whence = fl.l_whence; 476 error = copyout(&ofl, (void *)arg, sizeof(ofl)); 477 } else if (cmd == F_GETLK) { 478 error = copyout(&fl, (void *)arg, sizeof(fl)); 479 } 480 return (error); 481 } 482 483 struct flags_trans_elem { 484 u_int f; 485 u_int t; 486 }; 487 488 static u_int 489 flags_trans(const struct flags_trans_elem *ftes, int nitems, u_int from_flags) 490 { 491 u_int res; 492 int i; 493 494 res = 0; 495 for (i = 0; i < nitems; i++) { 496 if ((from_flags & ftes[i].f) != 0) 497 res |= ftes[i].t; 498 } 499 return (res); 500 } 501 502 static uint8_t 503 fd_to_fde_flags(int fd_flags) 504 { 505 static const struct flags_trans_elem fd_to_fde_flags_s[] = { 506 { .f = FD_CLOEXEC, .t = UF_EXCLOSE }, 507 { .f = FD_CLOFORK, .t = UF_FOCLOSE }, 508 { .f = FD_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH }, 509 }; 510 511 return (flags_trans(fd_to_fde_flags_s, nitems(fd_to_fde_flags_s), 512 fd_flags)); 513 } 514 515 static int 516 fde_to_fd_flags(uint8_t fde_flags) 517 { 518 static const struct flags_trans_elem fde_to_fd_flags_s[] = { 519 { .f = UF_EXCLOSE, .t = FD_CLOEXEC }, 520 { .f = UF_FOCLOSE, .t = FD_CLOFORK }, 521 { .f = UF_RESOLVE_BENEATH, .t = FD_RESOLVE_BENEATH }, 522 }; 523 524 return (flags_trans(fde_to_fd_flags_s, nitems(fde_to_fd_flags_s), 525 fde_flags)); 526 } 527 528 static uint8_t 529 fddup_to_fde_flags(int fddup_flags) 530 { 531 static const struct flags_trans_elem fddup_to_fde_flags_s[] = { 532 { .f = FDDUP_FLAG_CLOEXEC, .t = UF_EXCLOSE }, 533 { .f = FDDUP_FLAG_CLOFORK, .t = UF_FOCLOSE }, 534 }; 535 536 return (flags_trans(fddup_to_fde_flags_s, nitems(fddup_to_fde_flags_s), 537 fddup_flags)); 538 } 539 540 static uint8_t 541 close_range_to_fde_flags(int close_range_flags) 542 { 543 static const struct flags_trans_elem close_range_to_fde_flags_s[] = { 544 { .f = CLOSE_RANGE_CLOEXEC, .t = UF_EXCLOSE }, 545 { .f = CLOSE_RANGE_CLOFORK, .t = UF_FOCLOSE }, 546 }; 547 548 return (flags_trans(close_range_to_fde_flags_s, 549 nitems(close_range_to_fde_flags_s), close_range_flags)); 550 } 551 552 static uint8_t 553 open_to_fde_flags(int open_flags, bool sticky_orb) 554 { 555 static const struct flags_trans_elem open_to_fde_flags_s[] = { 556 { .f = O_CLOEXEC, .t = UF_EXCLOSE }, 557 { .f = O_CLOFORK, .t = UF_FOCLOSE }, 558 { .f = O_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH }, 559 }; 560 #if defined(__clang__) && __clang_major__ >= 19 561 _Static_assert(open_to_fde_flags_s[nitems(open_to_fde_flags_s) - 1].f == 562 O_RESOLVE_BENEATH, "O_RESOLVE_BENEATH must be last, for sticky_orb"); 563 #endif 564 565 return (flags_trans(open_to_fde_flags_s, nitems(open_to_fde_flags_s) - 566 (sticky_orb ? 0 : 1), open_flags)); 567 } 568 569 int 570 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 571 { 572 struct filedesc *fdp; 573 struct flock *flp; 574 struct file *fp, *fp2; 575 struct filedescent *fde; 576 struct proc *p; 577 struct vnode *vp; 578 struct mount *mp; 579 struct kinfo_file *kif; 580 int error, flg, kif_sz, seals, tmp, got_set, got_cleared; 581 uint64_t bsize; 582 off_t foffset; 583 int flags; 584 585 error = 0; 586 flg = F_POSIX; 587 p = td->td_proc; 588 fdp = p->p_fd; 589 590 AUDIT_ARG_FD(cmd); 591 AUDIT_ARG_CMD(cmd); 592 switch (cmd) { 593 case F_DUPFD: 594 tmp = arg; 595 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 596 break; 597 598 case F_DUPFD_CLOEXEC: 599 tmp = arg; 600 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 601 break; 602 603 case F_DUPFD_CLOFORK: 604 tmp = arg; 605 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOFORK, fd, tmp); 606 break; 607 608 case F_DUP2FD: 609 tmp = arg; 610 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 611 break; 612 613 case F_DUP2FD_CLOEXEC: 614 tmp = arg; 615 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 616 break; 617 618 case F_GETFD: 619 error = EBADF; 620 FILEDESC_SLOCK(fdp); 621 fde = fdeget_noref(fdp, fd); 622 if (fde != NULL) { 623 td->td_retval[0] = fde_to_fd_flags(fde->fde_flags); 624 error = 0; 625 } 626 FILEDESC_SUNLOCK(fdp); 627 break; 628 629 case F_SETFD: 630 error = EBADF; 631 FILEDESC_XLOCK(fdp); 632 fde = fdeget_noref(fdp, fd); 633 if (fde != NULL) { 634 /* 635 * UF_RESOLVE_BENEATH is sticky and cannot be cleared. 636 */ 637 fde->fde_flags = (fde->fde_flags & 638 ~(UF_EXCLOSE | UF_FOCLOSE)) | fd_to_fde_flags(arg); 639 error = 0; 640 } 641 FILEDESC_XUNLOCK(fdp); 642 break; 643 644 case F_GETFL: 645 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); 646 if (error != 0) 647 break; 648 td->td_retval[0] = OFLAGS(fp->f_flag); 649 fdrop(fp, td); 650 break; 651 652 case F_SETFL: 653 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); 654 if (error != 0) 655 break; 656 if (fp->f_ops == &path_fileops) { 657 fdrop(fp, td); 658 error = EBADF; 659 break; 660 } 661 fsetfl_lock(fp); 662 do { 663 tmp = flg = fp->f_flag; 664 tmp &= ~FCNTLFLAGS; 665 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 666 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 667 got_set = tmp & ~flg; 668 got_cleared = flg & ~tmp; 669 if (((got_set | got_cleared) & FNONBLOCK) != 0) { 670 tmp = fp->f_flag & FNONBLOCK; 671 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 672 if (error != 0) 673 goto revert_flags; 674 } 675 if (((got_set | got_cleared) & FASYNC) != 0) { 676 tmp = fp->f_flag & FASYNC; 677 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 678 if (error != 0) 679 goto revert_nonblock; 680 } 681 fsetfl_unlock(fp); 682 fdrop(fp, td); 683 break; 684 revert_nonblock: 685 if (((got_set | got_cleared) & FNONBLOCK) != 0) { 686 tmp = ~fp->f_flag & FNONBLOCK; 687 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 688 } 689 revert_flags: 690 do { 691 tmp = flg = fp->f_flag; 692 tmp &= ~FCNTLFLAGS; 693 tmp |= got_cleared; 694 tmp &= ~got_set; 695 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 696 fsetfl_unlock(fp); 697 fdrop(fp, td); 698 break; 699 700 case F_GETOWN: 701 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 702 if (error != 0) 703 break; 704 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 705 if (error == 0) 706 td->td_retval[0] = tmp; 707 fdrop(fp, td); 708 break; 709 710 case F_SETOWN: 711 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 712 if (error != 0) 713 break; 714 tmp = arg; 715 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 716 fdrop(fp, td); 717 break; 718 719 case F_SETLK_REMOTE: 720 error = priv_check(td, PRIV_NFS_LOCKD); 721 if (error != 0) 722 return (error); 723 flg = F_REMOTE; 724 goto do_setlk; 725 726 case F_SETLKW: 727 flg |= F_WAIT; 728 /* FALLTHROUGH F_SETLK */ 729 730 case F_SETLK: 731 do_setlk: 732 flp = (struct flock *)arg; 733 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { 734 error = EINVAL; 735 break; 736 } 737 738 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 739 if (error != 0) 740 break; 741 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 742 error = EBADF; 743 fdrop(fp, td); 744 break; 745 } 746 747 if (flp->l_whence == SEEK_CUR) { 748 foffset = foffset_get(fp); 749 if (foffset < 0 || 750 (flp->l_start > 0 && 751 foffset > OFF_MAX - flp->l_start)) { 752 error = EOVERFLOW; 753 fdrop(fp, td); 754 break; 755 } 756 flp->l_start += foffset; 757 } 758 759 vp = fp->f_vnode; 760 switch (flp->l_type) { 761 case F_RDLCK: 762 if ((fp->f_flag & FREAD) == 0) { 763 error = EBADF; 764 break; 765 } 766 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 767 PROC_LOCK(p->p_leader); 768 p->p_leader->p_flag |= P_ADVLOCK; 769 PROC_UNLOCK(p->p_leader); 770 } 771 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 772 flp, flg); 773 break; 774 case F_WRLCK: 775 if ((fp->f_flag & FWRITE) == 0) { 776 error = EBADF; 777 break; 778 } 779 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 780 PROC_LOCK(p->p_leader); 781 p->p_leader->p_flag |= P_ADVLOCK; 782 PROC_UNLOCK(p->p_leader); 783 } 784 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 785 flp, flg); 786 break; 787 case F_UNLCK: 788 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 789 flp, flg); 790 break; 791 case F_UNLCKSYS: 792 if (flg != F_REMOTE) { 793 error = EINVAL; 794 break; 795 } 796 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 797 F_UNLCKSYS, flp, flg); 798 break; 799 default: 800 error = EINVAL; 801 break; 802 } 803 if (error != 0 || flp->l_type == F_UNLCK || 804 flp->l_type == F_UNLCKSYS) { 805 fdrop(fp, td); 806 break; 807 } 808 809 /* 810 * Check for a race with close. 811 * 812 * The vnode is now advisory locked (or unlocked, but this case 813 * is not really important) as the caller requested. 814 * We had to drop the filedesc lock, so we need to recheck if 815 * the descriptor is still valid, because if it was closed 816 * in the meantime we need to remove advisory lock from the 817 * vnode - close on any descriptor leading to an advisory 818 * locked vnode, removes that lock. 819 * We will return 0 on purpose in that case, as the result of 820 * successful advisory lock might have been externally visible 821 * already. This is fine - effectively we pretend to the caller 822 * that the closing thread was a bit slower and that the 823 * advisory lock succeeded before the close. 824 */ 825 error = fget_unlocked(td, fd, &cap_no_rights, &fp2); 826 if (error != 0) { 827 fdrop(fp, td); 828 break; 829 } 830 if (fp != fp2) { 831 flp->l_whence = SEEK_SET; 832 flp->l_start = 0; 833 flp->l_len = 0; 834 flp->l_type = F_UNLCK; 835 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 836 F_UNLCK, flp, F_POSIX); 837 } 838 fdrop(fp, td); 839 fdrop(fp2, td); 840 break; 841 842 case F_GETLK: 843 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 844 if (error != 0) 845 break; 846 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 847 error = EBADF; 848 fdrop(fp, td); 849 break; 850 } 851 flp = (struct flock *)arg; 852 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 853 flp->l_type != F_UNLCK) { 854 error = EINVAL; 855 fdrop(fp, td); 856 break; 857 } 858 if (flp->l_whence == SEEK_CUR) { 859 foffset = foffset_get(fp); 860 if ((flp->l_start > 0 && 861 foffset > OFF_MAX - flp->l_start) || 862 (flp->l_start < 0 && 863 foffset < OFF_MIN - flp->l_start)) { 864 error = EOVERFLOW; 865 fdrop(fp, td); 866 break; 867 } 868 flp->l_start += foffset; 869 } 870 vp = fp->f_vnode; 871 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 872 F_POSIX); 873 fdrop(fp, td); 874 break; 875 876 case F_ADD_SEALS: 877 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 878 if (error != 0) 879 break; 880 error = fo_add_seals(fp, arg); 881 fdrop(fp, td); 882 break; 883 884 case F_GET_SEALS: 885 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 886 if (error != 0) 887 break; 888 if (fo_get_seals(fp, &seals) == 0) 889 td->td_retval[0] = seals; 890 else 891 error = EINVAL; 892 fdrop(fp, td); 893 break; 894 895 case F_RDAHEAD: 896 arg = arg ? 128 * 1024: 0; 897 /* FALLTHROUGH */ 898 case F_READAHEAD: 899 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 900 if (error != 0) 901 break; 902 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 903 fdrop(fp, td); 904 error = EBADF; 905 break; 906 } 907 vp = fp->f_vnode; 908 if (vp->v_type != VREG) { 909 fdrop(fp, td); 910 error = ENOTTY; 911 break; 912 } 913 914 /* 915 * Exclusive lock synchronizes against f_seqcount reads and 916 * writes in sequential_heuristic(). 917 */ 918 error = vn_lock(vp, LK_EXCLUSIVE); 919 if (error != 0) { 920 fdrop(fp, td); 921 break; 922 } 923 if (arg >= 0) { 924 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 925 arg = MIN(arg, INT_MAX - bsize + 1); 926 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, 927 (arg + bsize - 1) / bsize); 928 atomic_set_int(&fp->f_flag, FRDAHEAD); 929 } else { 930 atomic_clear_int(&fp->f_flag, FRDAHEAD); 931 } 932 VOP_UNLOCK(vp); 933 fdrop(fp, td); 934 break; 935 936 case F_ISUNIONSTACK: 937 /* 938 * Check if the vnode is part of a union stack (either the 939 * "union" flag from mount(2) or unionfs). 940 * 941 * Prior to introduction of this op libc's readdir would call 942 * fstatfs(2), in effect unnecessarily copying kilobytes of 943 * data just to check fs name and a mount flag. 944 * 945 * Fixing the code to handle everything in the kernel instead 946 * is a non-trivial endeavor and has low priority, thus this 947 * horrible kludge facilitates the current behavior in a much 948 * cheaper manner until someone(tm) sorts this out. 949 */ 950 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 951 if (error != 0) 952 break; 953 if (fp->f_type != DTYPE_VNODE) { 954 fdrop(fp, td); 955 error = EBADF; 956 break; 957 } 958 vp = fp->f_vnode; 959 /* 960 * Since we don't prevent dooming the vnode even non-null mp 961 * found can become immediately stale. This is tolerable since 962 * mount points are type-stable (providing safe memory access) 963 * and any vfs op on this vnode going forward will return an 964 * error (meaning return value in this case is meaningless). 965 */ 966 mp = atomic_load_ptr(&vp->v_mount); 967 if (__predict_false(mp == NULL)) { 968 fdrop(fp, td); 969 error = EBADF; 970 break; 971 } 972 td->td_retval[0] = 0; 973 if (mp->mnt_kern_flag & MNTK_UNIONFS || 974 mp->mnt_flag & MNT_UNION) 975 td->td_retval[0] = 1; 976 fdrop(fp, td); 977 break; 978 979 case F_KINFO: 980 #ifdef CAPABILITY_MODE 981 if (CAP_TRACING(td)) 982 ktrcapfail(CAPFAIL_SYSCALL, &cmd); 983 if (IN_CAPABILITY_MODE(td)) { 984 error = ECAPMODE; 985 break; 986 } 987 #endif 988 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz)); 989 if (error != 0) 990 break; 991 if (kif_sz != sizeof(*kif)) { 992 error = EINVAL; 993 break; 994 } 995 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO); 996 FILEDESC_SLOCK(fdp); 997 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL); 998 if (error == 0 && fhold(fp)) { 999 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0); 1000 FILEDESC_SUNLOCK(fdp); 1001 fdrop(fp, td); 1002 if ((kif->kf_status & KF_ATTR_VALID) != 0) { 1003 kif->kf_structsize = sizeof(*kif); 1004 error = copyout(kif, (void *)arg, sizeof(*kif)); 1005 } else { 1006 error = EBADF; 1007 } 1008 } else { 1009 FILEDESC_SUNLOCK(fdp); 1010 if (error == 0) 1011 error = EBADF; 1012 } 1013 free(kif, M_TEMP); 1014 break; 1015 1016 default: 1017 if ((cmd & ((1u << F_DUP3FD_SHIFT) - 1)) != F_DUP3FD) 1018 return (EXTERROR(EINVAL, "invalid fcntl cmd")); 1019 /* Handle F_DUP3FD */ 1020 flags = (cmd >> F_DUP3FD_SHIFT); 1021 if ((flags & ~(FD_CLOEXEC | FD_CLOFORK)) != 0) 1022 return (EXTERROR(EINVAL, "invalid flags for F_DUP3FD")); 1023 tmp = arg; 1024 error = kern_dup(td, FDDUP_FIXED, 1025 ((flags & FD_CLOEXEC) != 0 ? FDDUP_FLAG_CLOEXEC : 0) | 1026 ((flags & FD_CLOFORK) != 0 ? FDDUP_FLAG_CLOFORK : 0), 1027 fd, tmp); 1028 break; 1029 } 1030 return (error); 1031 } 1032 1033 static int 1034 getmaxfd(struct thread *td) 1035 { 1036 1037 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 1038 } 1039 1040 /* 1041 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 1042 */ 1043 int 1044 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 1045 { 1046 struct filedesc *fdp; 1047 struct filedescent *oldfde, *newfde; 1048 struct proc *p; 1049 struct file *delfp, *oldfp; 1050 u_long *oioctls, *nioctls; 1051 int error, maxfd; 1052 1053 p = td->td_proc; 1054 fdp = p->p_fd; 1055 oioctls = NULL; 1056 1057 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC | FDDUP_FLAG_CLOFORK)) == 0); 1058 MPASS(mode < FDDUP_LASTMODE); 1059 1060 AUDIT_ARG_FD(old); 1061 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 1062 1063 /* 1064 * Verify we have a valid descriptor to dup from and possibly to 1065 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 1066 * return EINVAL when the new descriptor is out of bounds. 1067 */ 1068 if (old < 0) 1069 return (EBADF); 1070 if (new < 0) 1071 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1072 maxfd = getmaxfd(td); 1073 if (new >= maxfd) 1074 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1075 1076 error = EBADF; 1077 FILEDESC_XLOCK(fdp); 1078 if (fget_noref(fdp, old) == NULL) 1079 goto unlock; 1080 if (mode == FDDUP_FIXED && old == new) { 1081 td->td_retval[0] = new; 1082 fdp->fd_ofiles[new].fde_flags |= fddup_to_fde_flags(flags); 1083 error = 0; 1084 goto unlock; 1085 } 1086 1087 oldfde = &fdp->fd_ofiles[old]; 1088 oldfp = oldfde->fde_file; 1089 if (!fhold(oldfp)) 1090 goto unlock; 1091 1092 /* 1093 * If the caller specified a file descriptor, make sure the file 1094 * table is large enough to hold it, and grab it. Otherwise, just 1095 * allocate a new descriptor the usual way. 1096 */ 1097 switch (mode) { 1098 case FDDUP_NORMAL: 1099 case FDDUP_FCNTL: 1100 if ((error = fdalloc(td, new, &new)) != 0) { 1101 fdrop(oldfp, td); 1102 goto unlock; 1103 } 1104 break; 1105 case FDDUP_FIXED: 1106 if (new >= fdp->fd_nfiles) { 1107 /* 1108 * The resource limits are here instead of e.g. 1109 * fdalloc(), because the file descriptor table may be 1110 * shared between processes, so we can't really use 1111 * racct_add()/racct_sub(). Instead of counting the 1112 * number of actually allocated descriptors, just put 1113 * the limit on the size of the file descriptor table. 1114 */ 1115 #ifdef RACCT 1116 if (RACCT_ENABLED()) { 1117 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); 1118 if (error != 0) { 1119 error = EMFILE; 1120 fdrop(oldfp, td); 1121 goto unlock; 1122 } 1123 } 1124 #endif 1125 fdgrowtable_exp(fdp, new + 1); 1126 } 1127 if (!fdisused(fdp, new)) 1128 fdused(fdp, new); 1129 break; 1130 default: 1131 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 1132 } 1133 1134 KASSERT(old != new, ("new fd is same as old")); 1135 1136 /* Refetch oldfde because the table may have grown and old one freed. */ 1137 oldfde = &fdp->fd_ofiles[old]; 1138 KASSERT(oldfp == oldfde->fde_file, 1139 ("fdt_ofiles shift from growth observed at fd %d", 1140 old)); 1141 1142 newfde = &fdp->fd_ofiles[new]; 1143 delfp = newfde->fde_file; 1144 1145 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 1146 1147 /* 1148 * Duplicate the source descriptor. 1149 */ 1150 #ifdef CAPABILITIES 1151 seqc_write_begin(&newfde->fde_seqc); 1152 #endif 1153 oioctls = filecaps_free_prep(&newfde->fde_caps); 1154 fde_copy(oldfde, newfde); 1155 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 1156 nioctls); 1157 newfde->fde_flags = (oldfde->fde_flags & ~(UF_EXCLOSE | UF_FOCLOSE)) | 1158 fddup_to_fde_flags(flags); 1159 #ifdef CAPABILITIES 1160 seqc_write_end(&newfde->fde_seqc); 1161 #endif 1162 td->td_retval[0] = new; 1163 1164 error = 0; 1165 1166 if (delfp != NULL) { 1167 (void) closefp(fdp, new, delfp, td, true, false); 1168 FILEDESC_UNLOCK_ASSERT(fdp); 1169 } else { 1170 unlock: 1171 FILEDESC_XUNLOCK(fdp); 1172 } 1173 1174 filecaps_free_finish(oioctls); 1175 return (error); 1176 } 1177 1178 static void 1179 sigiofree(struct sigio *sigio) 1180 { 1181 crfree(sigio->sio_ucred); 1182 free(sigio, M_SIGIO); 1183 } 1184 1185 static struct sigio * 1186 funsetown_locked(struct sigio *sigio) 1187 { 1188 struct proc *p; 1189 struct pgrp *pg; 1190 1191 SIGIO_ASSERT_LOCKED(); 1192 1193 if (sigio == NULL) 1194 return (NULL); 1195 *sigio->sio_myref = NULL; 1196 if (sigio->sio_pgid < 0) { 1197 pg = sigio->sio_pgrp; 1198 PGRP_LOCK(pg); 1199 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); 1200 PGRP_UNLOCK(pg); 1201 } else { 1202 p = sigio->sio_proc; 1203 PROC_LOCK(p); 1204 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 1205 PROC_UNLOCK(p); 1206 } 1207 return (sigio); 1208 } 1209 1210 /* 1211 * If sigio is on the list associated with a process or process group, 1212 * disable signalling from the device, remove sigio from the list and 1213 * free sigio. 1214 */ 1215 void 1216 funsetown(struct sigio **sigiop) 1217 { 1218 struct sigio *sigio; 1219 1220 /* Racy check, consumers must provide synchronization. */ 1221 if (*sigiop == NULL) 1222 return; 1223 1224 SIGIO_LOCK(); 1225 sigio = funsetown_locked(*sigiop); 1226 SIGIO_UNLOCK(); 1227 if (sigio != NULL) 1228 sigiofree(sigio); 1229 } 1230 1231 /* 1232 * Free a list of sigio structures. The caller must ensure that new sigio 1233 * structures cannot be added after this point. For process groups this is 1234 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves 1235 * as an interlock. 1236 */ 1237 void 1238 funsetownlst(struct sigiolst *sigiolst) 1239 { 1240 struct proc *p; 1241 struct pgrp *pg; 1242 struct sigio *sigio, *tmp; 1243 1244 /* Racy check. */ 1245 sigio = SLIST_FIRST(sigiolst); 1246 if (sigio == NULL) 1247 return; 1248 1249 p = NULL; 1250 pg = NULL; 1251 1252 SIGIO_LOCK(); 1253 sigio = SLIST_FIRST(sigiolst); 1254 if (sigio == NULL) { 1255 SIGIO_UNLOCK(); 1256 return; 1257 } 1258 1259 /* 1260 * Every entry of the list should belong to a single proc or pgrp. 1261 */ 1262 if (sigio->sio_pgid < 0) { 1263 pg = sigio->sio_pgrp; 1264 sx_assert(&proctree_lock, SX_XLOCKED); 1265 PGRP_LOCK(pg); 1266 } else /* if (sigio->sio_pgid > 0) */ { 1267 p = sigio->sio_proc; 1268 PROC_LOCK(p); 1269 KASSERT((p->p_flag & P_WEXIT) != 0, 1270 ("%s: process %p is not exiting", __func__, p)); 1271 } 1272 1273 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) { 1274 *sigio->sio_myref = NULL; 1275 if (pg != NULL) { 1276 KASSERT(sigio->sio_pgid < 0, 1277 ("Proc sigio in pgrp sigio list")); 1278 KASSERT(sigio->sio_pgrp == pg, 1279 ("Bogus pgrp in sigio list")); 1280 } else /* if (p != NULL) */ { 1281 KASSERT(sigio->sio_pgid > 0, 1282 ("Pgrp sigio in proc sigio list")); 1283 KASSERT(sigio->sio_proc == p, 1284 ("Bogus proc in sigio list")); 1285 } 1286 } 1287 1288 if (pg != NULL) 1289 PGRP_UNLOCK(pg); 1290 else 1291 PROC_UNLOCK(p); 1292 SIGIO_UNLOCK(); 1293 1294 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp) 1295 sigiofree(sigio); 1296 } 1297 1298 /* 1299 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1300 * 1301 * After permission checking, add a sigio structure to the sigio list for 1302 * the process or process group. 1303 */ 1304 int 1305 fsetown(pid_t pgid, struct sigio **sigiop) 1306 { 1307 struct proc *proc; 1308 struct pgrp *pgrp; 1309 struct sigio *osigio, *sigio; 1310 int ret; 1311 1312 if (pgid == 0) { 1313 funsetown(sigiop); 1314 return (0); 1315 } 1316 1317 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1318 sigio->sio_pgid = pgid; 1319 sigio->sio_ucred = crhold(curthread->td_ucred); 1320 sigio->sio_myref = sigiop; 1321 1322 ret = 0; 1323 if (pgid > 0) { 1324 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc); 1325 SIGIO_LOCK(); 1326 osigio = funsetown_locked(*sigiop); 1327 if (ret == 0) { 1328 PROC_LOCK(proc); 1329 _PRELE(proc); 1330 if ((proc->p_flag & P_WEXIT) != 0) { 1331 ret = ESRCH; 1332 } else if (proc->p_session != 1333 curthread->td_proc->p_session) { 1334 /* 1335 * Policy - Don't allow a process to FSETOWN a 1336 * process in another session. 1337 * 1338 * Remove this test to allow maximum flexibility 1339 * or restrict FSETOWN to the current process or 1340 * process group for maximum safety. 1341 */ 1342 ret = EPERM; 1343 } else { 1344 sigio->sio_proc = proc; 1345 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, 1346 sio_pgsigio); 1347 } 1348 PROC_UNLOCK(proc); 1349 } 1350 } else /* if (pgid < 0) */ { 1351 sx_slock(&proctree_lock); 1352 SIGIO_LOCK(); 1353 osigio = funsetown_locked(*sigiop); 1354 pgrp = pgfind(-pgid); 1355 if (pgrp == NULL) { 1356 ret = ESRCH; 1357 } else { 1358 if (pgrp->pg_session != curthread->td_proc->p_session) { 1359 /* 1360 * Policy - Don't allow a process to FSETOWN a 1361 * process in another session. 1362 * 1363 * Remove this test to allow maximum flexibility 1364 * or restrict FSETOWN to the current process or 1365 * process group for maximum safety. 1366 */ 1367 ret = EPERM; 1368 } else { 1369 sigio->sio_pgrp = pgrp; 1370 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, 1371 sio_pgsigio); 1372 } 1373 PGRP_UNLOCK(pgrp); 1374 } 1375 sx_sunlock(&proctree_lock); 1376 } 1377 if (ret == 0) 1378 *sigiop = sigio; 1379 SIGIO_UNLOCK(); 1380 if (osigio != NULL) 1381 sigiofree(osigio); 1382 return (ret); 1383 } 1384 1385 /* 1386 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1387 */ 1388 pid_t 1389 fgetown(struct sigio **sigiop) 1390 { 1391 pid_t pgid; 1392 1393 SIGIO_LOCK(); 1394 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1395 SIGIO_UNLOCK(); 1396 return (pgid); 1397 } 1398 1399 static int 1400 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1401 bool audit) 1402 { 1403 int error; 1404 1405 FILEDESC_XLOCK_ASSERT(fdp); 1406 1407 /* 1408 * We now hold the fp reference that used to be owned by the 1409 * descriptor array. We have to unlock the FILEDESC *AFTER* 1410 * knote_fdclose to prevent a race of the fd getting opened, a knote 1411 * added, and deleteing a knote for the new fd. 1412 */ 1413 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1414 knote_fdclose(td, fd); 1415 1416 /* 1417 * We need to notify mqueue if the object is of type mqueue. 1418 */ 1419 if (__predict_false(fp->f_type == DTYPE_MQUEUE)) 1420 mq_fdclose(td, fd, fp); 1421 FILEDESC_XUNLOCK(fdp); 1422 1423 #ifdef AUDIT 1424 if (AUDITING_TD(td) && audit) 1425 audit_sysclose(td, fd, fp); 1426 #endif 1427 error = closef(fp, td); 1428 1429 /* 1430 * All paths leading up to closefp() will have already removed or 1431 * replaced the fd in the filedesc table, so a restart would not 1432 * operate on the same file. 1433 */ 1434 if (error == ERESTART) 1435 error = EINTR; 1436 1437 return (error); 1438 } 1439 1440 static int 1441 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1442 bool holdleaders, bool audit) 1443 { 1444 int error; 1445 1446 FILEDESC_XLOCK_ASSERT(fdp); 1447 1448 if (holdleaders) { 1449 if (td->td_proc->p_fdtol != NULL) { 1450 /* 1451 * Ask fdfree() to sleep to ensure that all relevant 1452 * process leaders can be traversed in closef(). 1453 */ 1454 fdp->fd_holdleaderscount++; 1455 } else { 1456 holdleaders = false; 1457 } 1458 } 1459 1460 error = closefp_impl(fdp, fd, fp, td, audit); 1461 if (holdleaders) { 1462 FILEDESC_XLOCK(fdp); 1463 fdp->fd_holdleaderscount--; 1464 if (fdp->fd_holdleaderscount == 0 && 1465 fdp->fd_holdleaderswakeup != 0) { 1466 fdp->fd_holdleaderswakeup = 0; 1467 wakeup(&fdp->fd_holdleaderscount); 1468 } 1469 FILEDESC_XUNLOCK(fdp); 1470 } 1471 return (error); 1472 } 1473 1474 static int 1475 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1476 bool holdleaders, bool audit) 1477 { 1478 1479 FILEDESC_XLOCK_ASSERT(fdp); 1480 1481 if (__predict_false(td->td_proc->p_fdtol != NULL)) { 1482 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit)); 1483 } else { 1484 return (closefp_impl(fdp, fd, fp, td, audit)); 1485 } 1486 } 1487 1488 /* 1489 * Close a file descriptor. 1490 */ 1491 #ifndef _SYS_SYSPROTO_H_ 1492 struct close_args { 1493 int fd; 1494 }; 1495 #endif 1496 /* ARGSUSED */ 1497 int 1498 sys_close(struct thread *td, struct close_args *uap) 1499 { 1500 1501 return (kern_close(td, uap->fd)); 1502 } 1503 1504 int 1505 kern_close(struct thread *td, int fd) 1506 { 1507 struct filedesc *fdp; 1508 struct file *fp; 1509 1510 fdp = td->td_proc->p_fd; 1511 1512 FILEDESC_XLOCK(fdp); 1513 if ((fp = fget_noref(fdp, fd)) == NULL) { 1514 FILEDESC_XUNLOCK(fdp); 1515 return (EBADF); 1516 } 1517 fdfree(fdp, fd); 1518 1519 /* closefp() drops the FILEDESC lock for us. */ 1520 return (closefp(fdp, fd, fp, td, true, true)); 1521 } 1522 1523 static int 1524 close_range_flags(struct thread *td, u_int lowfd, u_int highfd, int flags) 1525 { 1526 struct filedesc *fdp; 1527 struct fdescenttbl *fdt; 1528 struct filedescent *fde; 1529 int fd, fde_flags; 1530 1531 fde_flags = close_range_to_fde_flags(flags); 1532 fdp = td->td_proc->p_fd; 1533 FILEDESC_XLOCK(fdp); 1534 fdt = atomic_load_ptr(&fdp->fd_files); 1535 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1536 fd = lowfd; 1537 if (__predict_false(fd > highfd)) { 1538 goto out_locked; 1539 } 1540 for (; fd <= highfd; fd++) { 1541 fde = &fdt->fdt_ofiles[fd]; 1542 if (fde->fde_file != NULL) 1543 fde->fde_flags |= fde_flags; 1544 } 1545 out_locked: 1546 FILEDESC_XUNLOCK(fdp); 1547 return (0); 1548 } 1549 1550 static int 1551 close_range_impl(struct thread *td, u_int lowfd, u_int highfd) 1552 { 1553 struct filedesc *fdp; 1554 const struct fdescenttbl *fdt; 1555 struct file *fp; 1556 int fd; 1557 1558 fdp = td->td_proc->p_fd; 1559 FILEDESC_XLOCK(fdp); 1560 fdt = atomic_load_ptr(&fdp->fd_files); 1561 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1562 fd = lowfd; 1563 if (__predict_false(fd > highfd)) { 1564 goto out_locked; 1565 } 1566 for (;;) { 1567 fp = fdt->fdt_ofiles[fd].fde_file; 1568 if (fp == NULL) { 1569 if (fd == highfd) 1570 goto out_locked; 1571 } else { 1572 fdfree(fdp, fd); 1573 (void) closefp(fdp, fd, fp, td, true, true); 1574 if (fd == highfd) 1575 goto out_unlocked; 1576 FILEDESC_XLOCK(fdp); 1577 fdt = atomic_load_ptr(&fdp->fd_files); 1578 } 1579 fd++; 1580 } 1581 out_locked: 1582 FILEDESC_XUNLOCK(fdp); 1583 out_unlocked: 1584 return (0); 1585 } 1586 1587 int 1588 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd) 1589 { 1590 1591 /* 1592 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 1593 * open should not be a usage error. From a close_range() perspective, 1594 * close_range(3, ~0U, 0) in the same scenario should also likely not 1595 * be a usage error as all fd above 3 are in-fact already closed. 1596 */ 1597 if (highfd < lowfd) { 1598 return (EINVAL); 1599 } 1600 1601 if ((flags & (CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0) 1602 return (close_range_flags(td, lowfd, highfd, flags)); 1603 1604 return (close_range_impl(td, lowfd, highfd)); 1605 } 1606 1607 #ifndef _SYS_SYSPROTO_H_ 1608 struct close_range_args { 1609 u_int lowfd; 1610 u_int highfd; 1611 int flags; 1612 }; 1613 #endif 1614 int 1615 sys_close_range(struct thread *td, struct close_range_args *uap) 1616 { 1617 1618 AUDIT_ARG_FD(uap->lowfd); 1619 AUDIT_ARG_CMD(uap->highfd); 1620 AUDIT_ARG_FFLAGS(uap->flags); 1621 1622 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0) 1623 return (EINVAL); 1624 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd)); 1625 } 1626 1627 #ifdef COMPAT_FREEBSD12 1628 /* 1629 * Close open file descriptors. 1630 */ 1631 #ifndef _SYS_SYSPROTO_H_ 1632 struct freebsd12_closefrom_args { 1633 int lowfd; 1634 }; 1635 #endif 1636 /* ARGSUSED */ 1637 int 1638 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) 1639 { 1640 u_int lowfd; 1641 1642 AUDIT_ARG_FD(uap->lowfd); 1643 1644 /* 1645 * Treat negative starting file descriptor values identical to 1646 * closefrom(0) which closes all files. 1647 */ 1648 lowfd = MAX(0, uap->lowfd); 1649 return (kern_close_range(td, 0, lowfd, ~0U)); 1650 } 1651 #endif /* COMPAT_FREEBSD12 */ 1652 1653 #if defined(COMPAT_43) 1654 /* 1655 * Return status information about a file descriptor. 1656 */ 1657 #ifndef _SYS_SYSPROTO_H_ 1658 struct ofstat_args { 1659 int fd; 1660 struct ostat *sb; 1661 }; 1662 #endif 1663 /* ARGSUSED */ 1664 int 1665 ofstat(struct thread *td, struct ofstat_args *uap) 1666 { 1667 struct ostat oub; 1668 struct stat ub; 1669 int error; 1670 1671 error = kern_fstat(td, uap->fd, &ub); 1672 if (error == 0) { 1673 cvtstat(&ub, &oub); 1674 error = copyout(&oub, uap->sb, sizeof(oub)); 1675 } 1676 return (error); 1677 } 1678 #endif /* COMPAT_43 */ 1679 1680 #if defined(COMPAT_FREEBSD11) 1681 int 1682 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1683 { 1684 struct stat sb; 1685 struct freebsd11_stat osb; 1686 int error; 1687 1688 error = kern_fstat(td, uap->fd, &sb); 1689 if (error != 0) 1690 return (error); 1691 error = freebsd11_cvtstat(&sb, &osb); 1692 if (error == 0) 1693 error = copyout(&osb, uap->sb, sizeof(osb)); 1694 return (error); 1695 } 1696 #endif /* COMPAT_FREEBSD11 */ 1697 1698 /* 1699 * Return status information about a file descriptor. 1700 */ 1701 #ifndef _SYS_SYSPROTO_H_ 1702 struct fstat_args { 1703 int fd; 1704 struct stat *sb; 1705 }; 1706 #endif 1707 /* ARGSUSED */ 1708 int 1709 sys_fstat(struct thread *td, struct fstat_args *uap) 1710 { 1711 struct stat ub; 1712 int error; 1713 1714 error = kern_fstat(td, uap->fd, &ub); 1715 if (error == 0) 1716 error = copyout(&ub, uap->sb, sizeof(ub)); 1717 return (error); 1718 } 1719 1720 int 1721 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1722 { 1723 struct file *fp; 1724 int error; 1725 1726 AUDIT_ARG_FD(fd); 1727 1728 error = fget(td, fd, &cap_fstat_rights, &fp); 1729 if (__predict_false(error != 0)) 1730 return (error); 1731 1732 AUDIT_ARG_FILE(td->td_proc, fp); 1733 1734 sbp->st_filerev = 0; 1735 sbp->st_bsdflags = 0; 1736 error = fo_stat(fp, sbp, td->td_ucred); 1737 fdrop(fp, td); 1738 #ifdef __STAT_TIME_T_EXT 1739 sbp->st_atim_ext = 0; 1740 sbp->st_mtim_ext = 0; 1741 sbp->st_ctim_ext = 0; 1742 sbp->st_btim_ext = 0; 1743 #endif 1744 #ifdef KTRACE 1745 if (KTRPOINT(td, KTR_STRUCT)) 1746 ktrstat_error(sbp, error); 1747 #endif 1748 return (error); 1749 } 1750 1751 #if defined(COMPAT_FREEBSD11) 1752 /* 1753 * Return status information about a file descriptor. 1754 */ 1755 #ifndef _SYS_SYSPROTO_H_ 1756 struct freebsd11_nfstat_args { 1757 int fd; 1758 struct nstat *sb; 1759 }; 1760 #endif 1761 /* ARGSUSED */ 1762 int 1763 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1764 { 1765 struct nstat nub; 1766 struct stat ub; 1767 int error; 1768 1769 error = kern_fstat(td, uap->fd, &ub); 1770 if (error != 0) 1771 return (error); 1772 error = freebsd11_cvtnstat(&ub, &nub); 1773 if (error != 0) 1774 error = copyout(&nub, uap->sb, sizeof(nub)); 1775 return (error); 1776 } 1777 #endif /* COMPAT_FREEBSD11 */ 1778 1779 /* 1780 * Return pathconf information about a file descriptor. 1781 */ 1782 #ifndef _SYS_SYSPROTO_H_ 1783 struct fpathconf_args { 1784 int fd; 1785 int name; 1786 }; 1787 #endif 1788 /* ARGSUSED */ 1789 int 1790 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1791 { 1792 long value; 1793 int error; 1794 1795 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1796 if (error == 0) 1797 td->td_retval[0] = value; 1798 return (error); 1799 } 1800 1801 int 1802 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1803 { 1804 struct file *fp; 1805 struct vnode *vp; 1806 int error; 1807 1808 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1809 if (error != 0) 1810 return (error); 1811 1812 if (name == _PC_ASYNC_IO) { 1813 *valuep = _POSIX_ASYNCHRONOUS_IO; 1814 goto out; 1815 } 1816 vp = fp->f_vnode; 1817 if (vp != NULL) { 1818 vn_lock(vp, LK_SHARED | LK_RETRY); 1819 error = VOP_PATHCONF(vp, name, valuep); 1820 VOP_UNLOCK(vp); 1821 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1822 if (name != _PC_PIPE_BUF) { 1823 error = EINVAL; 1824 } else { 1825 *valuep = PIPE_BUF; 1826 error = 0; 1827 } 1828 } else { 1829 error = EOPNOTSUPP; 1830 } 1831 out: 1832 fdrop(fp, td); 1833 return (error); 1834 } 1835 1836 /* 1837 * Copy filecaps structure allocating memory for ioctls array if needed. 1838 * 1839 * The last parameter indicates whether the fdtable is locked. If it is not and 1840 * ioctls are encountered, copying fails and the caller must lock the table. 1841 * 1842 * Note that if the table was not locked, the caller has to check the relevant 1843 * sequence counter to determine whether the operation was successful. 1844 */ 1845 bool 1846 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1847 { 1848 size_t size; 1849 1850 if (src->fc_ioctls != NULL && !locked) 1851 return (false); 1852 memcpy(dst, src, sizeof(*src)); 1853 if (src->fc_ioctls == NULL) 1854 return (true); 1855 1856 KASSERT(src->fc_nioctls > 0, 1857 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1858 1859 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1860 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1861 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1862 return (true); 1863 } 1864 1865 static u_long * 1866 filecaps_copy_prep(const struct filecaps *src) 1867 { 1868 u_long *ioctls; 1869 size_t size; 1870 1871 if (__predict_true(src->fc_ioctls == NULL)) 1872 return (NULL); 1873 1874 KASSERT(src->fc_nioctls > 0, 1875 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1876 1877 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1878 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1879 return (ioctls); 1880 } 1881 1882 static void 1883 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1884 u_long *ioctls) 1885 { 1886 size_t size; 1887 1888 *dst = *src; 1889 if (__predict_true(src->fc_ioctls == NULL)) { 1890 MPASS(ioctls == NULL); 1891 return; 1892 } 1893 1894 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1895 dst->fc_ioctls = ioctls; 1896 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1897 } 1898 1899 /* 1900 * Move filecaps structure to the new place and clear the old place. 1901 */ 1902 void 1903 filecaps_move(struct filecaps *src, struct filecaps *dst) 1904 { 1905 1906 *dst = *src; 1907 bzero(src, sizeof(*src)); 1908 } 1909 1910 /* 1911 * Fill the given filecaps structure with full rights. 1912 */ 1913 static void 1914 filecaps_fill(struct filecaps *fcaps) 1915 { 1916 1917 CAP_ALL(&fcaps->fc_rights); 1918 fcaps->fc_ioctls = NULL; 1919 fcaps->fc_nioctls = -1; 1920 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1921 } 1922 1923 /* 1924 * Free memory allocated within filecaps structure. 1925 */ 1926 static void 1927 filecaps_free_ioctl(struct filecaps *fcaps) 1928 { 1929 1930 free(fcaps->fc_ioctls, M_FILECAPS); 1931 fcaps->fc_ioctls = NULL; 1932 } 1933 1934 void 1935 filecaps_free(struct filecaps *fcaps) 1936 { 1937 1938 filecaps_free_ioctl(fcaps); 1939 bzero(fcaps, sizeof(*fcaps)); 1940 } 1941 1942 static u_long * 1943 filecaps_free_prep(struct filecaps *fcaps) 1944 { 1945 u_long *ioctls; 1946 1947 ioctls = fcaps->fc_ioctls; 1948 bzero(fcaps, sizeof(*fcaps)); 1949 return (ioctls); 1950 } 1951 1952 static void 1953 filecaps_free_finish(u_long *ioctls) 1954 { 1955 1956 free(ioctls, M_FILECAPS); 1957 } 1958 1959 /* 1960 * Validate the given filecaps structure. 1961 */ 1962 static void 1963 filecaps_validate(const struct filecaps *fcaps, const char *func) 1964 { 1965 1966 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1967 ("%s: invalid rights", func)); 1968 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1969 ("%s: invalid fcntls", func)); 1970 KASSERT(fcaps->fc_fcntls == 0 || 1971 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1972 ("%s: fcntls without CAP_FCNTL", func)); 1973 /* 1974 * open calls without WANTIOCTLCAPS free caps but leave the counter 1975 */ 1976 #if 0 1977 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1978 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1979 ("%s: invalid ioctls", func)); 1980 #endif 1981 KASSERT(fcaps->fc_nioctls == 0 || 1982 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1983 ("%s: ioctls without CAP_IOCTL", func)); 1984 } 1985 1986 static void 1987 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1988 { 1989 int nfd1; 1990 1991 FILEDESC_XLOCK_ASSERT(fdp); 1992 1993 nfd1 = fdp->fd_nfiles * 2; 1994 if (nfd1 < nfd) 1995 nfd1 = nfd; 1996 fdgrowtable(fdp, nfd1); 1997 } 1998 1999 /* 2000 * Grow the file table to accommodate (at least) nfd descriptors. 2001 */ 2002 static void 2003 fdgrowtable(struct filedesc *fdp, int nfd) 2004 { 2005 struct filedesc0 *fdp0; 2006 struct freetable *ft; 2007 struct fdescenttbl *ntable; 2008 struct fdescenttbl *otable; 2009 int nnfiles, onfiles; 2010 NDSLOTTYPE *nmap, *omap; 2011 2012 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 2013 2014 /* save old values */ 2015 onfiles = fdp->fd_nfiles; 2016 otable = fdp->fd_files; 2017 omap = fdp->fd_map; 2018 2019 /* compute the size of the new table */ 2020 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 2021 if (nnfiles <= onfiles) 2022 /* the table is already large enough */ 2023 return; 2024 2025 /* 2026 * Allocate a new table. We need enough space for the number of 2027 * entries, file entries themselves and the struct freetable we will use 2028 * when we decommission the table and place it on the freelist. 2029 * We place the struct freetable in the middle so we don't have 2030 * to worry about padding. 2031 */ 2032 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 2033 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 2034 sizeof(struct freetable), 2035 M_FILEDESC, M_ZERO | M_WAITOK); 2036 /* copy the old data */ 2037 ntable->fdt_nfiles = nnfiles; 2038 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 2039 onfiles * sizeof(ntable->fdt_ofiles[0])); 2040 2041 /* 2042 * Allocate a new map only if the old is not large enough. It will 2043 * grow at a slower rate than the table as it can map more 2044 * entries than the table can hold. 2045 */ 2046 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 2047 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 2048 M_ZERO | M_WAITOK); 2049 /* copy over the old data and update the pointer */ 2050 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 2051 fdp->fd_map = nmap; 2052 } 2053 2054 /* 2055 * Make sure that ntable is correctly initialized before we replace 2056 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 2057 * data. 2058 */ 2059 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 2060 2061 /* 2062 * Free the old file table when not shared by other threads or processes. 2063 * The old file table is considered to be shared when either are true: 2064 * - The process has more than one thread. 2065 * - The file descriptor table has been shared via fdshare(). 2066 * 2067 * When shared, the old file table will be placed on a freelist 2068 * which will be processed when the struct filedesc is released. 2069 * 2070 * Note that if onfiles == NDFILE, we're dealing with the original 2071 * static allocation contained within (struct filedesc0 *)fdp, 2072 * which must not be freed. 2073 */ 2074 if (onfiles > NDFILE) { 2075 /* 2076 * Note we may be called here from fdinit while allocating a 2077 * table for a new process in which case ->p_fd points 2078 * elsewhere. 2079 */ 2080 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 2081 free(otable, M_FILEDESC); 2082 } else { 2083 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 2084 fdp0 = (struct filedesc0 *)fdp; 2085 ft->ft_table = otable; 2086 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 2087 } 2088 } 2089 /* 2090 * The map does not have the same possibility of threads still 2091 * holding references to it. So always free it as long as it 2092 * does not reference the original static allocation. 2093 */ 2094 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 2095 free(omap, M_FILEDESC); 2096 } 2097 2098 /* 2099 * Allocate a file descriptor for the process. 2100 */ 2101 int 2102 fdalloc(struct thread *td, int minfd, int *result) 2103 { 2104 struct proc *p = td->td_proc; 2105 struct filedesc *fdp = p->p_fd; 2106 int fd, maxfd, allocfd; 2107 #ifdef RACCT 2108 int error; 2109 #endif 2110 2111 FILEDESC_XLOCK_ASSERT(fdp); 2112 2113 if (fdp->fd_freefile > minfd) 2114 minfd = fdp->fd_freefile; 2115 2116 maxfd = getmaxfd(td); 2117 2118 /* 2119 * Search the bitmap for a free descriptor starting at minfd. 2120 * If none is found, grow the file table. 2121 */ 2122 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 2123 if (__predict_false(fd >= maxfd)) 2124 return (EMFILE); 2125 if (__predict_false(fd >= fdp->fd_nfiles)) { 2126 allocfd = min(fd * 2, maxfd); 2127 #ifdef RACCT 2128 if (RACCT_ENABLED()) { 2129 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 2130 if (error != 0) 2131 return (EMFILE); 2132 } 2133 #endif 2134 /* 2135 * fd is already equal to first free descriptor >= minfd, so 2136 * we only need to grow the table and we are done. 2137 */ 2138 fdgrowtable_exp(fdp, allocfd); 2139 } 2140 2141 /* 2142 * Perform some sanity checks, then mark the file descriptor as 2143 * used and return it to the caller. 2144 */ 2145 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 2146 ("invalid descriptor %d", fd)); 2147 KASSERT(!fdisused(fdp, fd), 2148 ("fd_first_free() returned non-free descriptor")); 2149 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 2150 ("file descriptor isn't free")); 2151 fdused(fdp, fd); 2152 *result = fd; 2153 return (0); 2154 } 2155 2156 /* 2157 * Allocate n file descriptors for the process. 2158 */ 2159 int 2160 fdallocn(struct thread *td, int minfd, int *fds, int n) 2161 { 2162 struct proc *p = td->td_proc; 2163 struct filedesc *fdp = p->p_fd; 2164 int i; 2165 2166 FILEDESC_XLOCK_ASSERT(fdp); 2167 2168 for (i = 0; i < n; i++) 2169 if (fdalloc(td, 0, &fds[i]) != 0) 2170 break; 2171 2172 if (i < n) { 2173 for (i--; i >= 0; i--) 2174 fdunused(fdp, fds[i]); 2175 return (EMFILE); 2176 } 2177 2178 return (0); 2179 } 2180 2181 /* 2182 * Create a new open file structure and allocate a file descriptor for the 2183 * process that refers to it. We add one reference to the file for the 2184 * descriptor table and one reference for resultfp. This is to prevent us 2185 * being preempted and the entry in the descriptor table closed after we 2186 * release the FILEDESC lock. 2187 */ 2188 int 2189 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 2190 struct filecaps *fcaps) 2191 { 2192 struct file *fp; 2193 int error, fd; 2194 2195 MPASS(resultfp != NULL); 2196 MPASS(resultfd != NULL); 2197 2198 error = _falloc_noinstall(td, &fp, 2); 2199 if (__predict_false(error != 0)) { 2200 return (error); 2201 } 2202 2203 error = finstall_refed(td, fp, &fd, flags, fcaps); 2204 if (__predict_false(error != 0)) { 2205 falloc_abort(td, fp); 2206 return (error); 2207 } 2208 2209 *resultfp = fp; 2210 *resultfd = fd; 2211 2212 return (0); 2213 } 2214 2215 /* 2216 * Create a new open file structure without allocating a file descriptor. 2217 */ 2218 int 2219 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2220 { 2221 struct file *fp; 2222 int maxuserfiles = maxfiles - (maxfiles / 20); 2223 int openfiles_new; 2224 static struct timeval lastfail; 2225 static int curfail; 2226 2227 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2228 MPASS(n > 0); 2229 2230 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2231 if ((openfiles_new >= maxuserfiles && 2232 priv_check(td, PRIV_MAXFILES) != 0) || 2233 openfiles_new >= maxfiles) { 2234 atomic_subtract_int(&openfiles, 1); 2235 if (ppsratecheck(&lastfail, &curfail, 1)) { 2236 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2237 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2238 } 2239 return (ENFILE); 2240 } 2241 fp = uma_zalloc(file_zone, M_WAITOK); 2242 bzero(fp, sizeof(*fp)); 2243 refcount_init(&fp->f_count, n); 2244 fp->f_cred = crhold(td->td_ucred); 2245 fp->f_ops = &badfileops; 2246 *resultfp = fp; 2247 return (0); 2248 } 2249 2250 void 2251 falloc_abort(struct thread *td, struct file *fp) 2252 { 2253 2254 /* 2255 * For assertion purposes. 2256 */ 2257 refcount_init(&fp->f_count, 0); 2258 _fdrop(fp, td); 2259 } 2260 2261 /* 2262 * Install a file in a file descriptor table. 2263 */ 2264 void 2265 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2266 struct filecaps *fcaps) 2267 { 2268 struct filedescent *fde; 2269 2270 MPASS(fp != NULL); 2271 if (fcaps != NULL) 2272 filecaps_validate(fcaps, __func__); 2273 FILEDESC_XLOCK_ASSERT(fdp); 2274 2275 fde = &fdp->fd_ofiles[fd]; 2276 #ifdef CAPABILITIES 2277 seqc_write_begin(&fde->fde_seqc); 2278 #endif 2279 fde->fde_file = fp; 2280 fde->fde_flags = open_to_fde_flags(flags, true); 2281 if (fcaps != NULL) 2282 filecaps_move(fcaps, &fde->fde_caps); 2283 else 2284 filecaps_fill(&fde->fde_caps); 2285 #ifdef CAPABILITIES 2286 seqc_write_end(&fde->fde_seqc); 2287 #endif 2288 } 2289 2290 int 2291 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2292 struct filecaps *fcaps) 2293 { 2294 struct filedesc *fdp = td->td_proc->p_fd; 2295 int error; 2296 2297 MPASS(fd != NULL); 2298 2299 FILEDESC_XLOCK(fdp); 2300 error = fdalloc(td, 0, fd); 2301 if (__predict_true(error == 0)) { 2302 _finstall(fdp, fp, *fd, flags, fcaps); 2303 } 2304 FILEDESC_XUNLOCK(fdp); 2305 return (error); 2306 } 2307 2308 int 2309 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2310 struct filecaps *fcaps) 2311 { 2312 int error; 2313 2314 MPASS(fd != NULL); 2315 2316 if (!fhold(fp)) 2317 return (EBADF); 2318 error = finstall_refed(td, fp, fd, flags, fcaps); 2319 if (__predict_false(error != 0)) { 2320 fdrop(fp, td); 2321 } 2322 return (error); 2323 } 2324 2325 /* 2326 * Build a new filedesc structure from another. 2327 * 2328 * If fdp is not NULL, return with it shared locked. 2329 */ 2330 struct filedesc * 2331 fdinit(void) 2332 { 2333 struct filedesc0 *newfdp0; 2334 struct filedesc *newfdp; 2335 2336 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2337 newfdp = &newfdp0->fd_fd; 2338 2339 /* Create the file descriptor table. */ 2340 FILEDESC_LOCK_INIT(newfdp); 2341 refcount_init(&newfdp->fd_refcnt, 1); 2342 refcount_init(&newfdp->fd_holdcnt, 1); 2343 newfdp->fd_map = newfdp0->fd_dmap; 2344 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2345 newfdp->fd_files->fdt_nfiles = NDFILE; 2346 2347 return (newfdp); 2348 } 2349 2350 /* 2351 * Build a pwddesc structure from another. 2352 * Copy the current, root, and jail root vnode references. 2353 * 2354 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked. 2355 */ 2356 struct pwddesc * 2357 pdinit(struct pwddesc *pdp, bool keeplock) 2358 { 2359 struct pwddesc *newpdp; 2360 struct pwd *newpwd; 2361 2362 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2363 2364 PWDDESC_LOCK_INIT(newpdp); 2365 refcount_init(&newpdp->pd_refcount, 1); 2366 newpdp->pd_cmask = CMASK; 2367 2368 if (pdp == NULL) { 2369 newpwd = pwd_alloc(); 2370 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2371 return (newpdp); 2372 } 2373 2374 PWDDESC_XLOCK(pdp); 2375 newpwd = pwd_hold_pwddesc(pdp); 2376 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2377 if (!keeplock) 2378 PWDDESC_XUNLOCK(pdp); 2379 return (newpdp); 2380 } 2381 2382 /* 2383 * Hold either filedesc or pwddesc of the passed process. 2384 * 2385 * The process lock is used to synchronize against the target exiting and 2386 * freeing the data. 2387 * 2388 * Clearing can be ilustrated in 3 steps: 2389 * 1. set the pointer to NULL. Either routine can race against it, hence 2390 * atomic_load_ptr. 2391 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2392 * race to either still see the pointer or find NULL. It is still safe to 2393 * grab a reference as clearing is stalled. 2394 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2395 * guaranteed to see NULL, making it safe to finish clearing 2396 */ 2397 static struct filedesc * 2398 fdhold(struct proc *p) 2399 { 2400 struct filedesc *fdp; 2401 2402 PROC_LOCK_ASSERT(p, MA_OWNED); 2403 fdp = atomic_load_ptr(&p->p_fd); 2404 if (fdp != NULL) 2405 refcount_acquire(&fdp->fd_holdcnt); 2406 return (fdp); 2407 } 2408 2409 static struct pwddesc * 2410 pdhold(struct proc *p) 2411 { 2412 struct pwddesc *pdp; 2413 2414 PROC_LOCK_ASSERT(p, MA_OWNED); 2415 pdp = atomic_load_ptr(&p->p_pd); 2416 if (pdp != NULL) 2417 refcount_acquire(&pdp->pd_refcount); 2418 return (pdp); 2419 } 2420 2421 static void 2422 fddrop(struct filedesc *fdp) 2423 { 2424 2425 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2426 if (refcount_release(&fdp->fd_holdcnt) == 0) 2427 return; 2428 } 2429 2430 FILEDESC_LOCK_DESTROY(fdp); 2431 uma_zfree(filedesc0_zone, fdp); 2432 } 2433 2434 static void 2435 pddrop(struct pwddesc *pdp) 2436 { 2437 struct pwd *pwd; 2438 2439 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2440 return; 2441 2442 PWDDESC_XLOCK(pdp); 2443 if (refcount_release(&pdp->pd_refcount) == 0) { 2444 PWDDESC_XUNLOCK(pdp); 2445 return; 2446 } 2447 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2448 pwd_set(pdp, NULL); 2449 PWDDESC_XUNLOCK(pdp); 2450 pwd_drop(pwd); 2451 2452 PWDDESC_LOCK_DESTROY(pdp); 2453 free(pdp, M_PWDDESC); 2454 } 2455 2456 /* 2457 * Share a filedesc structure. 2458 */ 2459 struct filedesc * 2460 fdshare(struct filedesc *fdp) 2461 { 2462 2463 refcount_acquire(&fdp->fd_refcnt); 2464 return (fdp); 2465 } 2466 2467 /* 2468 * Share a pwddesc structure. 2469 */ 2470 struct pwddesc * 2471 pdshare(struct pwddesc *pdp) 2472 { 2473 refcount_acquire(&pdp->pd_refcount); 2474 return (pdp); 2475 } 2476 2477 /* 2478 * Unshare a filedesc structure, if necessary by making a copy 2479 */ 2480 void 2481 fdunshare(struct thread *td) 2482 { 2483 struct filedesc *tmp; 2484 struct proc *p = td->td_proc; 2485 2486 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2487 return; 2488 2489 tmp = fdcopy(p->p_fd); 2490 fdescfree(td); 2491 p->p_fd = tmp; 2492 } 2493 2494 /* 2495 * Unshare a pwddesc structure. 2496 */ 2497 void 2498 pdunshare(struct thread *td) 2499 { 2500 struct pwddesc *pdp; 2501 struct proc *p; 2502 2503 p = td->td_proc; 2504 /* Not shared. */ 2505 if (refcount_load(&p->p_pd->pd_refcount) == 1) 2506 return; 2507 2508 pdp = pdcopy(p->p_pd); 2509 pdescfree(td); 2510 p->p_pd = pdp; 2511 } 2512 2513 /* 2514 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2515 * this is to ease callers, not catch errors. 2516 */ 2517 struct filedesc * 2518 fdcopy(struct filedesc *fdp) 2519 { 2520 struct filedesc *newfdp; 2521 struct filedescent *nfde, *ofde; 2522 int i, lastfile; 2523 2524 MPASS(fdp != NULL); 2525 2526 newfdp = fdinit(); 2527 FILEDESC_SLOCK(fdp); 2528 for (;;) { 2529 lastfile = fdlastfile(fdp); 2530 if (lastfile < newfdp->fd_nfiles) 2531 break; 2532 FILEDESC_SUNLOCK(fdp); 2533 fdgrowtable(newfdp, lastfile + 1); 2534 FILEDESC_SLOCK(fdp); 2535 } 2536 /* copy all passable descriptors (i.e. not kqueue) */ 2537 newfdp->fd_freefile = fdp->fd_freefile; 2538 FILEDESC_FOREACH_FDE(fdp, i, ofde) { 2539 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 || 2540 (ofde->fde_flags & UF_FOCLOSE) != 0 || 2541 !fhold(ofde->fde_file)) { 2542 if (newfdp->fd_freefile == fdp->fd_freefile) 2543 newfdp->fd_freefile = i; 2544 continue; 2545 } 2546 nfde = &newfdp->fd_ofiles[i]; 2547 *nfde = *ofde; 2548 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2549 fdused_init(newfdp, i); 2550 } 2551 MPASS(newfdp->fd_freefile != -1); 2552 FILEDESC_SUNLOCK(fdp); 2553 return (newfdp); 2554 } 2555 2556 /* 2557 * Copy a pwddesc structure. 2558 */ 2559 struct pwddesc * 2560 pdcopy(struct pwddesc *pdp) 2561 { 2562 struct pwddesc *newpdp; 2563 2564 MPASS(pdp != NULL); 2565 2566 newpdp = pdinit(pdp, true); 2567 newpdp->pd_cmask = pdp->pd_cmask; 2568 PWDDESC_XUNLOCK(pdp); 2569 return (newpdp); 2570 } 2571 2572 /* 2573 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2574 * one of processes using it exits) and the table used to be shared. 2575 */ 2576 static void 2577 fdclearlocks(struct thread *td) 2578 { 2579 struct filedesc *fdp; 2580 struct filedesc_to_leader *fdtol; 2581 struct flock lf; 2582 struct file *fp; 2583 struct proc *p; 2584 struct vnode *vp; 2585 int i; 2586 2587 p = td->td_proc; 2588 fdp = p->p_fd; 2589 fdtol = p->p_fdtol; 2590 MPASS(fdtol != NULL); 2591 2592 FILEDESC_XLOCK(fdp); 2593 KASSERT(fdtol->fdl_refcount > 0, 2594 ("filedesc_to_refcount botch: fdl_refcount=%d", 2595 fdtol->fdl_refcount)); 2596 if (fdtol->fdl_refcount == 1 && 2597 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2598 FILEDESC_FOREACH_FP(fdp, i, fp) { 2599 if (fp->f_type != DTYPE_VNODE || 2600 !fhold(fp)) 2601 continue; 2602 FILEDESC_XUNLOCK(fdp); 2603 lf.l_whence = SEEK_SET; 2604 lf.l_start = 0; 2605 lf.l_len = 0; 2606 lf.l_type = F_UNLCK; 2607 vp = fp->f_vnode; 2608 (void) VOP_ADVLOCK(vp, 2609 (caddr_t)p->p_leader, F_UNLCK, 2610 &lf, F_POSIX); 2611 FILEDESC_XLOCK(fdp); 2612 fdrop(fp, td); 2613 } 2614 } 2615 retry: 2616 if (fdtol->fdl_refcount == 1) { 2617 if (fdp->fd_holdleaderscount > 0 && 2618 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2619 /* 2620 * close() or kern_dup() has cleared a reference 2621 * in a shared file descriptor table. 2622 */ 2623 fdp->fd_holdleaderswakeup = 1; 2624 sx_sleep(&fdp->fd_holdleaderscount, 2625 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2626 goto retry; 2627 } 2628 if (fdtol->fdl_holdcount > 0) { 2629 /* 2630 * Ensure that fdtol->fdl_leader remains 2631 * valid in closef(). 2632 */ 2633 fdtol->fdl_wakeup = 1; 2634 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2635 "fdlhold", 0); 2636 goto retry; 2637 } 2638 } 2639 fdtol->fdl_refcount--; 2640 if (fdtol->fdl_refcount == 0 && 2641 fdtol->fdl_holdcount == 0) { 2642 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2643 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2644 } else 2645 fdtol = NULL; 2646 p->p_fdtol = NULL; 2647 FILEDESC_XUNLOCK(fdp); 2648 if (fdtol != NULL) 2649 free(fdtol, M_FILEDESC_TO_LEADER); 2650 } 2651 2652 /* 2653 * Release a filedesc structure. 2654 */ 2655 static void 2656 fdescfree_fds(struct thread *td, struct filedesc *fdp) 2657 { 2658 struct filedesc0 *fdp0; 2659 struct freetable *ft, *tft; 2660 struct filedescent *fde; 2661 struct file *fp; 2662 int i; 2663 2664 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2665 ("%s: fd table %p carries references", __func__, fdp)); 2666 2667 /* 2668 * Serialize with threads iterating over the table, if any. 2669 */ 2670 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2671 FILEDESC_XLOCK(fdp); 2672 FILEDESC_XUNLOCK(fdp); 2673 } 2674 2675 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2676 fp = fde->fde_file; 2677 fdefree_last(fde); 2678 (void) closef(fp, td); 2679 } 2680 2681 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2682 free(fdp->fd_map, M_FILEDESC); 2683 if (fdp->fd_nfiles > NDFILE) 2684 free(fdp->fd_files, M_FILEDESC); 2685 2686 fdp0 = (struct filedesc0 *)fdp; 2687 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2688 free(ft->ft_table, M_FILEDESC); 2689 2690 fddrop(fdp); 2691 } 2692 2693 void 2694 fdescfree(struct thread *td) 2695 { 2696 struct proc *p; 2697 struct filedesc *fdp; 2698 2699 p = td->td_proc; 2700 fdp = p->p_fd; 2701 MPASS(fdp != NULL); 2702 2703 #ifdef RACCT 2704 if (RACCT_ENABLED()) 2705 racct_set_unlocked(p, RACCT_NOFILE, 0); 2706 #endif 2707 2708 if (p->p_fdtol != NULL) 2709 fdclearlocks(td); 2710 2711 /* 2712 * Check fdhold for an explanation. 2713 */ 2714 atomic_store_ptr(&p->p_fd, NULL); 2715 atomic_thread_fence_seq_cst(); 2716 PROC_WAIT_UNLOCKED(p); 2717 2718 if (refcount_release(&fdp->fd_refcnt) == 0) 2719 return; 2720 2721 fdescfree_fds(td, fdp); 2722 } 2723 2724 void 2725 pdescfree(struct thread *td) 2726 { 2727 struct proc *p; 2728 struct pwddesc *pdp; 2729 2730 p = td->td_proc; 2731 pdp = p->p_pd; 2732 MPASS(pdp != NULL); 2733 2734 /* 2735 * Check pdhold for an explanation. 2736 */ 2737 atomic_store_ptr(&p->p_pd, NULL); 2738 atomic_thread_fence_seq_cst(); 2739 PROC_WAIT_UNLOCKED(p); 2740 2741 pddrop(pdp); 2742 } 2743 2744 /* 2745 * For setugid programs, we don't want to people to use that setugidness 2746 * to generate error messages which write to a file which otherwise would 2747 * otherwise be off-limits to the process. We check for filesystems where 2748 * the vnode can change out from under us after execve (like [lin]procfs). 2749 * 2750 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2751 * sufficient. We also don't check for setugidness since we know we are. 2752 */ 2753 static bool 2754 is_unsafe(struct file *fp) 2755 { 2756 struct vnode *vp; 2757 2758 if (fp->f_type != DTYPE_VNODE) 2759 return (false); 2760 2761 vp = fp->f_vnode; 2762 return ((vp->v_vflag & VV_PROCDEP) != 0); 2763 } 2764 2765 /* 2766 * Make this setguid thing safe, if at all possible. 2767 */ 2768 void 2769 fdsetugidsafety(struct thread *td) 2770 { 2771 struct filedesc *fdp; 2772 struct file *fp; 2773 int i; 2774 2775 fdp = td->td_proc->p_fd; 2776 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2777 ("the fdtable should not be shared")); 2778 MPASS(fdp->fd_nfiles >= 3); 2779 for (i = 0; i <= 2; i++) { 2780 fp = fdp->fd_ofiles[i].fde_file; 2781 if (fp != NULL && is_unsafe(fp)) { 2782 FILEDESC_XLOCK(fdp); 2783 knote_fdclose(td, i); 2784 /* 2785 * NULL-out descriptor prior to close to avoid 2786 * a race while close blocks. 2787 */ 2788 fdfree(fdp, i); 2789 FILEDESC_XUNLOCK(fdp); 2790 (void) closef(fp, td); 2791 } 2792 } 2793 } 2794 2795 /* 2796 * If a specific file object occupies a specific file descriptor, close the 2797 * file descriptor entry and drop a reference on the file object. This is a 2798 * convenience function to handle a subsequent error in a function that calls 2799 * falloc() that handles the race that another thread might have closed the 2800 * file descriptor out from under the thread creating the file object. 2801 */ 2802 void 2803 fdclose(struct thread *td, struct file *fp, int idx) 2804 { 2805 struct filedesc *fdp = td->td_proc->p_fd; 2806 2807 FILEDESC_XLOCK(fdp); 2808 if (fdp->fd_ofiles[idx].fde_file == fp) { 2809 fdfree(fdp, idx); 2810 FILEDESC_XUNLOCK(fdp); 2811 fdrop(fp, td); 2812 } else 2813 FILEDESC_XUNLOCK(fdp); 2814 } 2815 2816 /* 2817 * Close any files on exec? 2818 */ 2819 void 2820 fdcloseexec(struct thread *td) 2821 { 2822 struct filedesc *fdp; 2823 struct filedescent *fde; 2824 struct file *fp; 2825 int i; 2826 2827 fdp = td->td_proc->p_fd; 2828 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2829 ("the fdtable should not be shared")); 2830 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2831 fp = fde->fde_file; 2832 if (fp->f_type == DTYPE_MQUEUE || 2833 (fde->fde_flags & UF_EXCLOSE)) { 2834 FILEDESC_XLOCK(fdp); 2835 fdfree(fdp, i); 2836 (void) closefp(fdp, i, fp, td, false, false); 2837 FILEDESC_UNLOCK_ASSERT(fdp); 2838 } else if (fde->fde_flags & UF_FOCLOSE) { 2839 /* 2840 * https://austingroupbugs.net/view.php?id=1851 2841 * FD_CLOFORK should not be preserved across exec 2842 */ 2843 fde->fde_flags &= ~UF_FOCLOSE; 2844 } 2845 } 2846 } 2847 2848 /* 2849 * It is unsafe for set[ug]id processes to be started with file 2850 * descriptors 0..2 closed, as these descriptors are given implicit 2851 * significance in the Standard C library. fdcheckstd() will create a 2852 * descriptor referencing /dev/null for each of stdin, stdout, and 2853 * stderr that is not already open. 2854 */ 2855 int 2856 fdcheckstd(struct thread *td) 2857 { 2858 struct filedesc *fdp; 2859 register_t save; 2860 int i, error, devnull; 2861 2862 fdp = td->td_proc->p_fd; 2863 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2864 ("the fdtable should not be shared")); 2865 MPASS(fdp->fd_nfiles >= 3); 2866 devnull = -1; 2867 for (i = 0; i <= 2; i++) { 2868 if (fdp->fd_ofiles[i].fde_file != NULL) 2869 continue; 2870 2871 save = td->td_retval[0]; 2872 if (devnull != -1) { 2873 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2874 } else { 2875 error = kern_openat(td, AT_FDCWD, "/dev/null", 2876 UIO_SYSSPACE, O_RDWR, 0); 2877 if (error == 0) { 2878 devnull = td->td_retval[0]; 2879 KASSERT(devnull == i, ("we didn't get our fd")); 2880 } 2881 } 2882 td->td_retval[0] = save; 2883 if (error != 0) 2884 return (error); 2885 } 2886 return (0); 2887 } 2888 2889 /* 2890 * Internal form of close. Decrement reference count on file structure. 2891 * Note: td may be NULL when closing a file that was being passed in a 2892 * message. 2893 */ 2894 int 2895 closef(struct file *fp, struct thread *td) 2896 { 2897 struct vnode *vp; 2898 struct flock lf; 2899 struct filedesc_to_leader *fdtol; 2900 struct filedesc *fdp; 2901 2902 MPASS(td != NULL); 2903 2904 /* 2905 * POSIX record locking dictates that any close releases ALL 2906 * locks owned by this process. This is handled by setting 2907 * a flag in the unlock to free ONLY locks obeying POSIX 2908 * semantics, and not to free BSD-style file locks. 2909 * If the descriptor was in a message, POSIX-style locks 2910 * aren't passed with the descriptor, and the thread pointer 2911 * will be NULL. Callers should be careful only to pass a 2912 * NULL thread pointer when there really is no owning 2913 * context that might have locks, or the locks will be 2914 * leaked. 2915 */ 2916 if (fp->f_type == DTYPE_VNODE) { 2917 vp = fp->f_vnode; 2918 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2919 lf.l_whence = SEEK_SET; 2920 lf.l_start = 0; 2921 lf.l_len = 0; 2922 lf.l_type = F_UNLCK; 2923 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2924 F_UNLCK, &lf, F_POSIX); 2925 } 2926 fdtol = td->td_proc->p_fdtol; 2927 if (fdtol != NULL) { 2928 /* 2929 * Handle special case where file descriptor table is 2930 * shared between multiple process leaders. 2931 */ 2932 fdp = td->td_proc->p_fd; 2933 FILEDESC_XLOCK(fdp); 2934 for (fdtol = fdtol->fdl_next; 2935 fdtol != td->td_proc->p_fdtol; 2936 fdtol = fdtol->fdl_next) { 2937 if ((fdtol->fdl_leader->p_flag & 2938 P_ADVLOCK) == 0) 2939 continue; 2940 fdtol->fdl_holdcount++; 2941 FILEDESC_XUNLOCK(fdp); 2942 lf.l_whence = SEEK_SET; 2943 lf.l_start = 0; 2944 lf.l_len = 0; 2945 lf.l_type = F_UNLCK; 2946 vp = fp->f_vnode; 2947 (void) VOP_ADVLOCK(vp, 2948 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2949 F_POSIX); 2950 FILEDESC_XLOCK(fdp); 2951 fdtol->fdl_holdcount--; 2952 if (fdtol->fdl_holdcount == 0 && 2953 fdtol->fdl_wakeup != 0) { 2954 fdtol->fdl_wakeup = 0; 2955 wakeup(fdtol); 2956 } 2957 } 2958 FILEDESC_XUNLOCK(fdp); 2959 } 2960 } 2961 return (fdrop_close(fp, td)); 2962 } 2963 2964 /* 2965 * Hack for file descriptor passing code. 2966 */ 2967 void 2968 closef_nothread(struct file *fp) 2969 { 2970 2971 fdrop(fp, NULL); 2972 } 2973 2974 /* 2975 * Initialize the file pointer with the specified properties. 2976 * 2977 * The ops are set with release semantics to be certain that the flags, type, 2978 * and data are visible when ops is. This is to prevent ops methods from being 2979 * called with bad data. 2980 */ 2981 void 2982 finit(struct file *fp, u_int flag, short type, void *data, 2983 const struct fileops *ops) 2984 { 2985 fp->f_data = data; 2986 fp->f_flag = flag; 2987 fp->f_type = type; 2988 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2989 } 2990 2991 void 2992 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops) 2993 { 2994 fp->f_seqcount[UIO_READ] = 1; 2995 fp->f_seqcount[UIO_WRITE] = 1; 2996 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 2997 data, ops); 2998 } 2999 3000 int 3001 fget_cap_noref(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3002 struct file **fpp, struct filecaps *havecapsp) 3003 { 3004 struct filedescent *fde; 3005 int error; 3006 3007 FILEDESC_LOCK_ASSERT(fdp); 3008 3009 *fpp = NULL; 3010 fde = fdeget_noref(fdp, fd); 3011 if (fde == NULL) { 3012 error = EBADF; 3013 goto out; 3014 } 3015 3016 #ifdef CAPABILITIES 3017 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 3018 if (error != 0) 3019 goto out; 3020 #endif 3021 3022 if (havecapsp != NULL) 3023 filecaps_copy(&fde->fde_caps, havecapsp, true); 3024 3025 *fpp = fde->fde_file; 3026 3027 error = 0; 3028 out: 3029 return (error); 3030 } 3031 3032 #ifdef CAPABILITIES 3033 int 3034 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3035 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3036 { 3037 struct filedesc *fdp = td->td_proc->p_fd; 3038 int error; 3039 struct file *fp; 3040 seqc_t seq; 3041 3042 *fpp = NULL; 3043 for (;;) { 3044 error = fget_unlocked_seq(td, fd, needrightsp, flagsp, &fp, 3045 &seq); 3046 if (error != 0) 3047 return (error); 3048 3049 if (havecapsp != NULL) { 3050 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 3051 havecapsp, false)) { 3052 fdrop(fp, td); 3053 goto get_locked; 3054 } 3055 } 3056 3057 if (!fd_modified(fdp, fd, seq)) 3058 break; 3059 fdrop(fp, td); 3060 } 3061 3062 *fpp = fp; 3063 return (0); 3064 3065 get_locked: 3066 FILEDESC_SLOCK(fdp); 3067 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp); 3068 if (error == 0 && !fhold(*fpp)) 3069 error = EBADF; 3070 FILEDESC_SUNLOCK(fdp); 3071 return (error); 3072 } 3073 #else 3074 int 3075 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3076 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3077 { 3078 int error; 3079 error = fget_unlocked_flags(td, fd, needrightsp, flagsp, fpp); 3080 if (havecapsp != NULL && error == 0) 3081 filecaps_fill(havecapsp); 3082 3083 return (error); 3084 } 3085 #endif 3086 3087 int 3088 fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp) 3089 { 3090 struct filedesc *fdp; 3091 struct file *fp; 3092 int error; 3093 3094 if (p == td->td_proc) /* curproc */ 3095 return (fget_unlocked(td, fd, &cap_no_rights, fpp)); 3096 3097 PROC_LOCK(p); 3098 fdp = fdhold(p); 3099 PROC_UNLOCK(p); 3100 if (fdp == NULL) 3101 return (ENOENT); 3102 FILEDESC_SLOCK(fdp); 3103 if (refcount_load(&fdp->fd_refcnt) != 0) { 3104 fp = fget_noref(fdp, fd); 3105 if (fp != NULL && fhold(fp)) { 3106 *fpp = fp; 3107 error = 0; 3108 } else { 3109 error = EBADF; 3110 } 3111 } else { 3112 error = ENOENT; 3113 } 3114 FILEDESC_SUNLOCK(fdp); 3115 fddrop(fdp); 3116 return (error); 3117 } 3118 3119 int 3120 fget_remote_foreach(struct thread *td, struct proc *p, 3121 int (*fn)(struct proc *, int, struct file *, void *), void *arg) 3122 { 3123 struct filedesc *fdp; 3124 struct fdescenttbl *fdt; 3125 struct file *fp; 3126 int error, error1, fd, highfd; 3127 3128 error = 0; 3129 PROC_LOCK(p); 3130 fdp = fdhold(p); 3131 PROC_UNLOCK(p); 3132 if (fdp == NULL) 3133 return (ENOENT); 3134 3135 FILEDESC_SLOCK(fdp); 3136 if (refcount_load(&fdp->fd_refcnt) != 0) { 3137 fdt = atomic_load_ptr(&fdp->fd_files); 3138 highfd = fdt->fdt_nfiles - 1; 3139 FILEDESC_SUNLOCK(fdp); 3140 } else { 3141 error = ENOENT; 3142 FILEDESC_SUNLOCK(fdp); 3143 goto out; 3144 } 3145 3146 for (fd = 0; fd <= highfd; fd++) { 3147 error1 = fget_remote(td, p, fd, &fp); 3148 if (error1 != 0) 3149 continue; 3150 error = fn(p, fd, fp, arg); 3151 fdrop(fp, td); 3152 if (error != 0) 3153 break; 3154 } 3155 out: 3156 fddrop(fdp); 3157 return (error); 3158 } 3159 3160 #ifdef CAPABILITIES 3161 int 3162 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3163 { 3164 const struct filedescent *fde; 3165 const struct fdescenttbl *fdt; 3166 struct filedesc *fdp; 3167 struct file *fp; 3168 struct vnode *vp; 3169 const cap_rights_t *haverights; 3170 cap_rights_t rights; 3171 seqc_t seq; 3172 int fd, flags; 3173 3174 VFS_SMR_ASSERT_ENTERED(); 3175 3176 fd = ndp->ni_dirfd; 3177 rights = *ndp->ni_rightsneeded; 3178 cap_rights_set_one(&rights, CAP_LOOKUP); 3179 3180 fdp = curproc->p_fd; 3181 fdt = fdp->fd_files; 3182 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3183 return (EBADF); 3184 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3185 fde = &fdt->fdt_ofiles[fd]; 3186 haverights = cap_rights_fde_inline(fde); 3187 fp = fde->fde_file; 3188 if (__predict_false(fp == NULL)) 3189 return (EAGAIN); 3190 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 3191 return (EAGAIN); 3192 flags = fp->f_flag & FSEARCH; 3193 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3194 O_RESOLVE_BENEATH : 0; 3195 vp = fp->f_vnode; 3196 if (__predict_false(vp == NULL)) { 3197 return (EAGAIN); 3198 } 3199 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 3200 return (EAGAIN); 3201 } 3202 /* 3203 * Use an acquire barrier to force re-reading of fdt so it is 3204 * refreshed for verification. 3205 */ 3206 atomic_thread_fence_acq(); 3207 fdt = fdp->fd_files; 3208 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3209 return (EAGAIN); 3210 /* 3211 * If file descriptor doesn't have all rights, 3212 * all lookups relative to it must also be 3213 * strictly relative. 3214 * 3215 * Not yet supported by fast path. 3216 */ 3217 CAP_ALL(&rights); 3218 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3219 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3220 ndp->ni_filecaps.fc_nioctls != -1) { 3221 #ifdef notyet 3222 ndp->ni_lcf |= NI_LCF_STRICTREL; 3223 #else 3224 return (EAGAIN); 3225 #endif 3226 } 3227 *vpp = vp; 3228 *flagsp = flags; 3229 return (0); 3230 } 3231 #else 3232 int 3233 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3234 { 3235 const struct filedescent *fde; 3236 const struct fdescenttbl *fdt; 3237 struct filedesc *fdp; 3238 struct file *fp; 3239 struct vnode *vp; 3240 int fd, flags; 3241 3242 VFS_SMR_ASSERT_ENTERED(); 3243 3244 fd = ndp->ni_dirfd; 3245 fdp = curproc->p_fd; 3246 fdt = fdp->fd_files; 3247 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3248 return (EBADF); 3249 fde = &fdt->fdt_ofiles[fd]; 3250 fp = fde->fde_file; 3251 if (__predict_false(fp == NULL)) 3252 return (EAGAIN); 3253 flags = fp->f_flag & FSEARCH; 3254 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3255 O_RESOLVE_BENEATH : 0; 3256 vp = fp->f_vnode; 3257 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3258 return (EAGAIN); 3259 } 3260 /* 3261 * Use an acquire barrier to force re-reading of fdt so it is 3262 * refreshed for verification. 3263 */ 3264 atomic_thread_fence_acq(); 3265 fdt = fdp->fd_files; 3266 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3267 return (EAGAIN); 3268 filecaps_fill(&ndp->ni_filecaps); 3269 *vpp = vp; 3270 *flagsp = flags; 3271 return (0); 3272 } 3273 #endif 3274 3275 int 3276 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp) 3277 { 3278 struct thread *td; 3279 struct file *fp; 3280 struct vnode *vp; 3281 struct componentname *cnp; 3282 cap_rights_t rights; 3283 int error; 3284 uint8_t flags; 3285 3286 td = curthread; 3287 rights = *ndp->ni_rightsneeded; 3288 cap_rights_set_one(&rights, CAP_LOOKUP); 3289 cnp = &ndp->ni_cnd; 3290 3291 error = fget_cap(td, ndp->ni_dirfd, &rights, &flags, &fp, 3292 &ndp->ni_filecaps); 3293 if (__predict_false(error != 0)) 3294 return (error); 3295 if (__predict_false(fp->f_ops == &badfileops)) { 3296 error = EBADF; 3297 goto out_free; 3298 } 3299 vp = fp->f_vnode; 3300 if (__predict_false(vp == NULL)) { 3301 error = ENOTDIR; 3302 goto out_free; 3303 } 3304 vrefact(vp); 3305 /* 3306 * XXX does not check for VDIR, handled by namei_setup 3307 */ 3308 if ((fp->f_flag & FSEARCH) != 0) 3309 cnp->cn_flags |= NOEXECCHECK; 3310 if ((flags & UF_RESOLVE_BENEATH) != 0) { 3311 cnp->cn_flags |= RBENEATH; 3312 ndp->ni_resflags |= NIRES_BENEATH; 3313 } 3314 fdrop(fp, td); 3315 3316 #ifdef CAPABILITIES 3317 /* 3318 * If file descriptor doesn't have all rights, 3319 * all lookups relative to it must also be 3320 * strictly relative. 3321 */ 3322 CAP_ALL(&rights); 3323 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3324 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3325 ndp->ni_filecaps.fc_nioctls != -1) { 3326 ndp->ni_lcf |= NI_LCF_STRICTREL; 3327 ndp->ni_resflags |= NIRES_STRICTREL; 3328 } 3329 #endif 3330 3331 /* 3332 * TODO: avoid copying ioctl caps if it can be helped to begin with 3333 */ 3334 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0) 3335 filecaps_free_ioctl(&ndp->ni_filecaps); 3336 3337 *vpp = vp; 3338 return (0); 3339 3340 out_free: 3341 filecaps_free(&ndp->ni_filecaps); 3342 fdrop(fp, td); 3343 return (error); 3344 } 3345 3346 /* 3347 * Fetch the descriptor locklessly. 3348 * 3349 * We avoid fdrop() races by never raising a refcount above 0. To accomplish 3350 * this we have to use a cmpset loop rather than an atomic_add. The descriptor 3351 * must be re-verified once we acquire a reference to be certain that the 3352 * identity is still correct and we did not lose a race due to preemption. 3353 * 3354 * Force a reload of fdt when looping. Another thread could reallocate 3355 * the table before this fd was closed, so it is possible that there is 3356 * a stale fp pointer in cached version. 3357 */ 3358 #ifdef CAPABILITIES 3359 static int 3360 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3361 uint8_t *flagsp, struct file **fpp, seqc_t *seqp) 3362 { 3363 struct filedesc *fdp; 3364 const struct filedescent *fde; 3365 const struct fdescenttbl *fdt; 3366 struct file *fp; 3367 seqc_t seq; 3368 cap_rights_t haverights; 3369 int error; 3370 uint8_t flags; 3371 3372 fdp = td->td_proc->p_fd; 3373 fdt = fdp->fd_files; 3374 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3375 return (EBADF); 3376 3377 for (;;) { 3378 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3379 fde = &fdt->fdt_ofiles[fd]; 3380 haverights = *cap_rights_fde_inline(fde); 3381 fp = fde->fde_file; 3382 flags = fde->fde_flags; 3383 if (__predict_false(fp == NULL)) { 3384 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3385 return (EBADF); 3386 fdt = atomic_load_ptr(&fdp->fd_files); 3387 continue; 3388 } 3389 error = cap_check_inline(&haverights, needrightsp); 3390 if (__predict_false(error != 0)) { 3391 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3392 return (error); 3393 fdt = atomic_load_ptr(&fdp->fd_files); 3394 continue; 3395 } 3396 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3397 fdt = atomic_load_ptr(&fdp->fd_files); 3398 continue; 3399 } 3400 /* 3401 * Use an acquire barrier to force re-reading of fdt so it is 3402 * refreshed for verification. 3403 */ 3404 atomic_thread_fence_acq(); 3405 fdt = fdp->fd_files; 3406 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)) 3407 break; 3408 fdrop(fp, td); 3409 } 3410 *fpp = fp; 3411 if (flagsp != NULL) 3412 *flagsp = flags; 3413 if (seqp != NULL) 3414 *seqp = seq; 3415 return (0); 3416 } 3417 #else 3418 static int 3419 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3420 uint8_t *flagsp, struct file **fpp, seqc_t *seqp __unused) 3421 { 3422 struct filedesc *fdp; 3423 const struct fdescenttbl *fdt; 3424 struct file *fp; 3425 uint8_t flags; 3426 3427 fdp = td->td_proc->p_fd; 3428 fdt = fdp->fd_files; 3429 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3430 return (EBADF); 3431 3432 for (;;) { 3433 fp = fdt->fdt_ofiles[fd].fde_file; 3434 flags = fdt->fdt_ofiles[fd].fde_flags; 3435 if (__predict_false(fp == NULL)) 3436 return (EBADF); 3437 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3438 fdt = atomic_load_ptr(&fdp->fd_files); 3439 continue; 3440 } 3441 /* 3442 * Use an acquire barrier to force re-reading of fdt so it is 3443 * refreshed for verification. 3444 */ 3445 atomic_thread_fence_acq(); 3446 fdt = fdp->fd_files; 3447 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file)) 3448 break; 3449 fdrop(fp, td); 3450 } 3451 if (flagsp != NULL) 3452 *flagsp = flags; 3453 *fpp = fp; 3454 return (0); 3455 } 3456 #endif 3457 3458 /* 3459 * See the comments in fget_unlocked_seq for an explanation of how this works. 3460 * 3461 * This is a simplified variant which bails out to the aforementioned routine 3462 * if anything goes wrong. In practice this only happens when userspace is 3463 * racing with itself. 3464 */ 3465 int 3466 fget_unlocked_flags(struct thread *td, int fd, const cap_rights_t *needrightsp, 3467 uint8_t *flagsp, struct file **fpp) 3468 { 3469 struct filedesc *fdp; 3470 #ifdef CAPABILITIES 3471 const struct filedescent *fde; 3472 #endif 3473 const struct fdescenttbl *fdt; 3474 struct file *fp; 3475 #ifdef CAPABILITIES 3476 seqc_t seq; 3477 const cap_rights_t *haverights; 3478 #endif 3479 uint8_t flags; 3480 3481 fdp = td->td_proc->p_fd; 3482 fdt = fdp->fd_files; 3483 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { 3484 *fpp = NULL; 3485 return (EBADF); 3486 } 3487 #ifdef CAPABILITIES 3488 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3489 fde = &fdt->fdt_ofiles[fd]; 3490 haverights = cap_rights_fde_inline(fde); 3491 fp = fde->fde_file; 3492 flags = fde->fde_flags; 3493 #else 3494 fp = fdt->fdt_ofiles[fd].fde_file; 3495 flags = fdt->fdt_ofiles[fd].fde_flags; 3496 #endif 3497 if (__predict_false(fp == NULL)) 3498 goto out_fallback; 3499 #ifdef CAPABILITIES 3500 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3501 goto out_fallback; 3502 #endif 3503 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3504 goto out_fallback; 3505 3506 /* 3507 * Use an acquire barrier to force re-reading of fdt so it is 3508 * refreshed for verification. 3509 */ 3510 atomic_thread_fence_acq(); 3511 fdt = fdp->fd_files; 3512 #ifdef CAPABILITIES 3513 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3514 #else 3515 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3516 #endif 3517 goto out_fdrop; 3518 *fpp = fp; 3519 if (flagsp != NULL) 3520 *flagsp = flags; 3521 return (0); 3522 out_fdrop: 3523 fdrop(fp, td); 3524 out_fallback: 3525 *fpp = NULL; 3526 return (fget_unlocked_seq(td, fd, needrightsp, flagsp, fpp, NULL)); 3527 } 3528 3529 int 3530 fget_unlocked(struct thread *td, int fd, const cap_rights_t *needrightsp, 3531 struct file **fpp) 3532 { 3533 return (fget_unlocked_flags(td, fd, needrightsp, NULL, fpp)); 3534 } 3535 3536 /* 3537 * Translate fd -> file when the caller guarantees the file descriptor table 3538 * can't be changed by others. 3539 * 3540 * Note this does not mean the file object itself is only visible to the caller, 3541 * merely that it wont disappear without having to be referenced. 3542 * 3543 * Must be paired with fput_only_user. 3544 */ 3545 #ifdef CAPABILITIES 3546 int 3547 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3548 struct file **fpp) 3549 { 3550 const struct filedescent *fde; 3551 const struct fdescenttbl *fdt; 3552 const cap_rights_t *haverights; 3553 struct file *fp; 3554 int error; 3555 3556 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3557 3558 *fpp = NULL; 3559 if (__predict_false(fd >= fdp->fd_nfiles)) 3560 return (EBADF); 3561 3562 fdt = fdp->fd_files; 3563 fde = &fdt->fdt_ofiles[fd]; 3564 fp = fde->fde_file; 3565 if (__predict_false(fp == NULL)) 3566 return (EBADF); 3567 MPASS(refcount_load(&fp->f_count) > 0); 3568 haverights = cap_rights_fde_inline(fde); 3569 error = cap_check_inline(haverights, needrightsp); 3570 if (__predict_false(error != 0)) 3571 return (error); 3572 *fpp = fp; 3573 return (0); 3574 } 3575 #else 3576 int 3577 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3578 struct file **fpp) 3579 { 3580 struct file *fp; 3581 3582 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3583 3584 *fpp = NULL; 3585 if (__predict_false(fd >= fdp->fd_nfiles)) 3586 return (EBADF); 3587 3588 fp = fdp->fd_ofiles[fd].fde_file; 3589 if (__predict_false(fp == NULL)) 3590 return (EBADF); 3591 3592 MPASS(refcount_load(&fp->f_count) > 0); 3593 *fpp = fp; 3594 return (0); 3595 } 3596 #endif 3597 3598 /* 3599 * Extract the file pointer associated with the specified descriptor for the 3600 * current user process. 3601 * 3602 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3603 * returned. 3604 * 3605 * File's rights will be checked against the capability rights mask. 3606 * 3607 * If an error occurred the non-zero error is returned and *fpp is set to 3608 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3609 * responsible for fdrop(). 3610 */ 3611 static __inline int 3612 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3613 const cap_rights_t *needrightsp) 3614 { 3615 struct file *fp; 3616 int error; 3617 3618 *fpp = NULL; 3619 error = fget_unlocked(td, fd, needrightsp, &fp); 3620 if (__predict_false(error != 0)) 3621 return (error); 3622 if (__predict_false(fp->f_ops == &badfileops)) { 3623 fdrop(fp, td); 3624 return (EBADF); 3625 } 3626 3627 /* 3628 * FREAD and FWRITE failure return EBADF as per POSIX. 3629 */ 3630 error = 0; 3631 switch (flags) { 3632 case FREAD: 3633 case FWRITE: 3634 if ((fp->f_flag & flags) == 0) 3635 error = EBADF; 3636 break; 3637 case FEXEC: 3638 if (fp->f_ops != &path_fileops && 3639 ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3640 (fp->f_flag & FWRITE) != 0)) 3641 error = EBADF; 3642 break; 3643 case 0: 3644 break; 3645 default: 3646 KASSERT(0, ("wrong flags")); 3647 } 3648 3649 if (error != 0) { 3650 fdrop(fp, td); 3651 return (error); 3652 } 3653 3654 *fpp = fp; 3655 return (0); 3656 } 3657 3658 int 3659 fget(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp) 3660 { 3661 3662 return (_fget(td, fd, fpp, 0, rightsp)); 3663 } 3664 3665 int 3666 fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp, 3667 vm_prot_t *maxprotp, struct file **fpp) 3668 { 3669 int error; 3670 #ifndef CAPABILITIES 3671 error = _fget(td, fd, fpp, 0, rightsp); 3672 if (maxprotp != NULL) 3673 *maxprotp = VM_PROT_ALL; 3674 return (error); 3675 #else 3676 cap_rights_t fdrights; 3677 struct filedesc *fdp; 3678 struct file *fp; 3679 seqc_t seq; 3680 3681 *fpp = NULL; 3682 fdp = td->td_proc->p_fd; 3683 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3684 for (;;) { 3685 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3686 if (__predict_false(error != 0)) 3687 return (error); 3688 if (__predict_false(fp->f_ops == &badfileops)) { 3689 fdrop(fp, td); 3690 return (EBADF); 3691 } 3692 if (maxprotp != NULL) 3693 fdrights = *cap_rights(fdp, fd); 3694 if (!fd_modified(fdp, fd, seq)) 3695 break; 3696 fdrop(fp, td); 3697 } 3698 3699 /* 3700 * If requested, convert capability rights to access flags. 3701 */ 3702 if (maxprotp != NULL) 3703 *maxprotp = cap_rights_to_vmprot(&fdrights); 3704 *fpp = fp; 3705 return (0); 3706 #endif 3707 } 3708 3709 int 3710 fget_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3711 struct file **fpp) 3712 { 3713 3714 return (_fget(td, fd, fpp, FREAD, rightsp)); 3715 } 3716 3717 int 3718 fget_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3719 struct file **fpp) 3720 { 3721 3722 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3723 } 3724 3725 int 3726 fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp, 3727 int needfcntl, struct file **fpp) 3728 { 3729 #ifndef CAPABILITIES 3730 return (fget_unlocked(td, fd, rightsp, fpp)); 3731 #else 3732 struct filedesc *fdp = td->td_proc->p_fd; 3733 struct file *fp; 3734 int error; 3735 seqc_t seq; 3736 3737 *fpp = NULL; 3738 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3739 for (;;) { 3740 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3741 if (error != 0) 3742 return (error); 3743 error = cap_fcntl_check(fdp, fd, needfcntl); 3744 if (!fd_modified(fdp, fd, seq)) 3745 break; 3746 fdrop(fp, td); 3747 } 3748 if (error != 0) { 3749 fdrop(fp, td); 3750 return (error); 3751 } 3752 *fpp = fp; 3753 return (0); 3754 #endif 3755 } 3756 3757 /* 3758 * Like fget() but loads the underlying vnode, or returns an error if the 3759 * descriptor does not represent a vnode. Note that pipes use vnodes but 3760 * never have VM objects. The returned vnode will be vref()'d. 3761 * 3762 * XXX: what about the unused flags ? 3763 */ 3764 static __inline int 3765 _fgetvp(struct thread *td, int fd, int flags, const cap_rights_t *needrightsp, 3766 struct vnode **vpp) 3767 { 3768 struct file *fp; 3769 int error; 3770 3771 *vpp = NULL; 3772 error = _fget(td, fd, &fp, flags, needrightsp); 3773 if (error != 0) 3774 return (error); 3775 if (fp->f_vnode == NULL) { 3776 error = EINVAL; 3777 } else { 3778 *vpp = fp->f_vnode; 3779 vrefact(*vpp); 3780 } 3781 fdrop(fp, td); 3782 3783 return (error); 3784 } 3785 3786 int 3787 fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp, 3788 struct vnode **vpp) 3789 { 3790 3791 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3792 } 3793 3794 int 3795 fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp, 3796 struct filecaps *havecaps, struct vnode **vpp) 3797 { 3798 struct filecaps caps; 3799 struct file *fp; 3800 int error; 3801 3802 error = fget_cap(td, fd, needrightsp, NULL, &fp, &caps); 3803 if (error != 0) 3804 return (error); 3805 if (fp->f_ops == &badfileops) { 3806 error = EBADF; 3807 goto out; 3808 } 3809 if (fp->f_vnode == NULL) { 3810 error = EINVAL; 3811 goto out; 3812 } 3813 3814 *havecaps = caps; 3815 *vpp = fp->f_vnode; 3816 vrefact(*vpp); 3817 fdrop(fp, td); 3818 3819 return (0); 3820 out: 3821 filecaps_free(&caps); 3822 fdrop(fp, td); 3823 return (error); 3824 } 3825 3826 int 3827 fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3828 struct vnode **vpp) 3829 { 3830 3831 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3832 } 3833 3834 int 3835 fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp, 3836 struct vnode **vpp) 3837 { 3838 3839 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3840 } 3841 3842 #ifdef notyet 3843 int 3844 fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3845 struct vnode **vpp) 3846 { 3847 3848 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3849 } 3850 #endif 3851 3852 /* 3853 * Handle the last reference to a file being closed. 3854 * 3855 * Without the noinline attribute clang keeps inlining the func thorough this 3856 * file when fdrop is used. 3857 */ 3858 int __noinline 3859 _fdrop(struct file *fp, struct thread *td) 3860 { 3861 int error; 3862 3863 KASSERT(refcount_load(&fp->f_count) == 0, 3864 ("fdrop: fp %p count %d", fp, refcount_load(&fp->f_count))); 3865 3866 error = fo_close(fp, td); 3867 atomic_subtract_int(&openfiles, 1); 3868 crfree(fp->f_cred); 3869 free(fp->f_advice, M_FADVISE); 3870 uma_zfree(file_zone, fp); 3871 3872 return (error); 3873 } 3874 3875 /* 3876 * Apply an advisory lock on a file descriptor. 3877 * 3878 * Just attempt to get a record lock of the requested type on the entire file 3879 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 3880 */ 3881 #ifndef _SYS_SYSPROTO_H_ 3882 struct flock_args { 3883 int fd; 3884 int how; 3885 }; 3886 #endif 3887 /* ARGSUSED */ 3888 int 3889 sys_flock(struct thread *td, struct flock_args *uap) 3890 { 3891 struct file *fp; 3892 struct vnode *vp; 3893 struct flock lf; 3894 int error; 3895 3896 error = fget(td, uap->fd, &cap_flock_rights, &fp); 3897 if (error != 0) 3898 return (error); 3899 error = EOPNOTSUPP; 3900 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 3901 goto done; 3902 } 3903 if (fp->f_ops == &path_fileops) { 3904 goto done; 3905 } 3906 3907 error = 0; 3908 vp = fp->f_vnode; 3909 lf.l_whence = SEEK_SET; 3910 lf.l_start = 0; 3911 lf.l_len = 0; 3912 if (uap->how & LOCK_UN) { 3913 lf.l_type = F_UNLCK; 3914 atomic_clear_int(&fp->f_flag, FHASLOCK); 3915 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3916 goto done; 3917 } 3918 if (uap->how & LOCK_EX) 3919 lf.l_type = F_WRLCK; 3920 else if (uap->how & LOCK_SH) 3921 lf.l_type = F_RDLCK; 3922 else { 3923 error = EBADF; 3924 goto done; 3925 } 3926 atomic_set_int(&fp->f_flag, FHASLOCK); 3927 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3928 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3929 done: 3930 fdrop(fp, td); 3931 return (error); 3932 } 3933 /* 3934 * Duplicate the specified descriptor to a free descriptor. 3935 */ 3936 int 3937 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3938 int openerror, int *indxp) 3939 { 3940 struct filedescent *newfde, *oldfde; 3941 struct file *fp; 3942 u_long *ioctls; 3943 int error, indx; 3944 3945 KASSERT(openerror == ENODEV || openerror == ENXIO, 3946 ("unexpected error %d in %s", openerror, __func__)); 3947 3948 /* 3949 * If the to-be-dup'd fd number is greater than the allowed number 3950 * of file descriptors, or the fd to be dup'd has already been 3951 * closed, then reject. 3952 */ 3953 FILEDESC_XLOCK(fdp); 3954 if ((fp = fget_noref(fdp, dfd)) == NULL) { 3955 FILEDESC_XUNLOCK(fdp); 3956 return (EBADF); 3957 } 3958 3959 error = fdalloc(td, 0, &indx); 3960 if (error != 0) { 3961 FILEDESC_XUNLOCK(fdp); 3962 return (error); 3963 } 3964 3965 /* 3966 * There are two cases of interest here. 3967 * 3968 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3969 * 3970 * For ENXIO steal away the file structure from (dfd) and store it in 3971 * (indx). (dfd) is effectively closed by this operation. 3972 */ 3973 switch (openerror) { 3974 case ENODEV: 3975 /* 3976 * Check that the mode the file is being opened for is a 3977 * subset of the mode of the existing descriptor. 3978 */ 3979 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3980 fdunused(fdp, indx); 3981 FILEDESC_XUNLOCK(fdp); 3982 return (EACCES); 3983 } 3984 if (!fhold(fp)) { 3985 fdunused(fdp, indx); 3986 FILEDESC_XUNLOCK(fdp); 3987 return (EBADF); 3988 } 3989 newfde = &fdp->fd_ofiles[indx]; 3990 oldfde = &fdp->fd_ofiles[dfd]; 3991 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3992 #ifdef CAPABILITIES 3993 seqc_write_begin(&newfde->fde_seqc); 3994 #endif 3995 fde_copy(oldfde, newfde); 3996 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3997 ioctls); 3998 #ifdef CAPABILITIES 3999 seqc_write_end(&newfde->fde_seqc); 4000 #endif 4001 break; 4002 case ENXIO: 4003 /* 4004 * Steal away the file pointer from dfd and stuff it into indx. 4005 */ 4006 newfde = &fdp->fd_ofiles[indx]; 4007 oldfde = &fdp->fd_ofiles[dfd]; 4008 #ifdef CAPABILITIES 4009 seqc_write_begin(&oldfde->fde_seqc); 4010 seqc_write_begin(&newfde->fde_seqc); 4011 #endif 4012 fde_copy(oldfde, newfde); 4013 oldfde->fde_file = NULL; 4014 fdunused(fdp, dfd); 4015 #ifdef CAPABILITIES 4016 seqc_write_end(&newfde->fde_seqc); 4017 seqc_write_end(&oldfde->fde_seqc); 4018 #endif 4019 break; 4020 } 4021 FILEDESC_XUNLOCK(fdp); 4022 *indxp = indx; 4023 return (0); 4024 } 4025 4026 /* 4027 * This sysctl determines if we will allow a process to chroot(2) if it 4028 * has a directory open: 4029 * 0: disallowed for all processes. 4030 * 1: allowed for processes that were not already chroot(2)'ed. 4031 * 2: allowed for all processes. 4032 */ 4033 4034 static int chroot_allow_open_directories = 1; 4035 4036 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 4037 &chroot_allow_open_directories, 0, 4038 "Allow a process to chroot(2) if it has a directory open"); 4039 4040 /* 4041 * Helper function for raised chroot(2) security function: Refuse if 4042 * any filedescriptors are open directories. 4043 */ 4044 static int 4045 chroot_refuse_vdir_fds(struct filedesc *fdp) 4046 { 4047 struct vnode *vp; 4048 struct file *fp; 4049 int i; 4050 4051 FILEDESC_LOCK_ASSERT(fdp); 4052 4053 FILEDESC_FOREACH_FP(fdp, i, fp) { 4054 if (fp->f_type == DTYPE_VNODE) { 4055 vp = fp->f_vnode; 4056 if (vp->v_type == VDIR) 4057 return (EPERM); 4058 } 4059 } 4060 return (0); 4061 } 4062 4063 static void 4064 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 4065 { 4066 4067 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 4068 vrefact(oldpwd->pwd_cdir); 4069 newpwd->pwd_cdir = oldpwd->pwd_cdir; 4070 } 4071 4072 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 4073 vrefact(oldpwd->pwd_rdir); 4074 newpwd->pwd_rdir = oldpwd->pwd_rdir; 4075 } 4076 4077 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 4078 vrefact(oldpwd->pwd_jdir); 4079 newpwd->pwd_jdir = oldpwd->pwd_jdir; 4080 } 4081 4082 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) { 4083 vrefact(oldpwd->pwd_adir); 4084 newpwd->pwd_adir = oldpwd->pwd_adir; 4085 } 4086 } 4087 4088 struct pwd * 4089 pwd_hold_pwddesc(struct pwddesc *pdp) 4090 { 4091 struct pwd *pwd; 4092 4093 PWDDESC_ASSERT_XLOCKED(pdp); 4094 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4095 if (pwd != NULL) 4096 refcount_acquire(&pwd->pwd_refcount); 4097 return (pwd); 4098 } 4099 4100 bool 4101 pwd_hold_smr(struct pwd *pwd) 4102 { 4103 4104 MPASS(pwd != NULL); 4105 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 4106 return (true); 4107 } 4108 return (false); 4109 } 4110 4111 struct pwd * 4112 pwd_hold(struct thread *td) 4113 { 4114 struct pwddesc *pdp; 4115 struct pwd *pwd; 4116 4117 pdp = td->td_proc->p_pd; 4118 4119 vfs_smr_enter(); 4120 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 4121 if (pwd_hold_smr(pwd)) { 4122 vfs_smr_exit(); 4123 return (pwd); 4124 } 4125 vfs_smr_exit(); 4126 PWDDESC_XLOCK(pdp); 4127 pwd = pwd_hold_pwddesc(pdp); 4128 MPASS(pwd != NULL); 4129 PWDDESC_XUNLOCK(pdp); 4130 return (pwd); 4131 } 4132 4133 struct pwd * 4134 pwd_hold_proc(struct proc *p) 4135 { 4136 struct pwddesc *pdp; 4137 struct pwd *pwd; 4138 4139 PROC_ASSERT_HELD(p); 4140 PROC_LOCK(p); 4141 pdp = pdhold(p); 4142 MPASS(pdp != NULL); 4143 PROC_UNLOCK(p); 4144 4145 PWDDESC_XLOCK(pdp); 4146 pwd = pwd_hold_pwddesc(pdp); 4147 MPASS(pwd != NULL); 4148 PWDDESC_XUNLOCK(pdp); 4149 pddrop(pdp); 4150 return (pwd); 4151 } 4152 4153 static struct pwd * 4154 pwd_alloc(void) 4155 { 4156 struct pwd *pwd; 4157 4158 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 4159 bzero(pwd, sizeof(*pwd)); 4160 refcount_init(&pwd->pwd_refcount, 1); 4161 return (pwd); 4162 } 4163 4164 void 4165 pwd_drop(struct pwd *pwd) 4166 { 4167 4168 if (!refcount_release(&pwd->pwd_refcount)) 4169 return; 4170 4171 if (pwd->pwd_cdir != NULL) 4172 vrele(pwd->pwd_cdir); 4173 if (pwd->pwd_rdir != NULL) 4174 vrele(pwd->pwd_rdir); 4175 if (pwd->pwd_jdir != NULL) 4176 vrele(pwd->pwd_jdir); 4177 if (pwd->pwd_adir != NULL) 4178 vrele(pwd->pwd_adir); 4179 uma_zfree_smr(pwd_zone, pwd); 4180 } 4181 4182 /* 4183 * The caller is responsible for invoking priv_check() and 4184 * mac_vnode_check_chroot() to authorize this operation. 4185 */ 4186 int 4187 pwd_chroot(struct thread *td, struct vnode *vp) 4188 { 4189 struct pwddesc *pdp; 4190 struct filedesc *fdp; 4191 struct pwd *newpwd, *oldpwd; 4192 int error; 4193 4194 fdp = td->td_proc->p_fd; 4195 pdp = td->td_proc->p_pd; 4196 newpwd = pwd_alloc(); 4197 FILEDESC_SLOCK(fdp); 4198 PWDDESC_XLOCK(pdp); 4199 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4200 if (chroot_allow_open_directories == 0 || 4201 (chroot_allow_open_directories == 1 && 4202 oldpwd->pwd_rdir != rootvnode)) { 4203 error = chroot_refuse_vdir_fds(fdp); 4204 FILEDESC_SUNLOCK(fdp); 4205 if (error != 0) { 4206 PWDDESC_XUNLOCK(pdp); 4207 pwd_drop(newpwd); 4208 return (error); 4209 } 4210 } else { 4211 FILEDESC_SUNLOCK(fdp); 4212 } 4213 4214 vrefact(vp); 4215 newpwd->pwd_rdir = vp; 4216 vrefact(vp); 4217 newpwd->pwd_adir = vp; 4218 if (oldpwd->pwd_jdir == NULL) { 4219 vrefact(vp); 4220 newpwd->pwd_jdir = vp; 4221 } 4222 pwd_fill(oldpwd, newpwd); 4223 pwd_set(pdp, newpwd); 4224 PWDDESC_XUNLOCK(pdp); 4225 pwd_drop(oldpwd); 4226 return (0); 4227 } 4228 4229 void 4230 pwd_chdir(struct thread *td, struct vnode *vp) 4231 { 4232 struct pwddesc *pdp; 4233 struct pwd *newpwd, *oldpwd; 4234 4235 VNPASS(vp->v_usecount > 0, vp); 4236 4237 newpwd = pwd_alloc(); 4238 pdp = td->td_proc->p_pd; 4239 PWDDESC_XLOCK(pdp); 4240 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4241 newpwd->pwd_cdir = vp; 4242 pwd_fill(oldpwd, newpwd); 4243 pwd_set(pdp, newpwd); 4244 PWDDESC_XUNLOCK(pdp); 4245 pwd_drop(oldpwd); 4246 } 4247 4248 /* 4249 * Process is transitioning to/from a non-native ABI. 4250 */ 4251 void 4252 pwd_altroot(struct thread *td, struct vnode *altroot_vp) 4253 { 4254 struct pwddesc *pdp; 4255 struct pwd *newpwd, *oldpwd; 4256 4257 newpwd = pwd_alloc(); 4258 pdp = td->td_proc->p_pd; 4259 PWDDESC_XLOCK(pdp); 4260 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4261 if (altroot_vp != NULL) { 4262 /* 4263 * Native process to a non-native ABI. 4264 */ 4265 4266 vrefact(altroot_vp); 4267 newpwd->pwd_adir = altroot_vp; 4268 } else { 4269 /* 4270 * Non-native process to the native ABI. 4271 */ 4272 4273 vrefact(oldpwd->pwd_rdir); 4274 newpwd->pwd_adir = oldpwd->pwd_rdir; 4275 } 4276 pwd_fill(oldpwd, newpwd); 4277 pwd_set(pdp, newpwd); 4278 PWDDESC_XUNLOCK(pdp); 4279 pwd_drop(oldpwd); 4280 } 4281 4282 /* 4283 * jail_attach(2) changes both root and working directories. 4284 */ 4285 int 4286 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 4287 { 4288 struct pwddesc *pdp; 4289 struct filedesc *fdp; 4290 struct pwd *newpwd, *oldpwd; 4291 int error; 4292 4293 fdp = td->td_proc->p_fd; 4294 pdp = td->td_proc->p_pd; 4295 newpwd = pwd_alloc(); 4296 FILEDESC_SLOCK(fdp); 4297 PWDDESC_XLOCK(pdp); 4298 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4299 error = chroot_refuse_vdir_fds(fdp); 4300 FILEDESC_SUNLOCK(fdp); 4301 if (error != 0) { 4302 PWDDESC_XUNLOCK(pdp); 4303 pwd_drop(newpwd); 4304 return (error); 4305 } 4306 4307 vrefact(vp); 4308 newpwd->pwd_rdir = vp; 4309 vrefact(vp); 4310 newpwd->pwd_cdir = vp; 4311 if (oldpwd->pwd_jdir == NULL) { 4312 vrefact(vp); 4313 newpwd->pwd_jdir = vp; 4314 } 4315 vrefact(vp); 4316 newpwd->pwd_adir = vp; 4317 pwd_fill(oldpwd, newpwd); 4318 pwd_set(pdp, newpwd); 4319 PWDDESC_XUNLOCK(pdp); 4320 pwd_drop(oldpwd); 4321 return (0); 4322 } 4323 4324 void 4325 pwd_ensure_dirs(void) 4326 { 4327 struct pwddesc *pdp; 4328 struct pwd *oldpwd, *newpwd; 4329 4330 pdp = curproc->p_pd; 4331 PWDDESC_XLOCK(pdp); 4332 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4333 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL && 4334 oldpwd->pwd_adir != NULL) { 4335 PWDDESC_XUNLOCK(pdp); 4336 return; 4337 } 4338 PWDDESC_XUNLOCK(pdp); 4339 4340 newpwd = pwd_alloc(); 4341 PWDDESC_XLOCK(pdp); 4342 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4343 pwd_fill(oldpwd, newpwd); 4344 if (newpwd->pwd_cdir == NULL) { 4345 vrefact(rootvnode); 4346 newpwd->pwd_cdir = rootvnode; 4347 } 4348 if (newpwd->pwd_rdir == NULL) { 4349 vrefact(rootvnode); 4350 newpwd->pwd_rdir = rootvnode; 4351 } 4352 if (newpwd->pwd_adir == NULL) { 4353 vrefact(rootvnode); 4354 newpwd->pwd_adir = rootvnode; 4355 } 4356 pwd_set(pdp, newpwd); 4357 PWDDESC_XUNLOCK(pdp); 4358 pwd_drop(oldpwd); 4359 } 4360 4361 void 4362 pwd_set_rootvnode(void) 4363 { 4364 struct pwddesc *pdp; 4365 struct pwd *oldpwd, *newpwd; 4366 4367 pdp = curproc->p_pd; 4368 4369 newpwd = pwd_alloc(); 4370 PWDDESC_XLOCK(pdp); 4371 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4372 vrefact(rootvnode); 4373 newpwd->pwd_cdir = rootvnode; 4374 vrefact(rootvnode); 4375 newpwd->pwd_rdir = rootvnode; 4376 vrefact(rootvnode); 4377 newpwd->pwd_adir = rootvnode; 4378 pwd_fill(oldpwd, newpwd); 4379 pwd_set(pdp, newpwd); 4380 PWDDESC_XUNLOCK(pdp); 4381 pwd_drop(oldpwd); 4382 } 4383 4384 /* 4385 * Scan all active processes and prisons to see if any of them have a current 4386 * or root directory of `olddp'. If so, replace them with the new mount point. 4387 */ 4388 void 4389 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 4390 { 4391 struct pwddesc *pdp; 4392 struct pwd *newpwd, *oldpwd; 4393 struct prison *pr; 4394 struct proc *p; 4395 int nrele; 4396 4397 if (vrefcnt(olddp) == 1) 4398 return; 4399 nrele = 0; 4400 newpwd = pwd_alloc(); 4401 sx_slock(&allproc_lock); 4402 FOREACH_PROC_IN_SYSTEM(p) { 4403 PROC_LOCK(p); 4404 pdp = pdhold(p); 4405 PROC_UNLOCK(p); 4406 if (pdp == NULL) 4407 continue; 4408 PWDDESC_XLOCK(pdp); 4409 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4410 if (oldpwd == NULL || 4411 (oldpwd->pwd_cdir != olddp && 4412 oldpwd->pwd_rdir != olddp && 4413 oldpwd->pwd_jdir != olddp && 4414 oldpwd->pwd_adir != olddp)) { 4415 PWDDESC_XUNLOCK(pdp); 4416 pddrop(pdp); 4417 continue; 4418 } 4419 if (oldpwd->pwd_cdir == olddp) { 4420 vrefact(newdp); 4421 newpwd->pwd_cdir = newdp; 4422 } 4423 if (oldpwd->pwd_rdir == olddp) { 4424 vrefact(newdp); 4425 newpwd->pwd_rdir = newdp; 4426 } 4427 if (oldpwd->pwd_jdir == olddp) { 4428 vrefact(newdp); 4429 newpwd->pwd_jdir = newdp; 4430 } 4431 if (oldpwd->pwd_adir == olddp) { 4432 vrefact(newdp); 4433 newpwd->pwd_adir = newdp; 4434 } 4435 pwd_fill(oldpwd, newpwd); 4436 pwd_set(pdp, newpwd); 4437 PWDDESC_XUNLOCK(pdp); 4438 pwd_drop(oldpwd); 4439 pddrop(pdp); 4440 newpwd = pwd_alloc(); 4441 } 4442 sx_sunlock(&allproc_lock); 4443 pwd_drop(newpwd); 4444 if (rootvnode == olddp) { 4445 vrefact(newdp); 4446 rootvnode = newdp; 4447 nrele++; 4448 } 4449 mtx_lock(&prison0.pr_mtx); 4450 if (prison0.pr_root == olddp) { 4451 vrefact(newdp); 4452 prison0.pr_root = newdp; 4453 nrele++; 4454 } 4455 mtx_unlock(&prison0.pr_mtx); 4456 sx_slock(&allprison_lock); 4457 TAILQ_FOREACH(pr, &allprison, pr_list) { 4458 mtx_lock(&pr->pr_mtx); 4459 if (pr->pr_root == olddp) { 4460 vrefact(newdp); 4461 pr->pr_root = newdp; 4462 nrele++; 4463 } 4464 mtx_unlock(&pr->pr_mtx); 4465 } 4466 sx_sunlock(&allprison_lock); 4467 while (nrele--) 4468 vrele(olddp); 4469 } 4470 4471 int 4472 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp) 4473 { 4474 struct file *fp; 4475 struct vnode *vp; 4476 int error, i; 4477 4478 error = 0; 4479 FILEDESC_SLOCK(fdp); 4480 FILEDESC_FOREACH_FP(fdp, i, fp) { 4481 if (fp->f_type != DTYPE_VNODE || 4482 (atomic_load_int(&fp->f_flag) & FWRITE) == 0) 4483 continue; 4484 vp = fp->f_vnode; 4485 if (vp->v_mount == mp) { 4486 error = EDEADLK; 4487 break; 4488 } 4489 } 4490 FILEDESC_SUNLOCK(fdp); 4491 return (error); 4492 } 4493 4494 struct filedesc_to_leader * 4495 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, 4496 struct proc *leader) 4497 { 4498 struct filedesc_to_leader *fdtol; 4499 4500 fdtol = malloc(sizeof(struct filedesc_to_leader), 4501 M_FILEDESC_TO_LEADER, M_WAITOK); 4502 fdtol->fdl_refcount = 1; 4503 fdtol->fdl_holdcount = 0; 4504 fdtol->fdl_wakeup = 0; 4505 fdtol->fdl_leader = leader; 4506 if (old != NULL) { 4507 FILEDESC_XLOCK(fdp); 4508 fdtol->fdl_next = old->fdl_next; 4509 fdtol->fdl_prev = old; 4510 old->fdl_next = fdtol; 4511 fdtol->fdl_next->fdl_prev = fdtol; 4512 FILEDESC_XUNLOCK(fdp); 4513 } else { 4514 fdtol->fdl_next = fdtol; 4515 fdtol->fdl_prev = fdtol; 4516 } 4517 return (fdtol); 4518 } 4519 4520 struct filedesc_to_leader * 4521 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp) 4522 { 4523 FILEDESC_XLOCK(fdp); 4524 fdtol->fdl_refcount++; 4525 FILEDESC_XUNLOCK(fdp); 4526 return (fdtol); 4527 } 4528 4529 static int 4530 filedesc_nfiles(struct filedesc *fdp) 4531 { 4532 NDSLOTTYPE *map; 4533 int count, off, minoff; 4534 4535 if (fdp == NULL) 4536 return (0); 4537 count = 0; 4538 FILEDESC_SLOCK(fdp); 4539 map = fdp->fd_map; 4540 off = NDSLOT(fdp->fd_nfiles - 1); 4541 for (minoff = NDSLOT(0); off >= minoff; --off) 4542 count += bitcountl(map[off]); 4543 FILEDESC_SUNLOCK(fdp); 4544 return (count); 4545 } 4546 4547 int 4548 proc_nfiles(struct proc *p) 4549 { 4550 struct filedesc *fdp; 4551 int res; 4552 4553 PROC_LOCK(p); 4554 fdp = fdhold(p); 4555 PROC_UNLOCK(p); 4556 res = filedesc_nfiles(fdp); 4557 fddrop(fdp); 4558 return (res); 4559 } 4560 4561 static int 4562 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4563 { 4564 u_int namelen; 4565 int count; 4566 4567 namelen = arg2; 4568 if (namelen != 1) 4569 return (EINVAL); 4570 4571 if (*(int *)arg1 != 0) 4572 return (EINVAL); 4573 4574 count = filedesc_nfiles(curproc->p_fd); 4575 return (SYSCTL_OUT(req, &count, sizeof(count))); 4576 } 4577 4578 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4579 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4580 "Number of open file descriptors"); 4581 4582 /* 4583 * Get file structures globally. 4584 */ 4585 static int 4586 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4587 { 4588 struct xfile xf; 4589 struct filedesc *fdp; 4590 struct file *fp; 4591 struct proc *p; 4592 int error, n; 4593 4594 error = sysctl_wire_old_buffer(req, 0); 4595 if (error != 0) 4596 return (error); 4597 if (req->oldptr == NULL) { 4598 n = 0; 4599 sx_slock(&allproc_lock); 4600 FOREACH_PROC_IN_SYSTEM(p) { 4601 PROC_LOCK(p); 4602 if (p->p_state == PRS_NEW) { 4603 PROC_UNLOCK(p); 4604 continue; 4605 } 4606 fdp = fdhold(p); 4607 PROC_UNLOCK(p); 4608 if (fdp == NULL) 4609 continue; 4610 /* overestimates sparse tables. */ 4611 n += fdp->fd_nfiles; 4612 fddrop(fdp); 4613 } 4614 sx_sunlock(&allproc_lock); 4615 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4616 } 4617 error = 0; 4618 bzero(&xf, sizeof(xf)); 4619 xf.xf_size = sizeof(xf); 4620 sx_slock(&allproc_lock); 4621 FOREACH_PROC_IN_SYSTEM(p) { 4622 PROC_LOCK(p); 4623 if (p->p_state == PRS_NEW) { 4624 PROC_UNLOCK(p); 4625 continue; 4626 } 4627 if (p_cansee(req->td, p) != 0) { 4628 PROC_UNLOCK(p); 4629 continue; 4630 } 4631 xf.xf_pid = p->p_pid; 4632 xf.xf_uid = p->p_ucred->cr_uid; 4633 fdp = fdhold(p); 4634 PROC_UNLOCK(p); 4635 if (fdp == NULL) 4636 continue; 4637 FILEDESC_SLOCK(fdp); 4638 if (refcount_load(&fdp->fd_refcnt) == 0) 4639 goto nextproc; 4640 FILEDESC_FOREACH_FP(fdp, n, fp) { 4641 xf.xf_fd = n; 4642 xf.xf_file = (uintptr_t)fp; 4643 xf.xf_data = (uintptr_t)fp->f_data; 4644 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4645 xf.xf_type = (uintptr_t)fp->f_type; 4646 xf.xf_count = refcount_load(&fp->f_count); 4647 xf.xf_msgcount = 0; 4648 xf.xf_offset = foffset_get(fp); 4649 xf.xf_flag = fp->f_flag; 4650 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4651 4652 /* 4653 * There is no need to re-check the fdtable refcount 4654 * here since the filedesc lock is not dropped in the 4655 * loop body. 4656 */ 4657 if (error != 0) 4658 break; 4659 } 4660 nextproc: 4661 FILEDESC_SUNLOCK(fdp); 4662 fddrop(fdp); 4663 if (error) 4664 break; 4665 } 4666 sx_sunlock(&allproc_lock); 4667 return (error); 4668 } 4669 4670 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4671 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4672 4673 #ifdef KINFO_FILE_SIZE 4674 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4675 #endif 4676 4677 static int 4678 xlate_fflags(int fflags) 4679 { 4680 static const struct { 4681 int fflag; 4682 int kf_fflag; 4683 } fflags_table[] = { 4684 { FAPPEND, KF_FLAG_APPEND }, 4685 { FASYNC, KF_FLAG_ASYNC }, 4686 { FFSYNC, KF_FLAG_FSYNC }, 4687 { FHASLOCK, KF_FLAG_HASLOCK }, 4688 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4689 { FREAD, KF_FLAG_READ }, 4690 { FWRITE, KF_FLAG_WRITE }, 4691 { O_CREAT, KF_FLAG_CREAT }, 4692 { O_DIRECT, KF_FLAG_DIRECT }, 4693 { O_EXCL, KF_FLAG_EXCL }, 4694 { O_EXEC, KF_FLAG_EXEC }, 4695 { O_EXLOCK, KF_FLAG_EXLOCK }, 4696 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4697 { O_SHLOCK, KF_FLAG_SHLOCK }, 4698 { O_TRUNC, KF_FLAG_TRUNC } 4699 }; 4700 unsigned int i; 4701 int kflags; 4702 4703 kflags = 0; 4704 for (i = 0; i < nitems(fflags_table); i++) 4705 if (fflags & fflags_table[i].fflag) 4706 kflags |= fflags_table[i].kf_fflag; 4707 return (kflags); 4708 } 4709 4710 /* Trim unused data from kf_path by truncating the structure size. */ 4711 void 4712 pack_kinfo(struct kinfo_file *kif) 4713 { 4714 4715 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4716 strlen(kif->kf_path) + 1; 4717 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4718 } 4719 4720 static void 4721 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4722 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4723 { 4724 int error; 4725 4726 bzero(kif, sizeof(*kif)); 4727 4728 /* Set a default type to allow for empty fill_kinfo() methods. */ 4729 kif->kf_type = KF_TYPE_UNKNOWN; 4730 kif->kf_flags = xlate_fflags(fp->f_flag); 4731 if (rightsp != NULL) 4732 kif->kf_cap_rights = *rightsp; 4733 else 4734 cap_rights_init_zero(&kif->kf_cap_rights); 4735 kif->kf_fd = fd; 4736 kif->kf_ref_count = refcount_load(&fp->f_count); 4737 kif->kf_offset = foffset_get(fp); 4738 4739 /* 4740 * This may drop the filedesc lock, so the 'fp' cannot be 4741 * accessed after this call. 4742 */ 4743 error = fo_fill_kinfo(fp, kif, fdp); 4744 if (error == 0) 4745 kif->kf_status |= KF_ATTR_VALID; 4746 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4747 pack_kinfo(kif); 4748 else 4749 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4750 } 4751 4752 static void 4753 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4754 struct kinfo_file *kif, int flags) 4755 { 4756 int error; 4757 4758 bzero(kif, sizeof(*kif)); 4759 4760 kif->kf_type = KF_TYPE_VNODE; 4761 error = vn_fill_kinfo_vnode(vp, kif); 4762 if (error == 0) 4763 kif->kf_status |= KF_ATTR_VALID; 4764 kif->kf_flags = xlate_fflags(fflags); 4765 cap_rights_init_zero(&kif->kf_cap_rights); 4766 kif->kf_fd = fd; 4767 kif->kf_ref_count = -1; 4768 kif->kf_offset = -1; 4769 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4770 pack_kinfo(kif); 4771 else 4772 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4773 vrele(vp); 4774 } 4775 4776 struct export_fd_buf { 4777 struct filedesc *fdp; 4778 struct pwddesc *pdp; 4779 struct sbuf *sb; 4780 ssize_t remainder; 4781 struct kinfo_file kif; 4782 int flags; 4783 }; 4784 4785 static int 4786 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4787 { 4788 struct kinfo_file *kif; 4789 4790 kif = &efbuf->kif; 4791 if (efbuf->remainder != -1) { 4792 if (efbuf->remainder < kif->kf_structsize) 4793 return (ENOMEM); 4794 efbuf->remainder -= kif->kf_structsize; 4795 } 4796 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0) 4797 return (sbuf_error(efbuf->sb)); 4798 return (0); 4799 } 4800 4801 static int 4802 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4803 struct export_fd_buf *efbuf) 4804 { 4805 int error; 4806 4807 if (efbuf->remainder == 0) 4808 return (ENOMEM); 4809 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4810 efbuf->flags); 4811 FILEDESC_SUNLOCK(efbuf->fdp); 4812 error = export_kinfo_to_sb(efbuf); 4813 FILEDESC_SLOCK(efbuf->fdp); 4814 return (error); 4815 } 4816 4817 static int 4818 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4819 struct export_fd_buf *efbuf) 4820 { 4821 int error; 4822 4823 if (efbuf->remainder == 0) 4824 return (ENOMEM); 4825 if (efbuf->pdp != NULL) 4826 PWDDESC_XUNLOCK(efbuf->pdp); 4827 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4828 error = export_kinfo_to_sb(efbuf); 4829 if (efbuf->pdp != NULL) 4830 PWDDESC_XLOCK(efbuf->pdp); 4831 return (error); 4832 } 4833 4834 /* 4835 * Store a process file descriptor information to sbuf. 4836 * 4837 * Takes a locked proc as argument, and returns with the proc unlocked. 4838 */ 4839 int 4840 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4841 int flags) 4842 { 4843 struct file *fp; 4844 struct filedesc *fdp; 4845 struct pwddesc *pdp; 4846 struct export_fd_buf *efbuf; 4847 struct vnode *cttyvp, *textvp, *tracevp; 4848 struct pwd *pwd; 4849 int error, i; 4850 cap_rights_t rights; 4851 4852 PROC_LOCK_ASSERT(p, MA_OWNED); 4853 4854 /* ktrace vnode */ 4855 tracevp = ktr_get_tracevp(p, true); 4856 /* text vnode */ 4857 textvp = p->p_textvp; 4858 if (textvp != NULL) 4859 vrefact(textvp); 4860 /* Controlling tty. */ 4861 cttyvp = NULL; 4862 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 4863 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 4864 if (cttyvp != NULL) 4865 vrefact(cttyvp); 4866 } 4867 fdp = fdhold(p); 4868 pdp = pdhold(p); 4869 PROC_UNLOCK(p); 4870 4871 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4872 efbuf->fdp = NULL; 4873 efbuf->pdp = NULL; 4874 efbuf->sb = sb; 4875 efbuf->remainder = maxlen; 4876 efbuf->flags = flags; 4877 4878 error = 0; 4879 if (tracevp != NULL) 4880 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, 4881 FREAD | FWRITE, efbuf); 4882 if (error == 0 && textvp != NULL) 4883 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, 4884 efbuf); 4885 if (error == 0 && cttyvp != NULL) 4886 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, 4887 FREAD | FWRITE, efbuf); 4888 if (error != 0 || pdp == NULL || fdp == NULL) 4889 goto fail; 4890 efbuf->fdp = fdp; 4891 efbuf->pdp = pdp; 4892 PWDDESC_XLOCK(pdp); 4893 pwd = pwd_hold_pwddesc(pdp); 4894 if (pwd != NULL) { 4895 /* working directory */ 4896 if (pwd->pwd_cdir != NULL) { 4897 vrefact(pwd->pwd_cdir); 4898 error = export_vnode_to_sb(pwd->pwd_cdir, 4899 KF_FD_TYPE_CWD, FREAD, efbuf); 4900 } 4901 /* root directory */ 4902 if (error == 0 && pwd->pwd_rdir != NULL) { 4903 vrefact(pwd->pwd_rdir); 4904 error = export_vnode_to_sb(pwd->pwd_rdir, 4905 KF_FD_TYPE_ROOT, FREAD, efbuf); 4906 } 4907 /* jail directory */ 4908 if (error == 0 && pwd->pwd_jdir != NULL) { 4909 vrefact(pwd->pwd_jdir); 4910 error = export_vnode_to_sb(pwd->pwd_jdir, 4911 KF_FD_TYPE_JAIL, FREAD, efbuf); 4912 } 4913 } 4914 PWDDESC_XUNLOCK(pdp); 4915 if (error != 0) 4916 goto fail; 4917 if (pwd != NULL) 4918 pwd_drop(pwd); 4919 FILEDESC_SLOCK(fdp); 4920 if (refcount_load(&fdp->fd_refcnt) == 0) 4921 goto skip; 4922 FILEDESC_FOREACH_FP(fdp, i, fp) { 4923 #ifdef CAPABILITIES 4924 rights = *cap_rights(fdp, i); 4925 #else /* !CAPABILITIES */ 4926 rights = cap_no_rights; 4927 #endif 4928 /* 4929 * Create sysctl entry. It is OK to drop the filedesc 4930 * lock inside of export_file_to_sb() as we will 4931 * re-validate and re-evaluate its properties when the 4932 * loop continues. 4933 */ 4934 error = export_file_to_sb(fp, i, &rights, efbuf); 4935 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4936 break; 4937 } 4938 skip: 4939 FILEDESC_SUNLOCK(fdp); 4940 fail: 4941 if (fdp != NULL) 4942 fddrop(fdp); 4943 if (pdp != NULL) 4944 pddrop(pdp); 4945 free(efbuf, M_TEMP); 4946 return (error); 4947 } 4948 4949 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 4950 4951 /* 4952 * Get per-process file descriptors for use by procstat(1), et al. 4953 */ 4954 static int 4955 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 4956 { 4957 struct sbuf sb; 4958 struct proc *p; 4959 ssize_t maxlen; 4960 u_int namelen; 4961 int error, error2, *name; 4962 4963 namelen = arg2; 4964 if (namelen != 1) 4965 return (EINVAL); 4966 4967 name = (int *)arg1; 4968 4969 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 4970 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4971 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4972 if (error != 0) { 4973 sbuf_delete(&sb); 4974 return (error); 4975 } 4976 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4977 error = kern_proc_filedesc_out(p, &sb, maxlen, 4978 KERN_FILEDESC_PACK_KINFO); 4979 error2 = sbuf_finish(&sb); 4980 sbuf_delete(&sb); 4981 return (error != 0 ? error : error2); 4982 } 4983 4984 #ifdef COMPAT_FREEBSD7 4985 #ifdef KINFO_OFILE_SIZE 4986 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 4987 #endif 4988 4989 static void 4990 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 4991 { 4992 4993 okif->kf_structsize = sizeof(*okif); 4994 okif->kf_type = kif->kf_type; 4995 okif->kf_fd = kif->kf_fd; 4996 okif->kf_ref_count = kif->kf_ref_count; 4997 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 4998 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 4999 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 5000 okif->kf_offset = kif->kf_offset; 5001 if (kif->kf_type == KF_TYPE_VNODE) 5002 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 5003 else 5004 okif->kf_vnode_type = KF_VTYPE_VNON; 5005 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 5006 if (kif->kf_type == KF_TYPE_SOCKET) { 5007 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 5008 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 5009 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 5010 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 5011 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 5012 } else { 5013 okif->kf_sa_local.ss_family = AF_UNSPEC; 5014 okif->kf_sa_peer.ss_family = AF_UNSPEC; 5015 } 5016 } 5017 5018 static int 5019 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 5020 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 5021 { 5022 int error; 5023 5024 vrefact(vp); 5025 PWDDESC_XUNLOCK(pdp); 5026 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 5027 kinfo_to_okinfo(kif, okif); 5028 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5029 PWDDESC_XLOCK(pdp); 5030 return (error); 5031 } 5032 5033 /* 5034 * Get per-process file descriptors for use by procstat(1), et al. 5035 */ 5036 static int 5037 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 5038 { 5039 struct kinfo_ofile *okif; 5040 struct kinfo_file *kif; 5041 struct filedesc *fdp; 5042 struct pwddesc *pdp; 5043 struct pwd *pwd; 5044 u_int namelen; 5045 int error, i, *name; 5046 struct file *fp; 5047 struct proc *p; 5048 5049 namelen = arg2; 5050 if (namelen != 1) 5051 return (EINVAL); 5052 5053 name = (int *)arg1; 5054 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5055 if (error != 0) 5056 return (error); 5057 fdp = fdhold(p); 5058 if (fdp != NULL) 5059 pdp = pdhold(p); 5060 PROC_UNLOCK(p); 5061 if (fdp == NULL || pdp == NULL) { 5062 if (fdp != NULL) 5063 fddrop(fdp); 5064 return (ENOENT); 5065 } 5066 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 5067 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 5068 PWDDESC_XLOCK(pdp); 5069 pwd = pwd_hold_pwddesc(pdp); 5070 if (pwd != NULL) { 5071 if (pwd->pwd_cdir != NULL) 5072 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 5073 okif, pdp, req); 5074 if (pwd->pwd_rdir != NULL) 5075 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 5076 okif, pdp, req); 5077 if (pwd->pwd_jdir != NULL) 5078 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 5079 okif, pdp, req); 5080 } 5081 PWDDESC_XUNLOCK(pdp); 5082 if (pwd != NULL) 5083 pwd_drop(pwd); 5084 FILEDESC_SLOCK(fdp); 5085 if (refcount_load(&fdp->fd_refcnt) == 0) 5086 goto skip; 5087 FILEDESC_FOREACH_FP(fdp, i, fp) { 5088 export_file_to_kinfo(fp, i, NULL, kif, fdp, 5089 KERN_FILEDESC_PACK_KINFO); 5090 FILEDESC_SUNLOCK(fdp); 5091 kinfo_to_okinfo(kif, okif); 5092 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5093 FILEDESC_SLOCK(fdp); 5094 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 5095 break; 5096 } 5097 skip: 5098 FILEDESC_SUNLOCK(fdp); 5099 fddrop(fdp); 5100 pddrop(pdp); 5101 free(kif, M_TEMP); 5102 free(okif, M_TEMP); 5103 return (0); 5104 } 5105 5106 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 5107 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 5108 "Process ofiledesc entries"); 5109 #endif /* COMPAT_FREEBSD7 */ 5110 5111 int 5112 vntype_to_kinfo(int vtype) 5113 { 5114 struct { 5115 int vtype; 5116 int kf_vtype; 5117 } vtypes_table[] = { 5118 { VBAD, KF_VTYPE_VBAD }, 5119 { VBLK, KF_VTYPE_VBLK }, 5120 { VCHR, KF_VTYPE_VCHR }, 5121 { VDIR, KF_VTYPE_VDIR }, 5122 { VFIFO, KF_VTYPE_VFIFO }, 5123 { VLNK, KF_VTYPE_VLNK }, 5124 { VNON, KF_VTYPE_VNON }, 5125 { VREG, KF_VTYPE_VREG }, 5126 { VSOCK, KF_VTYPE_VSOCK } 5127 }; 5128 unsigned int i; 5129 5130 /* 5131 * Perform vtype translation. 5132 */ 5133 for (i = 0; i < nitems(vtypes_table); i++) 5134 if (vtypes_table[i].vtype == vtype) 5135 return (vtypes_table[i].kf_vtype); 5136 5137 return (KF_VTYPE_UNKNOWN); 5138 } 5139 5140 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 5141 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 5142 "Process filedesc entries"); 5143 5144 /* 5145 * Store a process current working directory information to sbuf. 5146 * 5147 * Takes a locked proc as argument, and returns with the proc unlocked. 5148 */ 5149 int 5150 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 5151 { 5152 struct pwddesc *pdp; 5153 struct pwd *pwd; 5154 struct export_fd_buf *efbuf; 5155 struct vnode *cdir; 5156 int error; 5157 5158 PROC_LOCK_ASSERT(p, MA_OWNED); 5159 5160 pdp = pdhold(p); 5161 PROC_UNLOCK(p); 5162 if (pdp == NULL) 5163 return (EINVAL); 5164 5165 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 5166 efbuf->fdp = NULL; 5167 efbuf->pdp = pdp; 5168 efbuf->sb = sb; 5169 efbuf->remainder = maxlen; 5170 efbuf->flags = 0; 5171 5172 PWDDESC_XLOCK(pdp); 5173 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 5174 cdir = pwd->pwd_cdir; 5175 if (cdir == NULL) { 5176 error = EINVAL; 5177 } else { 5178 vrefact(cdir); 5179 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 5180 } 5181 PWDDESC_XUNLOCK(pdp); 5182 pddrop(pdp); 5183 free(efbuf, M_TEMP); 5184 return (error); 5185 } 5186 5187 /* 5188 * Get per-process current working directory. 5189 */ 5190 static int 5191 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 5192 { 5193 struct sbuf sb; 5194 struct proc *p; 5195 ssize_t maxlen; 5196 u_int namelen; 5197 int error, error2, *name; 5198 5199 namelen = arg2; 5200 if (namelen != 1) 5201 return (EINVAL); 5202 5203 name = (int *)arg1; 5204 5205 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 5206 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5207 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5208 if (error != 0) { 5209 sbuf_delete(&sb); 5210 return (error); 5211 } 5212 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5213 error = kern_proc_cwd_out(p, &sb, maxlen); 5214 error2 = sbuf_finish(&sb); 5215 sbuf_delete(&sb); 5216 return (error != 0 ? error : error2); 5217 } 5218 5219 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 5220 sysctl_kern_proc_cwd, "Process current working directory"); 5221 5222 #ifdef DDB 5223 /* 5224 * For the purposes of debugging, generate a human-readable string for the 5225 * file type. 5226 */ 5227 static const char * 5228 file_type_to_name(short type) 5229 { 5230 5231 switch (type) { 5232 case 0: 5233 return ("zero"); 5234 case DTYPE_VNODE: 5235 return ("vnode"); 5236 case DTYPE_SOCKET: 5237 return ("socket"); 5238 case DTYPE_PIPE: 5239 return ("pipe"); 5240 case DTYPE_FIFO: 5241 return ("fifo"); 5242 case DTYPE_KQUEUE: 5243 return ("kqueue"); 5244 case DTYPE_CRYPTO: 5245 return ("crypto"); 5246 case DTYPE_MQUEUE: 5247 return ("mqueue"); 5248 case DTYPE_SHM: 5249 return ("shm"); 5250 case DTYPE_SEM: 5251 return ("ksem"); 5252 case DTYPE_PTS: 5253 return ("pts"); 5254 case DTYPE_DEV: 5255 return ("dev"); 5256 case DTYPE_PROCDESC: 5257 return ("proc"); 5258 case DTYPE_EVENTFD: 5259 return ("eventfd"); 5260 case DTYPE_TIMERFD: 5261 return ("timerfd"); 5262 case DTYPE_JAILDESC: 5263 return ("jail"); 5264 default: 5265 return ("unkn"); 5266 } 5267 } 5268 5269 /* 5270 * For the purposes of debugging, identify a process (if any, perhaps one of 5271 * many) that references the passed file in its file descriptor array. Return 5272 * NULL if none. 5273 */ 5274 static struct proc * 5275 file_to_first_proc(struct file *fp) 5276 { 5277 struct filedesc *fdp; 5278 struct proc *p; 5279 int n; 5280 5281 FOREACH_PROC_IN_SYSTEM(p) { 5282 if (p->p_state == PRS_NEW) 5283 continue; 5284 fdp = p->p_fd; 5285 if (fdp == NULL) 5286 continue; 5287 for (n = 0; n < fdp->fd_nfiles; n++) { 5288 if (fp == fdp->fd_ofiles[n].fde_file) 5289 return (p); 5290 } 5291 } 5292 return (NULL); 5293 } 5294 5295 static void 5296 db_print_file(struct file *fp, int header) 5297 { 5298 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 5299 struct proc *p; 5300 5301 if (header) 5302 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 5303 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 5304 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 5305 "FCmd"); 5306 p = file_to_first_proc(fp); 5307 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 5308 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 5309 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 5310 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 5311 5312 #undef XPTRWIDTH 5313 } 5314 5315 DB_SHOW_COMMAND(file, db_show_file) 5316 { 5317 struct file *fp; 5318 5319 if (!have_addr) { 5320 db_printf("usage: show file <addr>\n"); 5321 return; 5322 } 5323 fp = (struct file *)addr; 5324 db_print_file(fp, 1); 5325 } 5326 5327 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE) 5328 { 5329 struct filedesc *fdp; 5330 struct file *fp; 5331 struct proc *p; 5332 int header; 5333 int n; 5334 5335 header = 1; 5336 FOREACH_PROC_IN_SYSTEM(p) { 5337 if (p->p_state == PRS_NEW) 5338 continue; 5339 if ((fdp = p->p_fd) == NULL) 5340 continue; 5341 for (n = 0; n < fdp->fd_nfiles; ++n) { 5342 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 5343 continue; 5344 db_print_file(fp, header); 5345 header = 0; 5346 } 5347 } 5348 } 5349 #endif 5350 5351 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, 5352 CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5353 &maxfilesperproc, 0, "Maximum files allowed open per process"); 5354 5355 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5356 &maxfiles, 0, "Maximum number of files"); 5357 5358 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 5359 &openfiles, 0, "System-wide number of open files"); 5360 5361 /* ARGSUSED*/ 5362 static void 5363 filelistinit(void *dummy) 5364 { 5365 5366 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 5367 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 5368 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 5369 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 5370 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 5371 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 5372 /* 5373 * XXXMJG this is a temporary hack due to boot ordering issues against 5374 * the vnode zone. 5375 */ 5376 vfs_smr = uma_zone_get_smr(pwd_zone); 5377 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 5378 } 5379 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 5380 5381 /*-------------------------------------------------------------------*/ 5382 5383 static int 5384 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 5385 int flags, struct thread *td) 5386 { 5387 5388 return (EBADF); 5389 } 5390 5391 static int 5392 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5393 struct thread *td) 5394 { 5395 5396 return (EINVAL); 5397 } 5398 5399 static int 5400 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 5401 struct thread *td) 5402 { 5403 5404 return (EBADF); 5405 } 5406 5407 static int 5408 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 5409 struct thread *td) 5410 { 5411 5412 return (0); 5413 } 5414 5415 static int 5416 badfo_kqfilter(struct file *fp, struct knote *kn) 5417 { 5418 5419 return (EBADF); 5420 } 5421 5422 static int 5423 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 5424 { 5425 5426 return (EBADF); 5427 } 5428 5429 static int 5430 badfo_close(struct file *fp, struct thread *td) 5431 { 5432 5433 return (0); 5434 } 5435 5436 static int 5437 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5438 struct thread *td) 5439 { 5440 5441 return (EBADF); 5442 } 5443 5444 static int 5445 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5446 struct thread *td) 5447 { 5448 5449 return (EBADF); 5450 } 5451 5452 static int 5453 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5454 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5455 struct thread *td) 5456 { 5457 5458 return (EBADF); 5459 } 5460 5461 static int 5462 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 5463 { 5464 5465 return (0); 5466 } 5467 5468 const struct fileops badfileops = { 5469 .fo_read = badfo_readwrite, 5470 .fo_write = badfo_readwrite, 5471 .fo_truncate = badfo_truncate, 5472 .fo_ioctl = badfo_ioctl, 5473 .fo_poll = badfo_poll, 5474 .fo_kqfilter = badfo_kqfilter, 5475 .fo_stat = badfo_stat, 5476 .fo_close = badfo_close, 5477 .fo_chmod = badfo_chmod, 5478 .fo_chown = badfo_chown, 5479 .fo_sendfile = badfo_sendfile, 5480 .fo_fill_kinfo = badfo_fill_kinfo, 5481 }; 5482 5483 static int 5484 path_poll(struct file *fp, int events, struct ucred *active_cred, 5485 struct thread *td) 5486 { 5487 return (POLLNVAL); 5488 } 5489 5490 static int 5491 path_close(struct file *fp, struct thread *td) 5492 { 5493 MPASS(fp->f_type == DTYPE_VNODE); 5494 fp->f_ops = &badfileops; 5495 vrele(fp->f_vnode); 5496 return (0); 5497 } 5498 5499 const struct fileops path_fileops = { 5500 .fo_read = badfo_readwrite, 5501 .fo_write = badfo_readwrite, 5502 .fo_truncate = badfo_truncate, 5503 .fo_ioctl = badfo_ioctl, 5504 .fo_poll = path_poll, 5505 .fo_kqfilter = vn_kqfilter_opath, 5506 .fo_stat = vn_statfile, 5507 .fo_close = path_close, 5508 .fo_chmod = badfo_chmod, 5509 .fo_chown = badfo_chown, 5510 .fo_sendfile = badfo_sendfile, 5511 .fo_fill_kinfo = vn_fill_kinfo, 5512 .fo_cmp = vn_cmp, 5513 .fo_flags = DFLAG_PASSABLE, 5514 }; 5515 5516 int 5517 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5518 int flags, struct thread *td) 5519 { 5520 5521 return (EOPNOTSUPP); 5522 } 5523 5524 int 5525 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5526 struct thread *td) 5527 { 5528 5529 return (EINVAL); 5530 } 5531 5532 int 5533 invfo_ioctl(struct file *fp, u_long com, void *data, 5534 struct ucred *active_cred, struct thread *td) 5535 { 5536 5537 return (ENOTTY); 5538 } 5539 5540 int 5541 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5542 struct thread *td) 5543 { 5544 5545 return (poll_no_poll(events)); 5546 } 5547 5548 int 5549 invfo_kqfilter(struct file *fp, struct knote *kn) 5550 { 5551 5552 return (EINVAL); 5553 } 5554 5555 int 5556 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5557 struct thread *td) 5558 { 5559 5560 return (EINVAL); 5561 } 5562 5563 int 5564 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5565 struct thread *td) 5566 { 5567 5568 return (EINVAL); 5569 } 5570 5571 int 5572 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5573 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5574 struct thread *td) 5575 { 5576 5577 return (EINVAL); 5578 } 5579 5580 /*-------------------------------------------------------------------*/ 5581 5582 /* 5583 * File Descriptor pseudo-device driver (/dev/fd/). 5584 * 5585 * Opening minor device N dup()s the file (if any) connected to file 5586 * descriptor N belonging to the calling process. Note that this driver 5587 * consists of only the ``open()'' routine, because all subsequent 5588 * references to this file will be direct to the other driver. 5589 * 5590 * XXX: we could give this one a cloning event handler if necessary. 5591 */ 5592 5593 /* ARGSUSED */ 5594 static int 5595 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5596 { 5597 5598 /* 5599 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5600 * the file descriptor being sought for duplication. The error 5601 * return ensures that the vnode for this device will be released 5602 * by vn_open. Open will detect this special error and take the 5603 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5604 * will simply report the error. 5605 */ 5606 td->td_dupfd = dev2unit(dev); 5607 return (ENODEV); 5608 } 5609 5610 static struct cdevsw fildesc_cdevsw = { 5611 .d_version = D_VERSION, 5612 .d_open = fdopen, 5613 .d_name = "FD", 5614 }; 5615 5616 static void 5617 fildesc_drvinit(void *unused) 5618 { 5619 struct cdev *dev; 5620 5621 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5622 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5623 make_dev_alias(dev, "stdin"); 5624 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5625 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5626 make_dev_alias(dev, "stdout"); 5627 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5628 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5629 make_dev_alias(dev, "stderr"); 5630 } 5631 5632 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5633