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 old, int new) 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(old); 1060 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 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 (old < 0) 1068 return (EBADF); 1069 if (new < 0) 1070 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1071 maxfd = getmaxfd(td); 1072 if (new >= maxfd) 1073 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 1074 1075 error = EBADF; 1076 FILEDESC_XLOCK(fdp); 1077 if (fget_noref(fdp, old) == NULL) 1078 goto unlock; 1079 if (mode == FDDUP_FIXED && old == new) { 1080 td->td_retval[0] = new; 1081 fdp->fd_ofiles[new].fde_flags |= fddup_to_fde_flags(flags); 1082 error = 0; 1083 goto unlock; 1084 } 1085 1086 oldfde = &fdp->fd_ofiles[old]; 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, new, &new)) != 0) { 1100 fdrop(oldfp, td); 1101 goto unlock; 1102 } 1103 break; 1104 case FDDUP_FIXED: 1105 if (new >= 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, new + 1); 1117 if (error != 0) { 1118 error = EMFILE; 1119 fdrop(oldfp, td); 1120 goto unlock; 1121 } 1122 } 1123 #endif 1124 fdgrowtable_exp(fdp, new + 1); 1125 } 1126 if (!fdisused(fdp, new)) 1127 fdused(fdp, new); 1128 break; 1129 default: 1130 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 1131 } 1132 1133 KASSERT(old != new, ("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[old]; 1137 KASSERT(oldfp == oldfde->fde_file, 1138 ("fdt_ofiles shift from growth observed at fd %d", 1139 old)); 1140 1141 newfde = &fdp->fd_ofiles[new]; 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] = new; 1162 1163 error = 0; 1164 1165 if (delfp != NULL) { 1166 (void) closefp(fdp, new, 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 static 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 static u_long * 1939 filecaps_free_prep(struct filecaps *fcaps) 1940 { 1941 u_long *ioctls; 1942 1943 ioctls = fcaps->fc_ioctls; 1944 bzero(fcaps, sizeof(*fcaps)); 1945 return (ioctls); 1946 } 1947 1948 static void 1949 filecaps_free_finish(u_long *ioctls) 1950 { 1951 1952 free(ioctls, M_FILECAPS); 1953 } 1954 1955 /* 1956 * Validate the given filecaps structure. 1957 */ 1958 static void 1959 filecaps_validate(const struct filecaps *fcaps, const char *func) 1960 { 1961 1962 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1963 ("%s: invalid rights", func)); 1964 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1965 ("%s: invalid fcntls", func)); 1966 KASSERT(fcaps->fc_fcntls == 0 || 1967 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1968 ("%s: fcntls without CAP_FCNTL", func)); 1969 /* 1970 * open calls without WANTIOCTLCAPS free caps but leave the counter 1971 */ 1972 #if 0 1973 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1974 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1975 ("%s: invalid ioctls", func)); 1976 #endif 1977 KASSERT(fcaps->fc_nioctls == 0 || 1978 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1979 ("%s: ioctls without CAP_IOCTL", func)); 1980 } 1981 1982 static void 1983 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1984 { 1985 int nfd1; 1986 1987 FILEDESC_XLOCK_ASSERT(fdp); 1988 1989 nfd1 = fdp->fd_nfiles * 2; 1990 if (nfd1 < nfd) 1991 nfd1 = nfd; 1992 fdgrowtable(fdp, nfd1); 1993 } 1994 1995 /* 1996 * Grow the file table to accommodate (at least) nfd descriptors. 1997 */ 1998 static void 1999 fdgrowtable(struct filedesc *fdp, int nfd) 2000 { 2001 struct filedesc0 *fdp0; 2002 struct freetable *ft; 2003 struct fdescenttbl *ntable; 2004 struct fdescenttbl *otable; 2005 int nnfiles, onfiles; 2006 NDSLOTTYPE *nmap, *omap; 2007 2008 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 2009 2010 /* save old values */ 2011 onfiles = fdp->fd_nfiles; 2012 otable = fdp->fd_files; 2013 omap = fdp->fd_map; 2014 2015 /* compute the size of the new table */ 2016 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 2017 if (nnfiles <= onfiles) 2018 /* the table is already large enough */ 2019 return; 2020 2021 /* 2022 * Allocate a new table. We need enough space for the number of 2023 * entries, file entries themselves and the struct freetable we will use 2024 * when we decommission the table and place it on the freelist. 2025 * We place the struct freetable in the middle so we don't have 2026 * to worry about padding. 2027 */ 2028 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 2029 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 2030 sizeof(struct freetable), 2031 M_FILEDESC, M_ZERO | M_WAITOK); 2032 /* copy the old data */ 2033 ntable->fdt_nfiles = nnfiles; 2034 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 2035 onfiles * sizeof(ntable->fdt_ofiles[0])); 2036 2037 /* 2038 * Allocate a new map only if the old is not large enough. It will 2039 * grow at a slower rate than the table as it can map more 2040 * entries than the table can hold. 2041 */ 2042 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 2043 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 2044 M_ZERO | M_WAITOK); 2045 /* copy over the old data and update the pointer */ 2046 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 2047 fdp->fd_map = nmap; 2048 } 2049 2050 /* 2051 * Make sure that ntable is correctly initialized before we replace 2052 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 2053 * data. 2054 */ 2055 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 2056 2057 /* 2058 * Free the old file table when not shared by other threads or processes. 2059 * The old file table is considered to be shared when either are true: 2060 * - The process has more than one thread. 2061 * - The file descriptor table has been shared via fdshare(). 2062 * 2063 * When shared, the old file table will be placed on a freelist 2064 * which will be processed when the struct filedesc is released. 2065 * 2066 * Note that if onfiles == NDFILE, we're dealing with the original 2067 * static allocation contained within (struct filedesc0 *)fdp, 2068 * which must not be freed. 2069 */ 2070 if (onfiles > NDFILE) { 2071 /* 2072 * Note we may be called here from fdinit while allocating a 2073 * table for a new process in which case ->p_fd points 2074 * elsewhere. 2075 */ 2076 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 2077 free(otable, M_FILEDESC); 2078 } else { 2079 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 2080 fdp0 = (struct filedesc0 *)fdp; 2081 ft->ft_table = otable; 2082 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 2083 } 2084 } 2085 /* 2086 * The map does not have the same possibility of threads still 2087 * holding references to it. So always free it as long as it 2088 * does not reference the original static allocation. 2089 */ 2090 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 2091 free(omap, M_FILEDESC); 2092 } 2093 2094 /* 2095 * Allocate a file descriptor for the process. 2096 */ 2097 int 2098 fdalloc(struct thread *td, int minfd, int *result) 2099 { 2100 struct proc *p = td->td_proc; 2101 struct filedesc *fdp = p->p_fd; 2102 int fd, maxfd, allocfd; 2103 #ifdef RACCT 2104 int error; 2105 #endif 2106 2107 FILEDESC_XLOCK_ASSERT(fdp); 2108 2109 if (fdp->fd_freefile > minfd) 2110 minfd = fdp->fd_freefile; 2111 2112 maxfd = getmaxfd(td); 2113 2114 /* 2115 * Search the bitmap for a free descriptor starting at minfd. 2116 * If none is found, grow the file table. 2117 */ 2118 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 2119 if (__predict_false(fd >= maxfd)) 2120 return (EMFILE); 2121 if (__predict_false(fd >= fdp->fd_nfiles)) { 2122 allocfd = min(fd * 2, maxfd); 2123 #ifdef RACCT 2124 if (RACCT_ENABLED()) { 2125 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 2126 if (error != 0) 2127 return (EMFILE); 2128 } 2129 #endif 2130 /* 2131 * fd is already equal to first free descriptor >= minfd, so 2132 * we only need to grow the table and we are done. 2133 */ 2134 fdgrowtable_exp(fdp, allocfd); 2135 } 2136 2137 /* 2138 * Perform some sanity checks, then mark the file descriptor as 2139 * used and return it to the caller. 2140 */ 2141 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 2142 ("invalid descriptor %d", fd)); 2143 KASSERT(!fdisused(fdp, fd), 2144 ("fd_first_free() returned non-free descriptor")); 2145 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 2146 ("file descriptor isn't free")); 2147 fdused(fdp, fd); 2148 *result = fd; 2149 return (0); 2150 } 2151 2152 /* 2153 * Allocate n file descriptors for the process. 2154 */ 2155 int 2156 fdallocn(struct thread *td, int minfd, int *fds, int n) 2157 { 2158 struct proc *p = td->td_proc; 2159 struct filedesc *fdp = p->p_fd; 2160 int i; 2161 2162 FILEDESC_XLOCK_ASSERT(fdp); 2163 2164 for (i = 0; i < n; i++) 2165 if (fdalloc(td, 0, &fds[i]) != 0) 2166 break; 2167 2168 if (i < n) { 2169 for (i--; i >= 0; i--) 2170 fdunused(fdp, fds[i]); 2171 return (EMFILE); 2172 } 2173 2174 return (0); 2175 } 2176 2177 /* 2178 * Create a new open file structure and allocate a file descriptor for the 2179 * process that refers to it. We add one reference to the file for the 2180 * descriptor table and one reference for resultfp. This is to prevent us 2181 * being preempted and the entry in the descriptor table closed after we 2182 * release the FILEDESC lock. 2183 */ 2184 int 2185 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 2186 struct filecaps *fcaps) 2187 { 2188 struct file *fp; 2189 int error, fd; 2190 2191 MPASS(resultfp != NULL); 2192 MPASS(resultfd != NULL); 2193 2194 error = _falloc_noinstall(td, &fp, 2); 2195 if (__predict_false(error != 0)) { 2196 return (error); 2197 } 2198 2199 error = finstall_refed(td, fp, &fd, flags, fcaps); 2200 if (__predict_false(error != 0)) { 2201 falloc_abort(td, fp); 2202 return (error); 2203 } 2204 2205 *resultfp = fp; 2206 *resultfd = fd; 2207 2208 return (0); 2209 } 2210 2211 /* 2212 * Create a new open file structure without allocating a file descriptor. 2213 */ 2214 int 2215 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2216 { 2217 struct file *fp; 2218 int maxuserfiles = maxfiles - (maxfiles / 20); 2219 int openfiles_new; 2220 static struct timeval lastfail; 2221 static int curfail; 2222 2223 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2224 MPASS(n > 0); 2225 2226 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2227 if ((openfiles_new >= maxuserfiles && 2228 priv_check(td, PRIV_MAXFILES) != 0) || 2229 openfiles_new >= maxfiles) { 2230 atomic_subtract_int(&openfiles, 1); 2231 if (ppsratecheck(&lastfail, &curfail, 1)) { 2232 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2233 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2234 } 2235 return (ENFILE); 2236 } 2237 fp = uma_zalloc(file_zone, M_WAITOK); 2238 bzero(fp, sizeof(*fp)); 2239 refcount_init(&fp->f_count, n); 2240 fp->f_cred = crhold(td->td_ucred); 2241 fp->f_ops = &badfileops; 2242 *resultfp = fp; 2243 return (0); 2244 } 2245 2246 void 2247 falloc_abort(struct thread *td, struct file *fp) 2248 { 2249 2250 /* 2251 * For assertion purposes. 2252 */ 2253 refcount_init(&fp->f_count, 0); 2254 _fdrop(fp, td); 2255 } 2256 2257 /* 2258 * Install a file in a file descriptor table. 2259 */ 2260 void 2261 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2262 struct filecaps *fcaps) 2263 { 2264 struct filedescent *fde; 2265 2266 MPASS(fp != NULL); 2267 if (fcaps != NULL) 2268 filecaps_validate(fcaps, __func__); 2269 FILEDESC_XLOCK_ASSERT(fdp); 2270 2271 fde = &fdp->fd_ofiles[fd]; 2272 #ifdef CAPABILITIES 2273 seqc_write_begin(&fde->fde_seqc); 2274 #endif 2275 fde->fde_file = fp; 2276 fde->fde_flags = open_to_fde_flags(flags, true); 2277 if (fcaps != NULL) 2278 filecaps_move(fcaps, &fde->fde_caps); 2279 else 2280 filecaps_fill(&fde->fde_caps); 2281 #ifdef CAPABILITIES 2282 seqc_write_end(&fde->fde_seqc); 2283 #endif 2284 } 2285 2286 int 2287 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2288 struct filecaps *fcaps) 2289 { 2290 struct filedesc *fdp = td->td_proc->p_fd; 2291 int error; 2292 2293 MPASS(fd != NULL); 2294 2295 FILEDESC_XLOCK(fdp); 2296 error = fdalloc(td, 0, fd); 2297 if (__predict_true(error == 0)) { 2298 _finstall(fdp, fp, *fd, flags, fcaps); 2299 } 2300 FILEDESC_XUNLOCK(fdp); 2301 return (error); 2302 } 2303 2304 int 2305 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2306 struct filecaps *fcaps) 2307 { 2308 int error; 2309 2310 MPASS(fd != NULL); 2311 2312 if (!fhold(fp)) 2313 return (EBADF); 2314 error = finstall_refed(td, fp, fd, flags, fcaps); 2315 if (__predict_false(error != 0)) { 2316 fdrop(fp, td); 2317 } 2318 return (error); 2319 } 2320 2321 /* 2322 * Build a new filedesc structure from another. 2323 * 2324 * If fdp is not NULL, return with it shared locked. 2325 */ 2326 struct filedesc * 2327 fdinit(void) 2328 { 2329 struct filedesc0 *newfdp0; 2330 struct filedesc *newfdp; 2331 2332 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2333 newfdp = &newfdp0->fd_fd; 2334 2335 /* Create the file descriptor table. */ 2336 FILEDESC_LOCK_INIT(newfdp); 2337 refcount_init(&newfdp->fd_refcnt, 1); 2338 refcount_init(&newfdp->fd_holdcnt, 1); 2339 newfdp->fd_map = newfdp0->fd_dmap; 2340 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2341 newfdp->fd_files->fdt_nfiles = NDFILE; 2342 2343 return (newfdp); 2344 } 2345 2346 /* 2347 * Build a pwddesc structure from another. 2348 * Copy the current, root, and jail root vnode references. 2349 * 2350 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked. 2351 */ 2352 struct pwddesc * 2353 pdinit(struct pwddesc *pdp, bool keeplock) 2354 { 2355 struct pwddesc *newpdp; 2356 struct pwd *newpwd; 2357 2358 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2359 2360 PWDDESC_LOCK_INIT(newpdp); 2361 refcount_init(&newpdp->pd_refcount, 1); 2362 newpdp->pd_cmask = CMASK; 2363 2364 if (pdp == NULL) { 2365 newpwd = pwd_alloc(); 2366 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2367 return (newpdp); 2368 } 2369 2370 PWDDESC_XLOCK(pdp); 2371 newpwd = pwd_hold_pwddesc(pdp); 2372 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2373 if (!keeplock) 2374 PWDDESC_XUNLOCK(pdp); 2375 return (newpdp); 2376 } 2377 2378 /* 2379 * Hold either filedesc or pwddesc of the passed process. 2380 * 2381 * The process lock is used to synchronize against the target exiting and 2382 * freeing the data. 2383 * 2384 * Clearing can be ilustrated in 3 steps: 2385 * 1. set the pointer to NULL. Either routine can race against it, hence 2386 * atomic_load_ptr. 2387 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2388 * race to either still see the pointer or find NULL. It is still safe to 2389 * grab a reference as clearing is stalled. 2390 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2391 * guaranteed to see NULL, making it safe to finish clearing 2392 */ 2393 static struct filedesc * 2394 fdhold(struct proc *p) 2395 { 2396 struct filedesc *fdp; 2397 2398 PROC_LOCK_ASSERT(p, MA_OWNED); 2399 fdp = atomic_load_ptr(&p->p_fd); 2400 if (fdp != NULL) 2401 refcount_acquire(&fdp->fd_holdcnt); 2402 return (fdp); 2403 } 2404 2405 static struct pwddesc * 2406 pdhold(struct proc *p) 2407 { 2408 struct pwddesc *pdp; 2409 2410 PROC_LOCK_ASSERT(p, MA_OWNED); 2411 pdp = atomic_load_ptr(&p->p_pd); 2412 if (pdp != NULL) 2413 refcount_acquire(&pdp->pd_refcount); 2414 return (pdp); 2415 } 2416 2417 static void 2418 fddrop(struct filedesc *fdp) 2419 { 2420 2421 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2422 if (refcount_release(&fdp->fd_holdcnt) == 0) 2423 return; 2424 } 2425 2426 FILEDESC_LOCK_DESTROY(fdp); 2427 uma_zfree(filedesc0_zone, fdp); 2428 } 2429 2430 static void 2431 pddrop(struct pwddesc *pdp) 2432 { 2433 struct pwd *pwd; 2434 2435 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2436 return; 2437 2438 PWDDESC_XLOCK(pdp); 2439 if (refcount_release(&pdp->pd_refcount) == 0) { 2440 PWDDESC_XUNLOCK(pdp); 2441 return; 2442 } 2443 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2444 pwd_set(pdp, NULL); 2445 PWDDESC_XUNLOCK(pdp); 2446 pwd_drop(pwd); 2447 2448 PWDDESC_LOCK_DESTROY(pdp); 2449 free(pdp, M_PWDDESC); 2450 } 2451 2452 /* 2453 * Share a filedesc structure. 2454 */ 2455 struct filedesc * 2456 fdshare(struct filedesc *fdp) 2457 { 2458 2459 refcount_acquire(&fdp->fd_refcnt); 2460 return (fdp); 2461 } 2462 2463 /* 2464 * Share a pwddesc structure. 2465 */ 2466 struct pwddesc * 2467 pdshare(struct pwddesc *pdp) 2468 { 2469 refcount_acquire(&pdp->pd_refcount); 2470 return (pdp); 2471 } 2472 2473 /* 2474 * Unshare a filedesc structure, if necessary by making a copy 2475 */ 2476 void 2477 fdunshare(struct thread *td) 2478 { 2479 struct filedesc *tmp; 2480 struct proc *p = td->td_proc; 2481 2482 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2483 return; 2484 2485 tmp = fdcopy(p->p_fd, p); 2486 fdescfree(td); 2487 p->p_fd = tmp; 2488 } 2489 2490 /* 2491 * Unshare a pwddesc structure. 2492 */ 2493 void 2494 pdunshare(struct thread *td) 2495 { 2496 struct pwddesc *pdp; 2497 struct proc *p; 2498 2499 p = td->td_proc; 2500 /* Not shared. */ 2501 if (refcount_load(&p->p_pd->pd_refcount) == 1) 2502 return; 2503 2504 pdp = pdcopy(p->p_pd); 2505 pdescfree(td); 2506 p->p_pd = pdp; 2507 } 2508 2509 /* 2510 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2511 * this is to ease callers, not catch errors. 2512 */ 2513 struct filedesc * 2514 fdcopy(struct filedesc *fdp, struct proc *p1) 2515 { 2516 struct filedesc *newfdp; 2517 struct filedescent *nfde, *ofde; 2518 struct file *fp; 2519 int i, lastfile; 2520 bool fork_pass; 2521 2522 MPASS(fdp != NULL); 2523 2524 fork_pass = false; 2525 newfdp = fdinit(); 2526 FILEDESC_SLOCK(fdp); 2527 for (;;) { 2528 lastfile = fdlastfile(fdp); 2529 if (lastfile < newfdp->fd_nfiles) 2530 break; 2531 FILEDESC_SUNLOCK(fdp); 2532 fdgrowtable(newfdp, lastfile + 1); 2533 FILEDESC_SLOCK(fdp); 2534 } 2535 2536 /* 2537 * Copy all passable descriptors (i.e. not kqueue), and 2538 * prepare to handle copyable but not passable descriptors 2539 * (kqueues). 2540 * 2541 * The pass to handle copying is performed after all passable 2542 * files are installed into the new file descriptor's table, 2543 * since kqueues need all referenced file descriptors already 2544 * valid, including other kqueues. For the same reason the 2545 * copying is done in two passes by itself, first installing 2546 * not fully initialized ('empty') copyable files into the new 2547 * fd table, and then giving the subsystems a second chance to 2548 * really fill the copied file backing structure with the 2549 * content. 2550 */ 2551 newfdp->fd_freefile = fdp->fd_freefile; 2552 FILEDESC_FOREACH_FDE(fdp, i, ofde) { 2553 const struct fileops *ops; 2554 2555 ops = ofde->fde_file->f_ops; 2556 fp = NULL; 2557 if ((ops->fo_flags & DFLAG_FORK) != 0 && 2558 (ofde->fde_flags & UF_FOCLOSE) == 0) { 2559 if (ops->fo_fork(newfdp, ofde->fde_file, &fp, p1, 2560 curthread) != 0) 2561 continue; 2562 fork_pass = true; 2563 } else if ((ops->fo_flags & DFLAG_PASSABLE) == 0 || 2564 (ofde->fde_flags & UF_FOCLOSE) != 0 || 2565 !fhold(ofde->fde_file)) { 2566 if (newfdp->fd_freefile == fdp->fd_freefile) 2567 newfdp->fd_freefile = i; 2568 continue; 2569 } 2570 nfde = &newfdp->fd_ofiles[i]; 2571 *nfde = *ofde; 2572 if (fp != NULL) 2573 nfde->fde_file = fp; 2574 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2575 fdused_init(newfdp, i); 2576 } 2577 MPASS(newfdp->fd_freefile != -1); 2578 FILEDESC_SUNLOCK(fdp); 2579 2580 /* 2581 * Now handle copying kqueues, since all fds, including 2582 * kqueues, are in place. 2583 */ 2584 if (__predict_false(fork_pass)) { 2585 FILEDESC_FOREACH_FDE(newfdp, i, nfde) { 2586 const struct fileops *ops; 2587 2588 ops = nfde->fde_file->f_ops; 2589 if ((ops->fo_flags & DFLAG_FORK) == 0 || 2590 nfde->fde_file == NULL) 2591 continue; 2592 ops->fo_fork(newfdp, NULL, &nfde->fde_file, p1, 2593 curthread); 2594 } 2595 } 2596 return (newfdp); 2597 } 2598 2599 /* 2600 * Copy a pwddesc structure. 2601 */ 2602 struct pwddesc * 2603 pdcopy(struct pwddesc *pdp) 2604 { 2605 struct pwddesc *newpdp; 2606 2607 MPASS(pdp != NULL); 2608 2609 newpdp = pdinit(pdp, true); 2610 newpdp->pd_cmask = pdp->pd_cmask; 2611 PWDDESC_XUNLOCK(pdp); 2612 return (newpdp); 2613 } 2614 2615 /* 2616 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2617 * one of processes using it exits) and the table used to be shared. 2618 */ 2619 static void 2620 fdclearlocks(struct thread *td) 2621 { 2622 struct filedesc *fdp; 2623 struct filedesc_to_leader *fdtol; 2624 struct flock lf; 2625 struct file *fp; 2626 struct proc *p; 2627 struct vnode *vp; 2628 int i; 2629 2630 p = td->td_proc; 2631 fdp = p->p_fd; 2632 fdtol = p->p_fdtol; 2633 MPASS(fdtol != NULL); 2634 2635 FILEDESC_XLOCK(fdp); 2636 KASSERT(fdtol->fdl_refcount > 0, 2637 ("filedesc_to_refcount botch: fdl_refcount=%d", 2638 fdtol->fdl_refcount)); 2639 if (fdtol->fdl_refcount == 1 && 2640 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2641 FILEDESC_FOREACH_FP(fdp, i, fp) { 2642 if (fp->f_type != DTYPE_VNODE || 2643 !fhold(fp)) 2644 continue; 2645 FILEDESC_XUNLOCK(fdp); 2646 lf.l_whence = SEEK_SET; 2647 lf.l_start = 0; 2648 lf.l_len = 0; 2649 lf.l_type = F_UNLCK; 2650 vp = fp->f_vnode; 2651 (void) VOP_ADVLOCK(vp, 2652 (caddr_t)p->p_leader, F_UNLCK, 2653 &lf, F_POSIX); 2654 FILEDESC_XLOCK(fdp); 2655 fdrop(fp, td); 2656 } 2657 } 2658 retry: 2659 if (fdtol->fdl_refcount == 1) { 2660 if (fdp->fd_holdleaderscount > 0 && 2661 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2662 /* 2663 * close() or kern_dup() has cleared a reference 2664 * in a shared file descriptor table. 2665 */ 2666 fdp->fd_holdleaderswakeup = 1; 2667 sx_sleep(&fdp->fd_holdleaderscount, 2668 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2669 goto retry; 2670 } 2671 if (fdtol->fdl_holdcount > 0) { 2672 /* 2673 * Ensure that fdtol->fdl_leader remains 2674 * valid in closef(). 2675 */ 2676 fdtol->fdl_wakeup = 1; 2677 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2678 "fdlhold", 0); 2679 goto retry; 2680 } 2681 } 2682 fdtol->fdl_refcount--; 2683 if (fdtol->fdl_refcount == 0 && 2684 fdtol->fdl_holdcount == 0) { 2685 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2686 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2687 } else 2688 fdtol = NULL; 2689 p->p_fdtol = NULL; 2690 FILEDESC_XUNLOCK(fdp); 2691 if (fdtol != NULL) 2692 free(fdtol, M_FILEDESC_TO_LEADER); 2693 } 2694 2695 /* 2696 * Release a filedesc structure. 2697 */ 2698 static void 2699 fdescfree_fds(struct thread *td, struct filedesc *fdp) 2700 { 2701 struct filedesc0 *fdp0; 2702 struct freetable *ft, *tft; 2703 struct filedescent *fde; 2704 struct file *fp; 2705 int i; 2706 2707 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2708 ("%s: fd table %p carries references", __func__, fdp)); 2709 2710 /* 2711 * Serialize with threads iterating over the table, if any. 2712 */ 2713 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2714 FILEDESC_XLOCK(fdp); 2715 FILEDESC_XUNLOCK(fdp); 2716 } 2717 2718 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2719 fp = fde->fde_file; 2720 fdefree_last(fde); 2721 (void) closef(fp, td); 2722 } 2723 2724 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2725 free(fdp->fd_map, M_FILEDESC); 2726 if (fdp->fd_nfiles > NDFILE) 2727 free(fdp->fd_files, M_FILEDESC); 2728 2729 fdp0 = (struct filedesc0 *)fdp; 2730 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2731 free(ft->ft_table, M_FILEDESC); 2732 2733 fddrop(fdp); 2734 } 2735 2736 void 2737 fdescfree(struct thread *td) 2738 { 2739 struct proc *p; 2740 struct filedesc *fdp; 2741 2742 p = td->td_proc; 2743 fdp = p->p_fd; 2744 MPASS(fdp != NULL); 2745 2746 #ifdef RACCT 2747 if (RACCT_ENABLED()) 2748 racct_set_unlocked(p, RACCT_NOFILE, 0); 2749 #endif 2750 2751 if (p->p_fdtol != NULL) 2752 fdclearlocks(td); 2753 2754 /* 2755 * Check fdhold for an explanation. 2756 */ 2757 atomic_store_ptr(&p->p_fd, NULL); 2758 atomic_thread_fence_seq_cst(); 2759 PROC_WAIT_UNLOCKED(p); 2760 2761 if (refcount_release(&fdp->fd_refcnt) == 0) 2762 return; 2763 2764 fdescfree_fds(td, fdp); 2765 } 2766 2767 void 2768 pdescfree(struct thread *td) 2769 { 2770 struct proc *p; 2771 struct pwddesc *pdp; 2772 2773 p = td->td_proc; 2774 pdp = p->p_pd; 2775 MPASS(pdp != NULL); 2776 2777 /* 2778 * Check pdhold for an explanation. 2779 */ 2780 atomic_store_ptr(&p->p_pd, NULL); 2781 atomic_thread_fence_seq_cst(); 2782 PROC_WAIT_UNLOCKED(p); 2783 2784 pddrop(pdp); 2785 } 2786 2787 /* 2788 * For setugid programs, we don't want to people to use that setugidness 2789 * to generate error messages which write to a file which otherwise would 2790 * otherwise be off-limits to the process. We check for filesystems where 2791 * the vnode can change out from under us after execve (like [lin]procfs). 2792 * 2793 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2794 * sufficient. We also don't check for setugidness since we know we are. 2795 */ 2796 static bool 2797 is_unsafe(struct file *fp) 2798 { 2799 struct vnode *vp; 2800 2801 if (fp->f_type != DTYPE_VNODE) 2802 return (false); 2803 2804 vp = fp->f_vnode; 2805 return ((vp->v_vflag & VV_PROCDEP) != 0); 2806 } 2807 2808 /* 2809 * Make this setguid thing safe, if at all possible. 2810 */ 2811 void 2812 fdsetugidsafety(struct thread *td) 2813 { 2814 struct filedesc *fdp; 2815 struct file *fp; 2816 int i; 2817 2818 fdp = td->td_proc->p_fd; 2819 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2820 ("the fdtable should not be shared")); 2821 MPASS(fdp->fd_nfiles >= 3); 2822 for (i = 0; i <= 2; i++) { 2823 fp = fdp->fd_ofiles[i].fde_file; 2824 if (fp != NULL && is_unsafe(fp)) { 2825 FILEDESC_XLOCK(fdp); 2826 knote_fdclose(td, i); 2827 /* 2828 * NULL-out descriptor prior to close to avoid 2829 * a race while close blocks. 2830 */ 2831 fdfree(fdp, i); 2832 FILEDESC_XUNLOCK(fdp); 2833 (void) closef(fp, td); 2834 } 2835 } 2836 } 2837 2838 /* 2839 * If a specific file object occupies a specific file descriptor, close the 2840 * file descriptor entry and drop a reference on the file object. This is a 2841 * convenience function to handle a subsequent error in a function that calls 2842 * falloc() that handles the race that another thread might have closed the 2843 * file descriptor out from under the thread creating the file object. 2844 */ 2845 void 2846 fdclose(struct thread *td, struct file *fp, int idx) 2847 { 2848 struct filedesc *fdp = td->td_proc->p_fd; 2849 2850 FILEDESC_XLOCK(fdp); 2851 if (fdp->fd_ofiles[idx].fde_file == fp) { 2852 fdfree(fdp, idx); 2853 FILEDESC_XUNLOCK(fdp); 2854 fdrop(fp, td); 2855 } else 2856 FILEDESC_XUNLOCK(fdp); 2857 } 2858 2859 /* 2860 * Close any files on exec? 2861 */ 2862 void 2863 fdcloseexec(struct thread *td) 2864 { 2865 struct filedesc *fdp; 2866 struct filedescent *fde; 2867 struct file *fp; 2868 int i; 2869 2870 fdp = td->td_proc->p_fd; 2871 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2872 ("the fdtable should not be shared")); 2873 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2874 fp = fde->fde_file; 2875 if (fp->f_type == DTYPE_MQUEUE || 2876 (fde->fde_flags & UF_EXCLOSE)) { 2877 FILEDESC_XLOCK(fdp); 2878 fdfree(fdp, i); 2879 (void) closefp(fdp, i, fp, td, false, false); 2880 FILEDESC_UNLOCK_ASSERT(fdp); 2881 } else if (fde->fde_flags & UF_FOCLOSE) { 2882 /* 2883 * https://austingroupbugs.net/view.php?id=1851 2884 * FD_CLOFORK should not be preserved across exec 2885 */ 2886 fde->fde_flags &= ~UF_FOCLOSE; 2887 } 2888 } 2889 } 2890 2891 /* 2892 * It is unsafe for set[ug]id processes to be started with file 2893 * descriptors 0..2 closed, as these descriptors are given implicit 2894 * significance in the Standard C library. fdcheckstd() will create a 2895 * descriptor referencing /dev/null for each of stdin, stdout, and 2896 * stderr that is not already open. 2897 */ 2898 int 2899 fdcheckstd(struct thread *td) 2900 { 2901 struct filedesc *fdp; 2902 register_t save; 2903 int i, error, devnull; 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 devnull = -1; 2910 for (i = 0; i <= 2; i++) { 2911 if (fdp->fd_ofiles[i].fde_file != NULL) 2912 continue; 2913 2914 save = td->td_retval[0]; 2915 if (devnull != -1) { 2916 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2917 } else { 2918 error = kern_openat(td, AT_FDCWD, "/dev/null", 2919 UIO_SYSSPACE, O_RDWR, 0); 2920 if (error == 0) { 2921 devnull = td->td_retval[0]; 2922 KASSERT(devnull == i, ("we didn't get our fd")); 2923 } 2924 } 2925 td->td_retval[0] = save; 2926 if (error != 0) 2927 return (error); 2928 } 2929 return (0); 2930 } 2931 2932 /* 2933 * Internal form of close. Decrement reference count on file structure. 2934 * Note: td may be NULL when closing a file that was being passed in a 2935 * message. 2936 */ 2937 int 2938 closef(struct file *fp, struct thread *td) 2939 { 2940 struct vnode *vp; 2941 struct flock lf; 2942 struct filedesc_to_leader *fdtol; 2943 struct filedesc *fdp; 2944 2945 MPASS(td != NULL); 2946 2947 /* 2948 * POSIX record locking dictates that any close releases ALL 2949 * locks owned by this process. This is handled by setting 2950 * a flag in the unlock to free ONLY locks obeying POSIX 2951 * semantics, and not to free BSD-style file locks. 2952 * If the descriptor was in a message, POSIX-style locks 2953 * aren't passed with the descriptor, and the thread pointer 2954 * will be NULL. Callers should be careful only to pass a 2955 * NULL thread pointer when there really is no owning 2956 * context that might have locks, or the locks will be 2957 * leaked. 2958 */ 2959 if (fp->f_type == DTYPE_VNODE) { 2960 vp = fp->f_vnode; 2961 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2962 lf.l_whence = SEEK_SET; 2963 lf.l_start = 0; 2964 lf.l_len = 0; 2965 lf.l_type = F_UNLCK; 2966 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2967 F_UNLCK, &lf, F_POSIX); 2968 } 2969 fdtol = td->td_proc->p_fdtol; 2970 if (fdtol != NULL) { 2971 /* 2972 * Handle special case where file descriptor table is 2973 * shared between multiple process leaders. 2974 */ 2975 fdp = td->td_proc->p_fd; 2976 FILEDESC_XLOCK(fdp); 2977 for (fdtol = fdtol->fdl_next; 2978 fdtol != td->td_proc->p_fdtol; 2979 fdtol = fdtol->fdl_next) { 2980 if ((fdtol->fdl_leader->p_flag & 2981 P_ADVLOCK) == 0) 2982 continue; 2983 fdtol->fdl_holdcount++; 2984 FILEDESC_XUNLOCK(fdp); 2985 lf.l_whence = SEEK_SET; 2986 lf.l_start = 0; 2987 lf.l_len = 0; 2988 lf.l_type = F_UNLCK; 2989 vp = fp->f_vnode; 2990 (void) VOP_ADVLOCK(vp, 2991 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2992 F_POSIX); 2993 FILEDESC_XLOCK(fdp); 2994 fdtol->fdl_holdcount--; 2995 if (fdtol->fdl_holdcount == 0 && 2996 fdtol->fdl_wakeup != 0) { 2997 fdtol->fdl_wakeup = 0; 2998 wakeup(fdtol); 2999 } 3000 } 3001 FILEDESC_XUNLOCK(fdp); 3002 } 3003 } 3004 return (fdrop_close(fp, td)); 3005 } 3006 3007 /* 3008 * Hack for file descriptor passing code. 3009 */ 3010 void 3011 closef_nothread(struct file *fp) 3012 { 3013 3014 fdrop(fp, NULL); 3015 } 3016 3017 /* 3018 * Initialize the file pointer with the specified properties. 3019 * 3020 * The ops are set with release semantics to be certain that the flags, type, 3021 * and data are visible when ops is. This is to prevent ops methods from being 3022 * called with bad data. 3023 */ 3024 void 3025 finit(struct file *fp, u_int flag, short type, void *data, 3026 const struct fileops *ops) 3027 { 3028 fp->f_data = data; 3029 fp->f_flag = flag; 3030 fp->f_type = type; 3031 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 3032 } 3033 3034 void 3035 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops) 3036 { 3037 fp->f_seqcount[UIO_READ] = 1; 3038 fp->f_seqcount[UIO_WRITE] = 1; 3039 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 3040 data, ops); 3041 } 3042 3043 int 3044 fget_cap_noref(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3045 struct file **fpp, struct filecaps *havecapsp) 3046 { 3047 struct filedescent *fde; 3048 int error; 3049 3050 FILEDESC_LOCK_ASSERT(fdp); 3051 3052 *fpp = NULL; 3053 fde = fdeget_noref(fdp, fd); 3054 if (fde == NULL) { 3055 error = EBADF; 3056 goto out; 3057 } 3058 3059 #ifdef CAPABILITIES 3060 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 3061 if (error != 0) 3062 goto out; 3063 #endif 3064 3065 if (havecapsp != NULL) 3066 filecaps_copy(&fde->fde_caps, havecapsp, true); 3067 3068 *fpp = fde->fde_file; 3069 3070 error = 0; 3071 out: 3072 return (error); 3073 } 3074 3075 #ifdef CAPABILITIES 3076 int 3077 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3078 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3079 { 3080 struct filedesc *fdp = td->td_proc->p_fd; 3081 int error; 3082 struct file *fp; 3083 seqc_t seq; 3084 3085 *fpp = NULL; 3086 for (;;) { 3087 error = fget_unlocked_seq(td, fd, needrightsp, flagsp, &fp, 3088 &seq); 3089 if (error != 0) 3090 return (error); 3091 3092 if (havecapsp != NULL) { 3093 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 3094 havecapsp, false)) { 3095 fdrop(fp, td); 3096 goto get_locked; 3097 } 3098 } 3099 3100 if (!fd_modified(fdp, fd, seq)) 3101 break; 3102 fdrop(fp, td); 3103 } 3104 3105 *fpp = fp; 3106 return (0); 3107 3108 get_locked: 3109 FILEDESC_SLOCK(fdp); 3110 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp); 3111 if (error == 0 && !fhold(*fpp)) 3112 error = EBADF; 3113 FILEDESC_SUNLOCK(fdp); 3114 return (error); 3115 } 3116 #else 3117 int 3118 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, 3119 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) 3120 { 3121 int error; 3122 error = fget_unlocked_flags(td, fd, needrightsp, flagsp, fpp); 3123 if (havecapsp != NULL && error == 0) 3124 filecaps_fill(havecapsp); 3125 3126 return (error); 3127 } 3128 #endif 3129 3130 int 3131 fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp) 3132 { 3133 struct filedesc *fdp; 3134 struct file *fp; 3135 int error; 3136 3137 if (p == td->td_proc) /* curproc */ 3138 return (fget_unlocked(td, fd, &cap_no_rights, fpp)); 3139 3140 PROC_LOCK(p); 3141 fdp = fdhold(p); 3142 PROC_UNLOCK(p); 3143 if (fdp == NULL) 3144 return (ENOENT); 3145 FILEDESC_SLOCK(fdp); 3146 if (refcount_load(&fdp->fd_refcnt) != 0) { 3147 fp = fget_noref(fdp, fd); 3148 if (fp != NULL && fhold(fp)) { 3149 *fpp = fp; 3150 error = 0; 3151 } else { 3152 error = EBADF; 3153 } 3154 } else { 3155 error = ENOENT; 3156 } 3157 FILEDESC_SUNLOCK(fdp); 3158 fddrop(fdp); 3159 return (error); 3160 } 3161 3162 int 3163 fget_remote_foreach(struct thread *td, struct proc *p, 3164 int (*fn)(struct proc *, int, struct file *, void *), void *arg) 3165 { 3166 struct filedesc *fdp; 3167 struct fdescenttbl *fdt; 3168 struct file *fp; 3169 int error, error1, fd, highfd; 3170 3171 error = 0; 3172 PROC_LOCK(p); 3173 fdp = fdhold(p); 3174 PROC_UNLOCK(p); 3175 if (fdp == NULL) 3176 return (ENOENT); 3177 3178 FILEDESC_SLOCK(fdp); 3179 if (refcount_load(&fdp->fd_refcnt) != 0) { 3180 fdt = atomic_load_ptr(&fdp->fd_files); 3181 highfd = fdt->fdt_nfiles - 1; 3182 FILEDESC_SUNLOCK(fdp); 3183 } else { 3184 error = ENOENT; 3185 FILEDESC_SUNLOCK(fdp); 3186 goto out; 3187 } 3188 3189 for (fd = 0; fd <= highfd; fd++) { 3190 error1 = fget_remote(td, p, fd, &fp); 3191 if (error1 != 0) 3192 continue; 3193 error = fn(p, fd, fp, arg); 3194 fdrop(fp, td); 3195 if (error != 0) 3196 break; 3197 } 3198 out: 3199 fddrop(fdp); 3200 return (error); 3201 } 3202 3203 #ifdef CAPABILITIES 3204 int 3205 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3206 { 3207 const struct filedescent *fde; 3208 const struct fdescenttbl *fdt; 3209 struct filedesc *fdp; 3210 struct file *fp; 3211 struct vnode *vp; 3212 const cap_rights_t *haverights; 3213 cap_rights_t rights; 3214 seqc_t seq; 3215 int fd, flags; 3216 3217 VFS_SMR_ASSERT_ENTERED(); 3218 3219 fd = ndp->ni_dirfd; 3220 rights = *ndp->ni_rightsneeded; 3221 cap_rights_set_one(&rights, CAP_LOOKUP); 3222 3223 fdp = curproc->p_fd; 3224 fdt = fdp->fd_files; 3225 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3226 return (EBADF); 3227 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3228 fde = &fdt->fdt_ofiles[fd]; 3229 haverights = cap_rights_fde_inline(fde); 3230 fp = fde->fde_file; 3231 if (__predict_false(fp == NULL)) 3232 return (EAGAIN); 3233 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 3234 return (EAGAIN); 3235 flags = fp->f_flag & FSEARCH; 3236 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3237 O_RESOLVE_BENEATH : 0; 3238 vp = fp->f_vnode; 3239 if (__predict_false(vp == NULL)) { 3240 return (EAGAIN); 3241 } 3242 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 3243 return (EAGAIN); 3244 } 3245 /* 3246 * Use an acquire barrier to force re-reading of fdt so it is 3247 * refreshed for verification. 3248 */ 3249 atomic_thread_fence_acq(); 3250 fdt = fdp->fd_files; 3251 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3252 return (EAGAIN); 3253 /* 3254 * If file descriptor doesn't have all rights, 3255 * all lookups relative to it must also be 3256 * strictly relative. 3257 * 3258 * Not yet supported by fast path. 3259 */ 3260 CAP_ALL(&rights); 3261 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3262 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3263 ndp->ni_filecaps.fc_nioctls != -1) { 3264 #ifdef notyet 3265 ndp->ni_lcf |= NI_LCF_STRICTREL; 3266 #else 3267 return (EAGAIN); 3268 #endif 3269 } 3270 *vpp = vp; 3271 *flagsp = flags; 3272 return (0); 3273 } 3274 #else 3275 int 3276 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp) 3277 { 3278 const struct filedescent *fde; 3279 const struct fdescenttbl *fdt; 3280 struct filedesc *fdp; 3281 struct file *fp; 3282 struct vnode *vp; 3283 int fd, flags; 3284 3285 VFS_SMR_ASSERT_ENTERED(); 3286 3287 fd = ndp->ni_dirfd; 3288 fdp = curproc->p_fd; 3289 fdt = fdp->fd_files; 3290 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3291 return (EBADF); 3292 fde = &fdt->fdt_ofiles[fd]; 3293 fp = fde->fde_file; 3294 if (__predict_false(fp == NULL)) 3295 return (EAGAIN); 3296 flags = fp->f_flag & FSEARCH; 3297 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? 3298 O_RESOLVE_BENEATH : 0; 3299 vp = fp->f_vnode; 3300 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3301 return (EAGAIN); 3302 } 3303 /* 3304 * Use an acquire barrier to force re-reading of fdt so it is 3305 * refreshed for verification. 3306 */ 3307 atomic_thread_fence_acq(); 3308 fdt = fdp->fd_files; 3309 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3310 return (EAGAIN); 3311 filecaps_fill(&ndp->ni_filecaps); 3312 *vpp = vp; 3313 *flagsp = flags; 3314 return (0); 3315 } 3316 #endif 3317 3318 int 3319 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp) 3320 { 3321 struct thread *td; 3322 struct file *fp; 3323 struct vnode *vp; 3324 struct componentname *cnp; 3325 cap_rights_t rights; 3326 int error; 3327 uint8_t flags; 3328 3329 td = curthread; 3330 rights = *ndp->ni_rightsneeded; 3331 cap_rights_set_one(&rights, CAP_LOOKUP); 3332 cnp = &ndp->ni_cnd; 3333 3334 error = fget_cap(td, ndp->ni_dirfd, &rights, &flags, &fp, 3335 &ndp->ni_filecaps); 3336 if (__predict_false(error != 0)) 3337 return (error); 3338 if (__predict_false(fp->f_ops == &badfileops)) { 3339 error = EBADF; 3340 goto out_free; 3341 } 3342 vp = fp->f_vnode; 3343 if (__predict_false(vp == NULL)) { 3344 error = ENOTDIR; 3345 goto out_free; 3346 } 3347 vrefact(vp); 3348 /* 3349 * XXX does not check for VDIR, handled by namei_setup 3350 */ 3351 if ((fp->f_flag & FSEARCH) != 0) 3352 cnp->cn_flags |= NOEXECCHECK; 3353 if ((flags & UF_RESOLVE_BENEATH) != 0) { 3354 cnp->cn_flags |= RBENEATH; 3355 ndp->ni_resflags |= NIRES_BENEATH; 3356 } 3357 fdrop(fp, td); 3358 3359 #ifdef CAPABILITIES 3360 /* 3361 * If file descriptor doesn't have all rights, 3362 * all lookups relative to it must also be 3363 * strictly relative. 3364 */ 3365 CAP_ALL(&rights); 3366 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3367 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3368 ndp->ni_filecaps.fc_nioctls != -1) { 3369 ndp->ni_lcf |= NI_LCF_STRICTREL; 3370 ndp->ni_resflags |= NIRES_STRICTREL; 3371 } 3372 #endif 3373 3374 /* 3375 * TODO: avoid copying ioctl caps if it can be helped to begin with 3376 */ 3377 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0) 3378 filecaps_free_ioctl(&ndp->ni_filecaps); 3379 3380 *vpp = vp; 3381 return (0); 3382 3383 out_free: 3384 filecaps_free(&ndp->ni_filecaps); 3385 fdrop(fp, td); 3386 return (error); 3387 } 3388 3389 /* 3390 * Fetch the descriptor locklessly. 3391 * 3392 * We avoid fdrop() races by never raising a refcount above 0. To accomplish 3393 * this we have to use a cmpset loop rather than an atomic_add. The descriptor 3394 * must be re-verified once we acquire a reference to be certain that the 3395 * identity is still correct and we did not lose a race due to preemption. 3396 * 3397 * Force a reload of fdt when looping. Another thread could reallocate 3398 * the table before this fd was closed, so it is possible that there is 3399 * a stale fp pointer in cached version. 3400 */ 3401 #ifdef CAPABILITIES 3402 static int 3403 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3404 uint8_t *flagsp, struct file **fpp, seqc_t *seqp) 3405 { 3406 struct filedesc *fdp; 3407 const struct filedescent *fde; 3408 const struct fdescenttbl *fdt; 3409 struct file *fp; 3410 seqc_t seq; 3411 cap_rights_t haverights; 3412 int error; 3413 uint8_t flags; 3414 3415 fdp = td->td_proc->p_fd; 3416 fdt = fdp->fd_files; 3417 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3418 return (EBADF); 3419 3420 for (;;) { 3421 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3422 fde = &fdt->fdt_ofiles[fd]; 3423 haverights = *cap_rights_fde_inline(fde); 3424 fp = fde->fde_file; 3425 flags = fde->fde_flags; 3426 if (__predict_false(fp == NULL)) { 3427 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3428 return (EBADF); 3429 fdt = atomic_load_ptr(&fdp->fd_files); 3430 continue; 3431 } 3432 error = cap_check_inline(&haverights, needrightsp); 3433 if (__predict_false(error != 0)) { 3434 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3435 return (error); 3436 fdt = atomic_load_ptr(&fdp->fd_files); 3437 continue; 3438 } 3439 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3440 fdt = atomic_load_ptr(&fdp->fd_files); 3441 continue; 3442 } 3443 /* 3444 * Use an acquire barrier to force re-reading of fdt so it is 3445 * refreshed for verification. 3446 */ 3447 atomic_thread_fence_acq(); 3448 fdt = fdp->fd_files; 3449 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)) 3450 break; 3451 fdrop(fp, td); 3452 } 3453 *fpp = fp; 3454 if (flagsp != NULL) 3455 *flagsp = flags; 3456 if (seqp != NULL) 3457 *seqp = seq; 3458 return (0); 3459 } 3460 #else 3461 static int 3462 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp, 3463 uint8_t *flagsp, struct file **fpp, seqc_t *seqp __unused) 3464 { 3465 struct filedesc *fdp; 3466 const struct fdescenttbl *fdt; 3467 struct file *fp; 3468 uint8_t flags; 3469 3470 fdp = td->td_proc->p_fd; 3471 fdt = fdp->fd_files; 3472 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3473 return (EBADF); 3474 3475 for (;;) { 3476 fp = fdt->fdt_ofiles[fd].fde_file; 3477 flags = fdt->fdt_ofiles[fd].fde_flags; 3478 if (__predict_false(fp == NULL)) 3479 return (EBADF); 3480 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3481 fdt = atomic_load_ptr(&fdp->fd_files); 3482 continue; 3483 } 3484 /* 3485 * Use an acquire barrier to force re-reading of fdt so it is 3486 * refreshed for verification. 3487 */ 3488 atomic_thread_fence_acq(); 3489 fdt = fdp->fd_files; 3490 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file)) 3491 break; 3492 fdrop(fp, td); 3493 } 3494 if (flagsp != NULL) 3495 *flagsp = flags; 3496 *fpp = fp; 3497 return (0); 3498 } 3499 #endif 3500 3501 /* 3502 * See the comments in fget_unlocked_seq for an explanation of how this works. 3503 * 3504 * This is a simplified variant which bails out to the aforementioned routine 3505 * if anything goes wrong. In practice this only happens when userspace is 3506 * racing with itself. 3507 */ 3508 int 3509 fget_unlocked_flags(struct thread *td, int fd, const cap_rights_t *needrightsp, 3510 uint8_t *flagsp, struct file **fpp) 3511 { 3512 struct filedesc *fdp; 3513 #ifdef CAPABILITIES 3514 const struct filedescent *fde; 3515 #endif 3516 const struct fdescenttbl *fdt; 3517 struct file *fp; 3518 #ifdef CAPABILITIES 3519 seqc_t seq; 3520 const cap_rights_t *haverights; 3521 #endif 3522 uint8_t flags; 3523 3524 fdp = td->td_proc->p_fd; 3525 fdt = fdp->fd_files; 3526 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { 3527 *fpp = NULL; 3528 return (EBADF); 3529 } 3530 #ifdef CAPABILITIES 3531 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3532 fde = &fdt->fdt_ofiles[fd]; 3533 haverights = cap_rights_fde_inline(fde); 3534 fp = fde->fde_file; 3535 flags = fde->fde_flags; 3536 #else 3537 fp = fdt->fdt_ofiles[fd].fde_file; 3538 flags = fdt->fdt_ofiles[fd].fde_flags; 3539 #endif 3540 if (__predict_false(fp == NULL)) 3541 goto out_fallback; 3542 #ifdef CAPABILITIES 3543 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3544 goto out_fallback; 3545 #endif 3546 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3547 goto out_fallback; 3548 3549 /* 3550 * Use an acquire barrier to force re-reading of fdt so it is 3551 * refreshed for verification. 3552 */ 3553 atomic_thread_fence_acq(); 3554 fdt = fdp->fd_files; 3555 #ifdef CAPABILITIES 3556 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3557 #else 3558 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3559 #endif 3560 goto out_fdrop; 3561 *fpp = fp; 3562 if (flagsp != NULL) 3563 *flagsp = flags; 3564 return (0); 3565 out_fdrop: 3566 fdrop(fp, td); 3567 out_fallback: 3568 *fpp = NULL; 3569 return (fget_unlocked_seq(td, fd, needrightsp, flagsp, fpp, NULL)); 3570 } 3571 3572 int 3573 fget_unlocked(struct thread *td, int fd, const cap_rights_t *needrightsp, 3574 struct file **fpp) 3575 { 3576 return (fget_unlocked_flags(td, fd, needrightsp, NULL, fpp)); 3577 } 3578 3579 /* 3580 * Translate fd -> file when the caller guarantees the file descriptor table 3581 * can't be changed by others. 3582 * 3583 * Note this does not mean the file object itself is only visible to the caller, 3584 * merely that it wont disappear without having to be referenced. 3585 * 3586 * Must be paired with fput_only_user. 3587 */ 3588 #ifdef CAPABILITIES 3589 int 3590 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3591 struct file **fpp) 3592 { 3593 const struct filedescent *fde; 3594 const struct fdescenttbl *fdt; 3595 const cap_rights_t *haverights; 3596 struct file *fp; 3597 int error; 3598 3599 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3600 3601 *fpp = NULL; 3602 if (__predict_false(fd >= fdp->fd_nfiles)) 3603 return (EBADF); 3604 3605 fdt = fdp->fd_files; 3606 fde = &fdt->fdt_ofiles[fd]; 3607 fp = fde->fde_file; 3608 if (__predict_false(fp == NULL)) 3609 return (EBADF); 3610 MPASS(refcount_load(&fp->f_count) > 0); 3611 haverights = cap_rights_fde_inline(fde); 3612 error = cap_check_inline(haverights, needrightsp); 3613 if (__predict_false(error != 0)) 3614 return (error); 3615 *fpp = fp; 3616 return (0); 3617 } 3618 #else 3619 int 3620 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, 3621 struct file **fpp) 3622 { 3623 struct file *fp; 3624 3625 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3626 3627 *fpp = NULL; 3628 if (__predict_false(fd >= fdp->fd_nfiles)) 3629 return (EBADF); 3630 3631 fp = fdp->fd_ofiles[fd].fde_file; 3632 if (__predict_false(fp == NULL)) 3633 return (EBADF); 3634 3635 MPASS(refcount_load(&fp->f_count) > 0); 3636 *fpp = fp; 3637 return (0); 3638 } 3639 #endif 3640 3641 /* 3642 * Extract the file pointer associated with the specified descriptor for the 3643 * current user process. 3644 * 3645 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3646 * returned. 3647 * 3648 * File's rights will be checked against the capability rights mask. 3649 * 3650 * If an error occurred the non-zero error is returned and *fpp is set to 3651 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3652 * responsible for fdrop(). 3653 */ 3654 static __inline int 3655 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3656 const cap_rights_t *needrightsp) 3657 { 3658 struct file *fp; 3659 int error; 3660 3661 *fpp = NULL; 3662 error = fget_unlocked(td, fd, needrightsp, &fp); 3663 if (__predict_false(error != 0)) 3664 return (error); 3665 if (__predict_false(fp->f_ops == &badfileops)) { 3666 fdrop(fp, td); 3667 return (EBADF); 3668 } 3669 3670 /* 3671 * FREAD and FWRITE failure return EBADF as per POSIX. 3672 */ 3673 error = 0; 3674 switch (flags) { 3675 case FREAD: 3676 case FWRITE: 3677 if ((fp->f_flag & flags) == 0) 3678 error = EBADF; 3679 break; 3680 case FEXEC: 3681 if (fp->f_ops != &path_fileops && 3682 ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3683 (fp->f_flag & FWRITE) != 0)) 3684 error = EBADF; 3685 break; 3686 case 0: 3687 break; 3688 default: 3689 KASSERT(0, ("wrong flags")); 3690 } 3691 3692 if (error != 0) { 3693 fdrop(fp, td); 3694 return (error); 3695 } 3696 3697 *fpp = fp; 3698 return (0); 3699 } 3700 3701 int 3702 fget(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp) 3703 { 3704 3705 return (_fget(td, fd, fpp, 0, rightsp)); 3706 } 3707 3708 int 3709 fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp, 3710 vm_prot_t *maxprotp, struct file **fpp) 3711 { 3712 int error; 3713 #ifndef CAPABILITIES 3714 error = _fget(td, fd, fpp, 0, rightsp); 3715 if (maxprotp != NULL) 3716 *maxprotp = VM_PROT_ALL; 3717 return (error); 3718 #else 3719 cap_rights_t fdrights; 3720 struct filedesc *fdp; 3721 struct file *fp; 3722 seqc_t seq; 3723 3724 *fpp = NULL; 3725 fdp = td->td_proc->p_fd; 3726 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3727 for (;;) { 3728 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3729 if (__predict_false(error != 0)) 3730 return (error); 3731 if (__predict_false(fp->f_ops == &badfileops)) { 3732 fdrop(fp, td); 3733 return (EBADF); 3734 } 3735 if (maxprotp != NULL) 3736 fdrights = *cap_rights(fdp, fd); 3737 if (!fd_modified(fdp, fd, seq)) 3738 break; 3739 fdrop(fp, td); 3740 } 3741 3742 /* 3743 * If requested, convert capability rights to access flags. 3744 */ 3745 if (maxprotp != NULL) 3746 *maxprotp = cap_rights_to_vmprot(&fdrights); 3747 *fpp = fp; 3748 return (0); 3749 #endif 3750 } 3751 3752 int 3753 fget_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3754 struct file **fpp) 3755 { 3756 3757 return (_fget(td, fd, fpp, FREAD, rightsp)); 3758 } 3759 3760 int 3761 fget_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3762 struct file **fpp) 3763 { 3764 3765 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3766 } 3767 3768 int 3769 fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp, 3770 int needfcntl, struct file **fpp) 3771 { 3772 #ifndef CAPABILITIES 3773 return (fget_unlocked(td, fd, rightsp, fpp)); 3774 #else 3775 struct filedesc *fdp = td->td_proc->p_fd; 3776 struct file *fp; 3777 int error; 3778 seqc_t seq; 3779 3780 *fpp = NULL; 3781 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3782 for (;;) { 3783 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq); 3784 if (error != 0) 3785 return (error); 3786 error = cap_fcntl_check(fdp, fd, needfcntl); 3787 if (!fd_modified(fdp, fd, seq)) 3788 break; 3789 fdrop(fp, td); 3790 } 3791 if (error != 0) { 3792 fdrop(fp, td); 3793 return (error); 3794 } 3795 *fpp = fp; 3796 return (0); 3797 #endif 3798 } 3799 3800 /* 3801 * Like fget() but loads the underlying vnode, or returns an error if the 3802 * descriptor does not represent a vnode. Note that pipes use vnodes but 3803 * never have VM objects. The returned vnode will be vref()'d. 3804 * 3805 * XXX: what about the unused flags ? 3806 */ 3807 static __inline int 3808 _fgetvp(struct thread *td, int fd, int flags, const cap_rights_t *needrightsp, 3809 struct vnode **vpp) 3810 { 3811 struct file *fp; 3812 int error; 3813 3814 *vpp = NULL; 3815 error = _fget(td, fd, &fp, flags, needrightsp); 3816 if (error != 0) 3817 return (error); 3818 if (fp->f_vnode == NULL) { 3819 error = EINVAL; 3820 } else { 3821 *vpp = fp->f_vnode; 3822 vrefact(*vpp); 3823 } 3824 fdrop(fp, td); 3825 3826 return (error); 3827 } 3828 3829 int 3830 fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp, 3831 struct vnode **vpp) 3832 { 3833 3834 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3835 } 3836 3837 int 3838 fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp, 3839 struct filecaps *havecaps, struct vnode **vpp) 3840 { 3841 struct filecaps caps; 3842 struct file *fp; 3843 int error; 3844 3845 error = fget_cap(td, fd, needrightsp, NULL, &fp, &caps); 3846 if (error != 0) 3847 return (error); 3848 if (fp->f_ops == &badfileops) { 3849 error = EBADF; 3850 goto out; 3851 } 3852 if (fp->f_vnode == NULL) { 3853 error = EINVAL; 3854 goto out; 3855 } 3856 3857 *havecaps = caps; 3858 *vpp = fp->f_vnode; 3859 vrefact(*vpp); 3860 fdrop(fp, td); 3861 3862 return (0); 3863 out: 3864 filecaps_free(&caps); 3865 fdrop(fp, td); 3866 return (error); 3867 } 3868 3869 int 3870 fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp, 3871 struct vnode **vpp) 3872 { 3873 3874 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3875 } 3876 3877 int 3878 fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp, 3879 struct vnode **vpp) 3880 { 3881 3882 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3883 } 3884 3885 #ifdef notyet 3886 int 3887 fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp, 3888 struct vnode **vpp) 3889 { 3890 3891 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3892 } 3893 #endif 3894 3895 /* 3896 * Handle the last reference to a file being closed. 3897 * 3898 * Without the noinline attribute clang keeps inlining the func thorough this 3899 * file when fdrop is used. 3900 */ 3901 int __noinline 3902 _fdrop(struct file *fp, struct thread *td) 3903 { 3904 int error; 3905 3906 KASSERT(refcount_load(&fp->f_count) == 0, 3907 ("fdrop: fp %p count %d", fp, refcount_load(&fp->f_count))); 3908 3909 error = fo_close(fp, td); 3910 atomic_subtract_int(&openfiles, 1); 3911 crfree(fp->f_cred); 3912 free(fp->f_advice, M_FADVISE); 3913 uma_zfree(file_zone, fp); 3914 3915 return (error); 3916 } 3917 3918 /* 3919 * Apply an advisory lock on a file descriptor. 3920 * 3921 * Just attempt to get a record lock of the requested type on the entire file 3922 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 3923 */ 3924 #ifndef _SYS_SYSPROTO_H_ 3925 struct flock_args { 3926 int fd; 3927 int how; 3928 }; 3929 #endif 3930 /* ARGSUSED */ 3931 int 3932 sys_flock(struct thread *td, struct flock_args *uap) 3933 { 3934 struct file *fp; 3935 struct vnode *vp; 3936 struct flock lf; 3937 int error; 3938 3939 error = fget(td, uap->fd, &cap_flock_rights, &fp); 3940 if (error != 0) 3941 return (error); 3942 error = EOPNOTSUPP; 3943 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 3944 goto done; 3945 } 3946 if (fp->f_ops == &path_fileops) { 3947 goto done; 3948 } 3949 3950 error = 0; 3951 vp = fp->f_vnode; 3952 lf.l_whence = SEEK_SET; 3953 lf.l_start = 0; 3954 lf.l_len = 0; 3955 if (uap->how & LOCK_UN) { 3956 lf.l_type = F_UNLCK; 3957 atomic_clear_int(&fp->f_flag, FHASLOCK); 3958 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3959 goto done; 3960 } 3961 if (uap->how & LOCK_EX) 3962 lf.l_type = F_WRLCK; 3963 else if (uap->how & LOCK_SH) 3964 lf.l_type = F_RDLCK; 3965 else { 3966 error = EBADF; 3967 goto done; 3968 } 3969 atomic_set_int(&fp->f_flag, FHASLOCK); 3970 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3971 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3972 done: 3973 fdrop(fp, td); 3974 return (error); 3975 } 3976 /* 3977 * Duplicate the specified descriptor to a free descriptor. 3978 */ 3979 int 3980 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3981 int openerror, int *indxp) 3982 { 3983 struct filedescent *newfde, *oldfde; 3984 struct file *fp; 3985 u_long *ioctls; 3986 int error, indx; 3987 3988 KASSERT(openerror == ENODEV || openerror == ENXIO, 3989 ("unexpected error %d in %s", openerror, __func__)); 3990 3991 /* 3992 * If the to-be-dup'd fd number is greater than the allowed number 3993 * of file descriptors, or the fd to be dup'd has already been 3994 * closed, then reject. 3995 */ 3996 FILEDESC_XLOCK(fdp); 3997 if ((fp = fget_noref(fdp, dfd)) == NULL) { 3998 FILEDESC_XUNLOCK(fdp); 3999 return (EBADF); 4000 } 4001 4002 error = fdalloc(td, 0, &indx); 4003 if (error != 0) { 4004 FILEDESC_XUNLOCK(fdp); 4005 return (error); 4006 } 4007 4008 /* 4009 * There are two cases of interest here. 4010 * 4011 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 4012 * 4013 * For ENXIO steal away the file structure from (dfd) and store it in 4014 * (indx). (dfd) is effectively closed by this operation. 4015 */ 4016 switch (openerror) { 4017 case ENODEV: 4018 /* 4019 * Check that the mode the file is being opened for is a 4020 * subset of the mode of the existing descriptor. 4021 */ 4022 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 4023 fdunused(fdp, indx); 4024 FILEDESC_XUNLOCK(fdp); 4025 return (EACCES); 4026 } 4027 if (!fhold(fp)) { 4028 fdunused(fdp, indx); 4029 FILEDESC_XUNLOCK(fdp); 4030 return (EBADF); 4031 } 4032 newfde = &fdp->fd_ofiles[indx]; 4033 oldfde = &fdp->fd_ofiles[dfd]; 4034 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 4035 #ifdef CAPABILITIES 4036 seqc_write_begin(&newfde->fde_seqc); 4037 #endif 4038 fde_copy(oldfde, newfde); 4039 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 4040 ioctls); 4041 #ifdef CAPABILITIES 4042 seqc_write_end(&newfde->fde_seqc); 4043 #endif 4044 break; 4045 case ENXIO: 4046 /* 4047 * Steal away the file pointer from dfd and stuff it into indx. 4048 */ 4049 newfde = &fdp->fd_ofiles[indx]; 4050 oldfde = &fdp->fd_ofiles[dfd]; 4051 #ifdef CAPABILITIES 4052 seqc_write_begin(&oldfde->fde_seqc); 4053 seqc_write_begin(&newfde->fde_seqc); 4054 #endif 4055 fde_copy(oldfde, newfde); 4056 oldfde->fde_file = NULL; 4057 fdunused(fdp, dfd); 4058 #ifdef CAPABILITIES 4059 seqc_write_end(&newfde->fde_seqc); 4060 seqc_write_end(&oldfde->fde_seqc); 4061 #endif 4062 break; 4063 } 4064 FILEDESC_XUNLOCK(fdp); 4065 *indxp = indx; 4066 return (0); 4067 } 4068 4069 /* 4070 * This sysctl determines if we will allow a process to chroot(2) if it 4071 * has a directory open: 4072 * 0: disallowed for all processes. 4073 * 1: allowed for processes that were not already chroot(2)'ed. 4074 * 2: allowed for all processes. 4075 */ 4076 4077 static int chroot_allow_open_directories = 1; 4078 4079 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 4080 &chroot_allow_open_directories, 0, 4081 "Allow a process to chroot(2) if it has a directory open"); 4082 4083 /* 4084 * Helper function for raised chroot(2) security function: Refuse if 4085 * any filedescriptors are open directories. 4086 */ 4087 static int 4088 chroot_refuse_vdir_fds(struct filedesc *fdp) 4089 { 4090 struct vnode *vp; 4091 struct file *fp; 4092 int i; 4093 4094 FILEDESC_LOCK_ASSERT(fdp); 4095 4096 FILEDESC_FOREACH_FP(fdp, i, fp) { 4097 if (fp->f_type == DTYPE_VNODE) { 4098 vp = fp->f_vnode; 4099 if (vp->v_type == VDIR) 4100 return (EPERM); 4101 } 4102 } 4103 return (0); 4104 } 4105 4106 static void 4107 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 4108 { 4109 4110 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 4111 vrefact(oldpwd->pwd_cdir); 4112 newpwd->pwd_cdir = oldpwd->pwd_cdir; 4113 } 4114 4115 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 4116 vrefact(oldpwd->pwd_rdir); 4117 newpwd->pwd_rdir = oldpwd->pwd_rdir; 4118 } 4119 4120 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 4121 vrefact(oldpwd->pwd_jdir); 4122 newpwd->pwd_jdir = oldpwd->pwd_jdir; 4123 } 4124 4125 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) { 4126 vrefact(oldpwd->pwd_adir); 4127 newpwd->pwd_adir = oldpwd->pwd_adir; 4128 } 4129 } 4130 4131 struct pwd * 4132 pwd_hold_pwddesc(struct pwddesc *pdp) 4133 { 4134 struct pwd *pwd; 4135 4136 PWDDESC_ASSERT_XLOCKED(pdp); 4137 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4138 if (pwd != NULL) 4139 refcount_acquire(&pwd->pwd_refcount); 4140 return (pwd); 4141 } 4142 4143 bool 4144 pwd_hold_smr(struct pwd *pwd) 4145 { 4146 4147 MPASS(pwd != NULL); 4148 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 4149 return (true); 4150 } 4151 return (false); 4152 } 4153 4154 struct pwd * 4155 pwd_hold(struct thread *td) 4156 { 4157 struct pwddesc *pdp; 4158 struct pwd *pwd; 4159 4160 pdp = td->td_proc->p_pd; 4161 4162 vfs_smr_enter(); 4163 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 4164 if (pwd_hold_smr(pwd)) { 4165 vfs_smr_exit(); 4166 return (pwd); 4167 } 4168 vfs_smr_exit(); 4169 PWDDESC_XLOCK(pdp); 4170 pwd = pwd_hold_pwddesc(pdp); 4171 MPASS(pwd != NULL); 4172 PWDDESC_XUNLOCK(pdp); 4173 return (pwd); 4174 } 4175 4176 struct pwd * 4177 pwd_hold_proc(struct proc *p) 4178 { 4179 struct pwddesc *pdp; 4180 struct pwd *pwd; 4181 4182 PROC_ASSERT_HELD(p); 4183 PROC_LOCK(p); 4184 pdp = pdhold(p); 4185 MPASS(pdp != NULL); 4186 PROC_UNLOCK(p); 4187 4188 PWDDESC_XLOCK(pdp); 4189 pwd = pwd_hold_pwddesc(pdp); 4190 MPASS(pwd != NULL); 4191 PWDDESC_XUNLOCK(pdp); 4192 pddrop(pdp); 4193 return (pwd); 4194 } 4195 4196 static struct pwd * 4197 pwd_alloc(void) 4198 { 4199 struct pwd *pwd; 4200 4201 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 4202 bzero(pwd, sizeof(*pwd)); 4203 refcount_init(&pwd->pwd_refcount, 1); 4204 return (pwd); 4205 } 4206 4207 void 4208 pwd_drop(struct pwd *pwd) 4209 { 4210 4211 if (!refcount_release(&pwd->pwd_refcount)) 4212 return; 4213 4214 if (pwd->pwd_cdir != NULL) 4215 vrele(pwd->pwd_cdir); 4216 if (pwd->pwd_rdir != NULL) 4217 vrele(pwd->pwd_rdir); 4218 if (pwd->pwd_jdir != NULL) 4219 vrele(pwd->pwd_jdir); 4220 if (pwd->pwd_adir != NULL) 4221 vrele(pwd->pwd_adir); 4222 uma_zfree_smr(pwd_zone, pwd); 4223 } 4224 4225 /* 4226 * The caller is responsible for invoking priv_check() and 4227 * mac_vnode_check_chroot() to authorize this operation. 4228 */ 4229 int 4230 pwd_chroot(struct thread *td, struct vnode *vp) 4231 { 4232 struct pwddesc *pdp; 4233 struct filedesc *fdp; 4234 struct pwd *newpwd, *oldpwd; 4235 int error; 4236 4237 fdp = td->td_proc->p_fd; 4238 pdp = td->td_proc->p_pd; 4239 newpwd = pwd_alloc(); 4240 FILEDESC_SLOCK(fdp); 4241 PWDDESC_XLOCK(pdp); 4242 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4243 if (chroot_allow_open_directories == 0 || 4244 (chroot_allow_open_directories == 1 && 4245 oldpwd->pwd_rdir != rootvnode)) { 4246 error = chroot_refuse_vdir_fds(fdp); 4247 FILEDESC_SUNLOCK(fdp); 4248 if (error != 0) { 4249 PWDDESC_XUNLOCK(pdp); 4250 pwd_drop(newpwd); 4251 return (error); 4252 } 4253 } else { 4254 FILEDESC_SUNLOCK(fdp); 4255 } 4256 4257 vrefact(vp); 4258 newpwd->pwd_rdir = vp; 4259 vrefact(vp); 4260 newpwd->pwd_adir = vp; 4261 if (oldpwd->pwd_jdir == NULL) { 4262 vrefact(vp); 4263 newpwd->pwd_jdir = vp; 4264 } 4265 pwd_fill(oldpwd, newpwd); 4266 pwd_set(pdp, newpwd); 4267 PWDDESC_XUNLOCK(pdp); 4268 pwd_drop(oldpwd); 4269 return (0); 4270 } 4271 4272 void 4273 pwd_chdir(struct thread *td, struct vnode *vp) 4274 { 4275 struct pwddesc *pdp; 4276 struct pwd *newpwd, *oldpwd; 4277 4278 VNPASS(vp->v_usecount > 0, vp); 4279 4280 newpwd = pwd_alloc(); 4281 pdp = td->td_proc->p_pd; 4282 PWDDESC_XLOCK(pdp); 4283 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4284 newpwd->pwd_cdir = vp; 4285 pwd_fill(oldpwd, newpwd); 4286 pwd_set(pdp, newpwd); 4287 PWDDESC_XUNLOCK(pdp); 4288 pwd_drop(oldpwd); 4289 } 4290 4291 /* 4292 * Process is transitioning to/from a non-native ABI. 4293 */ 4294 void 4295 pwd_altroot(struct thread *td, struct vnode *altroot_vp) 4296 { 4297 struct pwddesc *pdp; 4298 struct pwd *newpwd, *oldpwd; 4299 4300 newpwd = pwd_alloc(); 4301 pdp = td->td_proc->p_pd; 4302 PWDDESC_XLOCK(pdp); 4303 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4304 if (altroot_vp != NULL) { 4305 /* 4306 * Native process to a non-native ABI. 4307 */ 4308 4309 vrefact(altroot_vp); 4310 newpwd->pwd_adir = altroot_vp; 4311 } else { 4312 /* 4313 * Non-native process to the native ABI. 4314 */ 4315 4316 vrefact(oldpwd->pwd_rdir); 4317 newpwd->pwd_adir = oldpwd->pwd_rdir; 4318 } 4319 pwd_fill(oldpwd, newpwd); 4320 pwd_set(pdp, newpwd); 4321 PWDDESC_XUNLOCK(pdp); 4322 pwd_drop(oldpwd); 4323 } 4324 4325 /* 4326 * jail_attach(2) changes both root and working directories. 4327 */ 4328 int 4329 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 4330 { 4331 struct pwddesc *pdp; 4332 struct filedesc *fdp; 4333 struct pwd *newpwd, *oldpwd; 4334 int error; 4335 4336 fdp = td->td_proc->p_fd; 4337 pdp = td->td_proc->p_pd; 4338 newpwd = pwd_alloc(); 4339 FILEDESC_SLOCK(fdp); 4340 PWDDESC_XLOCK(pdp); 4341 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4342 error = chroot_refuse_vdir_fds(fdp); 4343 FILEDESC_SUNLOCK(fdp); 4344 if (error != 0) { 4345 PWDDESC_XUNLOCK(pdp); 4346 pwd_drop(newpwd); 4347 return (error); 4348 } 4349 4350 vrefact(vp); 4351 newpwd->pwd_rdir = vp; 4352 vrefact(vp); 4353 newpwd->pwd_cdir = vp; 4354 if (oldpwd->pwd_jdir == NULL) { 4355 vrefact(vp); 4356 newpwd->pwd_jdir = vp; 4357 } 4358 vrefact(vp); 4359 newpwd->pwd_adir = vp; 4360 pwd_fill(oldpwd, newpwd); 4361 pwd_set(pdp, newpwd); 4362 PWDDESC_XUNLOCK(pdp); 4363 pwd_drop(oldpwd); 4364 return (0); 4365 } 4366 4367 void 4368 pwd_ensure_dirs(void) 4369 { 4370 struct pwddesc *pdp; 4371 struct pwd *oldpwd, *newpwd; 4372 4373 pdp = curproc->p_pd; 4374 PWDDESC_XLOCK(pdp); 4375 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4376 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL && 4377 oldpwd->pwd_adir != NULL) { 4378 PWDDESC_XUNLOCK(pdp); 4379 return; 4380 } 4381 PWDDESC_XUNLOCK(pdp); 4382 4383 newpwd = pwd_alloc(); 4384 PWDDESC_XLOCK(pdp); 4385 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4386 pwd_fill(oldpwd, newpwd); 4387 if (newpwd->pwd_cdir == NULL) { 4388 vrefact(rootvnode); 4389 newpwd->pwd_cdir = rootvnode; 4390 } 4391 if (newpwd->pwd_rdir == NULL) { 4392 vrefact(rootvnode); 4393 newpwd->pwd_rdir = rootvnode; 4394 } 4395 if (newpwd->pwd_adir == NULL) { 4396 vrefact(rootvnode); 4397 newpwd->pwd_adir = rootvnode; 4398 } 4399 pwd_set(pdp, newpwd); 4400 PWDDESC_XUNLOCK(pdp); 4401 pwd_drop(oldpwd); 4402 } 4403 4404 void 4405 pwd_set_rootvnode(void) 4406 { 4407 struct pwddesc *pdp; 4408 struct pwd *oldpwd, *newpwd; 4409 4410 pdp = curproc->p_pd; 4411 4412 newpwd = pwd_alloc(); 4413 PWDDESC_XLOCK(pdp); 4414 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4415 vrefact(rootvnode); 4416 newpwd->pwd_cdir = rootvnode; 4417 vrefact(rootvnode); 4418 newpwd->pwd_rdir = rootvnode; 4419 vrefact(rootvnode); 4420 newpwd->pwd_adir = rootvnode; 4421 pwd_fill(oldpwd, newpwd); 4422 pwd_set(pdp, newpwd); 4423 PWDDESC_XUNLOCK(pdp); 4424 pwd_drop(oldpwd); 4425 } 4426 4427 /* 4428 * Scan all active processes and prisons to see if any of them have a current 4429 * or root directory of `olddp'. If so, replace them with the new mount point. 4430 */ 4431 void 4432 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 4433 { 4434 struct pwddesc *pdp; 4435 struct pwd *newpwd, *oldpwd; 4436 struct prison *pr; 4437 struct proc *p; 4438 int nrele; 4439 4440 if (vrefcnt(olddp) == 1) 4441 return; 4442 nrele = 0; 4443 newpwd = pwd_alloc(); 4444 sx_slock(&allproc_lock); 4445 FOREACH_PROC_IN_SYSTEM(p) { 4446 PROC_LOCK(p); 4447 pdp = pdhold(p); 4448 PROC_UNLOCK(p); 4449 if (pdp == NULL) 4450 continue; 4451 PWDDESC_XLOCK(pdp); 4452 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4453 if (oldpwd == NULL || 4454 (oldpwd->pwd_cdir != olddp && 4455 oldpwd->pwd_rdir != olddp && 4456 oldpwd->pwd_jdir != olddp && 4457 oldpwd->pwd_adir != olddp)) { 4458 PWDDESC_XUNLOCK(pdp); 4459 pddrop(pdp); 4460 continue; 4461 } 4462 if (oldpwd->pwd_cdir == olddp) { 4463 vrefact(newdp); 4464 newpwd->pwd_cdir = newdp; 4465 } 4466 if (oldpwd->pwd_rdir == olddp) { 4467 vrefact(newdp); 4468 newpwd->pwd_rdir = newdp; 4469 } 4470 if (oldpwd->pwd_jdir == olddp) { 4471 vrefact(newdp); 4472 newpwd->pwd_jdir = newdp; 4473 } 4474 if (oldpwd->pwd_adir == olddp) { 4475 vrefact(newdp); 4476 newpwd->pwd_adir = newdp; 4477 } 4478 pwd_fill(oldpwd, newpwd); 4479 pwd_set(pdp, newpwd); 4480 PWDDESC_XUNLOCK(pdp); 4481 pwd_drop(oldpwd); 4482 pddrop(pdp); 4483 newpwd = pwd_alloc(); 4484 } 4485 sx_sunlock(&allproc_lock); 4486 pwd_drop(newpwd); 4487 if (rootvnode == olddp) { 4488 vrefact(newdp); 4489 rootvnode = newdp; 4490 nrele++; 4491 } 4492 mtx_lock(&prison0.pr_mtx); 4493 if (prison0.pr_root == olddp) { 4494 vrefact(newdp); 4495 prison0.pr_root = newdp; 4496 nrele++; 4497 } 4498 mtx_unlock(&prison0.pr_mtx); 4499 sx_slock(&allprison_lock); 4500 TAILQ_FOREACH(pr, &allprison, pr_list) { 4501 mtx_lock(&pr->pr_mtx); 4502 if (pr->pr_root == olddp) { 4503 vrefact(newdp); 4504 pr->pr_root = newdp; 4505 nrele++; 4506 } 4507 mtx_unlock(&pr->pr_mtx); 4508 } 4509 sx_sunlock(&allprison_lock); 4510 while (nrele--) 4511 vrele(olddp); 4512 } 4513 4514 int 4515 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp) 4516 { 4517 struct file *fp; 4518 struct vnode *vp; 4519 int error, i; 4520 4521 error = 0; 4522 FILEDESC_SLOCK(fdp); 4523 FILEDESC_FOREACH_FP(fdp, i, fp) { 4524 if (fp->f_type != DTYPE_VNODE || 4525 (atomic_load_int(&fp->f_flag) & FWRITE) == 0) 4526 continue; 4527 vp = fp->f_vnode; 4528 if (vp->v_mount == mp) { 4529 error = EDEADLK; 4530 break; 4531 } 4532 } 4533 FILEDESC_SUNLOCK(fdp); 4534 return (error); 4535 } 4536 4537 struct filedesc_to_leader * 4538 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, 4539 struct proc *leader) 4540 { 4541 struct filedesc_to_leader *fdtol; 4542 4543 fdtol = malloc(sizeof(struct filedesc_to_leader), 4544 M_FILEDESC_TO_LEADER, M_WAITOK); 4545 fdtol->fdl_refcount = 1; 4546 fdtol->fdl_holdcount = 0; 4547 fdtol->fdl_wakeup = 0; 4548 fdtol->fdl_leader = leader; 4549 if (old != NULL) { 4550 FILEDESC_XLOCK(fdp); 4551 fdtol->fdl_next = old->fdl_next; 4552 fdtol->fdl_prev = old; 4553 old->fdl_next = fdtol; 4554 fdtol->fdl_next->fdl_prev = fdtol; 4555 FILEDESC_XUNLOCK(fdp); 4556 } else { 4557 fdtol->fdl_next = fdtol; 4558 fdtol->fdl_prev = fdtol; 4559 } 4560 return (fdtol); 4561 } 4562 4563 struct filedesc_to_leader * 4564 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp) 4565 { 4566 FILEDESC_XLOCK(fdp); 4567 fdtol->fdl_refcount++; 4568 FILEDESC_XUNLOCK(fdp); 4569 return (fdtol); 4570 } 4571 4572 static int 4573 filedesc_nfiles(struct filedesc *fdp) 4574 { 4575 NDSLOTTYPE *map; 4576 int count, off, minoff; 4577 4578 if (fdp == NULL) 4579 return (0); 4580 count = 0; 4581 FILEDESC_SLOCK(fdp); 4582 map = fdp->fd_map; 4583 off = NDSLOT(fdp->fd_nfiles - 1); 4584 for (minoff = NDSLOT(0); off >= minoff; --off) 4585 count += bitcountl(map[off]); 4586 FILEDESC_SUNLOCK(fdp); 4587 return (count); 4588 } 4589 4590 int 4591 proc_nfiles(struct proc *p) 4592 { 4593 struct filedesc *fdp; 4594 int res; 4595 4596 PROC_LOCK(p); 4597 fdp = fdhold(p); 4598 PROC_UNLOCK(p); 4599 res = filedesc_nfiles(fdp); 4600 fddrop(fdp); 4601 return (res); 4602 } 4603 4604 static int 4605 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4606 { 4607 u_int namelen; 4608 int count; 4609 4610 namelen = arg2; 4611 if (namelen != 1) 4612 return (EINVAL); 4613 4614 if (*(int *)arg1 != 0) 4615 return (EINVAL); 4616 4617 count = filedesc_nfiles(curproc->p_fd); 4618 return (SYSCTL_OUT(req, &count, sizeof(count))); 4619 } 4620 4621 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4622 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4623 "Number of open file descriptors"); 4624 4625 /* 4626 * Get file structures globally. 4627 */ 4628 static int 4629 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4630 { 4631 struct xfile xf; 4632 struct filedesc *fdp; 4633 struct file *fp; 4634 struct proc *p; 4635 int error, n; 4636 4637 error = sysctl_wire_old_buffer(req, 0); 4638 if (error != 0) 4639 return (error); 4640 if (req->oldptr == NULL) { 4641 n = 0; 4642 sx_slock(&allproc_lock); 4643 FOREACH_PROC_IN_SYSTEM(p) { 4644 PROC_LOCK(p); 4645 if (p->p_state == PRS_NEW) { 4646 PROC_UNLOCK(p); 4647 continue; 4648 } 4649 fdp = fdhold(p); 4650 PROC_UNLOCK(p); 4651 if (fdp == NULL) 4652 continue; 4653 /* overestimates sparse tables. */ 4654 n += fdp->fd_nfiles; 4655 fddrop(fdp); 4656 } 4657 sx_sunlock(&allproc_lock); 4658 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4659 } 4660 error = 0; 4661 bzero(&xf, sizeof(xf)); 4662 xf.xf_size = sizeof(xf); 4663 sx_slock(&allproc_lock); 4664 FOREACH_PROC_IN_SYSTEM(p) { 4665 PROC_LOCK(p); 4666 if (p->p_state == PRS_NEW) { 4667 PROC_UNLOCK(p); 4668 continue; 4669 } 4670 if (p_cansee(req->td, p) != 0) { 4671 PROC_UNLOCK(p); 4672 continue; 4673 } 4674 xf.xf_pid = p->p_pid; 4675 xf.xf_uid = p->p_ucred->cr_uid; 4676 fdp = fdhold(p); 4677 PROC_UNLOCK(p); 4678 if (fdp == NULL) 4679 continue; 4680 FILEDESC_SLOCK(fdp); 4681 if (refcount_load(&fdp->fd_refcnt) == 0) 4682 goto nextproc; 4683 FILEDESC_FOREACH_FP(fdp, n, fp) { 4684 xf.xf_fd = n; 4685 xf.xf_file = (uintptr_t)fp; 4686 xf.xf_data = (uintptr_t)fp->f_data; 4687 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4688 xf.xf_type = (uintptr_t)fp->f_type; 4689 xf.xf_count = refcount_load(&fp->f_count); 4690 xf.xf_msgcount = 0; 4691 xf.xf_offset = foffset_get(fp); 4692 xf.xf_flag = fp->f_flag; 4693 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4694 4695 /* 4696 * There is no need to re-check the fdtable refcount 4697 * here since the filedesc lock is not dropped in the 4698 * loop body. 4699 */ 4700 if (error != 0) 4701 break; 4702 } 4703 nextproc: 4704 FILEDESC_SUNLOCK(fdp); 4705 fddrop(fdp); 4706 if (error) 4707 break; 4708 } 4709 sx_sunlock(&allproc_lock); 4710 return (error); 4711 } 4712 4713 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4714 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4715 4716 #ifdef KINFO_FILE_SIZE 4717 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4718 #endif 4719 4720 static int 4721 xlate_fflags(int fflags) 4722 { 4723 static const struct { 4724 int fflag; 4725 int kf_fflag; 4726 } fflags_table[] = { 4727 { FAPPEND, KF_FLAG_APPEND }, 4728 { FASYNC, KF_FLAG_ASYNC }, 4729 { FFSYNC, KF_FLAG_FSYNC }, 4730 { FHASLOCK, KF_FLAG_HASLOCK }, 4731 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4732 { FREAD, KF_FLAG_READ }, 4733 { FWRITE, KF_FLAG_WRITE }, 4734 { O_CREAT, KF_FLAG_CREAT }, 4735 { O_DIRECT, KF_FLAG_DIRECT }, 4736 { O_EXCL, KF_FLAG_EXCL }, 4737 { O_EXEC, KF_FLAG_EXEC }, 4738 { O_EXLOCK, KF_FLAG_EXLOCK }, 4739 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4740 { O_SHLOCK, KF_FLAG_SHLOCK }, 4741 { O_TRUNC, KF_FLAG_TRUNC } 4742 }; 4743 unsigned int i; 4744 int kflags; 4745 4746 kflags = 0; 4747 for (i = 0; i < nitems(fflags_table); i++) 4748 if (fflags & fflags_table[i].fflag) 4749 kflags |= fflags_table[i].kf_fflag; 4750 return (kflags); 4751 } 4752 4753 /* Trim unused data from kf_path by truncating the structure size. */ 4754 void 4755 pack_kinfo(struct kinfo_file *kif) 4756 { 4757 4758 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4759 strlen(kif->kf_path) + 1; 4760 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4761 } 4762 4763 static void 4764 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4765 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4766 { 4767 int error; 4768 4769 bzero(kif, sizeof(*kif)); 4770 4771 /* Set a default type to allow for empty fill_kinfo() methods. */ 4772 kif->kf_type = KF_TYPE_UNKNOWN; 4773 kif->kf_flags = xlate_fflags(fp->f_flag); 4774 if (rightsp != NULL) 4775 kif->kf_cap_rights = *rightsp; 4776 else 4777 cap_rights_init_zero(&kif->kf_cap_rights); 4778 kif->kf_fd = fd; 4779 kif->kf_ref_count = refcount_load(&fp->f_count); 4780 kif->kf_offset = foffset_get(fp); 4781 4782 /* 4783 * This may drop the filedesc lock, so the 'fp' cannot be 4784 * accessed after this call. 4785 */ 4786 error = fo_fill_kinfo(fp, kif, fdp); 4787 if (error == 0) 4788 kif->kf_status |= KF_ATTR_VALID; 4789 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4790 pack_kinfo(kif); 4791 else 4792 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4793 } 4794 4795 static void 4796 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4797 struct kinfo_file *kif, int flags) 4798 { 4799 int error; 4800 4801 bzero(kif, sizeof(*kif)); 4802 4803 kif->kf_type = KF_TYPE_VNODE; 4804 error = vn_fill_kinfo_vnode(vp, kif); 4805 if (error == 0) 4806 kif->kf_status |= KF_ATTR_VALID; 4807 kif->kf_flags = xlate_fflags(fflags); 4808 cap_rights_init_zero(&kif->kf_cap_rights); 4809 kif->kf_fd = fd; 4810 kif->kf_ref_count = -1; 4811 kif->kf_offset = -1; 4812 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4813 pack_kinfo(kif); 4814 else 4815 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4816 vrele(vp); 4817 } 4818 4819 struct export_fd_buf { 4820 struct filedesc *fdp; 4821 struct pwddesc *pdp; 4822 struct sbuf *sb; 4823 ssize_t remainder; 4824 struct kinfo_file kif; 4825 int flags; 4826 }; 4827 4828 static int 4829 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4830 { 4831 struct kinfo_file *kif; 4832 4833 kif = &efbuf->kif; 4834 if (efbuf->remainder != -1) { 4835 if (efbuf->remainder < kif->kf_structsize) 4836 return (ENOMEM); 4837 efbuf->remainder -= kif->kf_structsize; 4838 } 4839 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0) 4840 return (sbuf_error(efbuf->sb)); 4841 return (0); 4842 } 4843 4844 static int 4845 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4846 struct export_fd_buf *efbuf) 4847 { 4848 int error; 4849 4850 if (efbuf->remainder == 0) 4851 return (ENOMEM); 4852 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4853 efbuf->flags); 4854 FILEDESC_SUNLOCK(efbuf->fdp); 4855 error = export_kinfo_to_sb(efbuf); 4856 FILEDESC_SLOCK(efbuf->fdp); 4857 return (error); 4858 } 4859 4860 static int 4861 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4862 struct export_fd_buf *efbuf) 4863 { 4864 int error; 4865 4866 if (efbuf->remainder == 0) 4867 return (ENOMEM); 4868 if (efbuf->pdp != NULL) 4869 PWDDESC_XUNLOCK(efbuf->pdp); 4870 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4871 error = export_kinfo_to_sb(efbuf); 4872 if (efbuf->pdp != NULL) 4873 PWDDESC_XLOCK(efbuf->pdp); 4874 return (error); 4875 } 4876 4877 /* 4878 * Store a process file descriptor information to sbuf. 4879 * 4880 * Takes a locked proc as argument, and returns with the proc unlocked. 4881 */ 4882 int 4883 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4884 int flags) 4885 { 4886 struct file *fp; 4887 struct filedesc *fdp; 4888 struct pwddesc *pdp; 4889 struct export_fd_buf *efbuf; 4890 struct vnode *cttyvp, *textvp, *tracevp; 4891 struct pwd *pwd; 4892 int error, i; 4893 cap_rights_t rights; 4894 4895 PROC_LOCK_ASSERT(p, MA_OWNED); 4896 4897 /* ktrace vnode */ 4898 tracevp = ktr_get_tracevp(p, true); 4899 /* text vnode */ 4900 textvp = p->p_textvp; 4901 if (textvp != NULL) 4902 vrefact(textvp); 4903 /* Controlling tty. */ 4904 cttyvp = NULL; 4905 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 4906 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 4907 if (cttyvp != NULL) 4908 vrefact(cttyvp); 4909 } 4910 fdp = fdhold(p); 4911 pdp = pdhold(p); 4912 PROC_UNLOCK(p); 4913 4914 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4915 efbuf->fdp = NULL; 4916 efbuf->pdp = NULL; 4917 efbuf->sb = sb; 4918 efbuf->remainder = maxlen; 4919 efbuf->flags = flags; 4920 4921 error = 0; 4922 if (tracevp != NULL) 4923 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, 4924 FREAD | FWRITE, efbuf); 4925 if (error == 0 && textvp != NULL) 4926 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, 4927 efbuf); 4928 if (error == 0 && cttyvp != NULL) 4929 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, 4930 FREAD | FWRITE, efbuf); 4931 if (error != 0 || pdp == NULL || fdp == NULL) 4932 goto fail; 4933 efbuf->fdp = fdp; 4934 efbuf->pdp = pdp; 4935 PWDDESC_XLOCK(pdp); 4936 pwd = pwd_hold_pwddesc(pdp); 4937 if (pwd != NULL) { 4938 /* working directory */ 4939 if (pwd->pwd_cdir != NULL) { 4940 vrefact(pwd->pwd_cdir); 4941 error = export_vnode_to_sb(pwd->pwd_cdir, 4942 KF_FD_TYPE_CWD, FREAD, efbuf); 4943 } 4944 /* root directory */ 4945 if (error == 0 && pwd->pwd_rdir != NULL) { 4946 vrefact(pwd->pwd_rdir); 4947 error = export_vnode_to_sb(pwd->pwd_rdir, 4948 KF_FD_TYPE_ROOT, FREAD, efbuf); 4949 } 4950 /* jail directory */ 4951 if (error == 0 && pwd->pwd_jdir != NULL) { 4952 vrefact(pwd->pwd_jdir); 4953 error = export_vnode_to_sb(pwd->pwd_jdir, 4954 KF_FD_TYPE_JAIL, FREAD, efbuf); 4955 } 4956 } 4957 PWDDESC_XUNLOCK(pdp); 4958 if (error != 0) 4959 goto fail; 4960 if (pwd != NULL) 4961 pwd_drop(pwd); 4962 FILEDESC_SLOCK(fdp); 4963 if (refcount_load(&fdp->fd_refcnt) == 0) 4964 goto skip; 4965 FILEDESC_FOREACH_FP(fdp, i, fp) { 4966 #ifdef CAPABILITIES 4967 rights = *cap_rights(fdp, i); 4968 #else /* !CAPABILITIES */ 4969 rights = cap_no_rights; 4970 #endif 4971 /* 4972 * Create sysctl entry. It is OK to drop the filedesc 4973 * lock inside of export_file_to_sb() as we will 4974 * re-validate and re-evaluate its properties when the 4975 * loop continues. 4976 */ 4977 error = export_file_to_sb(fp, i, &rights, efbuf); 4978 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4979 break; 4980 } 4981 skip: 4982 FILEDESC_SUNLOCK(fdp); 4983 fail: 4984 if (fdp != NULL) 4985 fddrop(fdp); 4986 if (pdp != NULL) 4987 pddrop(pdp); 4988 free(efbuf, M_TEMP); 4989 return (error); 4990 } 4991 4992 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 4993 4994 /* 4995 * Get per-process file descriptors for use by procstat(1), et al. 4996 */ 4997 static int 4998 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 4999 { 5000 struct sbuf sb; 5001 struct proc *p; 5002 ssize_t maxlen; 5003 u_int namelen; 5004 int error, error2, *name; 5005 5006 namelen = arg2; 5007 if (namelen != 1) 5008 return (EINVAL); 5009 5010 name = (int *)arg1; 5011 5012 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 5013 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5014 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5015 if (error != 0) { 5016 sbuf_delete(&sb); 5017 return (error); 5018 } 5019 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5020 error = kern_proc_filedesc_out(p, &sb, maxlen, 5021 KERN_FILEDESC_PACK_KINFO); 5022 error2 = sbuf_finish(&sb); 5023 sbuf_delete(&sb); 5024 return (error != 0 ? error : error2); 5025 } 5026 5027 #ifdef COMPAT_FREEBSD7 5028 #ifdef KINFO_OFILE_SIZE 5029 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 5030 #endif 5031 5032 static void 5033 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 5034 { 5035 5036 okif->kf_structsize = sizeof(*okif); 5037 okif->kf_type = kif->kf_type; 5038 okif->kf_fd = kif->kf_fd; 5039 okif->kf_ref_count = kif->kf_ref_count; 5040 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 5041 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 5042 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 5043 okif->kf_offset = kif->kf_offset; 5044 if (kif->kf_type == KF_TYPE_VNODE) 5045 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 5046 else 5047 okif->kf_vnode_type = KF_VTYPE_VNON; 5048 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 5049 if (kif->kf_type == KF_TYPE_SOCKET) { 5050 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 5051 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 5052 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 5053 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 5054 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 5055 } else { 5056 okif->kf_sa_local.ss_family = AF_UNSPEC; 5057 okif->kf_sa_peer.ss_family = AF_UNSPEC; 5058 } 5059 } 5060 5061 static int 5062 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 5063 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 5064 { 5065 int error; 5066 5067 vrefact(vp); 5068 PWDDESC_XUNLOCK(pdp); 5069 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 5070 kinfo_to_okinfo(kif, okif); 5071 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5072 PWDDESC_XLOCK(pdp); 5073 return (error); 5074 } 5075 5076 /* 5077 * Get per-process file descriptors for use by procstat(1), et al. 5078 */ 5079 static int 5080 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 5081 { 5082 struct kinfo_ofile *okif; 5083 struct kinfo_file *kif; 5084 struct filedesc *fdp; 5085 struct pwddesc *pdp; 5086 struct pwd *pwd; 5087 u_int namelen; 5088 int error, i, *name; 5089 struct file *fp; 5090 struct proc *p; 5091 5092 namelen = arg2; 5093 if (namelen != 1) 5094 return (EINVAL); 5095 5096 name = (int *)arg1; 5097 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5098 if (error != 0) 5099 return (error); 5100 fdp = fdhold(p); 5101 if (fdp != NULL) 5102 pdp = pdhold(p); 5103 PROC_UNLOCK(p); 5104 if (fdp == NULL || pdp == NULL) { 5105 if (fdp != NULL) 5106 fddrop(fdp); 5107 return (ENOENT); 5108 } 5109 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 5110 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 5111 PWDDESC_XLOCK(pdp); 5112 pwd = pwd_hold_pwddesc(pdp); 5113 if (pwd != NULL) { 5114 if (pwd->pwd_cdir != NULL) 5115 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 5116 okif, pdp, req); 5117 if (pwd->pwd_rdir != NULL) 5118 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 5119 okif, pdp, req); 5120 if (pwd->pwd_jdir != NULL) 5121 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 5122 okif, pdp, req); 5123 } 5124 PWDDESC_XUNLOCK(pdp); 5125 if (pwd != NULL) 5126 pwd_drop(pwd); 5127 FILEDESC_SLOCK(fdp); 5128 if (refcount_load(&fdp->fd_refcnt) == 0) 5129 goto skip; 5130 FILEDESC_FOREACH_FP(fdp, i, fp) { 5131 export_file_to_kinfo(fp, i, NULL, kif, fdp, 5132 KERN_FILEDESC_PACK_KINFO); 5133 FILEDESC_SUNLOCK(fdp); 5134 kinfo_to_okinfo(kif, okif); 5135 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 5136 FILEDESC_SLOCK(fdp); 5137 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 5138 break; 5139 } 5140 skip: 5141 FILEDESC_SUNLOCK(fdp); 5142 fddrop(fdp); 5143 pddrop(pdp); 5144 free(kif, M_TEMP); 5145 free(okif, M_TEMP); 5146 return (0); 5147 } 5148 5149 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 5150 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 5151 "Process ofiledesc entries"); 5152 #endif /* COMPAT_FREEBSD7 */ 5153 5154 int 5155 vntype_to_kinfo(int vtype) 5156 { 5157 struct { 5158 int vtype; 5159 int kf_vtype; 5160 } vtypes_table[] = { 5161 { VBAD, KF_VTYPE_VBAD }, 5162 { VBLK, KF_VTYPE_VBLK }, 5163 { VCHR, KF_VTYPE_VCHR }, 5164 { VDIR, KF_VTYPE_VDIR }, 5165 { VFIFO, KF_VTYPE_VFIFO }, 5166 { VLNK, KF_VTYPE_VLNK }, 5167 { VNON, KF_VTYPE_VNON }, 5168 { VREG, KF_VTYPE_VREG }, 5169 { VSOCK, KF_VTYPE_VSOCK } 5170 }; 5171 unsigned int i; 5172 5173 /* 5174 * Perform vtype translation. 5175 */ 5176 for (i = 0; i < nitems(vtypes_table); i++) 5177 if (vtypes_table[i].vtype == vtype) 5178 return (vtypes_table[i].kf_vtype); 5179 5180 return (KF_VTYPE_UNKNOWN); 5181 } 5182 5183 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 5184 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 5185 "Process filedesc entries"); 5186 5187 /* 5188 * Store a process current working directory information to sbuf. 5189 * 5190 * Takes a locked proc as argument, and returns with the proc unlocked. 5191 */ 5192 int 5193 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 5194 { 5195 struct pwddesc *pdp; 5196 struct pwd *pwd; 5197 struct export_fd_buf *efbuf; 5198 struct vnode *cdir; 5199 int error; 5200 5201 PROC_LOCK_ASSERT(p, MA_OWNED); 5202 5203 pdp = pdhold(p); 5204 PROC_UNLOCK(p); 5205 if (pdp == NULL) 5206 return (EINVAL); 5207 5208 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 5209 efbuf->fdp = NULL; 5210 efbuf->pdp = pdp; 5211 efbuf->sb = sb; 5212 efbuf->remainder = maxlen; 5213 efbuf->flags = 0; 5214 5215 PWDDESC_XLOCK(pdp); 5216 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 5217 cdir = pwd->pwd_cdir; 5218 if (cdir == NULL) { 5219 error = EINVAL; 5220 } else { 5221 vrefact(cdir); 5222 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 5223 } 5224 PWDDESC_XUNLOCK(pdp); 5225 pddrop(pdp); 5226 free(efbuf, M_TEMP); 5227 return (error); 5228 } 5229 5230 /* 5231 * Get per-process current working directory. 5232 */ 5233 static int 5234 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 5235 { 5236 struct sbuf sb; 5237 struct proc *p; 5238 ssize_t maxlen; 5239 u_int namelen; 5240 int error, error2, *name; 5241 5242 namelen = arg2; 5243 if (namelen != 1) 5244 return (EINVAL); 5245 5246 name = (int *)arg1; 5247 5248 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 5249 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 5250 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 5251 if (error != 0) { 5252 sbuf_delete(&sb); 5253 return (error); 5254 } 5255 maxlen = req->oldptr != NULL ? req->oldlen : -1; 5256 error = kern_proc_cwd_out(p, &sb, maxlen); 5257 error2 = sbuf_finish(&sb); 5258 sbuf_delete(&sb); 5259 return (error != 0 ? error : error2); 5260 } 5261 5262 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 5263 sysctl_kern_proc_cwd, "Process current working directory"); 5264 5265 #ifdef DDB 5266 /* 5267 * For the purposes of debugging, generate a human-readable string for the 5268 * file type. 5269 */ 5270 static const char * 5271 file_type_to_name(short type) 5272 { 5273 5274 switch (type) { 5275 case 0: 5276 return ("zero"); 5277 case DTYPE_VNODE: 5278 return ("vnode"); 5279 case DTYPE_SOCKET: 5280 return ("socket"); 5281 case DTYPE_PIPE: 5282 return ("pipe"); 5283 case DTYPE_FIFO: 5284 return ("fifo"); 5285 case DTYPE_KQUEUE: 5286 return ("kqueue"); 5287 case DTYPE_CRYPTO: 5288 return ("crypto"); 5289 case DTYPE_MQUEUE: 5290 return ("mqueue"); 5291 case DTYPE_SHM: 5292 return ("shm"); 5293 case DTYPE_SEM: 5294 return ("ksem"); 5295 case DTYPE_PTS: 5296 return ("pts"); 5297 case DTYPE_DEV: 5298 return ("dev"); 5299 case DTYPE_PROCDESC: 5300 return ("proc"); 5301 case DTYPE_EVENTFD: 5302 return ("eventfd"); 5303 case DTYPE_TIMERFD: 5304 return ("timerfd"); 5305 case DTYPE_JAILDESC: 5306 return ("jail"); 5307 default: 5308 return ("unkn"); 5309 } 5310 } 5311 5312 /* 5313 * For the purposes of debugging, identify a process (if any, perhaps one of 5314 * many) that references the passed file in its file descriptor array. Return 5315 * NULL if none. 5316 */ 5317 static struct proc * 5318 file_to_first_proc(struct file *fp) 5319 { 5320 struct filedesc *fdp; 5321 struct proc *p; 5322 int n; 5323 5324 FOREACH_PROC_IN_SYSTEM(p) { 5325 if (p->p_state == PRS_NEW) 5326 continue; 5327 fdp = p->p_fd; 5328 if (fdp == NULL) 5329 continue; 5330 for (n = 0; n < fdp->fd_nfiles; n++) { 5331 if (fp == fdp->fd_ofiles[n].fde_file) 5332 return (p); 5333 } 5334 } 5335 return (NULL); 5336 } 5337 5338 static void 5339 db_print_file(struct file *fp, int header) 5340 { 5341 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 5342 struct proc *p; 5343 5344 if (header) 5345 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 5346 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 5347 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 5348 "FCmd"); 5349 p = file_to_first_proc(fp); 5350 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 5351 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 5352 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 5353 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 5354 5355 #undef XPTRWIDTH 5356 } 5357 5358 DB_SHOW_COMMAND(file, db_show_file) 5359 { 5360 struct file *fp; 5361 5362 if (!have_addr) { 5363 db_printf("usage: show file <addr>\n"); 5364 return; 5365 } 5366 fp = (struct file *)addr; 5367 db_print_file(fp, 1); 5368 } 5369 5370 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE) 5371 { 5372 struct filedesc *fdp; 5373 struct file *fp; 5374 struct proc *p; 5375 int header; 5376 int n; 5377 5378 header = 1; 5379 FOREACH_PROC_IN_SYSTEM(p) { 5380 if (p->p_state == PRS_NEW) 5381 continue; 5382 if ((fdp = p->p_fd) == NULL) 5383 continue; 5384 for (n = 0; n < fdp->fd_nfiles; ++n) { 5385 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 5386 continue; 5387 db_print_file(fp, header); 5388 header = 0; 5389 } 5390 } 5391 } 5392 #endif 5393 5394 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, 5395 CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5396 &maxfilesperproc, 0, "Maximum files allowed open per process"); 5397 5398 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, 5399 &maxfiles, 0, "Maximum number of files"); 5400 5401 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 5402 &openfiles, 0, "System-wide number of open files"); 5403 5404 /* ARGSUSED*/ 5405 static void 5406 filelistinit(void *dummy) 5407 { 5408 5409 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 5410 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 5411 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 5412 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 5413 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 5414 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 5415 /* 5416 * XXXMJG this is a temporary hack due to boot ordering issues against 5417 * the vnode zone. 5418 */ 5419 vfs_smr = uma_zone_get_smr(pwd_zone); 5420 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 5421 } 5422 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 5423 5424 /*-------------------------------------------------------------------*/ 5425 5426 static int 5427 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 5428 int flags, struct thread *td) 5429 { 5430 5431 return (EBADF); 5432 } 5433 5434 static int 5435 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5436 struct thread *td) 5437 { 5438 5439 return (EINVAL); 5440 } 5441 5442 static int 5443 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 5444 struct thread *td) 5445 { 5446 5447 return (EBADF); 5448 } 5449 5450 static int 5451 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 5452 struct thread *td) 5453 { 5454 5455 return (0); 5456 } 5457 5458 static int 5459 badfo_kqfilter(struct file *fp, struct knote *kn) 5460 { 5461 5462 return (EBADF); 5463 } 5464 5465 static int 5466 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 5467 { 5468 5469 return (EBADF); 5470 } 5471 5472 static int 5473 badfo_close(struct file *fp, struct thread *td) 5474 { 5475 5476 return (0); 5477 } 5478 5479 static int 5480 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5481 struct thread *td) 5482 { 5483 5484 return (EBADF); 5485 } 5486 5487 static int 5488 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5489 struct thread *td) 5490 { 5491 5492 return (EBADF); 5493 } 5494 5495 static int 5496 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5497 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5498 struct thread *td) 5499 { 5500 5501 return (EBADF); 5502 } 5503 5504 static int 5505 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 5506 { 5507 5508 return (0); 5509 } 5510 5511 const struct fileops badfileops = { 5512 .fo_read = badfo_readwrite, 5513 .fo_write = badfo_readwrite, 5514 .fo_truncate = badfo_truncate, 5515 .fo_ioctl = badfo_ioctl, 5516 .fo_poll = badfo_poll, 5517 .fo_kqfilter = badfo_kqfilter, 5518 .fo_stat = badfo_stat, 5519 .fo_close = badfo_close, 5520 .fo_chmod = badfo_chmod, 5521 .fo_chown = badfo_chown, 5522 .fo_sendfile = badfo_sendfile, 5523 .fo_fill_kinfo = badfo_fill_kinfo, 5524 }; 5525 5526 static int 5527 path_poll(struct file *fp, int events, struct ucred *active_cred, 5528 struct thread *td) 5529 { 5530 return (POLLNVAL); 5531 } 5532 5533 static int 5534 path_close(struct file *fp, struct thread *td) 5535 { 5536 MPASS(fp->f_type == DTYPE_VNODE); 5537 fp->f_ops = &badfileops; 5538 vrele(fp->f_vnode); 5539 return (0); 5540 } 5541 5542 const struct fileops path_fileops = { 5543 .fo_read = badfo_readwrite, 5544 .fo_write = badfo_readwrite, 5545 .fo_truncate = badfo_truncate, 5546 .fo_ioctl = badfo_ioctl, 5547 .fo_poll = path_poll, 5548 .fo_kqfilter = vn_kqfilter_opath, 5549 .fo_stat = vn_statfile, 5550 .fo_close = path_close, 5551 .fo_chmod = badfo_chmod, 5552 .fo_chown = badfo_chown, 5553 .fo_sendfile = badfo_sendfile, 5554 .fo_fill_kinfo = vn_fill_kinfo, 5555 .fo_cmp = vn_cmp, 5556 .fo_flags = DFLAG_PASSABLE, 5557 }; 5558 5559 int 5560 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5561 int flags, struct thread *td) 5562 { 5563 5564 return (EOPNOTSUPP); 5565 } 5566 5567 int 5568 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5569 struct thread *td) 5570 { 5571 5572 return (EINVAL); 5573 } 5574 5575 int 5576 invfo_ioctl(struct file *fp, u_long com, void *data, 5577 struct ucred *active_cred, struct thread *td) 5578 { 5579 5580 return (ENOTTY); 5581 } 5582 5583 int 5584 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5585 struct thread *td) 5586 { 5587 5588 return (poll_no_poll(events)); 5589 } 5590 5591 int 5592 invfo_kqfilter(struct file *fp, struct knote *kn) 5593 { 5594 5595 return (EINVAL); 5596 } 5597 5598 int 5599 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5600 struct thread *td) 5601 { 5602 5603 return (EINVAL); 5604 } 5605 5606 int 5607 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5608 struct thread *td) 5609 { 5610 5611 return (EINVAL); 5612 } 5613 5614 int 5615 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5616 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5617 struct thread *td) 5618 { 5619 5620 return (EINVAL); 5621 } 5622 5623 /*-------------------------------------------------------------------*/ 5624 5625 /* 5626 * File Descriptor pseudo-device driver (/dev/fd/). 5627 * 5628 * Opening minor device N dup()s the file (if any) connected to file 5629 * descriptor N belonging to the calling process. Note that this driver 5630 * consists of only the ``open()'' routine, because all subsequent 5631 * references to this file will be direct to the other driver. 5632 * 5633 * XXX: we could give this one a cloning event handler if necessary. 5634 */ 5635 5636 /* ARGSUSED */ 5637 static int 5638 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5639 { 5640 5641 /* 5642 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5643 * the file descriptor being sought for duplication. The error 5644 * return ensures that the vnode for this device will be released 5645 * by vn_open. Open will detect this special error and take the 5646 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5647 * will simply report the error. 5648 */ 5649 td->td_dupfd = dev2unit(dev); 5650 return (ENODEV); 5651 } 5652 5653 static struct cdevsw fildesc_cdevsw = { 5654 .d_version = D_VERSION, 5655 .d_open = fdopen, 5656 .d_name = "FD", 5657 }; 5658 5659 static void 5660 fildesc_drvinit(void *unused) 5661 { 5662 struct cdev *dev; 5663 5664 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5665 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5666 make_dev_alias(dev, "stdin"); 5667 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5668 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5669 make_dev_alias(dev, "stdout"); 5670 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5671 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5672 make_dev_alias(dev, "stderr"); 5673 } 5674 5675 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5676