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 */ 197 198 /* 199 * If low >= size, just return low. Otherwise find the first zero bit in the 200 * given bitmap, starting at low and not exceeding size - 1. Return size if 201 * not found. 202 */ 203 static int 204 fd_first_free(struct filedesc *fdp, int low, int size) 205 { 206 NDSLOTTYPE *map = fdp->fd_map; 207 NDSLOTTYPE mask; 208 int off, maxoff; 209 210 if (low >= size) 211 return (low); 212 213 off = NDSLOT(low); 214 if (low % NDENTRIES) { 215 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 216 if ((mask &= ~map[off]) != 0UL) 217 return (off * NDENTRIES + ffsl(mask) - 1); 218 ++off; 219 } 220 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 221 if (map[off] != ~0UL) 222 return (off * NDENTRIES + ffsl(~map[off]) - 1); 223 return (size); 224 } 225 226 /* 227 * Find the last used fd. 228 * 229 * Call this variant if fdp can't be modified by anyone else (e.g, during exec). 230 * Otherwise use fdlastfile. 231 */ 232 int 233 fdlastfile_single(struct filedesc *fdp) 234 { 235 NDSLOTTYPE *map = fdp->fd_map; 236 int off, minoff; 237 238 off = NDSLOT(fdp->fd_nfiles - 1); 239 for (minoff = NDSLOT(0); off >= minoff; --off) 240 if (map[off] != 0) 241 return (off * NDENTRIES + flsl(map[off]) - 1); 242 return (-1); 243 } 244 245 int 246 fdlastfile(struct filedesc *fdp) 247 { 248 249 FILEDESC_LOCK_ASSERT(fdp); 250 return (fdlastfile_single(fdp)); 251 } 252 253 static int 254 fdisused(struct filedesc *fdp, int fd) 255 { 256 257 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 258 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 259 260 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 261 } 262 263 /* 264 * Mark a file descriptor as used. 265 */ 266 static void 267 fdused_init(struct filedesc *fdp, int fd) 268 { 269 270 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 271 272 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 273 } 274 275 static void 276 fdused(struct filedesc *fdp, int fd) 277 { 278 279 FILEDESC_XLOCK_ASSERT(fdp); 280 281 fdused_init(fdp, fd); 282 if (fd == fdp->fd_freefile) 283 fdp->fd_freefile++; 284 } 285 286 /* 287 * Mark a file descriptor as unused. 288 */ 289 static void 290 fdunused(struct filedesc *fdp, int fd) 291 { 292 293 FILEDESC_XLOCK_ASSERT(fdp); 294 295 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 296 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 297 ("fd=%d is still in use", fd)); 298 299 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 300 if (fd < fdp->fd_freefile) 301 fdp->fd_freefile = fd; 302 } 303 304 /* 305 * Free a file descriptor. 306 * 307 * Avoid some work if fdp is about to be destroyed. 308 */ 309 static inline void 310 fdefree_last(struct filedescent *fde) 311 { 312 313 filecaps_free(&fde->fde_caps); 314 } 315 316 static inline void 317 fdfree(struct filedesc *fdp, int fd) 318 { 319 struct filedescent *fde; 320 321 FILEDESC_XLOCK_ASSERT(fdp); 322 fde = &fdp->fd_ofiles[fd]; 323 #ifdef CAPABILITIES 324 seqc_write_begin(&fde->fde_seqc); 325 #endif 326 fde->fde_file = NULL; 327 #ifdef CAPABILITIES 328 seqc_write_end(&fde->fde_seqc); 329 #endif 330 fdefree_last(fde); 331 fdunused(fdp, fd); 332 } 333 334 /* 335 * System calls on descriptors. 336 */ 337 #ifndef _SYS_SYSPROTO_H_ 338 struct getdtablesize_args { 339 int dummy; 340 }; 341 #endif 342 /* ARGSUSED */ 343 int 344 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 345 { 346 #ifdef RACCT 347 uint64_t lim; 348 #endif 349 350 td->td_retval[0] = getmaxfd(td); 351 #ifdef RACCT 352 PROC_LOCK(td->td_proc); 353 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 354 PROC_UNLOCK(td->td_proc); 355 if (lim < td->td_retval[0]) 356 td->td_retval[0] = lim; 357 #endif 358 return (0); 359 } 360 361 /* 362 * Duplicate a file descriptor to a particular value. 363 * 364 * Note: keep in mind that a potential race condition exists when closing 365 * descriptors from a shared descriptor table (via rfork). 366 */ 367 #ifndef _SYS_SYSPROTO_H_ 368 struct dup2_args { 369 u_int from; 370 u_int to; 371 }; 372 #endif 373 /* ARGSUSED */ 374 int 375 sys_dup2(struct thread *td, struct dup2_args *uap) 376 { 377 378 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 379 } 380 381 /* 382 * Duplicate a file descriptor. 383 */ 384 #ifndef _SYS_SYSPROTO_H_ 385 struct dup_args { 386 u_int fd; 387 }; 388 #endif 389 /* ARGSUSED */ 390 int 391 sys_dup(struct thread *td, struct dup_args *uap) 392 { 393 394 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 395 } 396 397 /* 398 * The file control system call. 399 */ 400 #ifndef _SYS_SYSPROTO_H_ 401 struct fcntl_args { 402 int fd; 403 int cmd; 404 long arg; 405 }; 406 #endif 407 /* ARGSUSED */ 408 int 409 sys_fcntl(struct thread *td, struct fcntl_args *uap) 410 { 411 412 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 413 } 414 415 int 416 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, intptr_t arg) 417 { 418 struct flock fl; 419 struct __oflock ofl; 420 intptr_t arg1; 421 int error, newcmd; 422 423 error = 0; 424 newcmd = cmd; 425 switch (cmd) { 426 case F_OGETLK: 427 case F_OSETLK: 428 case F_OSETLKW: 429 /* 430 * Convert old flock structure to new. 431 */ 432 error = copyin((void *)arg, &ofl, sizeof(ofl)); 433 fl.l_start = ofl.l_start; 434 fl.l_len = ofl.l_len; 435 fl.l_pid = ofl.l_pid; 436 fl.l_type = ofl.l_type; 437 fl.l_whence = ofl.l_whence; 438 fl.l_sysid = 0; 439 440 switch (cmd) { 441 case F_OGETLK: 442 newcmd = F_GETLK; 443 break; 444 case F_OSETLK: 445 newcmd = F_SETLK; 446 break; 447 case F_OSETLKW: 448 newcmd = F_SETLKW; 449 break; 450 } 451 arg1 = (intptr_t)&fl; 452 break; 453 case F_GETLK: 454 case F_SETLK: 455 case F_SETLKW: 456 case F_SETLK_REMOTE: 457 error = copyin((void *)arg, &fl, sizeof(fl)); 458 arg1 = (intptr_t)&fl; 459 break; 460 default: 461 arg1 = arg; 462 break; 463 } 464 if (error) 465 return (error); 466 error = kern_fcntl(td, fd, newcmd, arg1); 467 if (error) 468 return (error); 469 if (cmd == F_OGETLK) { 470 ofl.l_start = fl.l_start; 471 ofl.l_len = fl.l_len; 472 ofl.l_pid = fl.l_pid; 473 ofl.l_type = fl.l_type; 474 ofl.l_whence = fl.l_whence; 475 error = copyout(&ofl, (void *)arg, sizeof(ofl)); 476 } else if (cmd == F_GETLK) { 477 error = copyout(&fl, (void *)arg, sizeof(fl)); 478 } 479 return (error); 480 } 481 482 struct flags_trans_elem { 483 u_int f; 484 u_int t; 485 }; 486 487 static u_int 488 flags_trans(const struct flags_trans_elem *ftes, int nitems, u_int from_flags) 489 { 490 u_int res; 491 int i; 492 493 res = 0; 494 for (i = 0; i < nitems; i++) { 495 if ((from_flags & ftes[i].f) != 0) 496 res |= ftes[i].t; 497 } 498 return (res); 499 } 500 501 static uint8_t 502 fd_to_fde_flags(int fd_flags) 503 { 504 static const struct flags_trans_elem fd_to_fde_flags_s[] = { 505 { .f = FD_CLOEXEC, .t = UF_EXCLOSE }, 506 { .f = FD_CLOFORK, .t = UF_FOCLOSE }, 507 { .f = FD_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH }, 508 }; 509 510 return (flags_trans(fd_to_fde_flags_s, nitems(fd_to_fde_flags_s), 511 fd_flags)); 512 } 513 514 static int 515 fde_to_fd_flags(uint8_t fde_flags) 516 { 517 static const struct flags_trans_elem fde_to_fd_flags_s[] = { 518 { .f = UF_EXCLOSE, .t = FD_CLOEXEC }, 519 { .f = UF_FOCLOSE, .t = FD_CLOFORK }, 520 { .f = UF_RESOLVE_BENEATH, .t = FD_RESOLVE_BENEATH }, 521 }; 522 523 return (flags_trans(fde_to_fd_flags_s, nitems(fde_to_fd_flags_s), 524 fde_flags)); 525 } 526 527 static uint8_t 528 fddup_to_fde_flags(int fddup_flags) 529 { 530 static const struct flags_trans_elem fddup_to_fde_flags_s[] = { 531 { .f = FDDUP_FLAG_CLOEXEC, .t = UF_EXCLOSE }, 532 { .f = FDDUP_FLAG_CLOFORK, .t = UF_FOCLOSE }, 533 }; 534 535 return (flags_trans(fddup_to_fde_flags_s, nitems(fddup_to_fde_flags_s), 536 fddup_flags)); 537 } 538 539 static uint8_t 540 close_range_to_fde_flags(int close_range_flags) 541 { 542 static const struct flags_trans_elem close_range_to_fde_flags_s[] = { 543 { .f = CLOSE_RANGE_CLOEXEC, .t = UF_EXCLOSE }, 544 { .f = CLOSE_RANGE_CLOFORK, .t = UF_FOCLOSE }, 545 }; 546 547 return (flags_trans(close_range_to_fde_flags_s, 548 nitems(close_range_to_fde_flags_s), close_range_flags)); 549 } 550 551 static uint8_t 552 open_to_fde_flags(int open_flags, bool sticky_orb) 553 { 554 static const struct flags_trans_elem open_to_fde_flags_s[] = { 555 { .f = O_CLOEXEC, .t = UF_EXCLOSE }, 556 { .f = O_CLOFORK, .t = UF_FOCLOSE }, 557 { .f = O_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH }, 558 }; 559 #if defined(__clang__) && __clang_major__ >= 19 560 _Static_assert(open_to_fde_flags_s[nitems(open_to_fde_flags_s) - 1].f == 561 O_RESOLVE_BENEATH, "O_RESOLVE_BENEATH must be last, for sticky_orb"); 562 #endif 563 564 return (flags_trans(open_to_fde_flags_s, nitems(open_to_fde_flags_s) - 565 (sticky_orb ? 0 : 1), open_flags)); 566 } 567 568 int 569 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 570 { 571 struct filedesc *fdp; 572 struct flock *flp; 573 struct file *fp, *fp2; 574 struct filedescent *fde; 575 struct proc *p; 576 struct vnode *vp; 577 struct mount *mp; 578 struct kinfo_file *kif; 579 int error, flg, kif_sz, seals, tmp, got_set, got_cleared; 580 uint64_t bsize; 581 off_t foffset; 582 int flags; 583 584 error = 0; 585 flg = F_POSIX; 586 p = td->td_proc; 587 fdp = p->p_fd; 588 589 AUDIT_ARG_FD(cmd); 590 AUDIT_ARG_CMD(cmd); 591 switch (cmd) { 592 case F_DUPFD: 593 tmp = arg; 594 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 595 break; 596 597 case F_DUPFD_CLOEXEC: 598 tmp = arg; 599 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 600 break; 601 602 case F_DUPFD_CLOFORK: 603 tmp = arg; 604 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOFORK, fd, tmp); 605 break; 606 607 case F_DUP2FD: 608 tmp = arg; 609 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 610 break; 611 612 case F_DUP2FD_CLOEXEC: 613 tmp = arg; 614 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 615 break; 616 617 case F_GETFD: 618 error = EBADF; 619 FILEDESC_SLOCK(fdp); 620 fde = fdeget_noref(fdp, fd); 621 if (fde != NULL) { 622 td->td_retval[0] = fde_to_fd_flags(fde->fde_flags); 623 error = 0; 624 } 625 FILEDESC_SUNLOCK(fdp); 626 break; 627 628 case F_SETFD: 629 error = EBADF; 630 FILEDESC_XLOCK(fdp); 631 fde = fdeget_noref(fdp, fd); 632 if (fde != NULL) { 633 /* 634 * UF_RESOLVE_BENEATH is sticky and cannot be cleared. 635 */ 636 fde->fde_flags = (fde->fde_flags & 637 ~(UF_EXCLOSE | UF_FOCLOSE)) | fd_to_fde_flags(arg); 638 error = 0; 639 } 640 FILEDESC_XUNLOCK(fdp); 641 break; 642 643 case F_GETFL: 644 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); 645 if (error != 0) 646 break; 647 td->td_retval[0] = OFLAGS(fp->f_flag); 648 fdrop(fp, td); 649 break; 650 651 case F_SETFL: 652 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); 653 if (error != 0) 654 break; 655 if (fp->f_ops == &path_fileops) { 656 fdrop(fp, td); 657 error = EBADF; 658 break; 659 } 660 fsetfl_lock(fp); 661 do { 662 tmp = flg = fp->f_flag; 663 tmp &= ~FCNTLFLAGS; 664 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 665 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 666 got_set = tmp & ~flg; 667 got_cleared = flg & ~tmp; 668 if (((got_set | got_cleared) & FNONBLOCK) != 0) { 669 tmp = fp->f_flag & FNONBLOCK; 670 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 671 if (error != 0) 672 goto revert_flags; 673 } 674 if (((got_set | got_cleared) & FASYNC) != 0) { 675 tmp = fp->f_flag & FASYNC; 676 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 677 if (error != 0) 678 goto revert_nonblock; 679 } 680 fsetfl_unlock(fp); 681 fdrop(fp, td); 682 break; 683 revert_nonblock: 684 if (((got_set | got_cleared) & FNONBLOCK) != 0) { 685 tmp = ~fp->f_flag & FNONBLOCK; 686 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 687 } 688 revert_flags: 689 do { 690 tmp = flg = fp->f_flag; 691 tmp &= ~FCNTLFLAGS; 692 tmp |= got_cleared; 693 tmp &= ~got_set; 694 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 695 fsetfl_unlock(fp); 696 fdrop(fp, td); 697 break; 698 699 case F_GETOWN: 700 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 701 if (error != 0) 702 break; 703 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 704 if (error == 0) 705 td->td_retval[0] = tmp; 706 fdrop(fp, td); 707 break; 708 709 case F_SETOWN: 710 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 711 if (error != 0) 712 break; 713 tmp = arg; 714 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 715 fdrop(fp, td); 716 break; 717 718 case F_SETLK_REMOTE: 719 error = priv_check(td, PRIV_NFS_LOCKD); 720 if (error != 0) 721 return (error); 722 flg = F_REMOTE; 723 goto do_setlk; 724 725 case F_SETLKW: 726 flg |= F_WAIT; 727 /* FALLTHROUGH F_SETLK */ 728 729 case F_SETLK: 730 do_setlk: 731 flp = (struct flock *)arg; 732 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { 733 error = EINVAL; 734 break; 735 } 736 737 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 738 if (error != 0) 739 break; 740 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 741 error = EBADF; 742 fdrop(fp, td); 743 break; 744 } 745 746 if (flp->l_whence == SEEK_CUR) { 747 foffset = foffset_get(fp); 748 if (foffset < 0 || 749 (flp->l_start > 0 && 750 foffset > OFF_MAX - flp->l_start)) { 751 error = EOVERFLOW; 752 fdrop(fp, td); 753 break; 754 } 755 flp->l_start += foffset; 756 } 757 758 vp = fp->f_vnode; 759 switch (flp->l_type) { 760 case F_RDLCK: 761 if ((fp->f_flag & FREAD) == 0) { 762 error = EBADF; 763 break; 764 } 765 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 766 PROC_LOCK(p->p_leader); 767 p->p_leader->p_flag |= P_ADVLOCK; 768 PROC_UNLOCK(p->p_leader); 769 } 770 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 771 flp, flg); 772 break; 773 case F_WRLCK: 774 if ((fp->f_flag & FWRITE) == 0) { 775 error = EBADF; 776 break; 777 } 778 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 779 PROC_LOCK(p->p_leader); 780 p->p_leader->p_flag |= P_ADVLOCK; 781 PROC_UNLOCK(p->p_leader); 782 } 783 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 784 flp, flg); 785 break; 786 case F_UNLCK: 787 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 788 flp, flg); 789 break; 790 case F_UNLCKSYS: 791 if (flg != F_REMOTE) { 792 error = EINVAL; 793 break; 794 } 795 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 796 F_UNLCKSYS, flp, flg); 797 break; 798 default: 799 error = EINVAL; 800 break; 801 } 802 if (error != 0 || flp->l_type == F_UNLCK || 803 flp->l_type == F_UNLCKSYS) { 804 fdrop(fp, td); 805 break; 806 } 807 808 /* 809 * Check for a race with close. 810 * 811 * The vnode is now advisory locked (or unlocked, but this case 812 * is not really important) as the caller requested. 813 * We had to drop the filedesc lock, so we need to recheck if 814 * the descriptor is still valid, because if it was closed 815 * in the meantime we need to remove advisory lock from the 816 * vnode - close on any descriptor leading to an advisory 817 * locked vnode, removes that lock. 818 * We will return 0 on purpose in that case, as the result of 819 * successful advisory lock might have been externally visible 820 * already. This is fine - effectively we pretend to the caller 821 * that the closing thread was a bit slower and that the 822 * advisory lock succeeded before the close. 823 */ 824 error = fget_unlocked(td, fd, &cap_no_rights, &fp2); 825 if (error != 0) { 826 fdrop(fp, td); 827 break; 828 } 829 if (fp != fp2) { 830 flp->l_whence = SEEK_SET; 831 flp->l_start = 0; 832 flp->l_len = 0; 833 flp->l_type = F_UNLCK; 834 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 835 F_UNLCK, flp, F_POSIX); 836 } 837 fdrop(fp, td); 838 fdrop(fp2, td); 839 break; 840 841 case F_GETLK: 842 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 843 if (error != 0) 844 break; 845 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 846 error = EBADF; 847 fdrop(fp, td); 848 break; 849 } 850 flp = (struct flock *)arg; 851 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 852 flp->l_type != F_UNLCK) { 853 error = EINVAL; 854 fdrop(fp, td); 855 break; 856 } 857 if (flp->l_whence == SEEK_CUR) { 858 foffset = foffset_get(fp); 859 if ((flp->l_start > 0 && 860 foffset > OFF_MAX - flp->l_start) || 861 (flp->l_start < 0 && 862 foffset < OFF_MIN - flp->l_start)) { 863 error = EOVERFLOW; 864 fdrop(fp, td); 865 break; 866 } 867 flp->l_start += foffset; 868 } 869 vp = fp->f_vnode; 870 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 871 F_POSIX); 872 fdrop(fp, td); 873 break; 874 875 case F_ADD_SEALS: 876 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 877 if (error != 0) 878 break; 879 error = fo_add_seals(fp, arg); 880 fdrop(fp, td); 881 break; 882 883 case F_GET_SEALS: 884 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 885 if (error != 0) 886 break; 887 if (fo_get_seals(fp, &seals) == 0) 888 td->td_retval[0] = seals; 889 else 890 error = EINVAL; 891 fdrop(fp, td); 892 break; 893 894 case F_RDAHEAD: 895 arg = arg ? 128 * 1024: 0; 896 /* FALLTHROUGH */ 897 case F_READAHEAD: 898 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 899 if (error != 0) 900 break; 901 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 902 fdrop(fp, td); 903 error = EBADF; 904 break; 905 } 906 vp = fp->f_vnode; 907 if (vp->v_type != VREG) { 908 fdrop(fp, td); 909 error = ENOTTY; 910 break; 911 } 912 913 /* 914 * Exclusive lock synchronizes against f_seqcount reads and 915 * writes in sequential_heuristic(). 916 */ 917 error = vn_lock(vp, LK_EXCLUSIVE); 918 if (error != 0) { 919 fdrop(fp, td); 920 break; 921 } 922 if (arg >= 0) { 923 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 924 arg = MIN(arg, INT_MAX - bsize + 1); 925 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, 926 (arg + bsize - 1) / bsize); 927 atomic_set_int(&fp->f_flag, FRDAHEAD); 928 } else { 929 atomic_clear_int(&fp->f_flag, FRDAHEAD); 930 } 931 VOP_UNLOCK(vp); 932 fdrop(fp, td); 933 break; 934 935 case F_ISUNIONSTACK: 936 /* 937 * Check if the vnode is part of a union stack (either the 938 * "union" flag from mount(2) or unionfs). 939 * 940 * Prior to introduction of this op libc's readdir would call 941 * fstatfs(2), in effect unnecessarily copying kilobytes of 942 * data just to check fs name and a mount flag. 943 * 944 * Fixing the code to handle everything in the kernel instead 945 * is a non-trivial endeavor and has low priority, thus this 946 * horrible kludge facilitates the current behavior in a much 947 * cheaper manner until someone(tm) sorts this out. 948 */ 949 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 950 if (error != 0) 951 break; 952 if (fp->f_type != DTYPE_VNODE) { 953 fdrop(fp, td); 954 error = EBADF; 955 break; 956 } 957 vp = fp->f_vnode; 958 /* 959 * Since we don't prevent dooming the vnode even non-null mp 960 * found can become immediately stale. This is tolerable since 961 * mount points are type-stable (providing safe memory access) 962 * and any vfs op on this vnode going forward will return an 963 * error (meaning return value in this case is meaningless). 964 */ 965 mp = atomic_load_ptr(&vp->v_mount); 966 if (__predict_false(mp == NULL)) { 967 fdrop(fp, td); 968 error = EBADF; 969 break; 970 } 971 td->td_retval[0] = 0; 972 if (mp->mnt_kern_flag & MNTK_UNIONFS || 973 mp->mnt_flag & MNT_UNION) 974 td->td_retval[0] = 1; 975 fdrop(fp, td); 976 break; 977 978 case F_KINFO: 979 #ifdef CAPABILITY_MODE 980 if (CAP_TRACING(td)) 981 ktrcapfail(CAPFAIL_SYSCALL, &cmd); 982 if (IN_CAPABILITY_MODE(td)) { 983 error = ECAPMODE; 984 break; 985 } 986 #endif 987 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz)); 988 if (error != 0) 989 break; 990 if (kif_sz != sizeof(*kif)) { 991 error = EINVAL; 992 break; 993 } 994 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO); 995 FILEDESC_SLOCK(fdp); 996 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL); 997 if (error == 0 && fhold(fp)) { 998 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0); 999 FILEDESC_SUNLOCK(fdp); 1000 fdrop(fp, td); 1001 if ((kif->kf_status & KF_ATTR_VALID) != 0) { 1002 kif->kf_structsize = sizeof(*kif); 1003 error = copyout(kif, (void *)arg, sizeof(*kif)); 1004 } else { 1005 error = EBADF; 1006 } 1007 } else { 1008 FILEDESC_SUNLOCK(fdp); 1009 if (error == 0) 1010 error = EBADF; 1011 } 1012 free(kif, M_TEMP); 1013 break; 1014 1015 default: 1016 if ((cmd & ((1u << F_DUP3FD_SHIFT) - 1)) != F_DUP3FD) 1017 return (EXTERROR(EINVAL, "invalid fcntl cmd")); 1018 /* Handle F_DUP3FD */ 1019 flags = (cmd >> F_DUP3FD_SHIFT); 1020 if ((flags & ~(FD_CLOEXEC | FD_CLOFORK)) != 0) 1021 return (EXTERROR(EINVAL, "invalid flags for F_DUP3FD")); 1022 tmp = arg; 1023 error = kern_dup(td, FDDUP_FIXED, 1024 ((flags & FD_CLOEXEC) != 0 ? FDDUP_FLAG_CLOEXEC : 0) | 1025 ((flags & FD_CLOFORK) != 0 ? FDDUP_FLAG_CLOFORK : 0), 1026 fd, tmp); 1027 break; 1028 } 1029 return (error); 1030 } 1031 1032 static int 1033 getmaxfd(struct thread *td) 1034 { 1035 1036 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 1037 } 1038 1039 /* 1040 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 1041 */ 1042 int 1043 kern_dup(struct thread *td, u_int mode, int flags, int oldd, int newd) 1044 { 1045 struct filedesc *fdp; 1046 struct filedescent *oldfde, *newfde; 1047 struct proc *p; 1048 struct file *delfp, *oldfp; 1049 u_long *oioctls, *nioctls; 1050 int error, maxfd; 1051 1052 p = td->td_proc; 1053 fdp = p->p_fd; 1054 oioctls = NULL; 1055 1056 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC | FDDUP_FLAG_CLOFORK)) == 0); 1057 MPASS(mode < FDDUP_LASTMODE); 1058 1059 AUDIT_ARG_FD(oldd); 1060 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(newd); */ 1061 1062 /* 1063 * Verify we have a valid descriptor to dup from and possibly to 1064 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 1065 * return EINVAL when the new descriptor is out of bounds. 1066 */ 1067 if (oldd < 0) 1068 return (EBADF); 1069 if (newd < 0) 1070 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1071 maxfd = getmaxfd(td); 1072 if (newd >= maxfd) 1073 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1074 1075 error = EBADF; 1076 FILEDESC_XLOCK(fdp); 1077 if (fget_noref(fdp, oldd) == NULL) 1078 goto unlock; 1079 if (mode == FDDUP_FIXED && oldd == newd) { 1080 td->td_retval[0] = newd; 1081 fdp->fd_ofiles[newd].fde_flags |= fddup_to_fde_flags(flags); 1082 error = 0; 1083 goto unlock; 1084 } 1085 1086 oldfde = &fdp->fd_ofiles[oldd]; 1087 oldfp = oldfde->fde_file; 1088 if (!fhold(oldfp)) 1089 goto unlock; 1090 1091 /* 1092 * If the caller specified a file descriptor, make sure the file 1093 * table is large enough to hold it, and grab it. Otherwise, just 1094 * allocate a new descriptor the usual way. 1095 */ 1096 switch (mode) { 1097 case FDDUP_NORMAL: 1098 case FDDUP_FCNTL: 1099 if ((error = fdalloc(td, newd, &newd)) != 0) { 1100 fdrop(oldfp, td); 1101 goto unlock; 1102 } 1103 break; 1104 case FDDUP_FIXED: 1105 if (newd >= fdp->fd_nfiles) { 1106 /* 1107 * The resource limits are here instead of e.g. 1108 * fdalloc(), because the file descriptor table may be 1109 * shared between processes, so we can't really use 1110 * racct_add()/racct_sub(). Instead of counting the 1111 * number of actually allocated descriptors, just put 1112 * the limit on the size of the file descriptor table. 1113 */ 1114 #ifdef RACCT 1115 if (RACCT_ENABLED()) { 1116 error = racct_set_unlocked(p, RACCT_NOFILE, newd + 1); 1117 if (error != 0) { 1118 error = EMFILE; 1119 fdrop(oldfp, td); 1120 goto unlock; 1121 } 1122 } 1123 #endif 1124 fdgrowtable_exp(fdp, newd + 1); 1125 } 1126 if (!fdisused(fdp, newd)) 1127 fdused(fdp, newd); 1128 break; 1129 default: 1130 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 1131 } 1132 1133 KASSERT(oldd != newd, ("new fd is same as old")); 1134 1135 /* Refetch oldfde because the table may have grown and old one freed. */ 1136 oldfde = &fdp->fd_ofiles[oldd]; 1137 KASSERT(oldfp == oldfde->fde_file, 1138 ("fdt_ofiles shift from growth observed at fd %d", 1139 oldd)); 1140 1141 newfde = &fdp->fd_ofiles[newd]; 1142 delfp = newfde->fde_file; 1143 1144 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 1145 1146 /* 1147 * Duplicate the source descriptor. 1148 */ 1149 #ifdef CAPABILITIES 1150 seqc_write_begin(&newfde->fde_seqc); 1151 #endif 1152 oioctls = filecaps_free_prep(&newfde->fde_caps); 1153 fde_copy(oldfde, newfde); 1154 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 1155 nioctls); 1156 newfde->fde_flags = (oldfde->fde_flags & ~(UF_EXCLOSE | UF_FOCLOSE)) | 1157 fddup_to_fde_flags(flags); 1158 #ifdef CAPABILITIES 1159 seqc_write_end(&newfde->fde_seqc); 1160 #endif 1161 td->td_retval[0] = newd; 1162 1163 error = 0; 1164 1165 if (delfp != NULL) { 1166 (void) closefp(fdp, newd, delfp, td, true, false); 1167 FILEDESC_UNLOCK_ASSERT(fdp); 1168 } else { 1169 unlock: 1170 FILEDESC_XUNLOCK(fdp); 1171 } 1172 1173 filecaps_free_finish(oioctls); 1174 return (error); 1175 } 1176 1177 static void 1178 sigiofree(struct sigio *sigio) 1179 { 1180 crfree(sigio->sio_ucred); 1181 free(sigio, M_SIGIO); 1182 } 1183 1184 static struct sigio * 1185 funsetown_locked(struct sigio *sigio) 1186 { 1187 struct proc *p; 1188 struct pgrp *pg; 1189 1190 SIGIO_ASSERT_LOCKED(); 1191 1192 if (sigio == NULL) 1193 return (NULL); 1194 *sigio->sio_myref = NULL; 1195 if (sigio->sio_pgid < 0) { 1196 pg = sigio->sio_pgrp; 1197 PGRP_LOCK(pg); 1198 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); 1199 PGRP_UNLOCK(pg); 1200 } else { 1201 p = sigio->sio_proc; 1202 PROC_LOCK(p); 1203 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 1204 PROC_UNLOCK(p); 1205 } 1206 return (sigio); 1207 } 1208 1209 /* 1210 * If sigio is on the list associated with a process or process group, 1211 * disable signalling from the device, remove sigio from the list and 1212 * free sigio. 1213 */ 1214 void 1215 funsetown(struct sigio **sigiop) 1216 { 1217 struct sigio *sigio; 1218 1219 /* Racy check, consumers must provide synchronization. */ 1220 if (*sigiop == NULL) 1221 return; 1222 1223 SIGIO_LOCK(); 1224 sigio = funsetown_locked(*sigiop); 1225 SIGIO_UNLOCK(); 1226 if (sigio != NULL) 1227 sigiofree(sigio); 1228 } 1229 1230 /* 1231 * Free a list of sigio structures. The caller must ensure that new sigio 1232 * structures cannot be added after this point. For process groups this is 1233 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves 1234 * as an interlock. 1235 */ 1236 void 1237 funsetownlst(struct sigiolst *sigiolst) 1238 { 1239 struct proc *p; 1240 struct pgrp *pg; 1241 struct sigio *sigio, *tmp; 1242 1243 /* Racy check. */ 1244 sigio = SLIST_FIRST(sigiolst); 1245 if (sigio == NULL) 1246 return; 1247 1248 p = NULL; 1249 pg = NULL; 1250 1251 SIGIO_LOCK(); 1252 sigio = SLIST_FIRST(sigiolst); 1253 if (sigio == NULL) { 1254 SIGIO_UNLOCK(); 1255 return; 1256 } 1257 1258 /* 1259 * Every entry of the list should belong to a single proc or pgrp. 1260 */ 1261 if (sigio->sio_pgid < 0) { 1262 pg = sigio->sio_pgrp; 1263 sx_assert(&proctree_lock, SX_XLOCKED); 1264 PGRP_LOCK(pg); 1265 } else /* if (sigio->sio_pgid > 0) */ { 1266 p = sigio->sio_proc; 1267 PROC_LOCK(p); 1268 KASSERT((p->p_flag & P_WEXIT) != 0, 1269 ("%s: process %p is not exiting", __func__, p)); 1270 } 1271 1272 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) { 1273 *sigio->sio_myref = NULL; 1274 if (pg != NULL) { 1275 KASSERT(sigio->sio_pgid < 0, 1276 ("Proc sigio in pgrp sigio list")); 1277 KASSERT(sigio->sio_pgrp == pg, 1278 ("Bogus pgrp in sigio list")); 1279 } else /* if (p != NULL) */ { 1280 KASSERT(sigio->sio_pgid > 0, 1281 ("Pgrp sigio in proc sigio list")); 1282 KASSERT(sigio->sio_proc == p, 1283 ("Bogus proc in sigio list")); 1284 } 1285 } 1286 1287 if (pg != NULL) 1288 PGRP_UNLOCK(pg); 1289 else 1290 PROC_UNLOCK(p); 1291 SIGIO_UNLOCK(); 1292 1293 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp) 1294 sigiofree(sigio); 1295 } 1296 1297 /* 1298 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1299 * 1300 * After permission checking, add a sigio structure to the sigio list for 1301 * the process or process group. 1302 */ 1303 int 1304 fsetown(pid_t pgid, struct sigio **sigiop) 1305 { 1306 struct proc *proc; 1307 struct pgrp *pgrp; 1308 struct sigio *osigio, *sigio; 1309 int ret; 1310 1311 if (pgid == 0) { 1312 funsetown(sigiop); 1313 return (0); 1314 } 1315 1316 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1317 sigio->sio_pgid = pgid; 1318 sigio->sio_ucred = crhold(curthread->td_ucred); 1319 sigio->sio_myref = sigiop; 1320 1321 ret = 0; 1322 if (pgid > 0) { 1323 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc); 1324 SIGIO_LOCK(); 1325 osigio = funsetown_locked(*sigiop); 1326 if (ret == 0) { 1327 PROC_LOCK(proc); 1328 _PRELE(proc); 1329 if ((proc->p_flag & P_WEXIT) != 0) { 1330 ret = ESRCH; 1331 } else if (proc->p_session != 1332 curthread->td_proc->p_session) { 1333 /* 1334 * Policy - Don't allow a process to FSETOWN a 1335 * process in another session. 1336 * 1337 * Remove this test to allow maximum flexibility 1338 * or restrict FSETOWN to the current process or 1339 * process group for maximum safety. 1340 */ 1341 ret = EPERM; 1342 } else { 1343 sigio->sio_proc = proc; 1344 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, 1345 sio_pgsigio); 1346 } 1347 PROC_UNLOCK(proc); 1348 } 1349 } else /* if (pgid < 0) */ { 1350 sx_slock(&proctree_lock); 1351 SIGIO_LOCK(); 1352 osigio = funsetown_locked(*sigiop); 1353 pgrp = pgfind(-pgid); 1354 if (pgrp == NULL) { 1355 ret = ESRCH; 1356 } else { 1357 if (pgrp->pg_session != curthread->td_proc->p_session) { 1358 /* 1359 * Policy - Don't allow a process to FSETOWN a 1360 * process in another session. 1361 * 1362 * Remove this test to allow maximum flexibility 1363 * or restrict FSETOWN to the current process or 1364 * process group for maximum safety. 1365 */ 1366 ret = EPERM; 1367 } else { 1368 sigio->sio_pgrp = pgrp; 1369 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, 1370 sio_pgsigio); 1371 } 1372 PGRP_UNLOCK(pgrp); 1373 } 1374 sx_sunlock(&proctree_lock); 1375 } 1376 if (ret == 0) 1377 *sigiop = sigio; 1378 SIGIO_UNLOCK(); 1379 if (osigio != NULL) 1380 sigiofree(osigio); 1381 return (ret); 1382 } 1383 1384 /* 1385 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1386 */ 1387 pid_t 1388 fgetown(struct sigio **sigiop) 1389 { 1390 pid_t pgid; 1391 1392 SIGIO_LOCK(); 1393 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1394 SIGIO_UNLOCK(); 1395 return (pgid); 1396 } 1397 1398 static int 1399 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1400 bool audit) 1401 { 1402 int error; 1403 1404 FILEDESC_XLOCK_ASSERT(fdp); 1405 1406 /* 1407 * We now hold the fp reference that used to be owned by the 1408 * descriptor array. We have to unlock the FILEDESC *AFTER* 1409 * knote_fdclose to prevent a race of the fd getting opened, a knote 1410 * added, and deleteing a knote for the new fd. 1411 */ 1412 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1413 knote_fdclose(td, fd); 1414 1415 if (fp->f_ops->fo_fdclose != NULL) 1416 fp->f_ops->fo_fdclose(fp, fd, td); 1417 FILEDESC_XUNLOCK(fdp); 1418 1419 #ifdef AUDIT 1420 if (AUDITING_TD(td) && audit) 1421 audit_sysclose(td, fd, fp); 1422 #endif 1423 error = closef(fp, td); 1424 1425 /* 1426 * All paths leading up to closefp() will have already removed or 1427 * replaced the fd in the filedesc table, so a restart would not 1428 * operate on the same file. 1429 */ 1430 if (error == ERESTART) 1431 error = EINTR; 1432 1433 return (error); 1434 } 1435 1436 static int 1437 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1438 bool holdleaders, bool audit) 1439 { 1440 int error; 1441 1442 FILEDESC_XLOCK_ASSERT(fdp); 1443 1444 if (holdleaders) { 1445 if (td->td_proc->p_fdtol != NULL) { 1446 /* 1447 * Ask fdfree() to sleep to ensure that all relevant 1448 * process leaders can be traversed in closef(). 1449 */ 1450 fdp->fd_holdleaderscount++; 1451 } else { 1452 holdleaders = false; 1453 } 1454 } 1455 1456 error = closefp_impl(fdp, fd, fp, td, audit); 1457 if (holdleaders) { 1458 FILEDESC_XLOCK(fdp); 1459 fdp->fd_holdleaderscount--; 1460 if (fdp->fd_holdleaderscount == 0 && 1461 fdp->fd_holdleaderswakeup != 0) { 1462 fdp->fd_holdleaderswakeup = 0; 1463 wakeup(&fdp->fd_holdleaderscount); 1464 } 1465 FILEDESC_XUNLOCK(fdp); 1466 } 1467 return (error); 1468 } 1469 1470 static int 1471 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1472 bool holdleaders, bool audit) 1473 { 1474 1475 FILEDESC_XLOCK_ASSERT(fdp); 1476 1477 if (__predict_false(td->td_proc->p_fdtol != NULL)) { 1478 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit)); 1479 } else { 1480 return (closefp_impl(fdp, fd, fp, td, audit)); 1481 } 1482 } 1483 1484 /* 1485 * Close a file descriptor. 1486 */ 1487 #ifndef _SYS_SYSPROTO_H_ 1488 struct close_args { 1489 int fd; 1490 }; 1491 #endif 1492 /* ARGSUSED */ 1493 int 1494 sys_close(struct thread *td, struct close_args *uap) 1495 { 1496 1497 return (kern_close(td, uap->fd)); 1498 } 1499 1500 int 1501 kern_close(struct thread *td, int fd) 1502 { 1503 struct filedesc *fdp; 1504 struct file *fp; 1505 1506 fdp = td->td_proc->p_fd; 1507 1508 FILEDESC_XLOCK(fdp); 1509 if ((fp = fget_noref(fdp, fd)) == NULL) { 1510 FILEDESC_XUNLOCK(fdp); 1511 return (EBADF); 1512 } 1513 fdfree(fdp, fd); 1514 1515 /* closefp() drops the FILEDESC lock for us. */ 1516 return (closefp(fdp, fd, fp, td, true, true)); 1517 } 1518 1519 static int 1520 close_range_flags(struct thread *td, u_int lowfd, u_int highfd, int flags) 1521 { 1522 struct filedesc *fdp; 1523 struct fdescenttbl *fdt; 1524 struct filedescent *fde; 1525 int fd, fde_flags; 1526 1527 fde_flags = close_range_to_fde_flags(flags); 1528 fdp = td->td_proc->p_fd; 1529 FILEDESC_XLOCK(fdp); 1530 fdt = atomic_load_ptr(&fdp->fd_files); 1531 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1532 fd = lowfd; 1533 if (__predict_false(fd > highfd)) { 1534 goto out_locked; 1535 } 1536 for (; fd <= highfd; fd++) { 1537 fde = &fdt->fdt_ofiles[fd]; 1538 if (fde->fde_file != NULL) 1539 fde->fde_flags |= fde_flags; 1540 } 1541 out_locked: 1542 FILEDESC_XUNLOCK(fdp); 1543 return (0); 1544 } 1545 1546 static int 1547 close_range_impl(struct thread *td, u_int lowfd, u_int highfd) 1548 { 1549 struct filedesc *fdp; 1550 const struct fdescenttbl *fdt; 1551 struct file *fp; 1552 int fd; 1553 1554 fdp = td->td_proc->p_fd; 1555 FILEDESC_XLOCK(fdp); 1556 fdt = atomic_load_ptr(&fdp->fd_files); 1557 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1558 fd = lowfd; 1559 if (__predict_false(fd > highfd)) { 1560 goto out_locked; 1561 } 1562 for (;;) { 1563 fp = fdt->fdt_ofiles[fd].fde_file; 1564 if (fp == NULL) { 1565 if (fd == highfd) 1566 goto out_locked; 1567 } else { 1568 fdfree(fdp, fd); 1569 (void) closefp(fdp, fd, fp, td, true, true); 1570 if (fd == highfd) 1571 goto out_unlocked; 1572 FILEDESC_XLOCK(fdp); 1573 fdt = atomic_load_ptr(&fdp->fd_files); 1574 } 1575 fd++; 1576 } 1577 out_locked: 1578 FILEDESC_XUNLOCK(fdp); 1579 out_unlocked: 1580 return (0); 1581 } 1582 1583 int 1584 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd) 1585 { 1586 1587 /* 1588 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 1589 * open should not be a usage error. From a close_range() perspective, 1590 * close_range(3, ~0U, 0) in the same scenario should also likely not 1591 * be a usage error as all fd above 3 are in-fact already closed. 1592 */ 1593 if (highfd < lowfd) { 1594 return (EINVAL); 1595 } 1596 1597 if ((flags & (CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0) 1598 return (close_range_flags(td, lowfd, highfd, flags)); 1599 1600 return (close_range_impl(td, lowfd, highfd)); 1601 } 1602 1603 #ifndef _SYS_SYSPROTO_H_ 1604 struct close_range_args { 1605 u_int lowfd; 1606 u_int highfd; 1607 int flags; 1608 }; 1609 #endif 1610 int 1611 sys_close_range(struct thread *td, struct close_range_args *uap) 1612 { 1613 1614 AUDIT_ARG_FD(uap->lowfd); 1615 AUDIT_ARG_CMD(uap->highfd); 1616 AUDIT_ARG_FFLAGS(uap->flags); 1617 1618 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0) 1619 return (EINVAL); 1620 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd)); 1621 } 1622 1623 #ifdef COMPAT_FREEBSD12 1624 /* 1625 * Close open file descriptors. 1626 */ 1627 #ifndef _SYS_SYSPROTO_H_ 1628 struct freebsd12_closefrom_args { 1629 int lowfd; 1630 }; 1631 #endif 1632 /* ARGSUSED */ 1633 int 1634 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) 1635 { 1636 u_int lowfd; 1637 1638 AUDIT_ARG_FD(uap->lowfd); 1639 1640 /* 1641 * Treat negative starting file descriptor values identical to 1642 * closefrom(0) which closes all files. 1643 */ 1644 lowfd = MAX(0, uap->lowfd); 1645 return (kern_close_range(td, 0, lowfd, ~0U)); 1646 } 1647 #endif /* COMPAT_FREEBSD12 */ 1648 1649 #if defined(COMPAT_43) 1650 /* 1651 * Return status information about a file descriptor. 1652 */ 1653 #ifndef _SYS_SYSPROTO_H_ 1654 struct ofstat_args { 1655 int fd; 1656 struct ostat *sb; 1657 }; 1658 #endif 1659 /* ARGSUSED */ 1660 int 1661 ofstat(struct thread *td, struct ofstat_args *uap) 1662 { 1663 struct ostat oub; 1664 struct stat ub; 1665 int error; 1666 1667 error = kern_fstat(td, uap->fd, &ub); 1668 if (error == 0) { 1669 cvtstat(&ub, &oub); 1670 error = copyout(&oub, uap->sb, sizeof(oub)); 1671 } 1672 return (error); 1673 } 1674 #endif /* COMPAT_43 */ 1675 1676 #if defined(COMPAT_FREEBSD11) 1677 int 1678 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1679 { 1680 struct stat sb; 1681 struct freebsd11_stat osb; 1682 int error; 1683 1684 error = kern_fstat(td, uap->fd, &sb); 1685 if (error != 0) 1686 return (error); 1687 error = freebsd11_cvtstat(&sb, &osb); 1688 if (error == 0) 1689 error = copyout(&osb, uap->sb, sizeof(osb)); 1690 return (error); 1691 } 1692 #endif /* COMPAT_FREEBSD11 */ 1693 1694 /* 1695 * Return status information about a file descriptor. 1696 */ 1697 #ifndef _SYS_SYSPROTO_H_ 1698 struct fstat_args { 1699 int fd; 1700 struct stat *sb; 1701 }; 1702 #endif 1703 /* ARGSUSED */ 1704 int 1705 sys_fstat(struct thread *td, struct fstat_args *uap) 1706 { 1707 struct stat ub; 1708 int error; 1709 1710 error = kern_fstat(td, uap->fd, &ub); 1711 if (error == 0) 1712 error = copyout(&ub, uap->sb, sizeof(ub)); 1713 return (error); 1714 } 1715 1716 int 1717 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1718 { 1719 struct file *fp; 1720 int error; 1721 1722 AUDIT_ARG_FD(fd); 1723 1724 error = fget(td, fd, &cap_fstat_rights, &fp); 1725 if (__predict_false(error != 0)) 1726 return (error); 1727 1728 AUDIT_ARG_FILE(td->td_proc, fp); 1729 1730 sbp->st_filerev = 0; 1731 sbp->st_bsdflags = 0; 1732 error = fo_stat(fp, sbp, td->td_ucred); 1733 fdrop(fp, td); 1734 #ifdef __STAT_TIME_T_EXT 1735 sbp->st_atim_ext = 0; 1736 sbp->st_mtim_ext = 0; 1737 sbp->st_ctim_ext = 0; 1738 sbp->st_btim_ext = 0; 1739 #endif 1740 #ifdef KTRACE 1741 if (KTRPOINT(td, KTR_STRUCT)) 1742 ktrstat_error(sbp, error); 1743 #endif 1744 return (error); 1745 } 1746 1747 #if defined(COMPAT_FREEBSD11) 1748 /* 1749 * Return status information about a file descriptor. 1750 */ 1751 #ifndef _SYS_SYSPROTO_H_ 1752 struct freebsd11_nfstat_args { 1753 int fd; 1754 struct nstat *sb; 1755 }; 1756 #endif 1757 /* ARGSUSED */ 1758 int 1759 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1760 { 1761 struct nstat nub; 1762 struct stat ub; 1763 int error; 1764 1765 error = kern_fstat(td, uap->fd, &ub); 1766 if (error != 0) 1767 return (error); 1768 error = freebsd11_cvtnstat(&ub, &nub); 1769 if (error != 0) 1770 error = copyout(&nub, uap->sb, sizeof(nub)); 1771 return (error); 1772 } 1773 #endif /* COMPAT_FREEBSD11 */ 1774 1775 /* 1776 * Return pathconf information about a file descriptor. 1777 */ 1778 #ifndef _SYS_SYSPROTO_H_ 1779 struct fpathconf_args { 1780 int fd; 1781 int name; 1782 }; 1783 #endif 1784 /* ARGSUSED */ 1785 int 1786 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1787 { 1788 long value; 1789 int error; 1790 1791 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1792 if (error == 0) 1793 td->td_retval[0] = value; 1794 return (error); 1795 } 1796 1797 int 1798 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1799 { 1800 struct file *fp; 1801 struct vnode *vp; 1802 int error; 1803 1804 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1805 if (error != 0) 1806 return (error); 1807 1808 if (name == _PC_ASYNC_IO) { 1809 *valuep = _POSIX_ASYNCHRONOUS_IO; 1810 goto out; 1811 } 1812 vp = fp->f_vnode; 1813 if (vp != NULL) { 1814 vn_lock(vp, LK_SHARED | LK_RETRY); 1815 error = VOP_PATHCONF(vp, name, valuep); 1816 VOP_UNLOCK(vp); 1817 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1818 if (name != _PC_PIPE_BUF) { 1819 error = EINVAL; 1820 } else { 1821 *valuep = PIPE_BUF; 1822 error = 0; 1823 } 1824 } else { 1825 error = EOPNOTSUPP; 1826 } 1827 out: 1828 fdrop(fp, td); 1829 return (error); 1830 } 1831 1832 /* 1833 * Copy filecaps structure allocating memory for ioctls array if needed. 1834 * 1835 * The last parameter indicates whether the fdtable is locked. If it is not and 1836 * ioctls are encountered, copying fails and the caller must lock the table. 1837 * 1838 * Note that if the table was not locked, the caller has to check the relevant 1839 * sequence counter to determine whether the operation was successful. 1840 */ 1841 bool 1842 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1843 { 1844 size_t size; 1845 1846 if (src->fc_ioctls != NULL && !locked) 1847 return (false); 1848 memcpy(dst, src, sizeof(*src)); 1849 if (src->fc_ioctls == NULL) 1850 return (true); 1851 1852 KASSERT(src->fc_nioctls > 0, 1853 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1854 1855 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1856 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1857 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1858 return (true); 1859 } 1860 1861 static u_long * 1862 filecaps_copy_prep(const struct filecaps *src) 1863 { 1864 u_long *ioctls; 1865 size_t size; 1866 1867 if (__predict_true(src->fc_ioctls == NULL)) 1868 return (NULL); 1869 1870 KASSERT(src->fc_nioctls > 0, 1871 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1872 1873 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1874 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1875 return (ioctls); 1876 } 1877 1878 static void 1879 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1880 u_long *ioctls) 1881 { 1882 size_t size; 1883 1884 *dst = *src; 1885 if (__predict_true(src->fc_ioctls == NULL)) { 1886 MPASS(ioctls == NULL); 1887 return; 1888 } 1889 1890 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1891 dst->fc_ioctls = ioctls; 1892 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1893 } 1894 1895 /* 1896 * Move filecaps structure to the new place and clear the old place. 1897 */ 1898 void 1899 filecaps_move(struct filecaps *src, struct filecaps *dst) 1900 { 1901 1902 *dst = *src; 1903 bzero(src, sizeof(*src)); 1904 } 1905 1906 /* 1907 * Fill the given filecaps structure with full rights. 1908 */ 1909 void 1910 filecaps_fill(struct filecaps *fcaps) 1911 { 1912 1913 CAP_ALL(&fcaps->fc_rights); 1914 fcaps->fc_ioctls = NULL; 1915 fcaps->fc_nioctls = -1; 1916 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1917 } 1918 1919 /* 1920 * Free memory allocated within filecaps structure. 1921 */ 1922 static void 1923 filecaps_free_ioctl(struct filecaps *fcaps) 1924 { 1925 1926 free(fcaps->fc_ioctls, M_FILECAPS); 1927 fcaps->fc_ioctls = NULL; 1928 } 1929 1930 void 1931 filecaps_free(struct filecaps *fcaps) 1932 { 1933 1934 filecaps_free_ioctl(fcaps); 1935 bzero(fcaps, sizeof(*fcaps)); 1936 } 1937 1938 bool 1939 filecaps_full(const struct filecaps *fcaps) 1940 { 1941 cap_rights_t allrights; 1942 1943 CAP_ALL(&allrights); 1944 return (cap_rights_contains(&fcaps->fc_rights, &allrights) && 1945 fcaps->fc_fcntls == CAP_FCNTL_ALL && fcaps->fc_nioctls == -1); 1946 } 1947 1948 /* 1949 * Find the intersection of two filecaps structures and store the result in the 1950 * first structure. This is a destructive operation on the src structure. 1951 */ 1952 void 1953 filecaps_intersect(struct filecaps *src, struct filecaps *dst) 1954 { 1955 1956 cap_rights_intersect(&dst->fc_rights, &src->fc_rights); 1957 dst->fc_fcntls &= src->fc_fcntls; 1958 if (dst->fc_nioctls == -1) { 1959 dst->fc_ioctls = src->fc_ioctls; 1960 dst->fc_nioctls = src->fc_nioctls; 1961 src->fc_ioctls = NULL; 1962 } else if (src->fc_nioctls != -1) { 1963 int count; 1964 1965 /* 1966 * ioctl lists are usually short, so this dumb merge is fine. 1967 * We could alternately sort both lists and walk them in 1968 * parallel. 1969 */ 1970 count = 0; 1971 for (int i = 0; i < dst->fc_nioctls; i++) { 1972 bool found; 1973 1974 found = false; 1975 for (int j = 0; j < src->fc_nioctls; j++) { 1976 if (dst->fc_ioctls[i] == src->fc_ioctls[j]) { 1977 count++; 1978 found = true; 1979 break; 1980 } 1981 } 1982 if (!found) { 1983 if (i != dst->fc_nioctls - 1) 1984 dst->fc_ioctls[i] = 1985 dst->fc_ioctls[dst->fc_nioctls - 1]; 1986 dst->fc_nioctls--; 1987 i--; 1988 } 1989 } 1990 dst->fc_nioctls = count; 1991 } 1992 if (dst->fc_nioctls == 0) 1993 filecaps_free_ioctl(dst); 1994 filecaps_free(src); 1995 } 1996 1997 static u_long * 1998 filecaps_free_prep(struct filecaps *fcaps) 1999 { 2000 u_long *ioctls; 2001 2002 ioctls = fcaps->fc_ioctls; 2003 bzero(fcaps, sizeof(*fcaps)); 2004 return (ioctls); 2005 } 2006 2007 static void 2008 filecaps_free_finish(u_long *ioctls) 2009 { 2010 2011 free(ioctls, M_FILECAPS); 2012 } 2013 2014 /* 2015 * Validate the given filecaps structure. 2016 */ 2017 static void 2018 filecaps_validate(const struct filecaps *fcaps, const char *func) 2019 { 2020 2021 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 2022 ("%s: invalid rights", func)); 2023 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 2024 ("%s: invalid fcntls", func)); 2025 KASSERT(fcaps->fc_fcntls == 0 || 2026 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 2027 ("%s: fcntls without CAP_FCNTL", func)); 2028 /* 2029 * open calls without WANTIOCTLCAPS free caps but leave the counter 2030 */ 2031 #if 0 2032 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 2033 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 2034 ("%s: invalid ioctls", func)); 2035 #endif 2036 KASSERT(fcaps->fc_nioctls == 0 || 2037 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 2038 ("%s: ioctls without CAP_IOCTL", func)); 2039 } 2040 2041 static void 2042 fdgrowtable_exp(struct filedesc *fdp, int nfd) 2043 { 2044 int nfd1; 2045 2046 FILEDESC_XLOCK_ASSERT(fdp); 2047 2048 nfd1 = fdp->fd_nfiles * 2; 2049 if (nfd1 < nfd) 2050 nfd1 = nfd; 2051 fdgrowtable(fdp, nfd1); 2052 } 2053 2054 /* 2055 * Grow the file table to accommodate (at least) nfd descriptors. 2056 */ 2057 static void 2058 fdgrowtable(struct filedesc *fdp, int nfd) 2059 { 2060 struct filedesc0 *fdp0; 2061 struct freetable *ft; 2062 struct fdescenttbl *ntable; 2063 struct fdescenttbl *otable; 2064 int nnfiles, onfiles; 2065 NDSLOTTYPE *nmap, *omap; 2066 2067 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 2068 KASSERT(fdp->fd_nfiles >= NDFILE, ("file table of length %d shorter " 2069 "than NDFILE (%d)", fdp->fd_nfiles, NDFILE)); 2070 KASSERT(fdp->fd_nfiles == NDFILE || fdp->fd_nfiles % NDENTRIES == 0, 2071 ("file table of length %d should be multiple of NDENTRIES (%zu)", 2072 fdp->fd_nfiles, NDENTRIES)); 2073 KASSERT((fdp->fd_nfiles == NDFILE) == ((intptr_t)fdp->fd_files - 2074 offsetof(struct filedesc0, fd_dfiles) == (intptr_t)fdp - 2075 offsetof(struct filedesc0, fd_fd)), ("file table of length %d " 2076 "should have %s table", fdp->fd_nfiles, fdp->fd_nfiles == NDFILE ? 2077 "initial" : "dynamic")); 2078 KASSERT((NDSLOTS(fdp->fd_nfiles) <= NDSLOTS(NDFILE)) == ((intptr_t) 2079 fdp->fd_map - offsetof(struct filedesc0, fd_dmap) == (intptr_t)fdp - 2080 offsetof(struct filedesc0, fd_fd)), ("file table of length %d " 2081 "should have %s map", fdp->fd_nfiles, NDSLOTS(fdp->fd_nfiles) <= 2082 NDSLOTS(NDFILE) ? "initial" : "dynamic")); 2083 2084 /* save old values */ 2085 onfiles = fdp->fd_nfiles; 2086 otable = fdp->fd_files; 2087 omap = fdp->fd_map; 2088 2089 /* compute the size of the new table */ 2090 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 2091 if (nnfiles <= onfiles) 2092 /* the table is already large enough */ 2093 return; 2094 2095 /* 2096 * Allocate a new table. We need enough space for the number of 2097 * entries, file entries themselves and the struct freetable we will use 2098 * when we decommission the table and place it on the freelist. 2099 * We place the struct freetable in the middle so we don't have 2100 * to worry about padding. 2101 */ 2102 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 2103 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 2104 sizeof(struct freetable), 2105 M_FILEDESC, M_ZERO | M_WAITOK); 2106 /* copy the old data */ 2107 ntable->fdt_nfiles = nnfiles; 2108 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 2109 onfiles * sizeof(ntable->fdt_ofiles[0])); 2110 2111 /* 2112 * Allocate a new map only if the old one is not large enough. 2113 * 2114 * The initial struct filedesc0 object contains a table and map sized 2115 * for NDFILE (20) entries which means the initial map can accomodate 2116 * up to NDENTRIES (32 or 64) before requiring reallocation. 2117 * 2118 * As the new table size (nnfiles) is always rounded up to a multiple 2119 * of NDENTRIES, the map will be fully utilised following the first 2120 * enlargement, whether it is still the initial map (which will be the 2121 * case if nnfiles == NDENTRIES) or if a new one that has has been 2122 * allocated (which will be the case if nnfiles == X*NDENTRIES for some 2123 * X > 1). In either case, subsequent enlargements will always allocate 2124 * a new map to go along with the new table. 2125 */ 2126 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 2127 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 2128 M_ZERO | M_WAITOK); 2129 /* copy over the old data and update the pointer */ 2130 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 2131 fdp->fd_map = nmap; 2132 } else { 2133 nmap = NULL; 2134 } 2135 2136 /* 2137 * Make sure that ntable is correctly initialized before we replace 2138 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 2139 * data. 2140 */ 2141 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 2142 2143 /* 2144 * Free the old file table when not shared by other threads or processes. 2145 * The old file table is considered to be shared when either are true: 2146 * - The process has more than one thread. 2147 * - The file descriptor table has been shared via fdshare(). 2148 * 2149 * When shared, the old file table will be placed on a freelist 2150 * which will be processed when the struct filedesc is released. 2151 * 2152 * Note that if onfiles == NDFILE, we're dealing with the original 2153 * static allocation contained within (struct filedesc0 *)fdp, 2154 * which must not be freed. 2155 */ 2156 if (onfiles > NDFILE) { 2157 /* 2158 * Note we may be called here from fdinit while allocating a 2159 * table for a new process in which case ->p_fd points 2160 * elsewhere. 2161 */ 2162 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 2163 free(otable, M_FILEDESC); 2164 } else { 2165 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 2166 fdp0 = (struct filedesc0 *)fdp; 2167 ft->ft_table = otable; 2168 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 2169 } 2170 } 2171 /* 2172 * The map does not have the same possibility of threads still 2173 * holding references to it. So always free it as long as it 2174 * does not reference the original static allocation and a new 2175 * map was allocated. 2176 */ 2177 if (nmap != NULL && NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 2178 free(omap, M_FILEDESC); 2179 } 2180 2181 /* 2182 * Allocate a file descriptor for the process. 2183 */ 2184 int 2185 fdalloc(struct thread *td, int minfd, int *result) 2186 { 2187 struct proc *p = td->td_proc; 2188 struct filedesc *fdp = p->p_fd; 2189 int fd, maxfd, allocfd; 2190 #ifdef RACCT 2191 int error; 2192 #endif 2193 2194 FILEDESC_XLOCK_ASSERT(fdp); 2195 2196 if (fdp->fd_freefile > minfd) 2197 minfd = fdp->fd_freefile; 2198 2199 maxfd = getmaxfd(td); 2200 2201 /* 2202 * Search the bitmap for a free descriptor starting at minfd. 2203 * If none is found, grow the file table. 2204 */ 2205 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 2206 if (__predict_false(fd >= maxfd)) 2207 return (EMFILE); 2208 if (__predict_false(fd >= fdp->fd_nfiles)) { 2209 allocfd = min(fd * 2, maxfd); 2210 #ifdef RACCT 2211 if (RACCT_ENABLED()) { 2212 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 2213 if (error != 0) 2214 return (EMFILE); 2215 } 2216 #endif 2217 /* 2218 * fd is already equal to first free descriptor >= minfd, so 2219 * we only need to grow the table and we are done. 2220 */ 2221 fdgrowtable_exp(fdp, allocfd); 2222 } 2223 2224 /* 2225 * Perform some sanity checks, then mark the file descriptor as 2226 * used and return it to the caller. 2227 */ 2228 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 2229 ("invalid descriptor %d", fd)); 2230 KASSERT(!fdisused(fdp, fd), 2231 ("fd_first_free() returned non-free descriptor")); 2232 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 2233 ("file descriptor isn't free")); 2234 fdused(fdp, fd); 2235 *result = fd; 2236 return (0); 2237 } 2238 2239 /* 2240 * Allocate n file descriptors for the process. 2241 */ 2242 int 2243 fdallocn(struct thread *td, int minfd, int *fds, int n) 2244 { 2245 struct proc *p = td->td_proc; 2246 struct filedesc *fdp = p->p_fd; 2247 int i; 2248 2249 FILEDESC_XLOCK_ASSERT(fdp); 2250 2251 for (i = 0; i < n; i++) 2252 if (fdalloc(td, 0, &fds[i]) != 0) 2253 break; 2254 2255 if (i < n) { 2256 for (i--; i >= 0; i--) 2257 fdunused(fdp, fds[i]); 2258 return (EMFILE); 2259 } 2260 2261 return (0); 2262 } 2263 2264 /* 2265 * Create a new open file structure and allocate a file descriptor for the 2266 * process that refers to it. We add one reference to the file for the 2267 * descriptor table and one reference for resultfp. This is to prevent us 2268 * being preempted and the entry in the descriptor table closed after we 2269 * release the FILEDESC lock. 2270 */ 2271 int 2272 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 2273 struct filecaps *fcaps) 2274 { 2275 struct file *fp; 2276 int error, fd; 2277 2278 MPASS(resultfp != NULL); 2279 MPASS(resultfd != NULL); 2280 2281 error = _falloc_noinstall(td, &fp, 2); 2282 if (__predict_false(error != 0)) { 2283 return (error); 2284 } 2285 2286 error = finstall_refed(td, fp, &fd, flags, fcaps); 2287 if (__predict_false(error != 0)) { 2288 falloc_abort(td, fp); 2289 return (error); 2290 } 2291 2292 *resultfp = fp; 2293 *resultfd = fd; 2294 2295 return (0); 2296 } 2297 2298 /* 2299 * Create a new open file structure without allocating a file descriptor. 2300 */ 2301 int 2302 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2303 { 2304 struct file *fp; 2305 int maxuserfiles = maxfiles - (maxfiles / 20); 2306 int openfiles_new; 2307 static struct timeval lastfail; 2308 static int curfail; 2309 2310 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2311 MPASS(n > 0); 2312 2313 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2314 if ((openfiles_new >= maxuserfiles && 2315 priv_check(td, PRIV_MAXFILES) != 0) || 2316 openfiles_new >= maxfiles) { 2317 atomic_subtract_int(&openfiles, 1); 2318 if (ppsratecheck(&lastfail, &curfail, 1)) { 2319 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2320 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2321 } 2322 return (ENFILE); 2323 } 2324 fp = uma_zalloc(file_zone, M_WAITOK); 2325 bzero(fp, sizeof(*fp)); 2326 refcount_init(&fp->f_count, n); 2327 fp->f_cred = crhold(td->td_ucred); 2328 fp->f_ops = &badfileops; 2329 *resultfp = fp; 2330 return (0); 2331 } 2332 2333 void 2334 falloc_abort(struct thread *td, struct file *fp) 2335 { 2336 2337 /* 2338 * For assertion purposes. 2339 */ 2340 refcount_init(&fp->f_count, 0); 2341 _fdrop(fp, td); 2342 } 2343 2344 /* 2345 * Install a file in a file descriptor table. 2346 */ 2347 void 2348 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2349 struct filecaps *fcaps) 2350 { 2351 struct filedescent *fde; 2352 2353 MPASS(fp != NULL); 2354 if (fcaps != NULL) 2355 filecaps_validate(fcaps, __func__); 2356 FILEDESC_XLOCK_ASSERT(fdp); 2357 2358 fde = &fdp->fd_ofiles[fd]; 2359 #ifdef CAPABILITIES 2360 seqc_write_begin(&fde->fde_seqc); 2361 #endif 2362 fde->fde_file = fp; 2363 fde->fde_flags = open_to_fde_flags(flags, true); 2364 if (fcaps != NULL) 2365 filecaps_move(fcaps, &fde->fde_caps); 2366 else 2367 filecaps_fill(&fde->fde_caps); 2368 #ifdef CAPABILITIES 2369 seqc_write_end(&fde->fde_seqc); 2370 #endif 2371 } 2372 2373 int 2374 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2375 struct filecaps *fcaps) 2376 { 2377 struct filedesc *fdp = td->td_proc->p_fd; 2378 int error; 2379 2380 MPASS(fd != NULL); 2381 2382 FILEDESC_XLOCK(fdp); 2383 error = fdalloc(td, 0, fd); 2384 if (__predict_true(error == 0)) { 2385 _finstall(fdp, fp, *fd, flags, fcaps); 2386 } 2387 FILEDESC_XUNLOCK(fdp); 2388 return (error); 2389 } 2390 2391 int 2392 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2393 struct filecaps *fcaps) 2394 { 2395 int error; 2396 2397 MPASS(fd != NULL); 2398 2399 if (!fhold(fp)) 2400 return (EBADF); 2401 error = finstall_refed(td, fp, fd, flags, fcaps); 2402 if (__predict_false(error != 0)) { 2403 fdrop(fp, td); 2404 } 2405 return (error); 2406 } 2407 2408 /* 2409 * Build a new filedesc structure from another. 2410 * 2411 * If fdp is not NULL, return with it shared locked. 2412 */ 2413 struct filedesc * 2414 fdinit(void) 2415 { 2416 struct filedesc0 *newfdp0; 2417 struct filedesc *newfdp; 2418 2419 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2420 newfdp = &newfdp0->fd_fd; 2421 2422 /* Create the file descriptor table. */ 2423 FILEDESC_LOCK_INIT(newfdp); 2424 refcount_init(&newfdp->fd_refcnt, 1); 2425 refcount_init(&newfdp->fd_holdcnt, 1); 2426 newfdp->fd_map = newfdp0->fd_dmap; 2427 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2428 newfdp->fd_files->fdt_nfiles = NDFILE; 2429 2430 return (newfdp); 2431 } 2432 2433 /* 2434 * Build a pwddesc structure from another. 2435 * Copy the current, root, and jail root vnode references. 2436 * 2437 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked. 2438 */ 2439 struct pwddesc * 2440 pdinit(struct pwddesc *pdp, bool keeplock) 2441 { 2442 struct pwddesc *newpdp; 2443 struct pwd *newpwd; 2444 2445 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2446 2447 PWDDESC_LOCK_INIT(newpdp); 2448 refcount_init(&newpdp->pd_refcount, 1); 2449 newpdp->pd_cmask = CMASK; 2450 2451 if (pdp == NULL) { 2452 newpwd = pwd_alloc(); 2453 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2454 return (newpdp); 2455 } 2456 2457 PWDDESC_XLOCK(pdp); 2458 newpwd = pwd_hold_pwddesc(pdp); 2459 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2460 if (!keeplock) 2461 PWDDESC_XUNLOCK(pdp); 2462 return (newpdp); 2463 } 2464 2465 /* 2466 * Hold either filedesc or pwddesc of the passed process. 2467 * 2468 * The process lock is used to synchronize against the target exiting and 2469 * freeing the data. 2470 * 2471 * Clearing can be ilustrated in 3 steps: 2472 * 1. set the pointer to NULL. Either routine can race against it, hence 2473 * atomic_load_ptr. 2474 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2475 * race to either still see the pointer or find NULL. It is still safe to 2476 * grab a reference as clearing is stalled. 2477 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2478 * guaranteed to see NULL, making it safe to finish clearing 2479 */ 2480 static struct filedesc * 2481 fdhold(struct proc *p) 2482 { 2483 struct filedesc *fdp; 2484 2485 PROC_LOCK_ASSERT(p, MA_OWNED); 2486 fdp = atomic_load_ptr(&p->p_fd); 2487 if (fdp != NULL) 2488 refcount_acquire(&fdp->fd_holdcnt); 2489 return (fdp); 2490 } 2491 2492 static struct pwddesc * 2493 pdhold(struct proc *p) 2494 { 2495 struct pwddesc *pdp; 2496 2497 PROC_LOCK_ASSERT(p, MA_OWNED); 2498 pdp = atomic_load_ptr(&p->p_pd); 2499 if (pdp != NULL) 2500 refcount_acquire(&pdp->pd_refcount); 2501 return (pdp); 2502 } 2503 2504 static void 2505 fddrop(struct filedesc *fdp) 2506 { 2507 2508 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2509 if (refcount_release(&fdp->fd_holdcnt) == 0) 2510 return; 2511 } 2512 2513 FILEDESC_LOCK_DESTROY(fdp); 2514 uma_zfree(filedesc0_zone, fdp); 2515 } 2516 2517 static void 2518 pddrop(struct pwddesc *pdp) 2519 { 2520 struct pwd *pwd; 2521 2522 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2523 return; 2524 2525 PWDDESC_XLOCK(pdp); 2526 if (refcount_release(&pdp->pd_refcount) == 0) { 2527 PWDDESC_XUNLOCK(pdp); 2528 return; 2529 } 2530 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2531 pwd_set(pdp, NULL); 2532 PWDDESC_XUNLOCK(pdp); 2533 pwd_drop(pwd); 2534 2535 PWDDESC_LOCK_DESTROY(pdp); 2536 free(pdp, M_PWDDESC); 2537 } 2538 2539 /* 2540 * Share a filedesc structure. 2541 */ 2542 struct filedesc * 2543 fdshare(struct filedesc *fdp) 2544 { 2545 2546 refcount_acquire(&fdp->fd_refcnt); 2547 return (fdp); 2548 } 2549 2550 /* 2551 * Share a pwddesc structure. 2552 */ 2553 struct pwddesc * 2554 pdshare(struct pwddesc *pdp) 2555 { 2556 refcount_acquire(&pdp->pd_refcount); 2557 return (pdp); 2558 } 2559 2560 /* 2561 * Unshare a filedesc structure, if necessary by making a copy 2562 */ 2563 void 2564 fdunshare(struct thread *td) 2565 { 2566 struct filedesc *tmp; 2567 struct proc *p = td->td_proc; 2568 2569 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2570 return; 2571 2572 tmp = fdcopy(p->p_fd, p); 2573 fdescfree(td); 2574 p->p_fd = tmp; 2575 } 2576 2577 /* 2578 * Unshare a pwddesc structure. 2579 */ 2580 void 2581 pdunshare(struct thread *td) 2582 { 2583 struct pwddesc *pdp; 2584 struct proc *p; 2585 2586 p = td->td_proc; 2587 /* Not shared. */ 2588 if (refcount_load(&p->p_pd->pd_refcount) == 1) 2589 return; 2590 2591 pdp = pdcopy(p->p_pd); 2592 pdescfree(td); 2593 p->p_pd = pdp; 2594 } 2595 2596 /* 2597 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2598 * this is to ease callers, not catch errors. 2599 */ 2600 struct filedesc * 2601 fdcopy(struct filedesc *fdp, struct proc *p1) 2602 { 2603 struct filedesc *newfdp; 2604 struct filedescent *nfde, *ofde; 2605 struct file *fp; 2606 int i, lastfile; 2607 bool fork_pass; 2608 2609 MPASS(fdp != NULL); 2610 2611 fork_pass = false; 2612 newfdp = fdinit(); 2613 FILEDESC_SLOCK(fdp); 2614 for (;;) { 2615 lastfile = fdlastfile(fdp); 2616 if (lastfile < newfdp->fd_nfiles) 2617 break; 2618 FILEDESC_SUNLOCK(fdp); 2619 fdgrowtable(newfdp, lastfile + 1); 2620 FILEDESC_SLOCK(fdp); 2621 } 2622 2623 /* 2624 * Copy all passable descriptors (i.e. not kqueue), and 2625 * prepare to handle copyable but not passable descriptors 2626 * (kqueues). 2627 * 2628 * The pass to handle copying is performed after all passable 2629 * files are installed into the new file descriptor's table, 2630 * since kqueues need all referenced file descriptors already 2631 * valid, including other kqueues. For the same reason the 2632 * copying is done in two passes by itself, first installing 2633 * not fully initialized ('empty') copyable files into the new 2634 * fd table, and then giving the subsystems a second chance to 2635 * really fill the copied file backing structure with the 2636 * content. 2637 */ 2638 newfdp->fd_freefile = fdp->fd_freefile; 2639 FILEDESC_FOREACH_FDE(fdp, i, ofde) { 2640 const struct fileops *ops; 2641 2642 ops = ofde->fde_file->f_ops; 2643 fp = NULL; 2644 if ((ops->fo_flags & DFLAG_FORK) != 0 && 2645 (ofde->fde_flags & UF_FOCLOSE) == 0) { 2646 if (ops->fo_fork(newfdp, ofde->fde_file, &fp, p1, 2647 curthread) != 0) 2648 continue; 2649 fork_pass = true; 2650 } else if ((ops->fo_flags & DFLAG_PASSABLE) == 0 || 2651 (ofde->fde_flags & UF_FOCLOSE) != 0 || 2652 !fhold(ofde->fde_file)) { 2653 if (newfdp->fd_freefile == fdp->fd_freefile) 2654 newfdp->fd_freefile = i; 2655 continue; 2656 } 2657 nfde = &newfdp->fd_ofiles[i]; 2658 *nfde = *ofde; 2659 if (fp != NULL) 2660 nfde->fde_file = fp; 2661 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2662 fdused_init(newfdp, i); 2663 } 2664 MPASS(newfdp->fd_freefile != -1); 2665 FILEDESC_SUNLOCK(fdp); 2666 2667 /* 2668 * Now handle copying kqueues, since all fds, including 2669 * kqueues, are in place. 2670 */ 2671 if (__predict_false(fork_pass)) { 2672 FILEDESC_FOREACH_FDE(newfdp, i, nfde) { 2673 const struct fileops *ops; 2674 2675 ops = nfde->fde_file->f_ops; 2676 if ((ops->fo_flags & DFLAG_FORK) == 0 || 2677 nfde->fde_file == NULL) 2678 continue; 2679 ops->fo_fork(newfdp, NULL, &nfde->fde_file, p1, 2680 curthread); 2681 } 2682 } 2683 return (newfdp); 2684 } 2685 2686 /* 2687 * Copy a pwddesc structure. 2688 */ 2689 struct pwddesc * 2690 pdcopy(struct pwddesc *pdp) 2691 { 2692 struct pwddesc *newpdp; 2693 2694 MPASS(pdp != NULL); 2695 2696 newpdp = pdinit(pdp, true); 2697 newpdp->pd_cmask = pdp->pd_cmask; 2698 PWDDESC_XUNLOCK(pdp); 2699 return (newpdp); 2700 } 2701 2702 /* 2703 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2704 * one of processes using it exits) and the table used to be shared. 2705 */ 2706 static void 2707 fdclearlocks(struct thread *td) 2708 { 2709 struct filedesc *fdp; 2710 struct filedesc_to_leader *fdtol; 2711 struct flock lf; 2712 struct file *fp; 2713 struct proc *p; 2714 struct vnode *vp; 2715 int i; 2716 2717 p = td->td_proc; 2718 fdp = p->p_fd; 2719 fdtol = p->p_fdtol; 2720 MPASS(fdtol != NULL); 2721 2722 FILEDESC_XLOCK(fdp); 2723 KASSERT(fdtol->fdl_refcount > 0, 2724 ("filedesc_to_refcount botch: fdl_refcount=%d", 2725 fdtol->fdl_refcount)); 2726 if (fdtol->fdl_refcount == 1 && 2727 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2728 FILEDESC_FOREACH_FP(fdp, i, fp) { 2729 if (fp->f_type != DTYPE_VNODE || 2730 !fhold(fp)) 2731 continue; 2732 FILEDESC_XUNLOCK(fdp); 2733 lf.l_whence = SEEK_SET; 2734 lf.l_start = 0; 2735 lf.l_len = 0; 2736 lf.l_type = F_UNLCK; 2737 vp = fp->f_vnode; 2738 (void) VOP_ADVLOCK(vp, 2739 (caddr_t)p->p_leader, F_UNLCK, 2740 &lf, F_POSIX); 2741 FILEDESC_XLOCK(fdp); 2742 fdrop(fp, td); 2743 } 2744 } 2745 retry: 2746 if (fdtol->fdl_refcount == 1) { 2747 if (fdp->fd_holdleaderscount > 0 && 2748 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2749 /* 2750 * close() or kern_dup() has cleared a reference 2751 * in a shared file descriptor table. 2752 */ 2753 fdp->fd_holdleaderswakeup = 1; 2754 sx_sleep(&fdp->fd_holdleaderscount, 2755 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2756 goto retry; 2757 } 2758 if (fdtol->fdl_holdcount > 0) { 2759 /* 2760 * Ensure that fdtol->fdl_leader remains 2761 * valid in closef(). 2762 */ 2763 fdtol->fdl_wakeup = 1; 2764 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2765 "fdlhold", 0); 2766 goto retry; 2767 } 2768 } 2769 fdtol->fdl_refcount--; 2770 if (fdtol->fdl_refcount == 0 && 2771 fdtol->fdl_holdcount == 0) { 2772 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2773 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2774 } else 2775 fdtol = NULL; 2776 p->p_fdtol = NULL; 2777 FILEDESC_XUNLOCK(fdp); 2778 if (fdtol != NULL) 2779 free(fdtol, M_FILEDESC_TO_LEADER); 2780 } 2781 2782 /* 2783 * Release a filedesc structure. 2784 */ 2785 static void 2786 fdescfree_fds(struct thread *td, struct filedesc *fdp) 2787 { 2788 struct filedesc0 *fdp0; 2789 struct freetable *ft, *tft; 2790 struct filedescent *fde; 2791 struct file *fp; 2792 int i; 2793 2794 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2795 ("%s: fd table %p carries references", __func__, fdp)); 2796 2797 /* 2798 * Serialize with threads iterating over the table, if any. 2799 */ 2800 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2801 FILEDESC_XLOCK(fdp); 2802 FILEDESC_XUNLOCK(fdp); 2803 } 2804 2805 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2806 fp = fde->fde_file; 2807 fdefree_last(fde); 2808 (void) closef(fp, td); 2809 } 2810 2811 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2812 free(fdp->fd_map, M_FILEDESC); 2813 if (fdp->fd_nfiles > NDFILE) 2814 free(fdp->fd_files, M_FILEDESC); 2815 2816 fdp0 = (struct filedesc0 *)fdp; 2817 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2818 free(ft->ft_table, M_FILEDESC); 2819 2820 fddrop(fdp); 2821 } 2822 2823 void 2824 fdescfree(struct thread *td) 2825 { 2826 struct proc *p; 2827 struct filedesc *fdp; 2828 2829 p = td->td_proc; 2830 fdp = p->p_fd; 2831 MPASS(fdp != NULL); 2832 2833 #ifdef RACCT 2834 if (RACCT_ENABLED()) 2835 racct_set_unlocked(p, RACCT_NOFILE, 0); 2836 #endif 2837 2838 if (p->p_fdtol != NULL) 2839 fdclearlocks(td); 2840 2841 /* 2842 * Check fdhold for an explanation. 2843 */ 2844 atomic_store_ptr(&p->p_fd, NULL); 2845 atomic_thread_fence_seq_cst(); 2846 PROC_WAIT_UNLOCKED(p); 2847 2848 if (refcount_release(&fdp->fd_refcnt) == 0) 2849 return; 2850 2851 fdescfree_fds(td, fdp); 2852 } 2853 2854 void 2855 pdescfree(struct thread *td) 2856 { 2857 struct proc *p; 2858 struct pwddesc *pdp; 2859 2860 p = td->td_proc; 2861 pdp = p->p_pd; 2862 MPASS(pdp != NULL); 2863 2864 /* 2865 * Check pdhold for an explanation. 2866 */ 2867 atomic_store_ptr(&p->p_pd, NULL); 2868 atomic_thread_fence_seq_cst(); 2869 PROC_WAIT_UNLOCKED(p); 2870 2871 pddrop(pdp); 2872 } 2873 2874 /* 2875 * For setugid programs, we don't want to people to use that setugidness 2876 * to generate error messages which write to a file which otherwise would 2877 * otherwise be off-limits to the process. We check for filesystems where 2878 * the vnode can change out from under us after execve (like [lin]procfs). 2879 * 2880 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2881 * sufficient. We also don't check for setugidness since we know we are. 2882 */ 2883 static bool 2884 is_unsafe(struct file *fp) 2885 { 2886 struct vnode *vp; 2887 2888 if (fp->f_type != DTYPE_VNODE) 2889 return (false); 2890 2891 vp = fp->f_vnode; 2892 return ((vp->v_vflag & VV_PROCDEP) != 0); 2893 } 2894 2895 /* 2896 * Make this setguid thing safe, if at all possible. 2897 */ 2898 void 2899 fdsetugidsafety(struct thread *td) 2900 { 2901 struct filedesc *fdp; 2902 struct file *fp; 2903 int i; 2904 2905 fdp = td->td_proc->p_fd; 2906 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2907 ("the fdtable should not be shared")); 2908 MPASS(fdp->fd_nfiles >= 3); 2909 for (i = 0; i <= 2; i++) { 2910 fp = fdp->fd_ofiles[i].fde_file; 2911 if (fp != NULL && is_unsafe(fp)) { 2912 FILEDESC_XLOCK(fdp); 2913 knote_fdclose(td, i); 2914 /* 2915 * NULL-out descriptor prior to close to avoid 2916 * a race while close blocks. 2917 */ 2918 fdfree(fdp, i); 2919 FILEDESC_XUNLOCK(fdp); 2920 (void) closef(fp, td); 2921 } 2922 } 2923 } 2924 2925 /* 2926 * If a specific file object occupies a specific file descriptor, close the 2927 * file descriptor entry and drop a reference on the file object. This is a 2928 * convenience function to handle a subsequent error in a function that calls 2929 * falloc() that handles the race that another thread might have closed the 2930 * file descriptor out from under the thread creating the file object. 2931 */ 2932 void 2933 fdclose(struct thread *td, struct file *fp, int idx) 2934 { 2935 struct filedesc *fdp = td->td_proc->p_fd; 2936 2937 FILEDESC_XLOCK(fdp); 2938 if (fdp->fd_ofiles[idx].fde_file == fp) { 2939 fdfree(fdp, idx); 2940 FILEDESC_XUNLOCK(fdp); 2941 fdrop(fp, td); 2942 } else 2943 FILEDESC_XUNLOCK(fdp); 2944 } 2945 2946 /* 2947 * Close any files on exec? 2948 */ 2949 void 2950 fdcloseexec(struct thread *td) 2951 { 2952 struct filedesc *fdp; 2953 struct filedescent *fde; 2954 struct file *fp; 2955 int i; 2956 2957 fdp = td->td_proc->p_fd; 2958 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2959 ("the fdtable should not be shared")); 2960 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2961 fp = fde->fde_file; 2962 if (fp->f_type == DTYPE_MQUEUE || 2963 (fde->fde_flags & UF_EXCLOSE)) { 2964 FILEDESC_XLOCK(fdp); 2965 fdfree(fdp, i); 2966 (void) closefp(fdp, i, fp, td, false, false); 2967 FILEDESC_UNLOCK_ASSERT(fdp); 2968 } else if (fde->fde_flags & UF_FOCLOSE) { 2969 /* 2970 * https://austingroupbugs.net/view.php?id=1851 2971 * FD_CLOFORK should not be preserved across exec 2972 */ 2973 fde->fde_flags &= ~UF_FOCLOSE; 2974 } 2975 } 2976 } 2977 2978 /* 2979 * It is unsafe for set[ug]id processes to be started with file 2980 * descriptors 0..2 closed, as these descriptors are given implicit 2981 * significance in the Standard C library. fdcheckstd() will create a 2982 * descriptor referencing /dev/null for each of stdin, stdout, and 2983 * stderr that is not already open. 2984 */ 2985 int 2986 fdcheckstd(struct thread *td) 2987 { 2988 struct filedesc *fdp; 2989 register_t save; 2990 int i, error, devnull; 2991 2992 fdp = td->td_proc->p_fd; 2993 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2994 ("the fdtable should not be shared")); 2995 MPASS(fdp->fd_nfiles >= 3); 2996 devnull = -1; 2997 for (i = 0; i <= 2; i++) { 2998 if (fdp->fd_ofiles[i].fde_file != NULL) 2999 continue; 3000 3001 save = td->td_retval[0]; 3002 if (devnull != -1) { 3003 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 3004 } else { 3005 error = kern_openat(td, AT_FDCWD, "/dev/null", 3006 UIO_SYSSPACE, O_RDWR, 0); 3007 if (error == 0) { 3008 devnull = td->td_retval[0]; 3009 KASSERT(devnull == i, ("we didn't get our fd")); 3010 } 3011 } 3012 td->td_retval[0] = save; 3013 if (error != 0) 3014 return (error); 3015 } 3016 return (0); 3017 } 3018 3019 /* 3020 * Internal form of close. Decrement reference count on file structure. 3021 * Note: td may be NULL when closing a file that was being passed in a 3022 * message. 3023 */ 3024 int 3025 closef(struct file *fp, struct thread *td) 3026 { 3027 struct vnode *vp; 3028 struct flock lf; 3029 struct filedesc_to_leader *fdtol; 3030 struct filedesc *fdp; 3031 3032 MPASS(td != NULL); 3033 3034 /* 3035 * POSIX record locking dictates that any close releases ALL 3036 * locks owned by this process. This is handled by setting 3037 * a flag in the unlock to free ONLY locks obeying POSIX 3038 * semantics, and not to free BSD-style file locks. 3039 * If the descriptor was in a message, POSIX-style locks 3040 * aren't passed with the descriptor, and the thread pointer 3041 * will be NULL. Callers should be careful only to pass a 3042 * NULL thread pointer when there really is no owning 3043 * context that might have locks, or the locks will be 3044 * leaked. 3045 */ 3046 if (fp->f_type == DTYPE_VNODE) { 3047 vp = fp->f_vnode; 3048 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 3049 lf.l_whence = SEEK_SET; 3050 lf.l_start = 0; 3051 lf.l_len = 0; 3052 lf.l_type = F_UNLCK; 3053 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 3054 F_UNLCK, &lf, F_POSIX); 3055 } 3056 fdtol = td->td_proc->p_fdtol; 3057 if (fdtol != NULL) { 3058 /* 3059 * Handle special case where file descriptor table is 3060 * shared between multiple process leaders. 3061 */ 3062 fdp = td->td_proc->p_fd; 3063 FILEDESC_XLOCK(fdp); 3064 for (fdtol = fdtol->fdl_next; 3065 fdtol != td->td_proc->p_fdtol; 3066 fdtol = fdtol->fdl_next) { 3067 if ((fdtol->fdl_leader->p_flag & 3068 P_ADVLOCK) == 0) 3069 continue; 3070 fdtol->fdl_holdcount++; 3071 FILEDESC_XUNLOCK(fdp); 3072 lf.l_whence = SEEK_SET; 3073 lf.l_start = 0; 3074 lf.l_len = 0; 3075 lf.l_type = F_UNLCK; 3076 vp = fp->f_vnode; 3077 (void) VOP_ADVLOCK(vp, 3078 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 3079 F_POSIX); 3080 FILEDESC_XLOCK(fdp); 3081 fdtol->fdl_holdcount--; 3082 if (fdtol->fdl_holdcount == 0 && 3083 fdtol->fdl_wakeup != 0) { 3084 fdtol->fdl_wakeup = 0; 3085 wakeup(fdtol); 3086 } 3087 } 3088 FILEDESC_XUNLOCK(fdp); 3089 } 3090 } 3091 return (fdrop_close(fp, td)); 3092 } 3093 3094 /* 3095 * Hack for file descriptor passing code. 3096 */ 3097 void 3098 closef_nothread(struct file *fp) 3099 { 3100 3101 fdrop(fp, NULL); 3102 } 3103 3104 /* 3105 * Initialize the file pointer with the specified properties. 3106 * 3107 * The ops are set with release semantics to be certain that the flags, type, 3108 * and data are visible when ops is. This is to prevent ops methods from being 3109 * called with bad data. 3110 */ 3111 void 3112 finit(struct file *fp, u_int flag, short type, void *data, 3113 const struct fileops *ops) 3114 { 3115 fp->f_data = data; 3116 fp->f_flag = flag; 3117 fp->f_type = type; 3118 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 3119 } 3120 3121 void 3122 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops) 3123 { 3124 fp->f_seqcount[UIO_READ] = 1; 3125 fp->f_seqcount[UIO_WRITE] = 1; 3126 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 3127 data, ops); 3128 } 3129 3130 int 3131 fget_cap_noref(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3132 struct file **fpp, struct filecaps *havecapsp) 3133 { 3134 struct filedescent *fde; 3135 int error; 3136 3137 FILEDESC_LOCK_ASSERT(fdp); 3138 3139 *fpp = NULL; 3140 fde = fdeget_noref(fdp, fd); 3141 if (fde == NULL) { 3142 error = EBADF; 3143 goto out; 3144 } 3145 3146 #ifdef CAPABILITIES 3147 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 3148 if (error != 0) 3149 goto out; 3150 #endif 3151 3152 if (havecapsp != NULL) 3153 filecaps_copy(&fde->fde_caps, havecapsp, true); 3154 3155 *fpp = fde->fde_file; 3156 3157 error = 0; 3158 out: 3159 return (error); 3160 } 3161 3162 #ifdef CAPABILITIES 3163 int 3164 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3165 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3166 { 3167 struct filedesc *fdp = td->td_proc->p_fd; 3168 int error; 3169 struct file *fp; 3170 seqc_t seq; 3171 3172 *fpp = NULL; 3173 for (;;) { 3174 error = fget_unlocked_seq(td, fd, needrightsp, flagsp, &fp, 3175 &seq); 3176 if (error != 0) 3177 return (error); 3178 3179 if (havecapsp != NULL) { 3180 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 3181 havecapsp, false)) { 3182 fdrop(fp, td); 3183 goto get_locked; 3184 } 3185 } 3186 3187 if (!fd_modified(fdp, fd, seq)) 3188 break; 3189 fdrop(fp, td); 3190 } 3191 3192 *fpp = fp; 3193 return (0); 3194 3195 get_locked: 3196 FILEDESC_SLOCK(fdp); 3197 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp); 3198 if (error == 0 && !fhold(*fpp)) 3199 error = EBADF; 3200 FILEDESC_SUNLOCK(fdp); 3201 return (error); 3202 } 3203 #else 3204 int 3205 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3206 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3207 { 3208 int error; 3209 error = fget_unlocked_flags(td, fd, needrightsp, flagsp, fpp); 3210 if (havecapsp != NULL && error == 0) 3211 filecaps_fill(havecapsp); 3212 3213 return (error); 3214 } 3215 #endif 3216 3217 int 3218 fget_remote(struct thread *td, struct proc *p, int fd, struct filecaps *fcaps, 3219 uint8_t *fd_flags, struct file **fpp) 3220 { 3221 struct filedesc *fdp; 3222 struct file *fp; 3223 int error; 3224 bool copied __diagused; 3225 3226 /* 3227 * Both fcaps and fd_flags must be either requested together, 3228 * or not at all. 3229 */ 3230 MPASS((!(fcaps == NULL) ^ (fd_flags == NULL))); 3231 3232 if (p == td->td_proc && fcaps == NULL) /* curproc */ 3233 return (fget_unlocked(td, fd, &cap_no_rights, fpp)); 3234 3235 PROC_LOCK(p); 3236 fdp = fdhold(p); 3237 PROC_UNLOCK(p); 3238 if (fdp == NULL) 3239 return (ENOENT); 3240 FILEDESC_SLOCK(fdp); 3241 if (refcount_load(&fdp->fd_refcnt) != 0) { 3242 fp = fget_noref(fdp, fd); 3243 if (fp != NULL && fhold(fp)) { 3244 *fpp = fp; 3245 if (fd_flags != NULL) { 3246 *fd_flags = fde_to_fd_flags(fdp->fd_ofiles[fd]. 3247 fde_flags); 3248 } 3249 if (fcaps != NULL) { 3250 copied = filecaps_copy( 3251 &fdp->fd_ofiles[fd].fde_caps, fcaps, true); 3252 MPASS(copied); 3253 } 3254 error = 0; 3255 } else { 3256 error = EBADF; 3257 } 3258 } else { 3259 error = ENOENT; 3260 } 3261 FILEDESC_SUNLOCK(fdp); 3262 fddrop(fdp); 3263 return (error); 3264 } 3265 3266 int 3267 fget_remote_foreach(struct thread *td, struct proc *p, 3268 int (*fn)(struct proc *, int, struct file *, void *), void *arg) 3269 { 3270 struct filedesc *fdp; 3271 struct fdescenttbl *fdt; 3272 struct file *fp; 3273 int error, error1, fd, highfd; 3274 3275 error = 0; 3276 PROC_LOCK(p); 3277 fdp = fdhold(p); 3278 PROC_UNLOCK(p); 3279 if (fdp == NULL) 3280 return (ENOENT); 3281 3282 FILEDESC_SLOCK(fdp); 3283 if (refcount_load(&fdp->fd_refcnt) != 0) { 3284 fdt = atomic_load_ptr(&fdp->fd_files); 3285 highfd = fdt->fdt_nfiles - 1; 3286 FILEDESC_SUNLOCK(fdp); 3287 } else { 3288 error = ENOENT; 3289 FILEDESC_SUNLOCK(fdp); 3290 goto out; 3291 } 3292 3293 for (fd = 0; fd <= highfd; fd++) { 3294 error1 = fget_remote(td, p, fd, NULL, NULL, &fp); 3295 if (error1 != 0) 3296 continue; 3297 error = fn(p, fd, fp, arg); 3298 fdrop(fp, td); 3299 if (error != 0) 3300 break; 3301 } 3302 out: 3303 fddrop(fdp); 3304 return (error); 3305 } 3306 3307 #ifdef CAPABILITIES 3308 int 3309 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3310 { 3311 const struct filedescent *fde; 3312 const struct fdescenttbl *fdt; 3313 struct filedesc *fdp; 3314 struct file *fp; 3315 struct vnode *vp; 3316 const cap_rights_t *haverights; 3317 cap_rights_t rights; 3318 seqc_t seq; 3319 int fd, flags; 3320 3321 VFS_SMR_ASSERT_ENTERED(); 3322 3323 fd = ndp->ni_dirfd; 3324 rights = *ndp->ni_rightsneeded; 3325 cap_rights_set_one(&rights, CAP_LOOKUP); 3326 3327 fdp = curproc->p_fd; 3328 fdt = fdp->fd_files; 3329 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3330 return (EBADF); 3331 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3332 fde = &fdt->fdt_ofiles[fd]; 3333 haverights = cap_rights_fde_inline(fde); 3334 fp = fde->fde_file; 3335 if (__predict_false(fp == NULL)) 3336 return (EAGAIN); 3337 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 3338 return (EAGAIN); 3339 flags = fp->f_flag & FSEARCH; 3340 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3341 O_RESOLVE_BENEATH : 0; 3342 vp = fp->f_vnode; 3343 if (__predict_false(vp == NULL)) { 3344 return (EAGAIN); 3345 } 3346 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 3347 return (EAGAIN); 3348 } 3349 /* 3350 * Use an acquire barrier to force re-reading of fdt so it is 3351 * refreshed for verification. 3352 */ 3353 atomic_thread_fence_acq(); 3354 fdt = fdp->fd_files; 3355 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3356 return (EAGAIN); 3357 /* 3358 * If file descriptor doesn't have all rights, 3359 * all lookups relative to it must also be 3360 * strictly relative. 3361 * 3362 * Not yet supported by fast path. 3363 */ 3364 if (!filecaps_full(&ndp->ni_filecaps)) { 3365 #ifdef notyet 3366 ndp->ni_lcf |= NI_LCF_STRICTREL; 3367 #else 3368 return (EAGAIN); 3369 #endif 3370 } 3371 *vpp = vp; 3372 *flagsp = flags; 3373 return (0); 3374 } 3375 #else 3376 int 3377 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3378 { 3379 const struct filedescent *fde; 3380 const struct fdescenttbl *fdt; 3381 struct filedesc *fdp; 3382 struct file *fp; 3383 struct vnode *vp; 3384 int fd, flags; 3385 3386 VFS_SMR_ASSERT_ENTERED(); 3387 3388 fd = ndp->ni_dirfd; 3389 fdp = curproc->p_fd; 3390 fdt = fdp->fd_files; 3391 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3392 return (EBADF); 3393 fde = &fdt->fdt_ofiles[fd]; 3394 fp = fde->fde_file; 3395 if (__predict_false(fp == NULL)) 3396 return (EAGAIN); 3397 flags = fp->f_flag & FSEARCH; 3398 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3399 O_RESOLVE_BENEATH : 0; 3400 vp = fp->f_vnode; 3401 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3402 return (EAGAIN); 3403 } 3404 /* 3405 * Use an acquire barrier to force re-reading of fdt so it is 3406 * refreshed for verification. 3407 */ 3408 atomic_thread_fence_acq(); 3409 fdt = fdp->fd_files; 3410 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3411 return (EAGAIN); 3412 filecaps_fill(&ndp->ni_filecaps); 3413 *vpp = vp; 3414 *flagsp = flags; 3415 return (0); 3416 } 3417 #endif 3418 3419 int 3420 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp) 3421 { 3422 struct thread *td; 3423 struct file *fp; 3424 struct vnode *vp; 3425 struct componentname *cnp; 3426 cap_rights_t rights; 3427 int error; 3428 uint8_t flags; 3429 3430 td = curthread; 3431 rights = *ndp->ni_rightsneeded; 3432 cap_rights_set_one(&rights, CAP_LOOKUP); 3433 cnp = &ndp->ni_cnd; 3434 3435 error = fget_cap(td, ndp->ni_dirfd, &rights, &flags, &fp, 3436 &ndp->ni_filecaps); 3437 if (__predict_false(error != 0)) 3438 return (error); 3439 if (__predict_false(fp->f_ops == &badfileops)) { 3440 error = EBADF; 3441 goto out_free; 3442 } 3443 vp = fp->f_vnode; 3444 if (__predict_false(vp == NULL)) { 3445 error = ENOTDIR; 3446 goto out_free; 3447 } 3448 vrefact(vp); 3449 /* 3450 * XXX does not check for VDIR, handled by namei_setup 3451 */ 3452 if ((fp->f_flag & FSEARCH) != 0) 3453 cnp->cn_flags |= NOEXECCHECK; 3454 if ((flags & UF_RESOLVE_BENEATH) != 0) { 3455 cnp->cn_flags |= RBENEATH; 3456 ndp->ni_resflags |= NIRES_BENEATH; 3457 } 3458 fdrop(fp, td); 3459 3460 #ifdef CAPABILITIES 3461 /* 3462 * If file descriptor doesn't have all rights, 3463 * all lookups relative to it must also be 3464 * strictly relative. 3465 */ 3466 if (!filecaps_full(&ndp->ni_filecaps)) { 3467 ndp->ni_lcf |= NI_LCF_STRICTREL; 3468 ndp->ni_resflags |= NIRES_STRICTREL; 3469 } 3470 #endif 3471 3472 /* 3473 * TODO: avoid copying ioctl caps if it can be helped to begin with 3474 */ 3475 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0) 3476 filecaps_free_ioctl(&ndp->ni_filecaps); 3477 3478 *vpp = vp; 3479 return (0); 3480 3481 out_free: 3482 filecaps_free(&ndp->ni_filecaps); 3483 fdrop(fp, td); 3484 return (error); 3485 } 3486 3487 /* 3488 * Fetch the descriptor locklessly. 3489 * 3490 * We avoid fdrop() races by never raising a refcount above 0. To accomplish 3491 * this we have to use a cmpset loop rather than an atomic_add. The descriptor 3492 * must be re-verified once we acquire a reference to be certain that the 3493 * identity is still correct and we did not lose a race due to preemption. 3494 * 3495 * Force a reload of fdt when looping. Another thread could reallocate 3496 * the table before this fd was closed, so it is possible that there is 3497 * a stale fp pointer in cached version. 3498 */ 3499 #ifdef CAPABILITIES 3500 static int 3501 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3502 uint8_t *flagsp, struct file **fpp, seqc_t *seqp) 3503 { 3504 struct filedesc *fdp; 3505 const struct filedescent *fde; 3506 const struct fdescenttbl *fdt; 3507 struct file *fp; 3508 seqc_t seq; 3509 cap_rights_t haverights; 3510 int error; 3511 uint8_t flags; 3512 3513 fdp = td->td_proc->p_fd; 3514 fdt = fdp->fd_files; 3515 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3516 return (EBADF); 3517 3518 for (;;) { 3519 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3520 fde = &fdt->fdt_ofiles[fd]; 3521 haverights = *cap_rights_fde_inline(fde); 3522 fp = fde->fde_file; 3523 flags = fde->fde_flags; 3524 if (__predict_false(fp == NULL)) { 3525 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3526 return (EBADF); 3527 fdt = atomic_load_ptr(&fdp->fd_files); 3528 continue; 3529 } 3530 error = cap_check_inline(&haverights, needrightsp); 3531 if (__predict_false(error != 0)) { 3532 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3533 return (error); 3534 fdt = atomic_load_ptr(&fdp->fd_files); 3535 continue; 3536 } 3537 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3538 fdt = atomic_load_ptr(&fdp->fd_files); 3539 continue; 3540 } 3541 /* 3542 * Use an acquire barrier to force re-reading of fdt so it is 3543 * refreshed for verification. 3544 */ 3545 atomic_thread_fence_acq(); 3546 fdt = fdp->fd_files; 3547 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)) 3548 break; 3549 fdrop(fp, td); 3550 } 3551 *fpp = fp; 3552 if (flagsp != NULL) 3553 *flagsp = flags; 3554 if (seqp != NULL) 3555 *seqp = seq; 3556 return (0); 3557 } 3558 #else 3559 static int 3560 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3561 uint8_t *flagsp, struct file **fpp, seqc_t *seqp __unused) 3562 { 3563 struct filedesc *fdp; 3564 const struct fdescenttbl *fdt; 3565 struct file *fp; 3566 uint8_t flags; 3567 3568 fdp = td->td_proc->p_fd; 3569 fdt = fdp->fd_files; 3570 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3571 return (EBADF); 3572 3573 for (;;) { 3574 fp = fdt->fdt_ofiles[fd].fde_file; 3575 flags = fdt->fdt_ofiles[fd].fde_flags; 3576 if (__predict_false(fp == NULL)) 3577 return (EBADF); 3578 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3579 fdt = atomic_load_ptr(&fdp->fd_files); 3580 continue; 3581 } 3582 /* 3583 * Use an acquire barrier to force re-reading of fdt so it is 3584 * refreshed for verification. 3585 */ 3586 atomic_thread_fence_acq(); 3587 fdt = fdp->fd_files; 3588 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file)) 3589 break; 3590 fdrop(fp, td); 3591 } 3592 if (flagsp != NULL) 3593 *flagsp = flags; 3594 *fpp = fp; 3595 return (0); 3596 } 3597 #endif 3598 3599 /* 3600 * See the comments in fget_unlocked_seq for an explanation of how this works. 3601 * 3602 * This is a simplified variant which bails out to the aforementioned routine 3603 * if anything goes wrong. In practice this only happens when userspace is 3604 * racing with itself. 3605 */ 3606 int 3607 fget_unlocked_flags(struct thread *td, int fd, const cap_rights_t *needrightsp, 3608 uint8_t *flagsp, struct file **fpp) 3609 { 3610 struct filedesc *fdp; 3611 #ifdef CAPABILITIES 3612 const struct filedescent *fde; 3613 #endif 3614 const struct fdescenttbl *fdt; 3615 struct file *fp; 3616 #ifdef CAPABILITIES 3617 seqc_t seq; 3618 const cap_rights_t *haverights; 3619 #endif 3620 uint8_t flags; 3621 3622 fdp = td->td_proc->p_fd; 3623 fdt = fdp->fd_files; 3624 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { 3625 *fpp = NULL; 3626 return (EBADF); 3627 } 3628 #ifdef CAPABILITIES 3629 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3630 fde = &fdt->fdt_ofiles[fd]; 3631 haverights = cap_rights_fde_inline(fde); 3632 fp = fde->fde_file; 3633 flags = fde->fde_flags; 3634 #else 3635 fp = fdt->fdt_ofiles[fd].fde_file; 3636 flags = fdt->fdt_ofiles[fd].fde_flags; 3637 #endif 3638 if (__predict_false(fp == NULL)) 3639 goto out_fallback; 3640 #ifdef CAPABILITIES 3641 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3642 goto out_fallback; 3643 #endif 3644 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3645 goto out_fallback; 3646 3647 /* 3648 * Use an acquire barrier to force re-reading of fdt so it is 3649 * refreshed for verification. 3650 */ 3651 atomic_thread_fence_acq(); 3652 fdt = fdp->fd_files; 3653 #ifdef CAPABILITIES 3654 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3655 #else 3656 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3657 #endif 3658 goto out_fdrop; 3659 *fpp = fp; 3660 if (flagsp != NULL) 3661 *flagsp = flags; 3662 return (0); 3663 out_fdrop: 3664 fdrop(fp, td); 3665 out_fallback: 3666 *fpp = NULL; 3667 return (fget_unlocked_seq(td, fd, needrightsp, flagsp, fpp, NULL)); 3668 } 3669 3670 int 3671 fget_unlocked(struct thread *td, int fd, const cap_rights_t *needrightsp, 3672 struct file **fpp) 3673 { 3674 return (fget_unlocked_flags(td, fd, needrightsp, NULL, fpp)); 3675 } 3676 3677 /* 3678 * Translate fd -> file when the caller guarantees the file descriptor table 3679 * can't be changed by others. 3680 * 3681 * Note this does not mean the file object itself is only visible to the caller, 3682 * merely that it wont disappear without having to be referenced. 3683 * 3684 * Must be paired with fput_only_user. 3685 */ 3686 #ifdef CAPABILITIES 3687 int 3688 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3689 struct file **fpp) 3690 { 3691 const struct filedescent *fde; 3692 const struct fdescenttbl *fdt; 3693 const cap_rights_t *haverights; 3694 struct file *fp; 3695 int error; 3696 3697 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3698 3699 *fpp = NULL; 3700 if (__predict_false(fd >= fdp->fd_nfiles)) 3701 return (EBADF); 3702 3703 fdt = fdp->fd_files; 3704 fde = &fdt->fdt_ofiles[fd]; 3705 fp = fde->fde_file; 3706 if (__predict_false(fp == NULL)) 3707 return (EBADF); 3708 MPASS(refcount_load(&fp->f_count) > 0); 3709 haverights = cap_rights_fde_inline(fde); 3710 error = cap_check_inline(haverights, needrightsp); 3711 if (__predict_false(error != 0)) 3712 return (error); 3713 *fpp = fp; 3714 return (0); 3715 } 3716 #else 3717 int 3718 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3719 struct file **fpp) 3720 { 3721 struct file *fp; 3722 3723 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3724 3725 *fpp = NULL; 3726 if (__predict_false(fd >= fdp->fd_nfiles)) 3727 return (EBADF); 3728 3729 fp = fdp->fd_ofiles[fd].fde_file; 3730 if (__predict_false(fp == NULL)) 3731 return (EBADF); 3732 3733 MPASS(refcount_load(&fp->f_count) > 0); 3734 *fpp = fp; 3735 return (0); 3736 } 3737 #endif 3738 3739 /* 3740 * Extract the file pointer associated with the specified descriptor for the 3741 * current user process. 3742 * 3743 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3744 * returned. 3745 * 3746 * File's rights will be checked against the capability rights mask. 3747 * 3748 * If an error occurred the non-zero error is returned and *fpp is set to 3749 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3750 * responsible for fdrop(). 3751 */ 3752 static __inline int 3753 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3754 const cap_rights_t *needrightsp) 3755 { 3756 struct file *fp; 3757 int error; 3758 3759 *fpp = NULL; 3760 error = fget_unlocked(td, fd, needrightsp, &fp); 3761 if (__predict_false(error != 0)) 3762 return (error); 3763 if (__predict_false(fp->f_ops == &badfileops)) { 3764 fdrop(fp, td); 3765 return (EBADF); 3766 } 3767 3768 /* 3769 * FREAD and FWRITE failure return EBADF as per POSIX. 3770 */ 3771 error = 0; 3772 switch (flags) { 3773 case FREAD: 3774 case FWRITE: 3775 if ((fp->f_flag & flags) == 0) 3776 error = EBADF; 3777 break; 3778 case FEXEC: 3779 if (fp->f_ops != &path_fileops && 3780 ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3781 (fp->f_flag & FWRITE) != 0)) 3782 error = EBADF; 3783 break; 3784 case 0: 3785 break; 3786 default: 3787 KASSERT(0, ("wrong flags")); 3788 } 3789 3790 if (error != 0) { 3791 fdrop(fp, td); 3792 return (error); 3793 } 3794 3795 *fpp = fp; 3796 return (0); 3797 } 3798 3799 int 3800 fget(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp) 3801 { 3802 3803 return (_fget(td, fd, fpp, 0, rightsp)); 3804 } 3805 3806 int 3807 fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp, 3808 vm_prot_t *maxprotp, struct file **fpp) 3809 { 3810 int error; 3811 #ifndef CAPABILITIES 3812 error = _fget(td, fd, fpp, 0, rightsp); 3813 if (maxprotp != NULL) 3814 *maxprotp = VM_PROT_ALL; 3815 return (error); 3816 #else 3817 cap_rights_t fdrights; 3818 struct filedesc *fdp; 3819 struct file *fp; 3820 seqc_t seq; 3821 3822 *fpp = NULL; 3823 fdp = td->td_proc->p_fd; 3824 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3825 for (;;) { 3826 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3827 if (__predict_false(error != 0)) 3828 return (error); 3829 if (__predict_false(fp->f_ops == &badfileops)) { 3830 fdrop(fp, td); 3831 return (EBADF); 3832 } 3833 if (maxprotp != NULL) 3834 fdrights = *cap_rights(fdp, fd); 3835 if (!fd_modified(fdp, fd, seq)) 3836 break; 3837 fdrop(fp, td); 3838 } 3839 3840 /* 3841 * If requested, convert capability rights to access flags. 3842 */ 3843 if (maxprotp != NULL) 3844 *maxprotp = cap_rights_to_vmprot(&fdrights); 3845 *fpp = fp; 3846 return (0); 3847 #endif 3848 } 3849 3850 int 3851 fget_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3852 struct file **fpp) 3853 { 3854 3855 return (_fget(td, fd, fpp, FREAD, rightsp)); 3856 } 3857 3858 int 3859 fget_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3860 struct file **fpp) 3861 { 3862 3863 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3864 } 3865 3866 int 3867 fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp, 3868 int needfcntl, struct file **fpp) 3869 { 3870 #ifndef CAPABILITIES 3871 return (fget_unlocked(td, fd, rightsp, fpp)); 3872 #else 3873 struct filedesc *fdp = td->td_proc->p_fd; 3874 struct file *fp; 3875 int error; 3876 seqc_t seq; 3877 3878 *fpp = NULL; 3879 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3880 for (;;) { 3881 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3882 if (error != 0) 3883 return (error); 3884 error = cap_fcntl_check(fdp, fd, needfcntl); 3885 if (!fd_modified(fdp, fd, seq)) 3886 break; 3887 fdrop(fp, td); 3888 } 3889 if (error != 0) { 3890 fdrop(fp, td); 3891 return (error); 3892 } 3893 *fpp = fp; 3894 return (0); 3895 #endif 3896 } 3897 3898 /* 3899 * Like fget() but loads the underlying vnode, or returns an error if the 3900 * descriptor does not represent a vnode. Note that pipes use vnodes but 3901 * never have VM objects. The returned vnode will be vref()'d. 3902 * 3903 * XXX: what about the unused flags ? 3904 */ 3905 static __inline int 3906 _fgetvp(struct thread *td, int fd, int flags, const cap_rights_t *needrightsp, 3907 struct vnode **vpp) 3908 { 3909 struct file *fp; 3910 int error; 3911 3912 *vpp = NULL; 3913 error = _fget(td, fd, &fp, flags, needrightsp); 3914 if (error != 0) 3915 return (error); 3916 if (fp->f_vnode == NULL) { 3917 error = EINVAL; 3918 } else { 3919 *vpp = fp->f_vnode; 3920 vrefact(*vpp); 3921 } 3922 fdrop(fp, td); 3923 3924 return (error); 3925 } 3926 3927 int 3928 fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp, 3929 struct vnode **vpp) 3930 { 3931 3932 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3933 } 3934 3935 int 3936 fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp, 3937 struct filecaps *havecaps, struct vnode **vpp) 3938 { 3939 struct filecaps caps; 3940 struct file *fp; 3941 int error; 3942 3943 error = fget_cap(td, fd, needrightsp, NULL, &fp, &caps); 3944 if (error != 0) 3945 return (error); 3946 if (fp->f_ops == &badfileops) { 3947 error = EBADF; 3948 goto out; 3949 } 3950 if (fp->f_vnode == NULL) { 3951 error = EINVAL; 3952 goto out; 3953 } 3954 3955 *havecaps = caps; 3956 *vpp = fp->f_vnode; 3957 vrefact(*vpp); 3958 fdrop(fp, td); 3959 3960 return (0); 3961 out: 3962 filecaps_free(&caps); 3963 fdrop(fp, td); 3964 return (error); 3965 } 3966 3967 int 3968 fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3969 struct vnode **vpp) 3970 { 3971 3972 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3973 } 3974 3975 int 3976 fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp, 3977 struct vnode **vpp) 3978 { 3979 3980 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3981 } 3982 3983 #ifdef notyet 3984 int 3985 fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3986 struct vnode **vpp) 3987 { 3988 3989 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3990 } 3991 #endif 3992 3993 /* 3994 * Handle the last reference to a file being closed. 3995 * 3996 * Without the noinline attribute clang keeps inlining the func thorough this 3997 * file when fdrop is used. 3998 */ 3999 int __noinline 4000 _fdrop(struct file *fp, struct thread *td) 4001 { 4002 int error; 4003 4004 KASSERT(refcount_load(&fp->f_count) == 0, 4005 ("fdrop: fp %p count %d", fp, refcount_load(&fp->f_count))); 4006 4007 error = fo_close(fp, td); 4008 atomic_subtract_int(&openfiles, 1); 4009 crfree(fp->f_cred); 4010 free(fp->f_advice, M_FADVISE); 4011 uma_zfree(file_zone, fp); 4012 4013 return (error); 4014 } 4015 4016 /* 4017 * Apply an advisory lock on a file descriptor. 4018 * 4019 * Just attempt to get a record lock of the requested type on the entire file 4020 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 4021 */ 4022 #ifndef _SYS_SYSPROTO_H_ 4023 struct flock_args { 4024 int fd; 4025 int how; 4026 }; 4027 #endif 4028 /* ARGSUSED */ 4029 int 4030 sys_flock(struct thread *td, struct flock_args *uap) 4031 { 4032 struct file *fp; 4033 struct vnode *vp; 4034 struct flock lf; 4035 int error; 4036 4037 error = fget(td, uap->fd, &cap_flock_rights, &fp); 4038 if (error != 0) 4039 return (error); 4040 error = EOPNOTSUPP; 4041 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 4042 goto done; 4043 } 4044 if (fp->f_ops == &path_fileops) { 4045 goto done; 4046 } 4047 4048 error = 0; 4049 vp = fp->f_vnode; 4050 lf.l_whence = SEEK_SET; 4051 lf.l_start = 0; 4052 lf.l_len = 0; 4053 if (uap->how & LOCK_UN) { 4054 lf.l_type = F_UNLCK; 4055 atomic_clear_int(&fp->f_flag, FHASLOCK); 4056 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 4057 goto done; 4058 } 4059 if (uap->how & LOCK_EX) 4060 lf.l_type = F_WRLCK; 4061 else if (uap->how & LOCK_SH) 4062 lf.l_type = F_RDLCK; 4063 else { 4064 error = EBADF; 4065 goto done; 4066 } 4067 atomic_set_int(&fp->f_flag, FHASLOCK); 4068 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 4069 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 4070 done: 4071 fdrop(fp, td); 4072 return (error); 4073 } 4074 /* 4075 * Duplicate the specified descriptor to a free descriptor. 4076 */ 4077 int 4078 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 4079 int openerror, int *indxp) 4080 { 4081 struct filedescent *newfde, *oldfde; 4082 struct file *fp; 4083 u_long *ioctls; 4084 int error, indx; 4085 4086 KASSERT(openerror == ENODEV || openerror == ENXIO, 4087 ("unexpected error %d in %s", openerror, __func__)); 4088 4089 /* 4090 * If the to-be-dup'd fd number is greater than the allowed number 4091 * of file descriptors, or the fd to be dup'd has already been 4092 * closed, then reject. 4093 */ 4094 FILEDESC_XLOCK(fdp); 4095 if ((fp = fget_noref(fdp, dfd)) == NULL) { 4096 FILEDESC_XUNLOCK(fdp); 4097 return (EBADF); 4098 } 4099 4100 error = fdalloc(td, 0, &indx); 4101 if (error != 0) { 4102 FILEDESC_XUNLOCK(fdp); 4103 return (error); 4104 } 4105 4106 /* 4107 * There are two cases of interest here. 4108 * 4109 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 4110 * 4111 * For ENXIO steal away the file structure from (dfd) and store it in 4112 * (indx). (dfd) is effectively closed by this operation. 4113 */ 4114 switch (openerror) { 4115 case ENODEV: 4116 /* 4117 * Check that the mode the file is being opened for is a 4118 * subset of the mode of the existing descriptor. 4119 */ 4120 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 4121 fdunused(fdp, indx); 4122 FILEDESC_XUNLOCK(fdp); 4123 return (EACCES); 4124 } 4125 if (!fhold(fp)) { 4126 fdunused(fdp, indx); 4127 FILEDESC_XUNLOCK(fdp); 4128 return (EBADF); 4129 } 4130 newfde = &fdp->fd_ofiles[indx]; 4131 oldfde = &fdp->fd_ofiles[dfd]; 4132 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 4133 #ifdef CAPABILITIES 4134 seqc_write_begin(&newfde->fde_seqc); 4135 #endif 4136 fde_copy(oldfde, newfde); 4137 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 4138 ioctls); 4139 #ifdef CAPABILITIES 4140 seqc_write_end(&newfde->fde_seqc); 4141 #endif 4142 break; 4143 case ENXIO: 4144 /* 4145 * Steal away the file pointer from dfd and stuff it into indx. 4146 */ 4147 newfde = &fdp->fd_ofiles[indx]; 4148 oldfde = &fdp->fd_ofiles[dfd]; 4149 #ifdef CAPABILITIES 4150 seqc_write_begin(&oldfde->fde_seqc); 4151 seqc_write_begin(&newfde->fde_seqc); 4152 #endif 4153 fde_copy(oldfde, newfde); 4154 oldfde->fde_file = NULL; 4155 fdunused(fdp, dfd); 4156 #ifdef CAPABILITIES 4157 seqc_write_end(&newfde->fde_seqc); 4158 seqc_write_end(&oldfde->fde_seqc); 4159 #endif 4160 break; 4161 } 4162 FILEDESC_XUNLOCK(fdp); 4163 *indxp = indx; 4164 return (0); 4165 } 4166 4167 /* 4168 * This sysctl determines if we will allow a process to chroot(2) if it 4169 * has a directory open: 4170 * 0: disallowed for all processes. 4171 * 1: allowed for processes that were not already chroot(2)'ed. 4172 * 2: allowed for all processes. 4173 */ 4174 4175 static int chroot_allow_open_directories = 1; 4176 4177 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 4178 &chroot_allow_open_directories, 0, 4179 "Allow a process to chroot(2) if it has a directory open"); 4180 4181 /* 4182 * Helper function for raised chroot(2) security function: Refuse if 4183 * any filedescriptors are open directories. 4184 */ 4185 static int 4186 chroot_refuse_vdir_fds(struct filedesc *fdp) 4187 { 4188 struct vnode *vp; 4189 struct file *fp; 4190 int i; 4191 4192 FILEDESC_LOCK_ASSERT(fdp); 4193 4194 FILEDESC_FOREACH_FP(fdp, i, fp) { 4195 if (fp->f_type == DTYPE_VNODE) { 4196 vp = fp->f_vnode; 4197 if (vp->v_type == VDIR) 4198 return (EPERM); 4199 } 4200 } 4201 return (0); 4202 } 4203 4204 static void 4205 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 4206 { 4207 4208 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 4209 vrefact(oldpwd->pwd_cdir); 4210 newpwd->pwd_cdir = oldpwd->pwd_cdir; 4211 } 4212 4213 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 4214 vrefact(oldpwd->pwd_rdir); 4215 newpwd->pwd_rdir = oldpwd->pwd_rdir; 4216 } 4217 4218 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 4219 vrefact(oldpwd->pwd_jdir); 4220 newpwd->pwd_jdir = oldpwd->pwd_jdir; 4221 } 4222 4223 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) { 4224 vrefact(oldpwd->pwd_adir); 4225 newpwd->pwd_adir = oldpwd->pwd_adir; 4226 } 4227 } 4228 4229 struct pwd * 4230 pwd_hold_pwddesc(struct pwddesc *pdp) 4231 { 4232 struct pwd *pwd; 4233 4234 PWDDESC_ASSERT_XLOCKED(pdp); 4235 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4236 if (pwd != NULL) 4237 refcount_acquire(&pwd->pwd_refcount); 4238 return (pwd); 4239 } 4240 4241 bool 4242 pwd_hold_smr(struct pwd *pwd) 4243 { 4244 4245 MPASS(pwd != NULL); 4246 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 4247 return (true); 4248 } 4249 return (false); 4250 } 4251 4252 struct pwd * 4253 pwd_hold(struct thread *td) 4254 { 4255 struct pwddesc *pdp; 4256 struct pwd *pwd; 4257 4258 pdp = td->td_proc->p_pd; 4259 4260 vfs_smr_enter(); 4261 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 4262 if (pwd_hold_smr(pwd)) { 4263 vfs_smr_exit(); 4264 return (pwd); 4265 } 4266 vfs_smr_exit(); 4267 PWDDESC_XLOCK(pdp); 4268 pwd = pwd_hold_pwddesc(pdp); 4269 MPASS(pwd != NULL); 4270 PWDDESC_XUNLOCK(pdp); 4271 return (pwd); 4272 } 4273 4274 struct pwd * 4275 pwd_hold_proc(struct proc *p) 4276 { 4277 struct pwddesc *pdp; 4278 struct pwd *pwd; 4279 4280 PROC_ASSERT_HELD(p); 4281 PROC_LOCK(p); 4282 pdp = pdhold(p); 4283 MPASS(pdp != NULL); 4284 PROC_UNLOCK(p); 4285 4286 PWDDESC_XLOCK(pdp); 4287 pwd = pwd_hold_pwddesc(pdp); 4288 MPASS(pwd != NULL); 4289 PWDDESC_XUNLOCK(pdp); 4290 pddrop(pdp); 4291 return (pwd); 4292 } 4293 4294 static struct pwd * 4295 pwd_alloc(void) 4296 { 4297 struct pwd *pwd; 4298 4299 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 4300 bzero(pwd, sizeof(*pwd)); 4301 refcount_init(&pwd->pwd_refcount, 1); 4302 return (pwd); 4303 } 4304 4305 void 4306 pwd_drop(struct pwd *pwd) 4307 { 4308 4309 if (!refcount_release(&pwd->pwd_refcount)) 4310 return; 4311 4312 if (pwd->pwd_cdir != NULL) 4313 vrele(pwd->pwd_cdir); 4314 if (pwd->pwd_rdir != NULL) 4315 vrele(pwd->pwd_rdir); 4316 if (pwd->pwd_jdir != NULL) 4317 vrele(pwd->pwd_jdir); 4318 if (pwd->pwd_adir != NULL) 4319 vrele(pwd->pwd_adir); 4320 uma_zfree_smr(pwd_zone, pwd); 4321 } 4322 4323 /* 4324 * The caller is responsible for invoking priv_check() and 4325 * mac_vnode_check_chroot() to authorize this operation. 4326 */ 4327 int 4328 pwd_chroot(struct thread *td, struct vnode *vp) 4329 { 4330 struct pwddesc *pdp; 4331 struct filedesc *fdp; 4332 struct pwd *newpwd, *oldpwd; 4333 int error; 4334 4335 fdp = td->td_proc->p_fd; 4336 pdp = td->td_proc->p_pd; 4337 newpwd = pwd_alloc(); 4338 FILEDESC_SLOCK(fdp); 4339 PWDDESC_XLOCK(pdp); 4340 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4341 if (chroot_allow_open_directories == 0 || 4342 (chroot_allow_open_directories == 1 && 4343 oldpwd->pwd_rdir != rootvnode)) { 4344 error = chroot_refuse_vdir_fds(fdp); 4345 FILEDESC_SUNLOCK(fdp); 4346 if (error != 0) { 4347 PWDDESC_XUNLOCK(pdp); 4348 pwd_drop(newpwd); 4349 return (error); 4350 } 4351 } else { 4352 FILEDESC_SUNLOCK(fdp); 4353 } 4354 4355 vrefact(vp); 4356 newpwd->pwd_rdir = vp; 4357 vrefact(vp); 4358 newpwd->pwd_adir = vp; 4359 if (oldpwd->pwd_jdir == NULL) { 4360 vrefact(vp); 4361 newpwd->pwd_jdir = vp; 4362 } 4363 pwd_fill(oldpwd, newpwd); 4364 pwd_set(pdp, newpwd); 4365 PWDDESC_XUNLOCK(pdp); 4366 pwd_drop(oldpwd); 4367 return (0); 4368 } 4369 4370 void 4371 pwd_chdir(struct thread *td, struct vnode *vp) 4372 { 4373 struct pwddesc *pdp; 4374 struct pwd *newpwd, *oldpwd; 4375 4376 VNPASS(vp->v_usecount > 0, vp); 4377 4378 newpwd = pwd_alloc(); 4379 pdp = td->td_proc->p_pd; 4380 PWDDESC_XLOCK(pdp); 4381 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4382 newpwd->pwd_cdir = vp; 4383 pwd_fill(oldpwd, newpwd); 4384 pwd_set(pdp, newpwd); 4385 PWDDESC_XUNLOCK(pdp); 4386 pwd_drop(oldpwd); 4387 } 4388 4389 /* 4390 * Process is transitioning to/from a non-native ABI. 4391 */ 4392 void 4393 pwd_altroot(struct thread *td, struct vnode *altroot_vp) 4394 { 4395 struct pwddesc *pdp; 4396 struct pwd *newpwd, *oldpwd; 4397 4398 newpwd = pwd_alloc(); 4399 pdp = td->td_proc->p_pd; 4400 PWDDESC_XLOCK(pdp); 4401 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4402 if (altroot_vp != NULL) { 4403 /* 4404 * Native process to a non-native ABI. 4405 */ 4406 4407 vrefact(altroot_vp); 4408 newpwd->pwd_adir = altroot_vp; 4409 } else { 4410 /* 4411 * Non-native process to the native ABI. 4412 */ 4413 4414 vrefact(oldpwd->pwd_rdir); 4415 newpwd->pwd_adir = oldpwd->pwd_rdir; 4416 } 4417 pwd_fill(oldpwd, newpwd); 4418 pwd_set(pdp, newpwd); 4419 PWDDESC_XUNLOCK(pdp); 4420 pwd_drop(oldpwd); 4421 } 4422 4423 /* 4424 * jail_attach(2) changes both root and working directories. 4425 */ 4426 int 4427 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 4428 { 4429 struct pwddesc *pdp; 4430 struct filedesc *fdp; 4431 struct pwd *newpwd, *oldpwd; 4432 int error; 4433 4434 fdp = td->td_proc->p_fd; 4435 pdp = td->td_proc->p_pd; 4436 newpwd = pwd_alloc(); 4437 FILEDESC_SLOCK(fdp); 4438 PWDDESC_XLOCK(pdp); 4439 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4440 error = chroot_refuse_vdir_fds(fdp); 4441 FILEDESC_SUNLOCK(fdp); 4442 if (error != 0) { 4443 PWDDESC_XUNLOCK(pdp); 4444 pwd_drop(newpwd); 4445 return (error); 4446 } 4447 4448 vrefact(vp); 4449 newpwd->pwd_rdir = vp; 4450 vrefact(vp); 4451 newpwd->pwd_cdir = vp; 4452 if (oldpwd->pwd_jdir == NULL) { 4453 vrefact(vp); 4454 newpwd->pwd_jdir = vp; 4455 } 4456 vrefact(vp); 4457 newpwd->pwd_adir = vp; 4458 pwd_fill(oldpwd, newpwd); 4459 pwd_set(pdp, newpwd); 4460 PWDDESC_XUNLOCK(pdp); 4461 pwd_drop(oldpwd); 4462 return (0); 4463 } 4464 4465 void 4466 pwd_ensure_dirs(void) 4467 { 4468 struct pwddesc *pdp; 4469 struct pwd *oldpwd, *newpwd; 4470 4471 pdp = curproc->p_pd; 4472 PWDDESC_XLOCK(pdp); 4473 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4474 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL && 4475 oldpwd->pwd_adir != NULL) { 4476 PWDDESC_XUNLOCK(pdp); 4477 return; 4478 } 4479 PWDDESC_XUNLOCK(pdp); 4480 4481 newpwd = pwd_alloc(); 4482 PWDDESC_XLOCK(pdp); 4483 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4484 pwd_fill(oldpwd, newpwd); 4485 if (newpwd->pwd_cdir == NULL) { 4486 vrefact(rootvnode); 4487 newpwd->pwd_cdir = rootvnode; 4488 } 4489 if (newpwd->pwd_rdir == NULL) { 4490 vrefact(rootvnode); 4491 newpwd->pwd_rdir = rootvnode; 4492 } 4493 if (newpwd->pwd_adir == NULL) { 4494 vrefact(rootvnode); 4495 newpwd->pwd_adir = rootvnode; 4496 } 4497 pwd_set(pdp, newpwd); 4498 PWDDESC_XUNLOCK(pdp); 4499 pwd_drop(oldpwd); 4500 } 4501 4502 void 4503 pwd_set_rootvnode(void) 4504 { 4505 struct pwddesc *pdp; 4506 struct pwd *oldpwd, *newpwd; 4507 4508 pdp = curproc->p_pd; 4509 4510 newpwd = pwd_alloc(); 4511 PWDDESC_XLOCK(pdp); 4512 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4513 vrefact(rootvnode); 4514 newpwd->pwd_cdir = rootvnode; 4515 vrefact(rootvnode); 4516 newpwd->pwd_rdir = rootvnode; 4517 vrefact(rootvnode); 4518 newpwd->pwd_adir = rootvnode; 4519 pwd_fill(oldpwd, newpwd); 4520 pwd_set(pdp, newpwd); 4521 PWDDESC_XUNLOCK(pdp); 4522 pwd_drop(oldpwd); 4523 } 4524 4525 /* 4526 * Scan all active processes and prisons to see if any of them have a current 4527 * or root directory of `olddp'. If so, replace them with the new mount point. 4528 */ 4529 void 4530 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 4531 { 4532 struct pwddesc *pdp; 4533 struct pwd *newpwd, *oldpwd; 4534 struct prison *pr; 4535 struct proc *p; 4536 int nrele; 4537 4538 if (vrefcnt(olddp) == 1) 4539 return; 4540 nrele = 0; 4541 newpwd = pwd_alloc(); 4542 sx_slock(&allproc_lock); 4543 FOREACH_PROC_IN_SYSTEM(p) { 4544 PROC_LOCK(p); 4545 pdp = pdhold(p); 4546 PROC_UNLOCK(p); 4547 if (pdp == NULL) 4548 continue; 4549 PWDDESC_XLOCK(pdp); 4550 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4551 if (oldpwd == NULL || 4552 (oldpwd->pwd_cdir != olddp && 4553 oldpwd->pwd_rdir != olddp && 4554 oldpwd->pwd_jdir != olddp && 4555 oldpwd->pwd_adir != olddp)) { 4556 PWDDESC_XUNLOCK(pdp); 4557 pddrop(pdp); 4558 continue; 4559 } 4560 if (oldpwd->pwd_cdir == olddp) { 4561 vrefact(newdp); 4562 newpwd->pwd_cdir = newdp; 4563 } 4564 if (oldpwd->pwd_rdir == olddp) { 4565 vrefact(newdp); 4566 newpwd->pwd_rdir = newdp; 4567 } 4568 if (oldpwd->pwd_jdir == olddp) { 4569 vrefact(newdp); 4570 newpwd->pwd_jdir = newdp; 4571 } 4572 if (oldpwd->pwd_adir == olddp) { 4573 vrefact(newdp); 4574 newpwd->pwd_adir = newdp; 4575 } 4576 pwd_fill(oldpwd, newpwd); 4577 pwd_set(pdp, newpwd); 4578 PWDDESC_XUNLOCK(pdp); 4579 pwd_drop(oldpwd); 4580 pddrop(pdp); 4581 newpwd = pwd_alloc(); 4582 } 4583 sx_sunlock(&allproc_lock); 4584 pwd_drop(newpwd); 4585 if (rootvnode == olddp) { 4586 vrefact(newdp); 4587 rootvnode = newdp; 4588 nrele++; 4589 } 4590 mtx_lock(&prison0.pr_mtx); 4591 if (prison0.pr_root == olddp) { 4592 vrefact(newdp); 4593 prison0.pr_root = newdp; 4594 nrele++; 4595 } 4596 mtx_unlock(&prison0.pr_mtx); 4597 sx_slock(&allprison_lock); 4598 TAILQ_FOREACH(pr, &allprison, pr_list) { 4599 mtx_lock(&pr->pr_mtx); 4600 if (pr->pr_root == olddp) { 4601 vrefact(newdp); 4602 pr->pr_root = newdp; 4603 nrele++; 4604 } 4605 mtx_unlock(&pr->pr_mtx); 4606 } 4607 sx_sunlock(&allprison_lock); 4608 while (nrele--) 4609 vrele(olddp); 4610 } 4611 4612 int 4613 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp) 4614 { 4615 struct file *fp; 4616 struct vnode *vp; 4617 int error, i; 4618 4619 error = 0; 4620 FILEDESC_SLOCK(fdp); 4621 FILEDESC_FOREACH_FP(fdp, i, fp) { 4622 if (fp->f_type != DTYPE_VNODE || 4623 (atomic_load_int(&fp->f_flag) & FWRITE) == 0) 4624 continue; 4625 vp = fp->f_vnode; 4626 if (vp->v_mount == mp) { 4627 error = EDEADLK; 4628 break; 4629 } 4630 } 4631 FILEDESC_SUNLOCK(fdp); 4632 return (error); 4633 } 4634 4635 struct filedesc_to_leader * 4636 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, 4637 struct proc *leader) 4638 { 4639 struct filedesc_to_leader *fdtol; 4640 4641 fdtol = malloc(sizeof(struct filedesc_to_leader), 4642 M_FILEDESC_TO_LEADER, M_WAITOK); 4643 fdtol->fdl_refcount = 1; 4644 fdtol->fdl_holdcount = 0; 4645 fdtol->fdl_wakeup = 0; 4646 fdtol->fdl_leader = leader; 4647 if (old != NULL) { 4648 FILEDESC_XLOCK(fdp); 4649 fdtol->fdl_next = old->fdl_next; 4650 fdtol->fdl_prev = old; 4651 old->fdl_next = fdtol; 4652 fdtol->fdl_next->fdl_prev = fdtol; 4653 FILEDESC_XUNLOCK(fdp); 4654 } else { 4655 fdtol->fdl_next = fdtol; 4656 fdtol->fdl_prev = fdtol; 4657 } 4658 return (fdtol); 4659 } 4660 4661 struct filedesc_to_leader * 4662 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp) 4663 { 4664 FILEDESC_XLOCK(fdp); 4665 fdtol->fdl_refcount++; 4666 FILEDESC_XUNLOCK(fdp); 4667 return (fdtol); 4668 } 4669 4670 static int 4671 filedesc_nfiles(struct filedesc *fdp) 4672 { 4673 NDSLOTTYPE *map; 4674 int count, off, minoff; 4675 4676 if (fdp == NULL) 4677 return (0); 4678 count = 0; 4679 FILEDESC_SLOCK(fdp); 4680 map = fdp->fd_map; 4681 off = NDSLOT(fdp->fd_nfiles - 1); 4682 for (minoff = NDSLOT(0); off >= minoff; --off) 4683 count += bitcountl(map[off]); 4684 FILEDESC_SUNLOCK(fdp); 4685 return (count); 4686 } 4687 4688 int 4689 proc_nfiles(struct proc *p) 4690 { 4691 struct filedesc *fdp; 4692 int res; 4693 4694 PROC_LOCK(p); 4695 fdp = fdhold(p); 4696 PROC_UNLOCK(p); 4697 res = filedesc_nfiles(fdp); 4698 fddrop(fdp); 4699 return (res); 4700 } 4701 4702 static int 4703 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4704 { 4705 u_int namelen; 4706 int count; 4707 4708 namelen = arg2; 4709 if (namelen != 1) 4710 return (EINVAL); 4711 4712 if (*(int *)arg1 != 0) 4713 return (EINVAL); 4714 4715 count = filedesc_nfiles(curproc->p_fd); 4716 return (SYSCTL_OUT(req, &count, sizeof(count))); 4717 } 4718 4719 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4720 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4721 "Number of open file descriptors"); 4722 4723 /* 4724 * Get file structures globally. 4725 */ 4726 static int 4727 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4728 { 4729 struct xfile xf; 4730 struct filedesc *fdp; 4731 struct file *fp; 4732 struct proc *p; 4733 int error, n; 4734 4735 error = sysctl_wire_old_buffer(req, 0); 4736 if (error != 0) 4737 return (error); 4738 if (req->oldptr == NULL) { 4739 n = 0; 4740 sx_slock(&allproc_lock); 4741 FOREACH_PROC_IN_SYSTEM(p) { 4742 PROC_LOCK(p); 4743 if (p->p_state == PRS_NEW) { 4744 PROC_UNLOCK(p); 4745 continue; 4746 } 4747 fdp = fdhold(p); 4748 PROC_UNLOCK(p); 4749 if (fdp == NULL) 4750 continue; 4751 /* overestimates sparse tables. */ 4752 n += fdp->fd_nfiles; 4753 fddrop(fdp); 4754 } 4755 sx_sunlock(&allproc_lock); 4756 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4757 } 4758 error = 0; 4759 bzero(&xf, sizeof(xf)); 4760 xf.xf_size = sizeof(xf); 4761 sx_slock(&allproc_lock); 4762 FOREACH_PROC_IN_SYSTEM(p) { 4763 PROC_LOCK(p); 4764 if (p->p_state == PRS_NEW) { 4765 PROC_UNLOCK(p); 4766 continue; 4767 } 4768 if (p_cansee(req->td, p) != 0) { 4769 PROC_UNLOCK(p); 4770 continue; 4771 } 4772 xf.xf_pid = p->p_pid; 4773 xf.xf_uid = p->p_ucred->cr_uid; 4774 fdp = fdhold(p); 4775 PROC_UNLOCK(p); 4776 if (fdp == NULL) 4777 continue; 4778 FILEDESC_SLOCK(fdp); 4779 if (refcount_load(&fdp->fd_refcnt) == 0) 4780 goto nextproc; 4781 FILEDESC_FOREACH_FP(fdp, n, fp) { 4782 xf.xf_fd = n; 4783 xf.xf_file = (uintptr_t)fp; 4784 xf.xf_data = (uintptr_t)fp->f_data; 4785 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4786 xf.xf_type = (uintptr_t)fp->f_type; 4787 xf.xf_count = refcount_load(&fp->f_count); 4788 xf.xf_msgcount = 0; 4789 xf.xf_offset = foffset_get(fp); 4790 xf.xf_flag = fp->f_flag; 4791 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4792 4793 /* 4794 * There is no need to re-check the fdtable refcount 4795 * here since the filedesc lock is not dropped in the 4796 * loop body. 4797 */ 4798 if (error != 0) 4799 break; 4800 } 4801 nextproc: 4802 FILEDESC_SUNLOCK(fdp); 4803 fddrop(fdp); 4804 if (error) 4805 break; 4806 } 4807 sx_sunlock(&allproc_lock); 4808 return (error); 4809 } 4810 4811 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4812 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4813 4814 #ifdef KINFO_FILE_SIZE 4815 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4816 #endif 4817 4818 static int 4819 xlate_fflags(int fflags) 4820 { 4821 static const struct { 4822 int fflag; 4823 int kf_fflag; 4824 } fflags_table[] = { 4825 { FAPPEND, KF_FLAG_APPEND }, 4826 { FASYNC, KF_FLAG_ASYNC }, 4827 { FFSYNC, KF_FLAG_FSYNC }, 4828 { FHASLOCK, KF_FLAG_HASLOCK }, 4829 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4830 { FREAD, KF_FLAG_READ }, 4831 { FWRITE, KF_FLAG_WRITE }, 4832 { O_CREAT, KF_FLAG_CREAT }, 4833 { O_DIRECT, KF_FLAG_DIRECT }, 4834 { O_EXCL, KF_FLAG_EXCL }, 4835 { O_EXEC, KF_FLAG_EXEC }, 4836 { O_EXLOCK, KF_FLAG_EXLOCK }, 4837 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4838 { O_SHLOCK, KF_FLAG_SHLOCK }, 4839 { O_TRUNC, KF_FLAG_TRUNC } 4840 }; 4841 unsigned int i; 4842 int kflags; 4843 4844 kflags = 0; 4845 for (i = 0; i < nitems(fflags_table); i++) 4846 if (fflags & fflags_table[i].fflag) 4847 kflags |= fflags_table[i].kf_fflag; 4848 return (kflags); 4849 } 4850 4851 /* Trim unused data from kf_path by truncating the structure size. */ 4852 void 4853 pack_kinfo(struct kinfo_file *kif) 4854 { 4855 4856 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4857 strlen(kif->kf_path) + 1; 4858 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4859 } 4860 4861 static void 4862 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4863 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4864 { 4865 int error; 4866 4867 bzero(kif, sizeof(*kif)); 4868 4869 /* Set a default type to allow for empty fill_kinfo() methods. */ 4870 kif->kf_type = KF_TYPE_UNKNOWN; 4871 kif->kf_flags = xlate_fflags(fp->f_flag); 4872 if (rightsp != NULL) 4873 kif->kf_cap_rights = *rightsp; 4874 else 4875 cap_rights_init_zero(&kif->kf_cap_rights); 4876 kif->kf_fd = fd; 4877 kif->kf_ref_count = refcount_load(&fp->f_count); 4878 kif->kf_offset = foffset_get(fp); 4879 4880 /* 4881 * This may drop the filedesc lock, so the 'fp' cannot be 4882 * accessed after this call. 4883 */ 4884 error = fo_fill_kinfo(fp, kif, fdp); 4885 if (error == 0) 4886 kif->kf_status |= KF_ATTR_VALID; 4887 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4888 pack_kinfo(kif); 4889 else 4890 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4891 } 4892 4893 static void 4894 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4895 struct kinfo_file *kif, int flags) 4896 { 4897 int error; 4898 4899 bzero(kif, sizeof(*kif)); 4900 4901 kif->kf_type = KF_TYPE_VNODE; 4902 error = vn_fill_kinfo_vnode(vp, kif); 4903 if (error == 0) 4904 kif->kf_status |= KF_ATTR_VALID; 4905 kif->kf_flags = xlate_fflags(fflags); 4906 cap_rights_init_zero(&kif->kf_cap_rights); 4907 kif->kf_fd = fd; 4908 kif->kf_ref_count = -1; 4909 kif->kf_offset = -1; 4910 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4911 pack_kinfo(kif); 4912 else 4913 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4914 vrele(vp); 4915 } 4916 4917 struct export_fd_buf { 4918 struct filedesc *fdp; 4919 struct pwddesc *pdp; 4920 struct sbuf *sb; 4921 ssize_t remainder; 4922 struct kinfo_file kif; 4923 int flags; 4924 }; 4925 4926 static int 4927 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4928 { 4929 struct kinfo_file *kif; 4930 4931 kif = &efbuf->kif; 4932 if (efbuf->remainder != -1) { 4933 if (efbuf->remainder < kif->kf_structsize) 4934 return (ENOMEM); 4935 efbuf->remainder -= kif->kf_structsize; 4936 } 4937 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0) 4938 return (sbuf_error(efbuf->sb)); 4939 return (0); 4940 } 4941 4942 static int 4943 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4944 struct export_fd_buf *efbuf) 4945 { 4946 int error; 4947 4948 if (efbuf->remainder == 0) 4949 return (ENOMEM); 4950 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4951 efbuf->flags); 4952 FILEDESC_SUNLOCK(efbuf->fdp); 4953 error = export_kinfo_to_sb(efbuf); 4954 FILEDESC_SLOCK(efbuf->fdp); 4955 return (error); 4956 } 4957 4958 static int 4959 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4960 struct export_fd_buf *efbuf) 4961 { 4962 int error; 4963 4964 if (efbuf->remainder == 0) 4965 return (ENOMEM); 4966 if (efbuf->pdp != NULL) 4967 PWDDESC_XUNLOCK(efbuf->pdp); 4968 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4969 error = export_kinfo_to_sb(efbuf); 4970 if (efbuf->pdp != NULL) 4971 PWDDESC_XLOCK(efbuf->pdp); 4972 return (error); 4973 } 4974 4975 /* 4976 * Store a process file descriptor information to sbuf. 4977 * 4978 * Takes a locked proc as argument, and returns with the proc unlocked. 4979 */ 4980 int 4981 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4982 int flags) 4983 { 4984 struct file *fp; 4985 struct filedesc *fdp; 4986 struct pwddesc *pdp; 4987 struct export_fd_buf *efbuf; 4988 struct vnode *cttyvp, *textvp, *tracevp; 4989 struct pwd *pwd; 4990 int error, i; 4991 cap_rights_t rights; 4992 4993 PROC_LOCK_ASSERT(p, MA_OWNED); 4994 4995 /* ktrace vnode */ 4996 tracevp = ktr_get_tracevp(p, true); 4997 /* text vnode */ 4998 textvp = p->p_textvp; 4999 if (textvp != NULL) 5000 vrefact(textvp); 5001 /* Controlling tty. */ 5002 cttyvp = NULL; 5003 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 5004 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 5005 if (cttyvp != NULL) 5006 vrefact(cttyvp); 5007 } 5008 fdp = fdhold(p); 5009 pdp = pdhold(p); 5010 PROC_UNLOCK(p); 5011 5012 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 5013 efbuf->fdp = NULL; 5014 efbuf->pdp = NULL; 5015 efbuf->sb = sb; 5016 efbuf->remainder = maxlen; 5017 efbuf->flags = flags; 5018 5019 error = 0; 5020 if (tracevp != NULL) 5021 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, 5022 FREAD | FWRITE, efbuf); 5023 if (error == 0 && textvp != NULL) 5024 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, 5025 efbuf); 5026 if (error == 0 && cttyvp != NULL) 5027 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, 5028 FREAD | FWRITE, efbuf); 5029 if (error != 0 || pdp == NULL || fdp == NULL) 5030 goto fail; 5031 efbuf->fdp = fdp; 5032 efbuf->pdp = pdp; 5033 PWDDESC_XLOCK(pdp); 5034 pwd = pwd_hold_pwddesc(pdp); 5035 if (pwd != NULL) { 5036 /* working directory */ 5037 if (pwd->pwd_cdir != NULL) { 5038 vrefact(pwd->pwd_cdir); 5039 error = export_vnode_to_sb(pwd->pwd_cdir, 5040 KF_FD_TYPE_CWD, FREAD, efbuf); 5041 } 5042 /* root directory */ 5043 if (error == 0 && pwd->pwd_rdir != NULL) { 5044 vrefact(pwd->pwd_rdir); 5045 error = export_vnode_to_sb(pwd->pwd_rdir, 5046 KF_FD_TYPE_ROOT, FREAD, efbuf); 5047 } 5048 /* jail directory */ 5049 if (error == 0 && pwd->pwd_jdir != NULL) { 5050 vrefact(pwd->pwd_jdir); 5051 error = export_vnode_to_sb(pwd->pwd_jdir, 5052 KF_FD_TYPE_JAIL, FREAD, efbuf); 5053 } 5054 } 5055 PWDDESC_XUNLOCK(pdp); 5056 if (error != 0) 5057 goto fail; 5058 if (pwd != NULL) 5059 pwd_drop(pwd); 5060 FILEDESC_SLOCK(fdp); 5061 if (refcount_load(&fdp->fd_refcnt) == 0) 5062 goto skip; 5063 FILEDESC_FOREACH_FP(fdp, i, fp) { 5064 #ifdef CAPABILITIES 5065 rights = *cap_rights(fdp, i); 5066 #else /* !CAPABILITIES */ 5067 rights = cap_no_rights; 5068 #endif 5069 /* 5070 * Create sysctl entry. It is OK to drop the filedesc 5071 * lock inside of export_file_to_sb() as we will 5072 * re-validate and re-evaluate its properties when the 5073 * loop continues. 5074 */ 5075 error = export_file_to_sb(fp, i, &rights, efbuf); 5076 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 5077 break; 5078 } 5079 skip: 5080 FILEDESC_SUNLOCK(fdp); 5081 fail: 5082 if (fdp != NULL) 5083 fddrop(fdp); 5084 if (pdp != NULL) 5085 pddrop(pdp); 5086 free(efbuf, M_TEMP); 5087 return (error); 5088 } 5089 5090 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 5091 5092 /* 5093 * Get per-process file descriptors for use by procstat(1), et al. 5094 */ 5095 static int 5096 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 5097 { 5098 struct sbuf sb; 5099 struct proc *p; 5100 ssize_t maxlen; 5101 u_int namelen; 5102 int error, error2, *name; 5103 5104 namelen = arg2; 5105 if (namelen != 1) 5106 return (EINVAL); 5107 5108 name = (int *)arg1; 5109 5110 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 5111 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5112 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5113 if (error != 0) { 5114 sbuf_delete(&sb); 5115 return (error); 5116 } 5117 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5118 error = kern_proc_filedesc_out(p, &sb, maxlen, 5119 KERN_FILEDESC_PACK_KINFO); 5120 error2 = sbuf_finish(&sb); 5121 sbuf_delete(&sb); 5122 return (error != 0 ? error : error2); 5123 } 5124 5125 #ifdef COMPAT_FREEBSD7 5126 #ifdef KINFO_OFILE_SIZE 5127 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 5128 #endif 5129 5130 static void 5131 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 5132 { 5133 5134 okif->kf_structsize = sizeof(*okif); 5135 okif->kf_type = kif->kf_type; 5136 okif->kf_fd = kif->kf_fd; 5137 okif->kf_ref_count = kif->kf_ref_count; 5138 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 5139 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 5140 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 5141 okif->kf_offset = kif->kf_offset; 5142 if (kif->kf_type == KF_TYPE_VNODE) 5143 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 5144 else 5145 okif->kf_vnode_type = KF_VTYPE_VNON; 5146 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 5147 if (kif->kf_type == KF_TYPE_SOCKET) { 5148 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 5149 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 5150 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 5151 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 5152 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 5153 } else { 5154 okif->kf_sa_local.ss_family = AF_UNSPEC; 5155 okif->kf_sa_peer.ss_family = AF_UNSPEC; 5156 } 5157 } 5158 5159 static int 5160 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 5161 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 5162 { 5163 int error; 5164 5165 vrefact(vp); 5166 PWDDESC_XUNLOCK(pdp); 5167 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 5168 kinfo_to_okinfo(kif, okif); 5169 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5170 PWDDESC_XLOCK(pdp); 5171 return (error); 5172 } 5173 5174 /* 5175 * Get per-process file descriptors for use by procstat(1), et al. 5176 */ 5177 static int 5178 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 5179 { 5180 struct kinfo_ofile *okif; 5181 struct kinfo_file *kif; 5182 struct filedesc *fdp; 5183 struct pwddesc *pdp; 5184 struct pwd *pwd; 5185 u_int namelen; 5186 int error, i, *name; 5187 struct file *fp; 5188 struct proc *p; 5189 5190 namelen = arg2; 5191 if (namelen != 1) 5192 return (EINVAL); 5193 5194 name = (int *)arg1; 5195 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5196 if (error != 0) 5197 return (error); 5198 fdp = fdhold(p); 5199 if (fdp != NULL) 5200 pdp = pdhold(p); 5201 PROC_UNLOCK(p); 5202 if (fdp == NULL || pdp == NULL) { 5203 if (fdp != NULL) 5204 fddrop(fdp); 5205 return (ENOENT); 5206 } 5207 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 5208 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK | M_ZERO); 5209 PWDDESC_XLOCK(pdp); 5210 pwd = pwd_hold_pwddesc(pdp); 5211 if (pwd != NULL) { 5212 if (pwd->pwd_cdir != NULL) 5213 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 5214 okif, pdp, req); 5215 if (pwd->pwd_rdir != NULL) 5216 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 5217 okif, pdp, req); 5218 if (pwd->pwd_jdir != NULL) 5219 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 5220 okif, pdp, req); 5221 } 5222 PWDDESC_XUNLOCK(pdp); 5223 if (pwd != NULL) 5224 pwd_drop(pwd); 5225 FILEDESC_SLOCK(fdp); 5226 if (refcount_load(&fdp->fd_refcnt) == 0) 5227 goto skip; 5228 FILEDESC_FOREACH_FP(fdp, i, fp) { 5229 export_file_to_kinfo(fp, i, NULL, kif, fdp, 5230 KERN_FILEDESC_PACK_KINFO); 5231 FILEDESC_SUNLOCK(fdp); 5232 kinfo_to_okinfo(kif, okif); 5233 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5234 FILEDESC_SLOCK(fdp); 5235 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 5236 break; 5237 } 5238 skip: 5239 FILEDESC_SUNLOCK(fdp); 5240 fddrop(fdp); 5241 pddrop(pdp); 5242 free(kif, M_TEMP); 5243 free(okif, M_TEMP); 5244 return (0); 5245 } 5246 5247 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 5248 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 5249 "Process ofiledesc entries"); 5250 #endif /* COMPAT_FREEBSD7 */ 5251 5252 int 5253 vntype_to_kinfo(int vtype) 5254 { 5255 struct { 5256 int vtype; 5257 int kf_vtype; 5258 } vtypes_table[] = { 5259 { VBAD, KF_VTYPE_VBAD }, 5260 { VBLK, KF_VTYPE_VBLK }, 5261 { VCHR, KF_VTYPE_VCHR }, 5262 { VDIR, KF_VTYPE_VDIR }, 5263 { VFIFO, KF_VTYPE_VFIFO }, 5264 { VLNK, KF_VTYPE_VLNK }, 5265 { VNON, KF_VTYPE_VNON }, 5266 { VREG, KF_VTYPE_VREG }, 5267 { VSOCK, KF_VTYPE_VSOCK } 5268 }; 5269 unsigned int i; 5270 5271 /* 5272 * Perform vtype translation. 5273 */ 5274 for (i = 0; i < nitems(vtypes_table); i++) 5275 if (vtypes_table[i].vtype == vtype) 5276 return (vtypes_table[i].kf_vtype); 5277 5278 return (KF_VTYPE_UNKNOWN); 5279 } 5280 5281 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 5282 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 5283 "Process filedesc entries"); 5284 5285 /* 5286 * Store a process current working directory information to sbuf. 5287 * 5288 * Takes a locked proc as argument, and returns with the proc unlocked. 5289 */ 5290 int 5291 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 5292 { 5293 struct pwddesc *pdp; 5294 struct pwd *pwd; 5295 struct export_fd_buf *efbuf; 5296 struct vnode *cdir; 5297 int error; 5298 5299 PROC_LOCK_ASSERT(p, MA_OWNED); 5300 5301 pdp = pdhold(p); 5302 PROC_UNLOCK(p); 5303 if (pdp == NULL) 5304 return (EINVAL); 5305 5306 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 5307 efbuf->fdp = NULL; 5308 efbuf->pdp = pdp; 5309 efbuf->sb = sb; 5310 efbuf->remainder = maxlen; 5311 efbuf->flags = 0; 5312 5313 PWDDESC_XLOCK(pdp); 5314 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 5315 cdir = pwd->pwd_cdir; 5316 if (cdir == NULL) { 5317 error = EINVAL; 5318 } else { 5319 vrefact(cdir); 5320 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 5321 } 5322 PWDDESC_XUNLOCK(pdp); 5323 pddrop(pdp); 5324 free(efbuf, M_TEMP); 5325 return (error); 5326 } 5327 5328 /* 5329 * Get per-process current working directory. 5330 */ 5331 static int 5332 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 5333 { 5334 struct sbuf sb; 5335 struct proc *p; 5336 ssize_t maxlen; 5337 u_int namelen; 5338 int error, error2, *name; 5339 5340 namelen = arg2; 5341 if (namelen != 1) 5342 return (EINVAL); 5343 5344 name = (int *)arg1; 5345 5346 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 5347 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5348 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5349 if (error != 0) { 5350 sbuf_delete(&sb); 5351 return (error); 5352 } 5353 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5354 error = kern_proc_cwd_out(p, &sb, maxlen); 5355 error2 = sbuf_finish(&sb); 5356 sbuf_delete(&sb); 5357 return (error != 0 ? error : error2); 5358 } 5359 5360 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 5361 sysctl_kern_proc_cwd, "Process current working directory"); 5362 5363 #ifdef DDB 5364 /* 5365 * For the purposes of debugging, generate a human-readable string for the 5366 * file type. 5367 */ 5368 static const char * 5369 file_type_to_name(short type) 5370 { 5371 5372 switch (type) { 5373 case 0: 5374 return ("zero"); 5375 case DTYPE_VNODE: 5376 return ("vnode"); 5377 case DTYPE_SOCKET: 5378 return ("socket"); 5379 case DTYPE_PIPE: 5380 return ("pipe"); 5381 case DTYPE_FIFO: 5382 return ("fifo"); 5383 case DTYPE_KQUEUE: 5384 return ("kqueue"); 5385 case DTYPE_CRYPTO: 5386 return ("crypto"); 5387 case DTYPE_MQUEUE: 5388 return ("mqueue"); 5389 case DTYPE_SHM: 5390 return ("shm"); 5391 case DTYPE_SEM: 5392 return ("ksem"); 5393 case DTYPE_PTS: 5394 return ("pts"); 5395 case DTYPE_DEV: 5396 return ("dev"); 5397 case DTYPE_PROCDESC: 5398 return ("proc"); 5399 case DTYPE_EVENTFD: 5400 return ("eventfd"); 5401 case DTYPE_TIMERFD: 5402 return ("timerfd"); 5403 case DTYPE_JAILDESC: 5404 return ("jail"); 5405 default: 5406 return ("unkn"); 5407 } 5408 } 5409 5410 /* 5411 * For the purposes of debugging, identify a process (if any, perhaps one of 5412 * many) that references the passed file in its file descriptor array. Return 5413 * NULL if none. 5414 */ 5415 static struct proc * 5416 file_to_first_proc(struct file *fp) 5417 { 5418 struct filedesc *fdp; 5419 struct proc *p; 5420 int n; 5421 5422 FOREACH_PROC_IN_SYSTEM(p) { 5423 if (p->p_state == PRS_NEW) 5424 continue; 5425 fdp = p->p_fd; 5426 if (fdp == NULL) 5427 continue; 5428 for (n = 0; n < fdp->fd_nfiles; n++) { 5429 if (fp == fdp->fd_ofiles[n].fde_file) 5430 return (p); 5431 } 5432 } 5433 return (NULL); 5434 } 5435 5436 static void 5437 db_print_file(struct file *fp, int header) 5438 { 5439 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 5440 struct proc *p; 5441 5442 if (header) 5443 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 5444 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 5445 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 5446 "FCmd"); 5447 p = file_to_first_proc(fp); 5448 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 5449 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 5450 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 5451 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 5452 5453 #undef XPTRWIDTH 5454 } 5455 5456 DB_SHOW_COMMAND(file, db_show_file) 5457 { 5458 struct file *fp; 5459 5460 if (!have_addr) { 5461 db_printf("usage: show file <addr>\n"); 5462 return; 5463 } 5464 fp = (struct file *)addr; 5465 db_print_file(fp, 1); 5466 } 5467 5468 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE) 5469 { 5470 struct filedesc *fdp; 5471 struct file *fp; 5472 struct proc *p; 5473 int header; 5474 int n; 5475 5476 header = 1; 5477 FOREACH_PROC_IN_SYSTEM(p) { 5478 if (p->p_state == PRS_NEW) 5479 continue; 5480 if ((fdp = p->p_fd) == NULL) 5481 continue; 5482 for (n = 0; n < fdp->fd_nfiles; ++n) { 5483 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 5484 continue; 5485 db_print_file(fp, header); 5486 header = 0; 5487 } 5488 } 5489 } 5490 #endif 5491 5492 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, 5493 CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5494 &maxfilesperproc, 0, "Maximum files allowed open per process"); 5495 5496 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5497 &maxfiles, 0, "Maximum number of files"); 5498 5499 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 5500 &openfiles, 0, "System-wide number of open files"); 5501 5502 /* ARGSUSED*/ 5503 static void 5504 filelistinit(void *dummy) 5505 { 5506 5507 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 5508 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 5509 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 5510 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 5511 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 5512 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 5513 /* 5514 * XXXMJG this is a temporary hack due to boot ordering issues against 5515 * the vnode zone. 5516 */ 5517 vfs_smr = uma_zone_get_smr(pwd_zone); 5518 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 5519 } 5520 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 5521 5522 /*-------------------------------------------------------------------*/ 5523 5524 static int 5525 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 5526 int flags, struct thread *td) 5527 { 5528 5529 return (EBADF); 5530 } 5531 5532 static int 5533 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5534 struct thread *td) 5535 { 5536 5537 return (EINVAL); 5538 } 5539 5540 static int 5541 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 5542 struct thread *td) 5543 { 5544 5545 return (EBADF); 5546 } 5547 5548 static int 5549 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 5550 struct thread *td) 5551 { 5552 5553 return (0); 5554 } 5555 5556 static int 5557 badfo_kqfilter(struct file *fp, struct knote *kn) 5558 { 5559 5560 return (EBADF); 5561 } 5562 5563 static int 5564 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 5565 { 5566 5567 return (EBADF); 5568 } 5569 5570 static int 5571 badfo_close(struct file *fp, struct thread *td) 5572 { 5573 5574 return (0); 5575 } 5576 5577 static int 5578 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5579 struct thread *td) 5580 { 5581 5582 return (EBADF); 5583 } 5584 5585 static int 5586 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5587 struct thread *td) 5588 { 5589 5590 return (EBADF); 5591 } 5592 5593 static int 5594 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5595 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5596 struct thread *td) 5597 { 5598 5599 return (EBADF); 5600 } 5601 5602 static int 5603 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 5604 { 5605 5606 return (0); 5607 } 5608 5609 const struct fileops badfileops = { 5610 .fo_read = badfo_readwrite, 5611 .fo_write = badfo_readwrite, 5612 .fo_truncate = badfo_truncate, 5613 .fo_ioctl = badfo_ioctl, 5614 .fo_poll = badfo_poll, 5615 .fo_kqfilter = badfo_kqfilter, 5616 .fo_stat = badfo_stat, 5617 .fo_close = badfo_close, 5618 .fo_chmod = badfo_chmod, 5619 .fo_chown = badfo_chown, 5620 .fo_sendfile = badfo_sendfile, 5621 .fo_fill_kinfo = badfo_fill_kinfo, 5622 }; 5623 5624 static int 5625 path_poll(struct file *fp, int events, struct ucred *active_cred, 5626 struct thread *td) 5627 { 5628 return (POLLNVAL); 5629 } 5630 5631 static int 5632 path_close(struct file *fp, struct thread *td) 5633 { 5634 MPASS(fp->f_type == DTYPE_VNODE); 5635 fp->f_ops = &badfileops; 5636 vrele(fp->f_vnode); 5637 return (0); 5638 } 5639 5640 const struct fileops path_fileops = { 5641 .fo_read = badfo_readwrite, 5642 .fo_write = badfo_readwrite, 5643 .fo_truncate = badfo_truncate, 5644 .fo_ioctl = badfo_ioctl, 5645 .fo_poll = path_poll, 5646 .fo_kqfilter = vn_kqfilter_opath, 5647 .fo_stat = vn_statfile, 5648 .fo_close = path_close, 5649 .fo_chmod = badfo_chmod, 5650 .fo_chown = badfo_chown, 5651 .fo_sendfile = badfo_sendfile, 5652 .fo_fill_kinfo = vn_fill_kinfo, 5653 .fo_cmp = vn_cmp, 5654 .fo_flags = DFLAG_PASSABLE, 5655 }; 5656 5657 int 5658 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5659 int flags, struct thread *td) 5660 { 5661 5662 return (EOPNOTSUPP); 5663 } 5664 5665 int 5666 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5667 struct thread *td) 5668 { 5669 5670 return (EINVAL); 5671 } 5672 5673 int 5674 invfo_ioctl(struct file *fp, u_long com, void *data, 5675 struct ucred *active_cred, struct thread *td) 5676 { 5677 5678 return (ENOTTY); 5679 } 5680 5681 int 5682 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5683 struct thread *td) 5684 { 5685 5686 return (poll_no_poll(events)); 5687 } 5688 5689 int 5690 invfo_kqfilter(struct file *fp, struct knote *kn) 5691 { 5692 5693 return (EINVAL); 5694 } 5695 5696 int 5697 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5698 struct thread *td) 5699 { 5700 5701 return (EINVAL); 5702 } 5703 5704 int 5705 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5706 struct thread *td) 5707 { 5708 5709 return (EINVAL); 5710 } 5711 5712 int 5713 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5714 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5715 struct thread *td) 5716 { 5717 5718 return (EINVAL); 5719 } 5720 5721 /*-------------------------------------------------------------------*/ 5722 5723 /* 5724 * File Descriptor pseudo-device driver (/dev/fd/). 5725 * 5726 * Opening minor device N dup()s the file (if any) connected to file 5727 * descriptor N belonging to the calling process. Note that this driver 5728 * consists of only the ``open()'' routine, because all subsequent 5729 * references to this file will be direct to the other driver. 5730 * 5731 * XXX: we could give this one a cloning event handler if necessary. 5732 */ 5733 5734 /* ARGSUSED */ 5735 static int 5736 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5737 { 5738 5739 /* 5740 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5741 * the file descriptor being sought for duplication. The error 5742 * return ensures that the vnode for this device will be released 5743 * by vn_open. Open will detect this special error and take the 5744 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5745 * will simply report the error. 5746 */ 5747 td->td_dupfd = dev2unit(dev); 5748 return (ENODEV); 5749 } 5750 5751 static struct cdevsw fildesc_cdevsw = { 5752 .d_version = D_VERSION, 5753 .d_open = fdopen, 5754 .d_name = "FD", 5755 }; 5756 5757 static void 5758 fildesc_drvinit(void *unused) 5759 { 5760 struct cdev *dev; 5761 5762 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5763 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5764 make_dev_alias(dev, "stdin"); 5765 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5766 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5767 make_dev_alias(dev, "stdout"); 5768 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5769 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5770 make_dev_alias(dev, "stderr"); 5771 } 5772 5773 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5774