1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_capsicum.h" 41 #include "opt_compat.h" 42 #include "opt_ddb.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 48 #include <sys/capsicum.h> 49 #include <sys/conf.h> 50 #include <sys/fcntl.h> 51 #include <sys/file.h> 52 #include <sys/filedesc.h> 53 #include <sys/filio.h> 54 #include <sys/jail.h> 55 #include <sys/kernel.h> 56 #include <sys/limits.h> 57 #include <sys/lock.h> 58 #include <sys/malloc.h> 59 #include <sys/mount.h> 60 #include <sys/mutex.h> 61 #include <sys/namei.h> 62 #include <sys/selinfo.h> 63 #include <sys/priv.h> 64 #include <sys/proc.h> 65 #include <sys/protosw.h> 66 #include <sys/racct.h> 67 #include <sys/resourcevar.h> 68 #include <sys/sbuf.h> 69 #include <sys/signalvar.h> 70 #include <sys/socketvar.h> 71 #include <sys/kdb.h> 72 #include <sys/stat.h> 73 #include <sys/sx.h> 74 #include <sys/syscallsubr.h> 75 #include <sys/sysctl.h> 76 #include <sys/sysproto.h> 77 #include <sys/unistd.h> 78 #include <sys/user.h> 79 #include <sys/vnode.h> 80 #ifdef KTRACE 81 #include <sys/ktrace.h> 82 #endif 83 84 #include <net/vnet.h> 85 86 #include <security/audit/audit.h> 87 88 #include <vm/uma.h> 89 #include <vm/vm.h> 90 91 #include <ddb/ddb.h> 92 93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 95 "file desc to leader structures"); 96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 98 99 MALLOC_DECLARE(M_FADVISE); 100 101 static uma_zone_t file_zone; 102 static uma_zone_t filedesc0_zone; 103 104 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 105 struct thread *td, int holdleaders); 106 static int fd_first_free(struct filedesc *fdp, int low, int size); 107 static int fd_last_used(struct filedesc *fdp, int size); 108 static void fdgrowtable(struct filedesc *fdp, int nfd); 109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 110 static void fdunused(struct filedesc *fdp, int fd); 111 static void fdused(struct filedesc *fdp, int fd); 112 static int getmaxfd(struct thread *td); 113 114 /* 115 * Each process has: 116 * 117 * - An array of open file descriptors (fd_ofiles) 118 * - An array of file flags (fd_ofileflags) 119 * - A bitmap recording which descriptors are in use (fd_map) 120 * 121 * A process starts out with NDFILE descriptors. The value of NDFILE has 122 * been selected based the historical limit of 20 open files, and an 123 * assumption that the majority of processes, especially short-lived 124 * processes like shells, will never need more. 125 * 126 * If this initial allocation is exhausted, a larger descriptor table and 127 * map are allocated dynamically, and the pointers in the process's struct 128 * filedesc are updated to point to those. This is repeated every time 129 * the process runs out of file descriptors (provided it hasn't hit its 130 * resource limit). 131 * 132 * Since threads may hold references to individual descriptor table 133 * entries, the tables are never freed. Instead, they are placed on a 134 * linked list and freed only when the struct filedesc is released. 135 */ 136 #define NDFILE 20 137 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 138 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 139 #define NDSLOT(x) ((x) / NDENTRIES) 140 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 141 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 142 143 /* 144 * SLIST entry used to keep track of ofiles which must be reclaimed when 145 * the process exits. 146 */ 147 struct freetable { 148 struct fdescenttbl *ft_table; 149 SLIST_ENTRY(freetable) ft_next; 150 }; 151 152 /* 153 * Initial allocation: a filedesc structure + the head of SLIST used to 154 * keep track of old ofiles + enough space for NDFILE descriptors. 155 */ 156 157 struct fdescenttbl0 { 158 int fdt_nfiles; 159 struct filedescent fdt_ofiles[NDFILE]; 160 }; 161 162 struct filedesc0 { 163 struct filedesc fd_fd; 164 SLIST_HEAD(, freetable) fd_free; 165 struct fdescenttbl0 fd_dfiles; 166 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 167 }; 168 169 /* 170 * Descriptor management. 171 */ 172 volatile int openfiles; /* actual number of open files */ 173 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 174 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 175 176 /* 177 * If low >= size, just return low. Otherwise find the first zero bit in the 178 * given bitmap, starting at low and not exceeding size - 1. Return size if 179 * not found. 180 */ 181 static int 182 fd_first_free(struct filedesc *fdp, int low, int size) 183 { 184 NDSLOTTYPE *map = fdp->fd_map; 185 NDSLOTTYPE mask; 186 int off, maxoff; 187 188 if (low >= size) 189 return (low); 190 191 off = NDSLOT(low); 192 if (low % NDENTRIES) { 193 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 194 if ((mask &= ~map[off]) != 0UL) 195 return (off * NDENTRIES + ffsl(mask) - 1); 196 ++off; 197 } 198 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 199 if (map[off] != ~0UL) 200 return (off * NDENTRIES + ffsl(~map[off]) - 1); 201 return (size); 202 } 203 204 /* 205 * Find the highest non-zero bit in the given bitmap, starting at 0 and 206 * not exceeding size - 1. Return -1 if not found. 207 */ 208 static int 209 fd_last_used(struct filedesc *fdp, int size) 210 { 211 NDSLOTTYPE *map = fdp->fd_map; 212 NDSLOTTYPE mask; 213 int off, minoff; 214 215 off = NDSLOT(size); 216 if (size % NDENTRIES) { 217 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 218 if ((mask &= map[off]) != 0) 219 return (off * NDENTRIES + flsl(mask) - 1); 220 --off; 221 } 222 for (minoff = NDSLOT(0); off >= minoff; --off) 223 if (map[off] != 0) 224 return (off * NDENTRIES + flsl(map[off]) - 1); 225 return (-1); 226 } 227 228 static int 229 fdisused(struct filedesc *fdp, int fd) 230 { 231 232 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 233 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 234 235 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 236 } 237 238 /* 239 * Mark a file descriptor as used. 240 */ 241 static void 242 fdused_init(struct filedesc *fdp, int fd) 243 { 244 245 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 246 247 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 248 } 249 250 static void 251 fdused(struct filedesc *fdp, int fd) 252 { 253 254 FILEDESC_XLOCK_ASSERT(fdp); 255 256 fdused_init(fdp, fd); 257 if (fd > fdp->fd_lastfile) 258 fdp->fd_lastfile = fd; 259 if (fd == fdp->fd_freefile) 260 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 261 } 262 263 /* 264 * Mark a file descriptor as unused. 265 */ 266 static void 267 fdunused(struct filedesc *fdp, int fd) 268 { 269 270 FILEDESC_XLOCK_ASSERT(fdp); 271 272 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 273 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 274 ("fd=%d is still in use", fd)); 275 276 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 277 if (fd < fdp->fd_freefile) 278 fdp->fd_freefile = fd; 279 if (fd == fdp->fd_lastfile) 280 fdp->fd_lastfile = fd_last_used(fdp, fd); 281 } 282 283 /* 284 * Free a file descriptor. 285 * 286 * Avoid some work if fdp is about to be destroyed. 287 */ 288 static inline void 289 fdefree_last(struct filedescent *fde) 290 { 291 292 filecaps_free(&fde->fde_caps); 293 } 294 295 static inline void 296 fdfree(struct filedesc *fdp, int fd) 297 { 298 struct filedescent *fde; 299 300 fde = &fdp->fd_ofiles[fd]; 301 #ifdef CAPABILITIES 302 seq_write_begin(&fde->fde_seq); 303 #endif 304 fdefree_last(fde); 305 fde->fde_file = NULL; 306 fdunused(fdp, fd); 307 #ifdef CAPABILITIES 308 seq_write_end(&fde->fde_seq); 309 #endif 310 } 311 312 void 313 pwd_ensure_dirs(void) 314 { 315 struct filedesc *fdp; 316 317 fdp = curproc->p_fd; 318 FILEDESC_XLOCK(fdp); 319 if (fdp->fd_cdir == NULL) { 320 fdp->fd_cdir = rootvnode; 321 VREF(rootvnode); 322 } 323 if (fdp->fd_rdir == NULL) { 324 fdp->fd_rdir = rootvnode; 325 VREF(rootvnode); 326 } 327 FILEDESC_XUNLOCK(fdp); 328 } 329 330 /* 331 * System calls on descriptors. 332 */ 333 #ifndef _SYS_SYSPROTO_H_ 334 struct getdtablesize_args { 335 int dummy; 336 }; 337 #endif 338 /* ARGSUSED */ 339 int 340 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 341 { 342 #ifdef RACCT 343 uint64_t lim; 344 #endif 345 346 td->td_retval[0] = 347 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); 348 #ifdef RACCT 349 PROC_LOCK(td->td_proc); 350 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 351 PROC_UNLOCK(td->td_proc); 352 if (lim < td->td_retval[0]) 353 td->td_retval[0] = lim; 354 #endif 355 return (0); 356 } 357 358 /* 359 * Duplicate a file descriptor to a particular value. 360 * 361 * Note: keep in mind that a potential race condition exists when closing 362 * descriptors from a shared descriptor table (via rfork). 363 */ 364 #ifndef _SYS_SYSPROTO_H_ 365 struct dup2_args { 366 u_int from; 367 u_int to; 368 }; 369 #endif 370 /* ARGSUSED */ 371 int 372 sys_dup2(struct thread *td, struct dup2_args *uap) 373 { 374 375 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 376 } 377 378 /* 379 * Duplicate a file descriptor. 380 */ 381 #ifndef _SYS_SYSPROTO_H_ 382 struct dup_args { 383 u_int fd; 384 }; 385 #endif 386 /* ARGSUSED */ 387 int 388 sys_dup(struct thread *td, struct dup_args *uap) 389 { 390 391 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 392 } 393 394 /* 395 * The file control system call. 396 */ 397 #ifndef _SYS_SYSPROTO_H_ 398 struct fcntl_args { 399 int fd; 400 int cmd; 401 long arg; 402 }; 403 #endif 404 /* ARGSUSED */ 405 int 406 sys_fcntl(struct thread *td, struct fcntl_args *uap) 407 { 408 409 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 410 } 411 412 int 413 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 414 { 415 struct flock fl; 416 struct __oflock ofl; 417 intptr_t arg1; 418 int error, newcmd; 419 420 error = 0; 421 newcmd = cmd; 422 switch (cmd) { 423 case F_OGETLK: 424 case F_OSETLK: 425 case F_OSETLKW: 426 /* 427 * Convert old flock structure to new. 428 */ 429 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 430 fl.l_start = ofl.l_start; 431 fl.l_len = ofl.l_len; 432 fl.l_pid = ofl.l_pid; 433 fl.l_type = ofl.l_type; 434 fl.l_whence = ofl.l_whence; 435 fl.l_sysid = 0; 436 437 switch (cmd) { 438 case F_OGETLK: 439 newcmd = F_GETLK; 440 break; 441 case F_OSETLK: 442 newcmd = F_SETLK; 443 break; 444 case F_OSETLKW: 445 newcmd = F_SETLKW; 446 break; 447 } 448 arg1 = (intptr_t)&fl; 449 break; 450 case F_GETLK: 451 case F_SETLK: 452 case F_SETLKW: 453 case F_SETLK_REMOTE: 454 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 455 arg1 = (intptr_t)&fl; 456 break; 457 default: 458 arg1 = arg; 459 break; 460 } 461 if (error) 462 return (error); 463 error = kern_fcntl(td, fd, newcmd, arg1); 464 if (error) 465 return (error); 466 if (cmd == F_OGETLK) { 467 ofl.l_start = fl.l_start; 468 ofl.l_len = fl.l_len; 469 ofl.l_pid = fl.l_pid; 470 ofl.l_type = fl.l_type; 471 ofl.l_whence = fl.l_whence; 472 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 473 } else if (cmd == F_GETLK) { 474 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 475 } 476 return (error); 477 } 478 479 int 480 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 481 { 482 struct filedesc *fdp; 483 struct flock *flp; 484 struct file *fp, *fp2; 485 struct filedescent *fde; 486 struct proc *p; 487 struct vnode *vp; 488 cap_rights_t rights; 489 int error, flg, tmp; 490 uint64_t bsize; 491 off_t foffset; 492 493 error = 0; 494 flg = F_POSIX; 495 p = td->td_proc; 496 fdp = p->p_fd; 497 498 switch (cmd) { 499 case F_DUPFD: 500 tmp = arg; 501 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 502 break; 503 504 case F_DUPFD_CLOEXEC: 505 tmp = arg; 506 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 507 break; 508 509 case F_DUP2FD: 510 tmp = arg; 511 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 512 break; 513 514 case F_DUP2FD_CLOEXEC: 515 tmp = arg; 516 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 517 break; 518 519 case F_GETFD: 520 FILEDESC_SLOCK(fdp); 521 if (fget_locked(fdp, fd) == NULL) { 522 FILEDESC_SUNLOCK(fdp); 523 error = EBADF; 524 break; 525 } 526 fde = &fdp->fd_ofiles[fd]; 527 td->td_retval[0] = 528 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 529 FILEDESC_SUNLOCK(fdp); 530 break; 531 532 case F_SETFD: 533 FILEDESC_XLOCK(fdp); 534 if (fget_locked(fdp, fd) == NULL) { 535 FILEDESC_XUNLOCK(fdp); 536 error = EBADF; 537 break; 538 } 539 fde = &fdp->fd_ofiles[fd]; 540 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 541 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 542 FILEDESC_XUNLOCK(fdp); 543 break; 544 545 case F_GETFL: 546 error = fget_fcntl(td, fd, 547 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 548 if (error != 0) 549 break; 550 td->td_retval[0] = OFLAGS(fp->f_flag); 551 fdrop(fp, td); 552 break; 553 554 case F_SETFL: 555 error = fget_fcntl(td, fd, 556 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 557 if (error != 0) 558 break; 559 do { 560 tmp = flg = fp->f_flag; 561 tmp &= ~FCNTLFLAGS; 562 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 563 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 564 tmp = fp->f_flag & FNONBLOCK; 565 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 566 if (error != 0) { 567 fdrop(fp, td); 568 break; 569 } 570 tmp = fp->f_flag & FASYNC; 571 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 572 if (error == 0) { 573 fdrop(fp, td); 574 break; 575 } 576 atomic_clear_int(&fp->f_flag, FNONBLOCK); 577 tmp = 0; 578 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 579 fdrop(fp, td); 580 break; 581 582 case F_GETOWN: 583 error = fget_fcntl(td, fd, 584 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 585 if (error != 0) 586 break; 587 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 588 if (error == 0) 589 td->td_retval[0] = tmp; 590 fdrop(fp, td); 591 break; 592 593 case F_SETOWN: 594 error = fget_fcntl(td, fd, 595 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 596 if (error != 0) 597 break; 598 tmp = arg; 599 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 600 fdrop(fp, td); 601 break; 602 603 case F_SETLK_REMOTE: 604 error = priv_check(td, PRIV_NFS_LOCKD); 605 if (error) 606 return (error); 607 flg = F_REMOTE; 608 goto do_setlk; 609 610 case F_SETLKW: 611 flg |= F_WAIT; 612 /* FALLTHROUGH F_SETLK */ 613 614 case F_SETLK: 615 do_setlk: 616 cap_rights_init(&rights, CAP_FLOCK); 617 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 618 if (error != 0) 619 break; 620 if (fp->f_type != DTYPE_VNODE) { 621 error = EBADF; 622 fdrop(fp, td); 623 break; 624 } 625 626 flp = (struct flock *)arg; 627 if (flp->l_whence == SEEK_CUR) { 628 foffset = foffset_get(fp); 629 if (foffset < 0 || 630 (flp->l_start > 0 && 631 foffset > OFF_MAX - flp->l_start)) { 632 error = EOVERFLOW; 633 fdrop(fp, td); 634 break; 635 } 636 flp->l_start += foffset; 637 } 638 639 vp = fp->f_vnode; 640 switch (flp->l_type) { 641 case F_RDLCK: 642 if ((fp->f_flag & FREAD) == 0) { 643 error = EBADF; 644 break; 645 } 646 PROC_LOCK(p->p_leader); 647 p->p_leader->p_flag |= P_ADVLOCK; 648 PROC_UNLOCK(p->p_leader); 649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 650 flp, flg); 651 break; 652 case F_WRLCK: 653 if ((fp->f_flag & FWRITE) == 0) { 654 error = EBADF; 655 break; 656 } 657 PROC_LOCK(p->p_leader); 658 p->p_leader->p_flag |= P_ADVLOCK; 659 PROC_UNLOCK(p->p_leader); 660 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 661 flp, flg); 662 break; 663 case F_UNLCK: 664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 665 flp, flg); 666 break; 667 case F_UNLCKSYS: 668 /* 669 * Temporary api for testing remote lock 670 * infrastructure. 671 */ 672 if (flg != F_REMOTE) { 673 error = EINVAL; 674 break; 675 } 676 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 677 F_UNLCKSYS, flp, flg); 678 break; 679 default: 680 error = EINVAL; 681 break; 682 } 683 if (error != 0 || flp->l_type == F_UNLCK || 684 flp->l_type == F_UNLCKSYS) { 685 fdrop(fp, td); 686 break; 687 } 688 689 /* 690 * Check for a race with close. 691 * 692 * The vnode is now advisory locked (or unlocked, but this case 693 * is not really important) as the caller requested. 694 * We had to drop the filedesc lock, so we need to recheck if 695 * the descriptor is still valid, because if it was closed 696 * in the meantime we need to remove advisory lock from the 697 * vnode - close on any descriptor leading to an advisory 698 * locked vnode, removes that lock. 699 * We will return 0 on purpose in that case, as the result of 700 * successful advisory lock might have been externally visible 701 * already. This is fine - effectively we pretend to the caller 702 * that the closing thread was a bit slower and that the 703 * advisory lock succeeded before the close. 704 */ 705 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 706 if (error != 0) { 707 fdrop(fp, td); 708 break; 709 } 710 if (fp != fp2) { 711 flp->l_whence = SEEK_SET; 712 flp->l_start = 0; 713 flp->l_len = 0; 714 flp->l_type = F_UNLCK; 715 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 716 F_UNLCK, flp, F_POSIX); 717 } 718 fdrop(fp, td); 719 fdrop(fp2, td); 720 break; 721 722 case F_GETLK: 723 error = fget_unlocked(fdp, fd, 724 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 725 if (error != 0) 726 break; 727 if (fp->f_type != DTYPE_VNODE) { 728 error = EBADF; 729 fdrop(fp, td); 730 break; 731 } 732 flp = (struct flock *)arg; 733 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 734 flp->l_type != F_UNLCK) { 735 error = EINVAL; 736 fdrop(fp, td); 737 break; 738 } 739 if (flp->l_whence == SEEK_CUR) { 740 foffset = foffset_get(fp); 741 if ((flp->l_start > 0 && 742 foffset > OFF_MAX - flp->l_start) || 743 (flp->l_start < 0 && 744 foffset < OFF_MIN - flp->l_start)) { 745 error = EOVERFLOW; 746 fdrop(fp, td); 747 break; 748 } 749 flp->l_start += foffset; 750 } 751 vp = fp->f_vnode; 752 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 753 F_POSIX); 754 fdrop(fp, td); 755 break; 756 757 case F_RDAHEAD: 758 arg = arg ? 128 * 1024: 0; 759 /* FALLTHROUGH */ 760 case F_READAHEAD: 761 error = fget_unlocked(fdp, fd, 762 cap_rights_init(&rights), &fp, NULL); 763 if (error != 0) 764 break; 765 if (fp->f_type != DTYPE_VNODE) { 766 fdrop(fp, td); 767 error = EBADF; 768 break; 769 } 770 vp = fp->f_vnode; 771 /* 772 * Exclusive lock synchronizes against f_seqcount reads and 773 * writes in sequential_heuristic(). 774 */ 775 error = vn_lock(vp, LK_EXCLUSIVE); 776 if (error != 0) { 777 fdrop(fp, td); 778 break; 779 } 780 if (arg >= 0) { 781 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 782 fp->f_seqcount = (arg + bsize - 1) / bsize; 783 atomic_set_int(&fp->f_flag, FRDAHEAD); 784 } else { 785 atomic_clear_int(&fp->f_flag, FRDAHEAD); 786 } 787 VOP_UNLOCK(vp, 0); 788 fdrop(fp, td); 789 break; 790 791 default: 792 error = EINVAL; 793 break; 794 } 795 return (error); 796 } 797 798 static int 799 getmaxfd(struct thread *td) 800 { 801 802 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 803 } 804 805 /* 806 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 807 */ 808 int 809 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 810 { 811 struct filedesc *fdp; 812 struct filedescent *oldfde, *newfde; 813 struct proc *p; 814 struct file *delfp; 815 int error, maxfd; 816 817 p = td->td_proc; 818 fdp = p->p_fd; 819 820 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 821 MPASS(mode < FDDUP_LASTMODE); 822 823 /* 824 * Verify we have a valid descriptor to dup from and possibly to 825 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 826 * return EINVAL when the new descriptor is out of bounds. 827 */ 828 if (old < 0) 829 return (EBADF); 830 if (new < 0) 831 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 832 maxfd = getmaxfd(td); 833 if (new >= maxfd) 834 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 835 836 FILEDESC_XLOCK(fdp); 837 if (fget_locked(fdp, old) == NULL) { 838 FILEDESC_XUNLOCK(fdp); 839 return (EBADF); 840 } 841 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 842 td->td_retval[0] = new; 843 if (flags & FDDUP_FLAG_CLOEXEC) 844 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 845 FILEDESC_XUNLOCK(fdp); 846 return (0); 847 } 848 849 /* 850 * If the caller specified a file descriptor, make sure the file 851 * table is large enough to hold it, and grab it. Otherwise, just 852 * allocate a new descriptor the usual way. 853 */ 854 switch (mode) { 855 case FDDUP_NORMAL: 856 case FDDUP_FCNTL: 857 if ((error = fdalloc(td, new, &new)) != 0) { 858 FILEDESC_XUNLOCK(fdp); 859 return (error); 860 } 861 break; 862 case FDDUP_MUSTREPLACE: 863 /* Target file descriptor must exist. */ 864 if (fget_locked(fdp, new) == NULL) { 865 FILEDESC_XUNLOCK(fdp); 866 return (EBADF); 867 } 868 break; 869 case FDDUP_FIXED: 870 if (new >= fdp->fd_nfiles) { 871 /* 872 * The resource limits are here instead of e.g. 873 * fdalloc(), because the file descriptor table may be 874 * shared between processes, so we can't really use 875 * racct_add()/racct_sub(). Instead of counting the 876 * number of actually allocated descriptors, just put 877 * the limit on the size of the file descriptor table. 878 */ 879 #ifdef RACCT 880 if (racct_enable) { 881 PROC_LOCK(p); 882 error = racct_set(p, RACCT_NOFILE, new + 1); 883 PROC_UNLOCK(p); 884 if (error != 0) { 885 FILEDESC_XUNLOCK(fdp); 886 return (EMFILE); 887 } 888 } 889 #endif 890 fdgrowtable_exp(fdp, new + 1); 891 } 892 if (!fdisused(fdp, new)) 893 fdused(fdp, new); 894 break; 895 default: 896 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 897 } 898 899 KASSERT(old != new, ("new fd is same as old")); 900 901 oldfde = &fdp->fd_ofiles[old]; 902 fhold(oldfde->fde_file); 903 newfde = &fdp->fd_ofiles[new]; 904 delfp = newfde->fde_file; 905 906 /* 907 * Duplicate the source descriptor. 908 */ 909 #ifdef CAPABILITIES 910 seq_write_begin(&newfde->fde_seq); 911 #endif 912 filecaps_free(&newfde->fde_caps); 913 memcpy(newfde, oldfde, fde_change_size); 914 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); 915 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 916 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 917 else 918 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 919 #ifdef CAPABILITIES 920 seq_write_end(&newfde->fde_seq); 921 #endif 922 td->td_retval[0] = new; 923 924 if (delfp != NULL) { 925 (void) closefp(fdp, new, delfp, td, 1); 926 /* closefp() drops the FILEDESC lock for us. */ 927 } else { 928 FILEDESC_XUNLOCK(fdp); 929 } 930 931 return (0); 932 } 933 934 /* 935 * If sigio is on the list associated with a process or process group, 936 * disable signalling from the device, remove sigio from the list and 937 * free sigio. 938 */ 939 void 940 funsetown(struct sigio **sigiop) 941 { 942 struct sigio *sigio; 943 944 SIGIO_LOCK(); 945 sigio = *sigiop; 946 if (sigio == NULL) { 947 SIGIO_UNLOCK(); 948 return; 949 } 950 *(sigio->sio_myref) = NULL; 951 if ((sigio)->sio_pgid < 0) { 952 struct pgrp *pg = (sigio)->sio_pgrp; 953 PGRP_LOCK(pg); 954 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 955 sigio, sio_pgsigio); 956 PGRP_UNLOCK(pg); 957 } else { 958 struct proc *p = (sigio)->sio_proc; 959 PROC_LOCK(p); 960 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 961 sigio, sio_pgsigio); 962 PROC_UNLOCK(p); 963 } 964 SIGIO_UNLOCK(); 965 crfree(sigio->sio_ucred); 966 free(sigio, M_SIGIO); 967 } 968 969 /* 970 * Free a list of sigio structures. 971 * We only need to lock the SIGIO_LOCK because we have made ourselves 972 * inaccessible to callers of fsetown and therefore do not need to lock 973 * the proc or pgrp struct for the list manipulation. 974 */ 975 void 976 funsetownlst(struct sigiolst *sigiolst) 977 { 978 struct proc *p; 979 struct pgrp *pg; 980 struct sigio *sigio; 981 982 sigio = SLIST_FIRST(sigiolst); 983 if (sigio == NULL) 984 return; 985 p = NULL; 986 pg = NULL; 987 988 /* 989 * Every entry of the list should belong 990 * to a single proc or pgrp. 991 */ 992 if (sigio->sio_pgid < 0) { 993 pg = sigio->sio_pgrp; 994 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 995 } else /* if (sigio->sio_pgid > 0) */ { 996 p = sigio->sio_proc; 997 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 998 } 999 1000 SIGIO_LOCK(); 1001 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1002 *(sigio->sio_myref) = NULL; 1003 if (pg != NULL) { 1004 KASSERT(sigio->sio_pgid < 0, 1005 ("Proc sigio in pgrp sigio list")); 1006 KASSERT(sigio->sio_pgrp == pg, 1007 ("Bogus pgrp in sigio list")); 1008 PGRP_LOCK(pg); 1009 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1010 sio_pgsigio); 1011 PGRP_UNLOCK(pg); 1012 } else /* if (p != NULL) */ { 1013 KASSERT(sigio->sio_pgid > 0, 1014 ("Pgrp sigio in proc sigio list")); 1015 KASSERT(sigio->sio_proc == p, 1016 ("Bogus proc in sigio list")); 1017 PROC_LOCK(p); 1018 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1019 sio_pgsigio); 1020 PROC_UNLOCK(p); 1021 } 1022 SIGIO_UNLOCK(); 1023 crfree(sigio->sio_ucred); 1024 free(sigio, M_SIGIO); 1025 SIGIO_LOCK(); 1026 } 1027 SIGIO_UNLOCK(); 1028 } 1029 1030 /* 1031 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1032 * 1033 * After permission checking, add a sigio structure to the sigio list for 1034 * the process or process group. 1035 */ 1036 int 1037 fsetown(pid_t pgid, struct sigio **sigiop) 1038 { 1039 struct proc *proc; 1040 struct pgrp *pgrp; 1041 struct sigio *sigio; 1042 int ret; 1043 1044 if (pgid == 0) { 1045 funsetown(sigiop); 1046 return (0); 1047 } 1048 1049 ret = 0; 1050 1051 /* Allocate and fill in the new sigio out of locks. */ 1052 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1053 sigio->sio_pgid = pgid; 1054 sigio->sio_ucred = crhold(curthread->td_ucred); 1055 sigio->sio_myref = sigiop; 1056 1057 sx_slock(&proctree_lock); 1058 if (pgid > 0) { 1059 proc = pfind(pgid); 1060 if (proc == NULL) { 1061 ret = ESRCH; 1062 goto fail; 1063 } 1064 1065 /* 1066 * Policy - Don't allow a process to FSETOWN a process 1067 * in another session. 1068 * 1069 * Remove this test to allow maximum flexibility or 1070 * restrict FSETOWN to the current process or process 1071 * group for maximum safety. 1072 */ 1073 PROC_UNLOCK(proc); 1074 if (proc->p_session != curthread->td_proc->p_session) { 1075 ret = EPERM; 1076 goto fail; 1077 } 1078 1079 pgrp = NULL; 1080 } else /* if (pgid < 0) */ { 1081 pgrp = pgfind(-pgid); 1082 if (pgrp == NULL) { 1083 ret = ESRCH; 1084 goto fail; 1085 } 1086 PGRP_UNLOCK(pgrp); 1087 1088 /* 1089 * Policy - Don't allow a process to FSETOWN a process 1090 * in another session. 1091 * 1092 * Remove this test to allow maximum flexibility or 1093 * restrict FSETOWN to the current process or process 1094 * group for maximum safety. 1095 */ 1096 if (pgrp->pg_session != curthread->td_proc->p_session) { 1097 ret = EPERM; 1098 goto fail; 1099 } 1100 1101 proc = NULL; 1102 } 1103 funsetown(sigiop); 1104 if (pgid > 0) { 1105 PROC_LOCK(proc); 1106 /* 1107 * Since funsetownlst() is called without the proctree 1108 * locked, we need to check for P_WEXIT. 1109 * XXX: is ESRCH correct? 1110 */ 1111 if ((proc->p_flag & P_WEXIT) != 0) { 1112 PROC_UNLOCK(proc); 1113 ret = ESRCH; 1114 goto fail; 1115 } 1116 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1117 sigio->sio_proc = proc; 1118 PROC_UNLOCK(proc); 1119 } else { 1120 PGRP_LOCK(pgrp); 1121 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1122 sigio->sio_pgrp = pgrp; 1123 PGRP_UNLOCK(pgrp); 1124 } 1125 sx_sunlock(&proctree_lock); 1126 SIGIO_LOCK(); 1127 *sigiop = sigio; 1128 SIGIO_UNLOCK(); 1129 return (0); 1130 1131 fail: 1132 sx_sunlock(&proctree_lock); 1133 crfree(sigio->sio_ucred); 1134 free(sigio, M_SIGIO); 1135 return (ret); 1136 } 1137 1138 /* 1139 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1140 */ 1141 pid_t 1142 fgetown(sigiop) 1143 struct sigio **sigiop; 1144 { 1145 pid_t pgid; 1146 1147 SIGIO_LOCK(); 1148 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1149 SIGIO_UNLOCK(); 1150 return (pgid); 1151 } 1152 1153 /* 1154 * Function drops the filedesc lock on return. 1155 */ 1156 static int 1157 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1158 int holdleaders) 1159 { 1160 int error; 1161 1162 FILEDESC_XLOCK_ASSERT(fdp); 1163 1164 if (holdleaders) { 1165 if (td->td_proc->p_fdtol != NULL) { 1166 /* 1167 * Ask fdfree() to sleep to ensure that all relevant 1168 * process leaders can be traversed in closef(). 1169 */ 1170 fdp->fd_holdleaderscount++; 1171 } else { 1172 holdleaders = 0; 1173 } 1174 } 1175 1176 /* 1177 * We now hold the fp reference that used to be owned by the 1178 * descriptor array. We have to unlock the FILEDESC *AFTER* 1179 * knote_fdclose to prevent a race of the fd getting opened, a knote 1180 * added, and deleteing a knote for the new fd. 1181 */ 1182 knote_fdclose(td, fd); 1183 1184 /* 1185 * We need to notify mqueue if the object is of type mqueue. 1186 */ 1187 if (fp->f_type == DTYPE_MQUEUE) 1188 mq_fdclose(td, fd, fp); 1189 FILEDESC_XUNLOCK(fdp); 1190 1191 error = closef(fp, td); 1192 if (holdleaders) { 1193 FILEDESC_XLOCK(fdp); 1194 fdp->fd_holdleaderscount--; 1195 if (fdp->fd_holdleaderscount == 0 && 1196 fdp->fd_holdleaderswakeup != 0) { 1197 fdp->fd_holdleaderswakeup = 0; 1198 wakeup(&fdp->fd_holdleaderscount); 1199 } 1200 FILEDESC_XUNLOCK(fdp); 1201 } 1202 return (error); 1203 } 1204 1205 /* 1206 * Close a file descriptor. 1207 */ 1208 #ifndef _SYS_SYSPROTO_H_ 1209 struct close_args { 1210 int fd; 1211 }; 1212 #endif 1213 /* ARGSUSED */ 1214 int 1215 sys_close(struct thread *td, struct close_args *uap) 1216 { 1217 1218 return (kern_close(td, uap->fd)); 1219 } 1220 1221 int 1222 kern_close(struct thread *td, int fd) 1223 { 1224 struct filedesc *fdp; 1225 struct file *fp; 1226 1227 fdp = td->td_proc->p_fd; 1228 1229 AUDIT_SYSCLOSE(td, fd); 1230 1231 FILEDESC_XLOCK(fdp); 1232 if ((fp = fget_locked(fdp, fd)) == NULL) { 1233 FILEDESC_XUNLOCK(fdp); 1234 return (EBADF); 1235 } 1236 fdfree(fdp, fd); 1237 1238 /* closefp() drops the FILEDESC lock for us. */ 1239 return (closefp(fdp, fd, fp, td, 1)); 1240 } 1241 1242 /* 1243 * Close open file descriptors. 1244 */ 1245 #ifndef _SYS_SYSPROTO_H_ 1246 struct closefrom_args { 1247 int lowfd; 1248 }; 1249 #endif 1250 /* ARGSUSED */ 1251 int 1252 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1253 { 1254 struct filedesc *fdp; 1255 int fd; 1256 1257 fdp = td->td_proc->p_fd; 1258 AUDIT_ARG_FD(uap->lowfd); 1259 1260 /* 1261 * Treat negative starting file descriptor values identical to 1262 * closefrom(0) which closes all files. 1263 */ 1264 if (uap->lowfd < 0) 1265 uap->lowfd = 0; 1266 FILEDESC_SLOCK(fdp); 1267 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1268 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1269 FILEDESC_SUNLOCK(fdp); 1270 (void)kern_close(td, fd); 1271 FILEDESC_SLOCK(fdp); 1272 } 1273 } 1274 FILEDESC_SUNLOCK(fdp); 1275 return (0); 1276 } 1277 1278 #if defined(COMPAT_43) 1279 /* 1280 * Return status information about a file descriptor. 1281 */ 1282 #ifndef _SYS_SYSPROTO_H_ 1283 struct ofstat_args { 1284 int fd; 1285 struct ostat *sb; 1286 }; 1287 #endif 1288 /* ARGSUSED */ 1289 int 1290 ofstat(struct thread *td, struct ofstat_args *uap) 1291 { 1292 struct ostat oub; 1293 struct stat ub; 1294 int error; 1295 1296 error = kern_fstat(td, uap->fd, &ub); 1297 if (error == 0) { 1298 cvtstat(&ub, &oub); 1299 error = copyout(&oub, uap->sb, sizeof(oub)); 1300 } 1301 return (error); 1302 } 1303 #endif /* COMPAT_43 */ 1304 1305 /* 1306 * Return status information about a file descriptor. 1307 */ 1308 #ifndef _SYS_SYSPROTO_H_ 1309 struct fstat_args { 1310 int fd; 1311 struct stat *sb; 1312 }; 1313 #endif 1314 /* ARGSUSED */ 1315 int 1316 sys_fstat(struct thread *td, struct fstat_args *uap) 1317 { 1318 struct stat ub; 1319 int error; 1320 1321 error = kern_fstat(td, uap->fd, &ub); 1322 if (error == 0) 1323 error = copyout(&ub, uap->sb, sizeof(ub)); 1324 return (error); 1325 } 1326 1327 int 1328 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1329 { 1330 struct file *fp; 1331 cap_rights_t rights; 1332 int error; 1333 1334 AUDIT_ARG_FD(fd); 1335 1336 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1337 if (error != 0) 1338 return (error); 1339 1340 AUDIT_ARG_FILE(td->td_proc, fp); 1341 1342 error = fo_stat(fp, sbp, td->td_ucred, td); 1343 fdrop(fp, td); 1344 #ifdef KTRACE 1345 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1346 ktrstat(sbp); 1347 #endif 1348 return (error); 1349 } 1350 1351 /* 1352 * Return status information about a file descriptor. 1353 */ 1354 #ifndef _SYS_SYSPROTO_H_ 1355 struct nfstat_args { 1356 int fd; 1357 struct nstat *sb; 1358 }; 1359 #endif 1360 /* ARGSUSED */ 1361 int 1362 sys_nfstat(struct thread *td, struct nfstat_args *uap) 1363 { 1364 struct nstat nub; 1365 struct stat ub; 1366 int error; 1367 1368 error = kern_fstat(td, uap->fd, &ub); 1369 if (error == 0) { 1370 cvtnstat(&ub, &nub); 1371 error = copyout(&nub, uap->sb, sizeof(nub)); 1372 } 1373 return (error); 1374 } 1375 1376 /* 1377 * Return pathconf information about a file descriptor. 1378 */ 1379 #ifndef _SYS_SYSPROTO_H_ 1380 struct fpathconf_args { 1381 int fd; 1382 int name; 1383 }; 1384 #endif 1385 /* ARGSUSED */ 1386 int 1387 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1388 { 1389 struct file *fp; 1390 struct vnode *vp; 1391 cap_rights_t rights; 1392 int error; 1393 1394 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1395 if (error != 0) 1396 return (error); 1397 1398 if (uap->name == _PC_ASYNC_IO) { 1399 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO; 1400 goto out; 1401 } 1402 vp = fp->f_vnode; 1403 if (vp != NULL) { 1404 vn_lock(vp, LK_SHARED | LK_RETRY); 1405 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1406 VOP_UNLOCK(vp, 0); 1407 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1408 if (uap->name != _PC_PIPE_BUF) { 1409 error = EINVAL; 1410 } else { 1411 td->td_retval[0] = PIPE_BUF; 1412 error = 0; 1413 } 1414 } else { 1415 error = EOPNOTSUPP; 1416 } 1417 out: 1418 fdrop(fp, td); 1419 return (error); 1420 } 1421 1422 /* 1423 * Initialize filecaps structure. 1424 */ 1425 void 1426 filecaps_init(struct filecaps *fcaps) 1427 { 1428 1429 bzero(fcaps, sizeof(*fcaps)); 1430 fcaps->fc_nioctls = -1; 1431 } 1432 1433 /* 1434 * Copy filecaps structure allocating memory for ioctls array if needed. 1435 * 1436 * The last parameter indicates whether the fdtable is locked. If it is not and 1437 * ioctls are encountered, copying fails and the caller must lock the table. 1438 * 1439 * Note that if the table was not locked, the caller has to check the relevant 1440 * sequence counter to determine whether the operation was successful. 1441 */ 1442 int 1443 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1444 { 1445 size_t size; 1446 1447 *dst = *src; 1448 if (src->fc_ioctls == NULL) 1449 return (0); 1450 if (!locked) 1451 return (1); 1452 1453 KASSERT(src->fc_nioctls > 0, 1454 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1455 1456 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1457 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1458 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1459 return (0); 1460 } 1461 1462 /* 1463 * Move filecaps structure to the new place and clear the old place. 1464 */ 1465 void 1466 filecaps_move(struct filecaps *src, struct filecaps *dst) 1467 { 1468 1469 *dst = *src; 1470 bzero(src, sizeof(*src)); 1471 } 1472 1473 /* 1474 * Fill the given filecaps structure with full rights. 1475 */ 1476 static void 1477 filecaps_fill(struct filecaps *fcaps) 1478 { 1479 1480 CAP_ALL(&fcaps->fc_rights); 1481 fcaps->fc_ioctls = NULL; 1482 fcaps->fc_nioctls = -1; 1483 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1484 } 1485 1486 /* 1487 * Free memory allocated within filecaps structure. 1488 */ 1489 void 1490 filecaps_free(struct filecaps *fcaps) 1491 { 1492 1493 free(fcaps->fc_ioctls, M_FILECAPS); 1494 bzero(fcaps, sizeof(*fcaps)); 1495 } 1496 1497 /* 1498 * Validate the given filecaps structure. 1499 */ 1500 static void 1501 filecaps_validate(const struct filecaps *fcaps, const char *func) 1502 { 1503 1504 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1505 ("%s: invalid rights", func)); 1506 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1507 ("%s: invalid fcntls", func)); 1508 KASSERT(fcaps->fc_fcntls == 0 || 1509 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1510 ("%s: fcntls without CAP_FCNTL", func)); 1511 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1512 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1513 ("%s: invalid ioctls", func)); 1514 KASSERT(fcaps->fc_nioctls == 0 || 1515 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1516 ("%s: ioctls without CAP_IOCTL", func)); 1517 } 1518 1519 static void 1520 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1521 { 1522 int nfd1; 1523 1524 FILEDESC_XLOCK_ASSERT(fdp); 1525 1526 nfd1 = fdp->fd_nfiles * 2; 1527 if (nfd1 < nfd) 1528 nfd1 = nfd; 1529 fdgrowtable(fdp, nfd1); 1530 } 1531 1532 /* 1533 * Grow the file table to accommodate (at least) nfd descriptors. 1534 */ 1535 static void 1536 fdgrowtable(struct filedesc *fdp, int nfd) 1537 { 1538 struct filedesc0 *fdp0; 1539 struct freetable *ft; 1540 struct fdescenttbl *ntable; 1541 struct fdescenttbl *otable; 1542 int nnfiles, onfiles; 1543 NDSLOTTYPE *nmap, *omap; 1544 1545 /* 1546 * If lastfile is -1 this struct filedesc was just allocated and we are 1547 * growing it to accommodate for the one we are going to copy from. There 1548 * is no need to have a lock on this one as it's not visible to anyone. 1549 */ 1550 if (fdp->fd_lastfile != -1) 1551 FILEDESC_XLOCK_ASSERT(fdp); 1552 1553 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1554 1555 /* save old values */ 1556 onfiles = fdp->fd_nfiles; 1557 otable = fdp->fd_files; 1558 omap = fdp->fd_map; 1559 1560 /* compute the size of the new table */ 1561 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1562 if (nnfiles <= onfiles) 1563 /* the table is already large enough */ 1564 return; 1565 1566 /* 1567 * Allocate a new table. We need enough space for the number of 1568 * entries, file entries themselves and the struct freetable we will use 1569 * when we decommission the table and place it on the freelist. 1570 * We place the struct freetable in the middle so we don't have 1571 * to worry about padding. 1572 */ 1573 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1574 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1575 sizeof(struct freetable), 1576 M_FILEDESC, M_ZERO | M_WAITOK); 1577 /* copy the old data */ 1578 ntable->fdt_nfiles = nnfiles; 1579 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1580 onfiles * sizeof(ntable->fdt_ofiles[0])); 1581 1582 /* 1583 * Allocate a new map only if the old is not large enough. It will 1584 * grow at a slower rate than the table as it can map more 1585 * entries than the table can hold. 1586 */ 1587 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1588 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1589 M_ZERO | M_WAITOK); 1590 /* copy over the old data and update the pointer */ 1591 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1592 fdp->fd_map = nmap; 1593 } 1594 1595 /* 1596 * Make sure that ntable is correctly initialized before we replace 1597 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1598 * data. 1599 */ 1600 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1601 1602 /* 1603 * Do not free the old file table, as some threads may still 1604 * reference entries within it. Instead, place it on a freelist 1605 * which will be processed when the struct filedesc is released. 1606 * 1607 * Note that if onfiles == NDFILE, we're dealing with the original 1608 * static allocation contained within (struct filedesc0 *)fdp, 1609 * which must not be freed. 1610 */ 1611 if (onfiles > NDFILE) { 1612 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1613 fdp0 = (struct filedesc0 *)fdp; 1614 ft->ft_table = otable; 1615 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1616 } 1617 /* 1618 * The map does not have the same possibility of threads still 1619 * holding references to it. So always free it as long as it 1620 * does not reference the original static allocation. 1621 */ 1622 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1623 free(omap, M_FILEDESC); 1624 } 1625 1626 /* 1627 * Allocate a file descriptor for the process. 1628 */ 1629 int 1630 fdalloc(struct thread *td, int minfd, int *result) 1631 { 1632 struct proc *p = td->td_proc; 1633 struct filedesc *fdp = p->p_fd; 1634 int fd, maxfd, allocfd; 1635 #ifdef RACCT 1636 int error; 1637 #endif 1638 1639 FILEDESC_XLOCK_ASSERT(fdp); 1640 1641 if (fdp->fd_freefile > minfd) 1642 minfd = fdp->fd_freefile; 1643 1644 maxfd = getmaxfd(td); 1645 1646 /* 1647 * Search the bitmap for a free descriptor starting at minfd. 1648 * If none is found, grow the file table. 1649 */ 1650 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1651 if (fd >= maxfd) 1652 return (EMFILE); 1653 if (fd >= fdp->fd_nfiles) { 1654 allocfd = min(fd * 2, maxfd); 1655 #ifdef RACCT 1656 if (racct_enable) { 1657 PROC_LOCK(p); 1658 error = racct_set(p, RACCT_NOFILE, allocfd); 1659 PROC_UNLOCK(p); 1660 if (error != 0) 1661 return (EMFILE); 1662 } 1663 #endif 1664 /* 1665 * fd is already equal to first free descriptor >= minfd, so 1666 * we only need to grow the table and we are done. 1667 */ 1668 fdgrowtable_exp(fdp, allocfd); 1669 } 1670 1671 /* 1672 * Perform some sanity checks, then mark the file descriptor as 1673 * used and return it to the caller. 1674 */ 1675 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1676 ("invalid descriptor %d", fd)); 1677 KASSERT(!fdisused(fdp, fd), 1678 ("fd_first_free() returned non-free descriptor")); 1679 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1680 ("file descriptor isn't free")); 1681 fdused(fdp, fd); 1682 *result = fd; 1683 return (0); 1684 } 1685 1686 /* 1687 * Allocate n file descriptors for the process. 1688 */ 1689 int 1690 fdallocn(struct thread *td, int minfd, int *fds, int n) 1691 { 1692 struct proc *p = td->td_proc; 1693 struct filedesc *fdp = p->p_fd; 1694 int i; 1695 1696 FILEDESC_XLOCK_ASSERT(fdp); 1697 1698 for (i = 0; i < n; i++) 1699 if (fdalloc(td, 0, &fds[i]) != 0) 1700 break; 1701 1702 if (i < n) { 1703 for (i--; i >= 0; i--) 1704 fdunused(fdp, fds[i]); 1705 return (EMFILE); 1706 } 1707 1708 return (0); 1709 } 1710 1711 /* 1712 * Create a new open file structure and allocate a file descriptor for the 1713 * process that refers to it. We add one reference to the file for the 1714 * descriptor table and one reference for resultfp. This is to prevent us 1715 * being preempted and the entry in the descriptor table closed after we 1716 * release the FILEDESC lock. 1717 */ 1718 int 1719 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1720 struct filecaps *fcaps) 1721 { 1722 struct file *fp; 1723 int error, fd; 1724 1725 error = falloc_noinstall(td, &fp); 1726 if (error) 1727 return (error); /* no reference held on error */ 1728 1729 error = finstall(td, fp, &fd, flags, fcaps); 1730 if (error) { 1731 fdrop(fp, td); /* one reference (fp only) */ 1732 return (error); 1733 } 1734 1735 if (resultfp != NULL) 1736 *resultfp = fp; /* copy out result */ 1737 else 1738 fdrop(fp, td); /* release local reference */ 1739 1740 if (resultfd != NULL) 1741 *resultfd = fd; 1742 1743 return (0); 1744 } 1745 1746 /* 1747 * Create a new open file structure without allocating a file descriptor. 1748 */ 1749 int 1750 falloc_noinstall(struct thread *td, struct file **resultfp) 1751 { 1752 struct file *fp; 1753 int maxuserfiles = maxfiles - (maxfiles / 20); 1754 static struct timeval lastfail; 1755 static int curfail; 1756 1757 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1758 1759 if ((openfiles >= maxuserfiles && 1760 priv_check(td, PRIV_MAXFILES) != 0) || 1761 openfiles >= maxfiles) { 1762 if (ppsratecheck(&lastfail, &curfail, 1)) { 1763 printf("kern.maxfiles limit exceeded by uid %i, " 1764 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1765 } 1766 return (ENFILE); 1767 } 1768 atomic_add_int(&openfiles, 1); 1769 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1770 refcount_init(&fp->f_count, 1); 1771 fp->f_cred = crhold(td->td_ucred); 1772 fp->f_ops = &badfileops; 1773 *resultfp = fp; 1774 return (0); 1775 } 1776 1777 /* 1778 * Install a file in a file descriptor table. 1779 */ 1780 void 1781 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1782 struct filecaps *fcaps) 1783 { 1784 struct filedescent *fde; 1785 1786 MPASS(fp != NULL); 1787 if (fcaps != NULL) 1788 filecaps_validate(fcaps, __func__); 1789 FILEDESC_XLOCK_ASSERT(fdp); 1790 1791 fde = &fdp->fd_ofiles[fd]; 1792 #ifdef CAPABILITIES 1793 seq_write_begin(&fde->fde_seq); 1794 #endif 1795 fde->fde_file = fp; 1796 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1797 if (fcaps != NULL) 1798 filecaps_move(fcaps, &fde->fde_caps); 1799 else 1800 filecaps_fill(&fde->fde_caps); 1801 #ifdef CAPABILITIES 1802 seq_write_end(&fde->fde_seq); 1803 #endif 1804 } 1805 1806 int 1807 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1808 struct filecaps *fcaps) 1809 { 1810 struct filedesc *fdp = td->td_proc->p_fd; 1811 int error; 1812 1813 MPASS(fd != NULL); 1814 1815 FILEDESC_XLOCK(fdp); 1816 if ((error = fdalloc(td, 0, fd))) { 1817 FILEDESC_XUNLOCK(fdp); 1818 return (error); 1819 } 1820 fhold(fp); 1821 _finstall(fdp, fp, *fd, flags, fcaps); 1822 FILEDESC_XUNLOCK(fdp); 1823 return (0); 1824 } 1825 1826 /* 1827 * Build a new filedesc structure from another. 1828 * Copy the current, root, and jail root vnode references. 1829 * 1830 * If fdp is not NULL, return with it shared locked. 1831 */ 1832 struct filedesc * 1833 fdinit(struct filedesc *fdp, bool prepfiles) 1834 { 1835 struct filedesc0 *newfdp0; 1836 struct filedesc *newfdp; 1837 1838 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1839 newfdp = &newfdp0->fd_fd; 1840 1841 /* Create the file descriptor table. */ 1842 FILEDESC_LOCK_INIT(newfdp); 1843 refcount_init(&newfdp->fd_refcnt, 1); 1844 refcount_init(&newfdp->fd_holdcnt, 1); 1845 newfdp->fd_cmask = CMASK; 1846 newfdp->fd_map = newfdp0->fd_dmap; 1847 newfdp->fd_lastfile = -1; 1848 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1849 newfdp->fd_files->fdt_nfiles = NDFILE; 1850 1851 if (fdp == NULL) 1852 return (newfdp); 1853 1854 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1855 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1856 1857 FILEDESC_SLOCK(fdp); 1858 newfdp->fd_cdir = fdp->fd_cdir; 1859 if (newfdp->fd_cdir) 1860 VREF(newfdp->fd_cdir); 1861 newfdp->fd_rdir = fdp->fd_rdir; 1862 if (newfdp->fd_rdir) 1863 VREF(newfdp->fd_rdir); 1864 newfdp->fd_jdir = fdp->fd_jdir; 1865 if (newfdp->fd_jdir) 1866 VREF(newfdp->fd_jdir); 1867 1868 if (!prepfiles) { 1869 FILEDESC_SUNLOCK(fdp); 1870 } else { 1871 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1872 FILEDESC_SUNLOCK(fdp); 1873 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1874 FILEDESC_SLOCK(fdp); 1875 } 1876 } 1877 1878 return (newfdp); 1879 } 1880 1881 static struct filedesc * 1882 fdhold(struct proc *p) 1883 { 1884 struct filedesc *fdp; 1885 1886 PROC_LOCK_ASSERT(p, MA_OWNED); 1887 fdp = p->p_fd; 1888 if (fdp != NULL) 1889 refcount_acquire(&fdp->fd_holdcnt); 1890 return (fdp); 1891 } 1892 1893 static void 1894 fddrop(struct filedesc *fdp) 1895 { 1896 1897 if (fdp->fd_holdcnt > 1) { 1898 if (refcount_release(&fdp->fd_holdcnt) == 0) 1899 return; 1900 } 1901 1902 FILEDESC_LOCK_DESTROY(fdp); 1903 uma_zfree(filedesc0_zone, fdp); 1904 } 1905 1906 /* 1907 * Share a filedesc structure. 1908 */ 1909 struct filedesc * 1910 fdshare(struct filedesc *fdp) 1911 { 1912 1913 refcount_acquire(&fdp->fd_refcnt); 1914 return (fdp); 1915 } 1916 1917 /* 1918 * Unshare a filedesc structure, if necessary by making a copy 1919 */ 1920 void 1921 fdunshare(struct thread *td) 1922 { 1923 struct filedesc *tmp; 1924 struct proc *p = td->td_proc; 1925 1926 if (p->p_fd->fd_refcnt == 1) 1927 return; 1928 1929 tmp = fdcopy(p->p_fd); 1930 fdescfree(td); 1931 p->p_fd = tmp; 1932 } 1933 1934 void 1935 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 1936 { 1937 1938 fdescfree(td); 1939 td->td_proc->p_fd = fdp; 1940 } 1941 1942 /* 1943 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1944 * this is to ease callers, not catch errors. 1945 */ 1946 struct filedesc * 1947 fdcopy(struct filedesc *fdp) 1948 { 1949 struct filedesc *newfdp; 1950 struct filedescent *nfde, *ofde; 1951 int i; 1952 1953 MPASS(fdp != NULL); 1954 1955 newfdp = fdinit(fdp, true); 1956 /* copy all passable descriptors (i.e. not kqueue) */ 1957 newfdp->fd_freefile = -1; 1958 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1959 ofde = &fdp->fd_ofiles[i]; 1960 if (ofde->fde_file == NULL || 1961 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1962 if (newfdp->fd_freefile == -1) 1963 newfdp->fd_freefile = i; 1964 continue; 1965 } 1966 nfde = &newfdp->fd_ofiles[i]; 1967 *nfde = *ofde; 1968 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 1969 fhold(nfde->fde_file); 1970 fdused_init(newfdp, i); 1971 newfdp->fd_lastfile = i; 1972 } 1973 if (newfdp->fd_freefile == -1) 1974 newfdp->fd_freefile = i; 1975 newfdp->fd_cmask = fdp->fd_cmask; 1976 FILEDESC_SUNLOCK(fdp); 1977 return (newfdp); 1978 } 1979 1980 /* 1981 * Copies a filedesc structure, while remapping all file descriptors 1982 * stored inside using a translation table. 1983 * 1984 * File descriptors are copied over to the new file descriptor table, 1985 * regardless of whether the close-on-exec flag is set. 1986 */ 1987 int 1988 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 1989 struct filedesc **ret) 1990 { 1991 struct filedesc *newfdp; 1992 struct filedescent *nfde, *ofde; 1993 int error, i; 1994 1995 MPASS(fdp != NULL); 1996 1997 newfdp = fdinit(fdp, true); 1998 if (nfds > fdp->fd_lastfile + 1) { 1999 /* New table cannot be larger than the old one. */ 2000 error = E2BIG; 2001 goto bad; 2002 } 2003 /* Copy all passable descriptors (i.e. not kqueue). */ 2004 newfdp->fd_freefile = nfds; 2005 for (i = 0; i < nfds; ++i) { 2006 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { 2007 /* File descriptor out of bounds. */ 2008 error = EBADF; 2009 goto bad; 2010 } 2011 ofde = &fdp->fd_ofiles[fds[i]]; 2012 if (ofde->fde_file == NULL) { 2013 /* Unused file descriptor. */ 2014 error = EBADF; 2015 goto bad; 2016 } 2017 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2018 /* File descriptor cannot be passed. */ 2019 error = EINVAL; 2020 goto bad; 2021 } 2022 nfde = &newfdp->fd_ofiles[i]; 2023 *nfde = *ofde; 2024 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2025 fhold(nfde->fde_file); 2026 fdused_init(newfdp, i); 2027 newfdp->fd_lastfile = i; 2028 } 2029 newfdp->fd_cmask = fdp->fd_cmask; 2030 FILEDESC_SUNLOCK(fdp); 2031 *ret = newfdp; 2032 return (0); 2033 bad: 2034 FILEDESC_SUNLOCK(fdp); 2035 fdescfree_remapped(newfdp); 2036 return (error); 2037 } 2038 2039 /* 2040 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2041 * one of processes using it exits) and the table used to be shared. 2042 */ 2043 static void 2044 fdclearlocks(struct thread *td) 2045 { 2046 struct filedesc *fdp; 2047 struct filedesc_to_leader *fdtol; 2048 struct flock lf; 2049 struct file *fp; 2050 struct proc *p; 2051 struct vnode *vp; 2052 int i; 2053 2054 p = td->td_proc; 2055 fdp = p->p_fd; 2056 fdtol = p->p_fdtol; 2057 MPASS(fdtol != NULL); 2058 2059 FILEDESC_XLOCK(fdp); 2060 KASSERT(fdtol->fdl_refcount > 0, 2061 ("filedesc_to_refcount botch: fdl_refcount=%d", 2062 fdtol->fdl_refcount)); 2063 if (fdtol->fdl_refcount == 1 && 2064 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2065 for (i = 0; i <= fdp->fd_lastfile; i++) { 2066 fp = fdp->fd_ofiles[i].fde_file; 2067 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2068 continue; 2069 fhold(fp); 2070 FILEDESC_XUNLOCK(fdp); 2071 lf.l_whence = SEEK_SET; 2072 lf.l_start = 0; 2073 lf.l_len = 0; 2074 lf.l_type = F_UNLCK; 2075 vp = fp->f_vnode; 2076 (void) VOP_ADVLOCK(vp, 2077 (caddr_t)p->p_leader, F_UNLCK, 2078 &lf, F_POSIX); 2079 FILEDESC_XLOCK(fdp); 2080 fdrop(fp, td); 2081 } 2082 } 2083 retry: 2084 if (fdtol->fdl_refcount == 1) { 2085 if (fdp->fd_holdleaderscount > 0 && 2086 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2087 /* 2088 * close() or kern_dup() has cleared a reference 2089 * in a shared file descriptor table. 2090 */ 2091 fdp->fd_holdleaderswakeup = 1; 2092 sx_sleep(&fdp->fd_holdleaderscount, 2093 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2094 goto retry; 2095 } 2096 if (fdtol->fdl_holdcount > 0) { 2097 /* 2098 * Ensure that fdtol->fdl_leader remains 2099 * valid in closef(). 2100 */ 2101 fdtol->fdl_wakeup = 1; 2102 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2103 "fdlhold", 0); 2104 goto retry; 2105 } 2106 } 2107 fdtol->fdl_refcount--; 2108 if (fdtol->fdl_refcount == 0 && 2109 fdtol->fdl_holdcount == 0) { 2110 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2111 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2112 } else 2113 fdtol = NULL; 2114 p->p_fdtol = NULL; 2115 FILEDESC_XUNLOCK(fdp); 2116 if (fdtol != NULL) 2117 free(fdtol, M_FILEDESC_TO_LEADER); 2118 } 2119 2120 /* 2121 * Release a filedesc structure. 2122 */ 2123 static void 2124 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2125 { 2126 struct filedesc0 *fdp0; 2127 struct freetable *ft, *tft; 2128 struct filedescent *fde; 2129 struct file *fp; 2130 int i; 2131 2132 for (i = 0; i <= fdp->fd_lastfile; i++) { 2133 fde = &fdp->fd_ofiles[i]; 2134 fp = fde->fde_file; 2135 if (fp != NULL) { 2136 fdefree_last(fde); 2137 if (needclose) 2138 (void) closef(fp, td); 2139 else 2140 fdrop(fp, td); 2141 } 2142 } 2143 2144 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2145 free(fdp->fd_map, M_FILEDESC); 2146 if (fdp->fd_nfiles > NDFILE) 2147 free(fdp->fd_files, M_FILEDESC); 2148 2149 fdp0 = (struct filedesc0 *)fdp; 2150 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2151 free(ft->ft_table, M_FILEDESC); 2152 2153 fddrop(fdp); 2154 } 2155 2156 void 2157 fdescfree(struct thread *td) 2158 { 2159 struct proc *p; 2160 struct filedesc *fdp; 2161 struct vnode *cdir, *jdir, *rdir; 2162 2163 p = td->td_proc; 2164 fdp = p->p_fd; 2165 MPASS(fdp != NULL); 2166 2167 #ifdef RACCT 2168 if (racct_enable) { 2169 PROC_LOCK(p); 2170 racct_set(p, RACCT_NOFILE, 0); 2171 PROC_UNLOCK(p); 2172 } 2173 #endif 2174 2175 if (p->p_fdtol != NULL) 2176 fdclearlocks(td); 2177 2178 PROC_LOCK(p); 2179 p->p_fd = NULL; 2180 PROC_UNLOCK(p); 2181 2182 if (refcount_release(&fdp->fd_refcnt) == 0) 2183 return; 2184 2185 FILEDESC_XLOCK(fdp); 2186 cdir = fdp->fd_cdir; 2187 fdp->fd_cdir = NULL; 2188 rdir = fdp->fd_rdir; 2189 fdp->fd_rdir = NULL; 2190 jdir = fdp->fd_jdir; 2191 fdp->fd_jdir = NULL; 2192 FILEDESC_XUNLOCK(fdp); 2193 2194 if (cdir != NULL) 2195 vrele(cdir); 2196 if (rdir != NULL) 2197 vrele(rdir); 2198 if (jdir != NULL) 2199 vrele(jdir); 2200 2201 fdescfree_fds(td, fdp, 1); 2202 } 2203 2204 void 2205 fdescfree_remapped(struct filedesc *fdp) 2206 { 2207 2208 if (fdp->fd_cdir != NULL) 2209 vrele(fdp->fd_cdir); 2210 if (fdp->fd_rdir != NULL) 2211 vrele(fdp->fd_rdir); 2212 if (fdp->fd_jdir != NULL) 2213 vrele(fdp->fd_jdir); 2214 2215 fdescfree_fds(curthread, fdp, 0); 2216 } 2217 2218 /* 2219 * For setugid programs, we don't want to people to use that setugidness 2220 * to generate error messages which write to a file which otherwise would 2221 * otherwise be off-limits to the process. We check for filesystems where 2222 * the vnode can change out from under us after execve (like [lin]procfs). 2223 * 2224 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2225 * sufficient. We also don't check for setugidness since we know we are. 2226 */ 2227 static bool 2228 is_unsafe(struct file *fp) 2229 { 2230 struct vnode *vp; 2231 2232 if (fp->f_type != DTYPE_VNODE) 2233 return (false); 2234 2235 vp = fp->f_vnode; 2236 return ((vp->v_vflag & VV_PROCDEP) != 0); 2237 } 2238 2239 /* 2240 * Make this setguid thing safe, if at all possible. 2241 */ 2242 void 2243 fdsetugidsafety(struct thread *td) 2244 { 2245 struct filedesc *fdp; 2246 struct file *fp; 2247 int i; 2248 2249 fdp = td->td_proc->p_fd; 2250 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2251 MPASS(fdp->fd_nfiles >= 3); 2252 for (i = 0; i <= 2; i++) { 2253 fp = fdp->fd_ofiles[i].fde_file; 2254 if (fp != NULL && is_unsafe(fp)) { 2255 FILEDESC_XLOCK(fdp); 2256 knote_fdclose(td, i); 2257 /* 2258 * NULL-out descriptor prior to close to avoid 2259 * a race while close blocks. 2260 */ 2261 fdfree(fdp, i); 2262 FILEDESC_XUNLOCK(fdp); 2263 (void) closef(fp, td); 2264 } 2265 } 2266 } 2267 2268 /* 2269 * If a specific file object occupies a specific file descriptor, close the 2270 * file descriptor entry and drop a reference on the file object. This is a 2271 * convenience function to handle a subsequent error in a function that calls 2272 * falloc() that handles the race that another thread might have closed the 2273 * file descriptor out from under the thread creating the file object. 2274 */ 2275 void 2276 fdclose(struct thread *td, struct file *fp, int idx) 2277 { 2278 struct filedesc *fdp = td->td_proc->p_fd; 2279 2280 FILEDESC_XLOCK(fdp); 2281 if (fdp->fd_ofiles[idx].fde_file == fp) { 2282 fdfree(fdp, idx); 2283 FILEDESC_XUNLOCK(fdp); 2284 fdrop(fp, td); 2285 } else 2286 FILEDESC_XUNLOCK(fdp); 2287 } 2288 2289 /* 2290 * Close any files on exec? 2291 */ 2292 void 2293 fdcloseexec(struct thread *td) 2294 { 2295 struct filedesc *fdp; 2296 struct filedescent *fde; 2297 struct file *fp; 2298 int i; 2299 2300 fdp = td->td_proc->p_fd; 2301 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2302 for (i = 0; i <= fdp->fd_lastfile; i++) { 2303 fde = &fdp->fd_ofiles[i]; 2304 fp = fde->fde_file; 2305 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2306 (fde->fde_flags & UF_EXCLOSE))) { 2307 FILEDESC_XLOCK(fdp); 2308 fdfree(fdp, i); 2309 (void) closefp(fdp, i, fp, td, 0); 2310 FILEDESC_UNLOCK_ASSERT(fdp); 2311 } 2312 } 2313 } 2314 2315 /* 2316 * It is unsafe for set[ug]id processes to be started with file 2317 * descriptors 0..2 closed, as these descriptors are given implicit 2318 * significance in the Standard C library. fdcheckstd() will create a 2319 * descriptor referencing /dev/null for each of stdin, stdout, and 2320 * stderr that is not already open. 2321 */ 2322 int 2323 fdcheckstd(struct thread *td) 2324 { 2325 struct filedesc *fdp; 2326 register_t save; 2327 int i, error, devnull; 2328 2329 fdp = td->td_proc->p_fd; 2330 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2331 MPASS(fdp->fd_nfiles >= 3); 2332 devnull = -1; 2333 for (i = 0; i <= 2; i++) { 2334 if (fdp->fd_ofiles[i].fde_file != NULL) 2335 continue; 2336 2337 save = td->td_retval[0]; 2338 if (devnull != -1) { 2339 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2340 } else { 2341 error = kern_openat(td, AT_FDCWD, "/dev/null", 2342 UIO_SYSSPACE, O_RDWR, 0); 2343 if (error == 0) { 2344 devnull = td->td_retval[0]; 2345 KASSERT(devnull == i, ("we didn't get our fd")); 2346 } 2347 } 2348 td->td_retval[0] = save; 2349 if (error != 0) 2350 return (error); 2351 } 2352 return (0); 2353 } 2354 2355 /* 2356 * Internal form of close. Decrement reference count on file structure. 2357 * Note: td may be NULL when closing a file that was being passed in a 2358 * message. 2359 * 2360 * XXXRW: Giant is not required for the caller, but often will be held; this 2361 * makes it moderately likely the Giant will be recursed in the VFS case. 2362 */ 2363 int 2364 closef(struct file *fp, struct thread *td) 2365 { 2366 struct vnode *vp; 2367 struct flock lf; 2368 struct filedesc_to_leader *fdtol; 2369 struct filedesc *fdp; 2370 2371 /* 2372 * POSIX record locking dictates that any close releases ALL 2373 * locks owned by this process. This is handled by setting 2374 * a flag in the unlock to free ONLY locks obeying POSIX 2375 * semantics, and not to free BSD-style file locks. 2376 * If the descriptor was in a message, POSIX-style locks 2377 * aren't passed with the descriptor, and the thread pointer 2378 * will be NULL. Callers should be careful only to pass a 2379 * NULL thread pointer when there really is no owning 2380 * context that might have locks, or the locks will be 2381 * leaked. 2382 */ 2383 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2384 vp = fp->f_vnode; 2385 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2386 lf.l_whence = SEEK_SET; 2387 lf.l_start = 0; 2388 lf.l_len = 0; 2389 lf.l_type = F_UNLCK; 2390 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2391 F_UNLCK, &lf, F_POSIX); 2392 } 2393 fdtol = td->td_proc->p_fdtol; 2394 if (fdtol != NULL) { 2395 /* 2396 * Handle special case where file descriptor table is 2397 * shared between multiple process leaders. 2398 */ 2399 fdp = td->td_proc->p_fd; 2400 FILEDESC_XLOCK(fdp); 2401 for (fdtol = fdtol->fdl_next; 2402 fdtol != td->td_proc->p_fdtol; 2403 fdtol = fdtol->fdl_next) { 2404 if ((fdtol->fdl_leader->p_flag & 2405 P_ADVLOCK) == 0) 2406 continue; 2407 fdtol->fdl_holdcount++; 2408 FILEDESC_XUNLOCK(fdp); 2409 lf.l_whence = SEEK_SET; 2410 lf.l_start = 0; 2411 lf.l_len = 0; 2412 lf.l_type = F_UNLCK; 2413 vp = fp->f_vnode; 2414 (void) VOP_ADVLOCK(vp, 2415 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2416 F_POSIX); 2417 FILEDESC_XLOCK(fdp); 2418 fdtol->fdl_holdcount--; 2419 if (fdtol->fdl_holdcount == 0 && 2420 fdtol->fdl_wakeup != 0) { 2421 fdtol->fdl_wakeup = 0; 2422 wakeup(fdtol); 2423 } 2424 } 2425 FILEDESC_XUNLOCK(fdp); 2426 } 2427 } 2428 return (fdrop(fp, td)); 2429 } 2430 2431 /* 2432 * Initialize the file pointer with the specified properties. 2433 * 2434 * The ops are set with release semantics to be certain that the flags, type, 2435 * and data are visible when ops is. This is to prevent ops methods from being 2436 * called with bad data. 2437 */ 2438 void 2439 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2440 { 2441 fp->f_data = data; 2442 fp->f_flag = flag; 2443 fp->f_type = type; 2444 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2445 } 2446 2447 int 2448 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2449 struct file **fpp, seq_t *seqp) 2450 { 2451 #ifdef CAPABILITIES 2452 struct filedescent *fde; 2453 #endif 2454 struct fdescenttbl *fdt; 2455 struct file *fp; 2456 u_int count; 2457 #ifdef CAPABILITIES 2458 seq_t seq; 2459 cap_rights_t haverights; 2460 int error; 2461 #endif 2462 2463 fdt = fdp->fd_files; 2464 if ((u_int)fd >= fdt->fdt_nfiles) 2465 return (EBADF); 2466 /* 2467 * Fetch the descriptor locklessly. We avoid fdrop() races by 2468 * never raising a refcount above 0. To accomplish this we have 2469 * to use a cmpset loop rather than an atomic_add. The descriptor 2470 * must be re-verified once we acquire a reference to be certain 2471 * that the identity is still correct and we did not lose a race 2472 * due to preemption. 2473 */ 2474 for (;;) { 2475 #ifdef CAPABILITIES 2476 seq = seq_read(fd_seq(fdt, fd)); 2477 fde = &fdt->fdt_ofiles[fd]; 2478 haverights = *cap_rights_fde(fde); 2479 fp = fde->fde_file; 2480 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2481 cpu_spinwait(); 2482 continue; 2483 } 2484 #else 2485 fp = fdt->fdt_ofiles[fd].fde_file; 2486 #endif 2487 if (fp == NULL) 2488 return (EBADF); 2489 #ifdef CAPABILITIES 2490 error = cap_check(&haverights, needrightsp); 2491 if (error != 0) 2492 return (error); 2493 #endif 2494 retry: 2495 count = fp->f_count; 2496 if (count == 0) { 2497 /* 2498 * Force a reload. Other thread could reallocate the 2499 * table before this fd was closed, so it possible that 2500 * there is a stale fp pointer in cached version. 2501 */ 2502 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2503 continue; 2504 } 2505 /* 2506 * Use an acquire barrier to force re-reading of fdt so it is 2507 * refreshed for verification. 2508 */ 2509 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2510 goto retry; 2511 fdt = fdp->fd_files; 2512 #ifdef CAPABILITIES 2513 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2514 #else 2515 if (fp == fdt->fdt_ofiles[fd].fde_file) 2516 #endif 2517 break; 2518 fdrop(fp, curthread); 2519 } 2520 *fpp = fp; 2521 if (seqp != NULL) { 2522 #ifdef CAPABILITIES 2523 *seqp = seq; 2524 #endif 2525 } 2526 return (0); 2527 } 2528 2529 /* 2530 * Extract the file pointer associated with the specified descriptor for the 2531 * current user process. 2532 * 2533 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2534 * returned. 2535 * 2536 * File's rights will be checked against the capability rights mask. 2537 * 2538 * If an error occurred the non-zero error is returned and *fpp is set to 2539 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2540 * responsible for fdrop(). 2541 */ 2542 static __inline int 2543 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2544 cap_rights_t *needrightsp, seq_t *seqp) 2545 { 2546 struct filedesc *fdp; 2547 struct file *fp; 2548 int error; 2549 2550 *fpp = NULL; 2551 fdp = td->td_proc->p_fd; 2552 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2553 if (error != 0) 2554 return (error); 2555 if (fp->f_ops == &badfileops) { 2556 fdrop(fp, td); 2557 return (EBADF); 2558 } 2559 2560 /* 2561 * FREAD and FWRITE failure return EBADF as per POSIX. 2562 */ 2563 error = 0; 2564 switch (flags) { 2565 case FREAD: 2566 case FWRITE: 2567 if ((fp->f_flag & flags) == 0) 2568 error = EBADF; 2569 break; 2570 case FEXEC: 2571 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2572 ((fp->f_flag & FWRITE) != 0)) 2573 error = EBADF; 2574 break; 2575 case 0: 2576 break; 2577 default: 2578 KASSERT(0, ("wrong flags")); 2579 } 2580 2581 if (error != 0) { 2582 fdrop(fp, td); 2583 return (error); 2584 } 2585 2586 *fpp = fp; 2587 return (0); 2588 } 2589 2590 int 2591 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2592 { 2593 2594 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2595 } 2596 2597 int 2598 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2599 struct file **fpp) 2600 { 2601 int error; 2602 #ifndef CAPABILITIES 2603 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2604 if (maxprotp != NULL) 2605 *maxprotp = VM_PROT_ALL; 2606 #else 2607 struct filedesc *fdp = td->td_proc->p_fd; 2608 seq_t seq; 2609 2610 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2611 for (;;) { 2612 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2613 if (error != 0) 2614 return (error); 2615 /* 2616 * If requested, convert capability rights to access flags. 2617 */ 2618 if (maxprotp != NULL) 2619 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2620 if (!fd_modified(fdp, fd, seq)) 2621 break; 2622 fdrop(*fpp, td); 2623 } 2624 #endif 2625 return (error); 2626 } 2627 2628 int 2629 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2630 { 2631 2632 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2633 } 2634 2635 int 2636 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2637 { 2638 2639 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2640 } 2641 2642 int 2643 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2644 struct file **fpp) 2645 { 2646 struct filedesc *fdp = td->td_proc->p_fd; 2647 #ifndef CAPABILITIES 2648 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2649 #else 2650 int error; 2651 seq_t seq; 2652 2653 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2654 for (;;) { 2655 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2656 if (error != 0) 2657 return (error); 2658 error = cap_fcntl_check(fdp, fd, needfcntl); 2659 if (!fd_modified(fdp, fd, seq)) 2660 break; 2661 fdrop(*fpp, td); 2662 } 2663 if (error != 0) { 2664 fdrop(*fpp, td); 2665 *fpp = NULL; 2666 } 2667 return (error); 2668 #endif 2669 } 2670 2671 /* 2672 * Like fget() but loads the underlying vnode, or returns an error if the 2673 * descriptor does not represent a vnode. Note that pipes use vnodes but 2674 * never have VM objects. The returned vnode will be vref()'d. 2675 * 2676 * XXX: what about the unused flags ? 2677 */ 2678 static __inline int 2679 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2680 struct vnode **vpp) 2681 { 2682 struct file *fp; 2683 int error; 2684 2685 *vpp = NULL; 2686 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2687 if (error != 0) 2688 return (error); 2689 if (fp->f_vnode == NULL) { 2690 error = EINVAL; 2691 } else { 2692 *vpp = fp->f_vnode; 2693 vref(*vpp); 2694 } 2695 fdrop(fp, td); 2696 2697 return (error); 2698 } 2699 2700 int 2701 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2702 { 2703 2704 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2705 } 2706 2707 int 2708 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2709 struct filecaps *havecaps, struct vnode **vpp) 2710 { 2711 struct filedesc *fdp; 2712 struct file *fp; 2713 #ifdef CAPABILITIES 2714 int error; 2715 #endif 2716 2717 fdp = td->td_proc->p_fd; 2718 fp = fget_locked(fdp, fd); 2719 if (fp == NULL || fp->f_ops == &badfileops) 2720 return (EBADF); 2721 2722 #ifdef CAPABILITIES 2723 error = cap_check(cap_rights(fdp, fd), needrightsp); 2724 if (error != 0) 2725 return (error); 2726 #endif 2727 2728 if (fp->f_vnode == NULL) 2729 return (EINVAL); 2730 2731 *vpp = fp->f_vnode; 2732 vref(*vpp); 2733 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps, true); 2734 2735 return (0); 2736 } 2737 2738 int 2739 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2740 { 2741 2742 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2743 } 2744 2745 int 2746 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2747 { 2748 2749 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2750 } 2751 2752 #ifdef notyet 2753 int 2754 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2755 struct vnode **vpp) 2756 { 2757 2758 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2759 } 2760 #endif 2761 2762 /* 2763 * Like fget() but loads the underlying socket, or returns an error if the 2764 * descriptor does not represent a socket. 2765 * 2766 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2767 * in the future. 2768 * 2769 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2770 * on their file descriptor reference to prevent the socket from being free'd 2771 * during use. 2772 */ 2773 int 2774 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2775 u_int *fflagp) 2776 { 2777 struct file *fp; 2778 int error; 2779 2780 *spp = NULL; 2781 if (fflagp != NULL) 2782 *fflagp = 0; 2783 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2784 return (error); 2785 if (fp->f_type != DTYPE_SOCKET) { 2786 error = ENOTSOCK; 2787 } else { 2788 *spp = fp->f_data; 2789 if (fflagp) 2790 *fflagp = fp->f_flag; 2791 SOCK_LOCK(*spp); 2792 soref(*spp); 2793 SOCK_UNLOCK(*spp); 2794 } 2795 fdrop(fp, td); 2796 2797 return (error); 2798 } 2799 2800 /* 2801 * Drop the reference count on the socket and XXX release the SX lock in the 2802 * future. The last reference closes the socket. 2803 * 2804 * Note: fputsock() is deprecated, see comment for fgetsock(). 2805 */ 2806 void 2807 fputsock(struct socket *so) 2808 { 2809 2810 ACCEPT_LOCK(); 2811 SOCK_LOCK(so); 2812 CURVNET_SET(so->so_vnet); 2813 sorele(so); 2814 CURVNET_RESTORE(); 2815 } 2816 2817 /* 2818 * Handle the last reference to a file being closed. 2819 */ 2820 int 2821 _fdrop(struct file *fp, struct thread *td) 2822 { 2823 int error; 2824 2825 if (fp->f_count != 0) 2826 panic("fdrop: count %d", fp->f_count); 2827 error = fo_close(fp, td); 2828 atomic_subtract_int(&openfiles, 1); 2829 crfree(fp->f_cred); 2830 free(fp->f_advice, M_FADVISE); 2831 uma_zfree(file_zone, fp); 2832 2833 return (error); 2834 } 2835 2836 /* 2837 * Apply an advisory lock on a file descriptor. 2838 * 2839 * Just attempt to get a record lock of the requested type on the entire file 2840 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2841 */ 2842 #ifndef _SYS_SYSPROTO_H_ 2843 struct flock_args { 2844 int fd; 2845 int how; 2846 }; 2847 #endif 2848 /* ARGSUSED */ 2849 int 2850 sys_flock(struct thread *td, struct flock_args *uap) 2851 { 2852 struct file *fp; 2853 struct vnode *vp; 2854 struct flock lf; 2855 cap_rights_t rights; 2856 int error; 2857 2858 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2859 if (error != 0) 2860 return (error); 2861 if (fp->f_type != DTYPE_VNODE) { 2862 fdrop(fp, td); 2863 return (EOPNOTSUPP); 2864 } 2865 2866 vp = fp->f_vnode; 2867 lf.l_whence = SEEK_SET; 2868 lf.l_start = 0; 2869 lf.l_len = 0; 2870 if (uap->how & LOCK_UN) { 2871 lf.l_type = F_UNLCK; 2872 atomic_clear_int(&fp->f_flag, FHASLOCK); 2873 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2874 goto done2; 2875 } 2876 if (uap->how & LOCK_EX) 2877 lf.l_type = F_WRLCK; 2878 else if (uap->how & LOCK_SH) 2879 lf.l_type = F_RDLCK; 2880 else { 2881 error = EBADF; 2882 goto done2; 2883 } 2884 atomic_set_int(&fp->f_flag, FHASLOCK); 2885 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2886 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2887 done2: 2888 fdrop(fp, td); 2889 return (error); 2890 } 2891 /* 2892 * Duplicate the specified descriptor to a free descriptor. 2893 */ 2894 int 2895 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2896 int openerror, int *indxp) 2897 { 2898 struct filedescent *newfde, *oldfde; 2899 struct file *fp; 2900 int error, indx; 2901 2902 KASSERT(openerror == ENODEV || openerror == ENXIO, 2903 ("unexpected error %d in %s", openerror, __func__)); 2904 2905 /* 2906 * If the to-be-dup'd fd number is greater than the allowed number 2907 * of file descriptors, or the fd to be dup'd has already been 2908 * closed, then reject. 2909 */ 2910 FILEDESC_XLOCK(fdp); 2911 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2912 FILEDESC_XUNLOCK(fdp); 2913 return (EBADF); 2914 } 2915 2916 error = fdalloc(td, 0, &indx); 2917 if (error != 0) { 2918 FILEDESC_XUNLOCK(fdp); 2919 return (error); 2920 } 2921 2922 /* 2923 * There are two cases of interest here. 2924 * 2925 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2926 * 2927 * For ENXIO steal away the file structure from (dfd) and store it in 2928 * (indx). (dfd) is effectively closed by this operation. 2929 */ 2930 switch (openerror) { 2931 case ENODEV: 2932 /* 2933 * Check that the mode the file is being opened for is a 2934 * subset of the mode of the existing descriptor. 2935 */ 2936 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2937 fdunused(fdp, indx); 2938 FILEDESC_XUNLOCK(fdp); 2939 return (EACCES); 2940 } 2941 fhold(fp); 2942 newfde = &fdp->fd_ofiles[indx]; 2943 oldfde = &fdp->fd_ofiles[dfd]; 2944 #ifdef CAPABILITIES 2945 seq_write_begin(&newfde->fde_seq); 2946 #endif 2947 memcpy(newfde, oldfde, fde_change_size); 2948 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); 2949 #ifdef CAPABILITIES 2950 seq_write_end(&newfde->fde_seq); 2951 #endif 2952 break; 2953 case ENXIO: 2954 /* 2955 * Steal away the file pointer from dfd and stuff it into indx. 2956 */ 2957 newfde = &fdp->fd_ofiles[indx]; 2958 oldfde = &fdp->fd_ofiles[dfd]; 2959 #ifdef CAPABILITIES 2960 seq_write_begin(&newfde->fde_seq); 2961 #endif 2962 memcpy(newfde, oldfde, fde_change_size); 2963 oldfde->fde_file = NULL; 2964 fdunused(fdp, dfd); 2965 #ifdef CAPABILITIES 2966 seq_write_end(&newfde->fde_seq); 2967 #endif 2968 break; 2969 } 2970 FILEDESC_XUNLOCK(fdp); 2971 *indxp = indx; 2972 return (0); 2973 } 2974 2975 /* 2976 * This sysctl determines if we will allow a process to chroot(2) if it 2977 * has a directory open: 2978 * 0: disallowed for all processes. 2979 * 1: allowed for processes that were not already chroot(2)'ed. 2980 * 2: allowed for all processes. 2981 */ 2982 2983 static int chroot_allow_open_directories = 1; 2984 2985 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 2986 &chroot_allow_open_directories, 0, 2987 "Allow a process to chroot(2) if it has a directory open"); 2988 2989 /* 2990 * Helper function for raised chroot(2) security function: Refuse if 2991 * any filedescriptors are open directories. 2992 */ 2993 static int 2994 chroot_refuse_vdir_fds(struct filedesc *fdp) 2995 { 2996 struct vnode *vp; 2997 struct file *fp; 2998 int fd; 2999 3000 FILEDESC_LOCK_ASSERT(fdp); 3001 3002 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 3003 fp = fget_locked(fdp, fd); 3004 if (fp == NULL) 3005 continue; 3006 if (fp->f_type == DTYPE_VNODE) { 3007 vp = fp->f_vnode; 3008 if (vp->v_type == VDIR) 3009 return (EPERM); 3010 } 3011 } 3012 return (0); 3013 } 3014 3015 /* 3016 * Common routine for kern_chroot() and jail_attach(). The caller is 3017 * responsible for invoking priv_check() and mac_vnode_check_chroot() to 3018 * authorize this operation. 3019 */ 3020 int 3021 pwd_chroot(struct thread *td, struct vnode *vp) 3022 { 3023 struct filedesc *fdp; 3024 struct vnode *oldvp; 3025 int error; 3026 3027 fdp = td->td_proc->p_fd; 3028 FILEDESC_XLOCK(fdp); 3029 if (chroot_allow_open_directories == 0 || 3030 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 3031 error = chroot_refuse_vdir_fds(fdp); 3032 if (error != 0) { 3033 FILEDESC_XUNLOCK(fdp); 3034 return (error); 3035 } 3036 } 3037 oldvp = fdp->fd_rdir; 3038 VREF(vp); 3039 fdp->fd_rdir = vp; 3040 if (fdp->fd_jdir == NULL) { 3041 VREF(vp); 3042 fdp->fd_jdir = vp; 3043 } 3044 FILEDESC_XUNLOCK(fdp); 3045 vrele(oldvp); 3046 return (0); 3047 } 3048 3049 void 3050 pwd_chdir(struct thread *td, struct vnode *vp) 3051 { 3052 struct filedesc *fdp; 3053 struct vnode *oldvp; 3054 3055 fdp = td->td_proc->p_fd; 3056 FILEDESC_XLOCK(fdp); 3057 VNASSERT(vp->v_usecount > 0, vp, 3058 ("chdir to a vnode with zero usecount")); 3059 oldvp = fdp->fd_cdir; 3060 fdp->fd_cdir = vp; 3061 FILEDESC_XUNLOCK(fdp); 3062 vrele(oldvp); 3063 } 3064 3065 /* 3066 * Scan all active processes and prisons to see if any of them have a current 3067 * or root directory of `olddp'. If so, replace them with the new mount point. 3068 */ 3069 void 3070 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3071 { 3072 struct filedesc *fdp; 3073 struct prison *pr; 3074 struct proc *p; 3075 int nrele; 3076 3077 if (vrefcnt(olddp) == 1) 3078 return; 3079 nrele = 0; 3080 sx_slock(&allproc_lock); 3081 FOREACH_PROC_IN_SYSTEM(p) { 3082 PROC_LOCK(p); 3083 fdp = fdhold(p); 3084 PROC_UNLOCK(p); 3085 if (fdp == NULL) 3086 continue; 3087 FILEDESC_XLOCK(fdp); 3088 if (fdp->fd_cdir == olddp) { 3089 vref(newdp); 3090 fdp->fd_cdir = newdp; 3091 nrele++; 3092 } 3093 if (fdp->fd_rdir == olddp) { 3094 vref(newdp); 3095 fdp->fd_rdir = newdp; 3096 nrele++; 3097 } 3098 if (fdp->fd_jdir == olddp) { 3099 vref(newdp); 3100 fdp->fd_jdir = newdp; 3101 nrele++; 3102 } 3103 FILEDESC_XUNLOCK(fdp); 3104 fddrop(fdp); 3105 } 3106 sx_sunlock(&allproc_lock); 3107 if (rootvnode == olddp) { 3108 vref(newdp); 3109 rootvnode = newdp; 3110 nrele++; 3111 } 3112 mtx_lock(&prison0.pr_mtx); 3113 if (prison0.pr_root == olddp) { 3114 vref(newdp); 3115 prison0.pr_root = newdp; 3116 nrele++; 3117 } 3118 mtx_unlock(&prison0.pr_mtx); 3119 sx_slock(&allprison_lock); 3120 TAILQ_FOREACH(pr, &allprison, pr_list) { 3121 mtx_lock(&pr->pr_mtx); 3122 if (pr->pr_root == olddp) { 3123 vref(newdp); 3124 pr->pr_root = newdp; 3125 nrele++; 3126 } 3127 mtx_unlock(&pr->pr_mtx); 3128 } 3129 sx_sunlock(&allprison_lock); 3130 while (nrele--) 3131 vrele(olddp); 3132 } 3133 3134 struct filedesc_to_leader * 3135 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 3136 { 3137 struct filedesc_to_leader *fdtol; 3138 3139 fdtol = malloc(sizeof(struct filedesc_to_leader), 3140 M_FILEDESC_TO_LEADER, M_WAITOK); 3141 fdtol->fdl_refcount = 1; 3142 fdtol->fdl_holdcount = 0; 3143 fdtol->fdl_wakeup = 0; 3144 fdtol->fdl_leader = leader; 3145 if (old != NULL) { 3146 FILEDESC_XLOCK(fdp); 3147 fdtol->fdl_next = old->fdl_next; 3148 fdtol->fdl_prev = old; 3149 old->fdl_next = fdtol; 3150 fdtol->fdl_next->fdl_prev = fdtol; 3151 FILEDESC_XUNLOCK(fdp); 3152 } else { 3153 fdtol->fdl_next = fdtol; 3154 fdtol->fdl_prev = fdtol; 3155 } 3156 return (fdtol); 3157 } 3158 3159 static int 3160 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 3161 { 3162 struct filedesc *fdp; 3163 int i, count, slots; 3164 3165 if (*(int *)arg1 != 0) 3166 return (EINVAL); 3167 3168 fdp = curproc->p_fd; 3169 count = 0; 3170 FILEDESC_SLOCK(fdp); 3171 slots = NDSLOTS(fdp->fd_lastfile + 1); 3172 for (i = 0; i < slots; i++) 3173 count += bitcountl(fdp->fd_map[i]); 3174 FILEDESC_SUNLOCK(fdp); 3175 3176 return (SYSCTL_OUT(req, &count, sizeof(count))); 3177 } 3178 3179 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 3180 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 3181 "Number of open file descriptors"); 3182 3183 /* 3184 * Get file structures globally. 3185 */ 3186 static int 3187 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 3188 { 3189 struct xfile xf; 3190 struct filedesc *fdp; 3191 struct file *fp; 3192 struct proc *p; 3193 int error, n; 3194 3195 error = sysctl_wire_old_buffer(req, 0); 3196 if (error != 0) 3197 return (error); 3198 if (req->oldptr == NULL) { 3199 n = 0; 3200 sx_slock(&allproc_lock); 3201 FOREACH_PROC_IN_SYSTEM(p) { 3202 PROC_LOCK(p); 3203 if (p->p_state == PRS_NEW) { 3204 PROC_UNLOCK(p); 3205 continue; 3206 } 3207 fdp = fdhold(p); 3208 PROC_UNLOCK(p); 3209 if (fdp == NULL) 3210 continue; 3211 /* overestimates sparse tables. */ 3212 if (fdp->fd_lastfile > 0) 3213 n += fdp->fd_lastfile; 3214 fddrop(fdp); 3215 } 3216 sx_sunlock(&allproc_lock); 3217 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3218 } 3219 error = 0; 3220 bzero(&xf, sizeof(xf)); 3221 xf.xf_size = sizeof(xf); 3222 sx_slock(&allproc_lock); 3223 FOREACH_PROC_IN_SYSTEM(p) { 3224 PROC_LOCK(p); 3225 if (p->p_state == PRS_NEW) { 3226 PROC_UNLOCK(p); 3227 continue; 3228 } 3229 if (p_cansee(req->td, p) != 0) { 3230 PROC_UNLOCK(p); 3231 continue; 3232 } 3233 xf.xf_pid = p->p_pid; 3234 xf.xf_uid = p->p_ucred->cr_uid; 3235 fdp = fdhold(p); 3236 PROC_UNLOCK(p); 3237 if (fdp == NULL) 3238 continue; 3239 FILEDESC_SLOCK(fdp); 3240 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3241 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3242 continue; 3243 xf.xf_fd = n; 3244 xf.xf_file = fp; 3245 xf.xf_data = fp->f_data; 3246 xf.xf_vnode = fp->f_vnode; 3247 xf.xf_type = fp->f_type; 3248 xf.xf_count = fp->f_count; 3249 xf.xf_msgcount = 0; 3250 xf.xf_offset = foffset_get(fp); 3251 xf.xf_flag = fp->f_flag; 3252 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3253 if (error) 3254 break; 3255 } 3256 FILEDESC_SUNLOCK(fdp); 3257 fddrop(fdp); 3258 if (error) 3259 break; 3260 } 3261 sx_sunlock(&allproc_lock); 3262 return (error); 3263 } 3264 3265 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3266 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3267 3268 #ifdef KINFO_FILE_SIZE 3269 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3270 #endif 3271 3272 static int 3273 xlate_fflags(int fflags) 3274 { 3275 static const struct { 3276 int fflag; 3277 int kf_fflag; 3278 } fflags_table[] = { 3279 { FAPPEND, KF_FLAG_APPEND }, 3280 { FASYNC, KF_FLAG_ASYNC }, 3281 { FFSYNC, KF_FLAG_FSYNC }, 3282 { FHASLOCK, KF_FLAG_HASLOCK }, 3283 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3284 { FREAD, KF_FLAG_READ }, 3285 { FWRITE, KF_FLAG_WRITE }, 3286 { O_CREAT, KF_FLAG_CREAT }, 3287 { O_DIRECT, KF_FLAG_DIRECT }, 3288 { O_EXCL, KF_FLAG_EXCL }, 3289 { O_EXEC, KF_FLAG_EXEC }, 3290 { O_EXLOCK, KF_FLAG_EXLOCK }, 3291 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3292 { O_SHLOCK, KF_FLAG_SHLOCK }, 3293 { O_TRUNC, KF_FLAG_TRUNC } 3294 }; 3295 unsigned int i; 3296 int kflags; 3297 3298 kflags = 0; 3299 for (i = 0; i < nitems(fflags_table); i++) 3300 if (fflags & fflags_table[i].fflag) 3301 kflags |= fflags_table[i].kf_fflag; 3302 return (kflags); 3303 } 3304 3305 /* Trim unused data from kf_path by truncating the structure size. */ 3306 static void 3307 pack_kinfo(struct kinfo_file *kif) 3308 { 3309 3310 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3311 strlen(kif->kf_path) + 1; 3312 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3313 } 3314 3315 static void 3316 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3317 struct kinfo_file *kif, struct filedesc *fdp, int flags) 3318 { 3319 int error; 3320 3321 bzero(kif, sizeof(*kif)); 3322 3323 /* Set a default type to allow for empty fill_kinfo() methods. */ 3324 kif->kf_type = KF_TYPE_UNKNOWN; 3325 kif->kf_flags = xlate_fflags(fp->f_flag); 3326 if (rightsp != NULL) 3327 kif->kf_cap_rights = *rightsp; 3328 else 3329 cap_rights_init(&kif->kf_cap_rights); 3330 kif->kf_fd = fd; 3331 kif->kf_ref_count = fp->f_count; 3332 kif->kf_offset = foffset_get(fp); 3333 3334 /* 3335 * This may drop the filedesc lock, so the 'fp' cannot be 3336 * accessed after this call. 3337 */ 3338 error = fo_fill_kinfo(fp, kif, fdp); 3339 if (error == 0) 3340 kif->kf_status |= KF_ATTR_VALID; 3341 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3342 pack_kinfo(kif); 3343 else 3344 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3345 } 3346 3347 static void 3348 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3349 struct kinfo_file *kif, int flags) 3350 { 3351 int error; 3352 3353 bzero(kif, sizeof(*kif)); 3354 3355 kif->kf_type = KF_TYPE_VNODE; 3356 error = vn_fill_kinfo_vnode(vp, kif); 3357 if (error == 0) 3358 kif->kf_status |= KF_ATTR_VALID; 3359 kif->kf_flags = xlate_fflags(fflags); 3360 cap_rights_init(&kif->kf_cap_rights); 3361 kif->kf_fd = fd; 3362 kif->kf_ref_count = -1; 3363 kif->kf_offset = -1; 3364 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3365 pack_kinfo(kif); 3366 else 3367 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3368 vrele(vp); 3369 } 3370 3371 struct export_fd_buf { 3372 struct filedesc *fdp; 3373 struct sbuf *sb; 3374 ssize_t remainder; 3375 struct kinfo_file kif; 3376 int flags; 3377 }; 3378 3379 static int 3380 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3381 { 3382 struct kinfo_file *kif; 3383 3384 kif = &efbuf->kif; 3385 if (efbuf->remainder != -1) { 3386 if (efbuf->remainder < kif->kf_structsize) { 3387 /* Terminate export. */ 3388 efbuf->remainder = 0; 3389 return (0); 3390 } 3391 efbuf->remainder -= kif->kf_structsize; 3392 } 3393 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3394 } 3395 3396 static int 3397 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3398 struct export_fd_buf *efbuf) 3399 { 3400 int error; 3401 3402 if (efbuf->remainder == 0) 3403 return (0); 3404 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 3405 efbuf->flags); 3406 FILEDESC_SUNLOCK(efbuf->fdp); 3407 error = export_kinfo_to_sb(efbuf); 3408 FILEDESC_SLOCK(efbuf->fdp); 3409 return (error); 3410 } 3411 3412 static int 3413 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3414 struct export_fd_buf *efbuf) 3415 { 3416 int error; 3417 3418 if (efbuf->remainder == 0) 3419 return (0); 3420 if (efbuf->fdp != NULL) 3421 FILEDESC_SUNLOCK(efbuf->fdp); 3422 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 3423 error = export_kinfo_to_sb(efbuf); 3424 if (efbuf->fdp != NULL) 3425 FILEDESC_SLOCK(efbuf->fdp); 3426 return (error); 3427 } 3428 3429 /* 3430 * Store a process file descriptor information to sbuf. 3431 * 3432 * Takes a locked proc as argument, and returns with the proc unlocked. 3433 */ 3434 int 3435 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 3436 int flags) 3437 { 3438 struct file *fp; 3439 struct filedesc *fdp; 3440 struct export_fd_buf *efbuf; 3441 struct vnode *cttyvp, *textvp, *tracevp; 3442 int error, i; 3443 cap_rights_t rights; 3444 3445 PROC_LOCK_ASSERT(p, MA_OWNED); 3446 3447 /* ktrace vnode */ 3448 tracevp = p->p_tracevp; 3449 if (tracevp != NULL) 3450 vref(tracevp); 3451 /* text vnode */ 3452 textvp = p->p_textvp; 3453 if (textvp != NULL) 3454 vref(textvp); 3455 /* Controlling tty. */ 3456 cttyvp = NULL; 3457 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3458 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3459 if (cttyvp != NULL) 3460 vref(cttyvp); 3461 } 3462 fdp = fdhold(p); 3463 PROC_UNLOCK(p); 3464 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3465 efbuf->fdp = NULL; 3466 efbuf->sb = sb; 3467 efbuf->remainder = maxlen; 3468 efbuf->flags = flags; 3469 if (tracevp != NULL) 3470 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3471 efbuf); 3472 if (textvp != NULL) 3473 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3474 if (cttyvp != NULL) 3475 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3476 efbuf); 3477 error = 0; 3478 if (fdp == NULL) 3479 goto fail; 3480 efbuf->fdp = fdp; 3481 FILEDESC_SLOCK(fdp); 3482 /* working directory */ 3483 if (fdp->fd_cdir != NULL) { 3484 vref(fdp->fd_cdir); 3485 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3486 } 3487 /* root directory */ 3488 if (fdp->fd_rdir != NULL) { 3489 vref(fdp->fd_rdir); 3490 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3491 } 3492 /* jail directory */ 3493 if (fdp->fd_jdir != NULL) { 3494 vref(fdp->fd_jdir); 3495 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3496 } 3497 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3498 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3499 continue; 3500 #ifdef CAPABILITIES 3501 rights = *cap_rights(fdp, i); 3502 #else /* !CAPABILITIES */ 3503 cap_rights_init(&rights); 3504 #endif 3505 /* 3506 * Create sysctl entry. It is OK to drop the filedesc 3507 * lock inside of export_file_to_sb() as we will 3508 * re-validate and re-evaluate its properties when the 3509 * loop continues. 3510 */ 3511 error = export_file_to_sb(fp, i, &rights, efbuf); 3512 if (error != 0 || efbuf->remainder == 0) 3513 break; 3514 } 3515 FILEDESC_SUNLOCK(fdp); 3516 fddrop(fdp); 3517 fail: 3518 free(efbuf, M_TEMP); 3519 return (error); 3520 } 3521 3522 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3523 3524 /* 3525 * Get per-process file descriptors for use by procstat(1), et al. 3526 */ 3527 static int 3528 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3529 { 3530 struct sbuf sb; 3531 struct proc *p; 3532 ssize_t maxlen; 3533 int error, error2, *name; 3534 3535 name = (int *)arg1; 3536 3537 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3538 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3539 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3540 if (error != 0) { 3541 sbuf_delete(&sb); 3542 return (error); 3543 } 3544 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3545 error = kern_proc_filedesc_out(p, &sb, maxlen, 3546 KERN_FILEDESC_PACK_KINFO); 3547 error2 = sbuf_finish(&sb); 3548 sbuf_delete(&sb); 3549 return (error != 0 ? error : error2); 3550 } 3551 3552 #ifdef KINFO_OFILE_SIZE 3553 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3554 #endif 3555 3556 #ifdef COMPAT_FREEBSD7 3557 static void 3558 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3559 { 3560 3561 okif->kf_structsize = sizeof(*okif); 3562 okif->kf_type = kif->kf_type; 3563 okif->kf_fd = kif->kf_fd; 3564 okif->kf_ref_count = kif->kf_ref_count; 3565 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3566 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3567 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3568 okif->kf_offset = kif->kf_offset; 3569 okif->kf_vnode_type = kif->kf_vnode_type; 3570 okif->kf_sock_domain = kif->kf_sock_domain; 3571 okif->kf_sock_type = kif->kf_sock_type; 3572 okif->kf_sock_protocol = kif->kf_sock_protocol; 3573 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3574 okif->kf_sa_local = kif->kf_sa_local; 3575 okif->kf_sa_peer = kif->kf_sa_peer; 3576 } 3577 3578 static int 3579 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3580 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3581 { 3582 int error; 3583 3584 vref(vp); 3585 FILEDESC_SUNLOCK(fdp); 3586 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 3587 kinfo_to_okinfo(kif, okif); 3588 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3589 FILEDESC_SLOCK(fdp); 3590 return (error); 3591 } 3592 3593 /* 3594 * Get per-process file descriptors for use by procstat(1), et al. 3595 */ 3596 static int 3597 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3598 { 3599 struct kinfo_ofile *okif; 3600 struct kinfo_file *kif; 3601 struct filedesc *fdp; 3602 int error, i, *name; 3603 struct file *fp; 3604 struct proc *p; 3605 3606 name = (int *)arg1; 3607 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3608 if (error != 0) 3609 return (error); 3610 fdp = fdhold(p); 3611 PROC_UNLOCK(p); 3612 if (fdp == NULL) 3613 return (ENOENT); 3614 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3615 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3616 FILEDESC_SLOCK(fdp); 3617 if (fdp->fd_cdir != NULL) 3618 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3619 okif, fdp, req); 3620 if (fdp->fd_rdir != NULL) 3621 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3622 okif, fdp, req); 3623 if (fdp->fd_jdir != NULL) 3624 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3625 okif, fdp, req); 3626 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3627 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3628 continue; 3629 export_file_to_kinfo(fp, i, NULL, kif, fdp, 3630 KERN_FILEDESC_PACK_KINFO); 3631 FILEDESC_SUNLOCK(fdp); 3632 kinfo_to_okinfo(kif, okif); 3633 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3634 FILEDESC_SLOCK(fdp); 3635 if (error) 3636 break; 3637 } 3638 FILEDESC_SUNLOCK(fdp); 3639 fddrop(fdp); 3640 free(kif, M_TEMP); 3641 free(okif, M_TEMP); 3642 return (0); 3643 } 3644 3645 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3646 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3647 "Process ofiledesc entries"); 3648 #endif /* COMPAT_FREEBSD7 */ 3649 3650 int 3651 vntype_to_kinfo(int vtype) 3652 { 3653 struct { 3654 int vtype; 3655 int kf_vtype; 3656 } vtypes_table[] = { 3657 { VBAD, KF_VTYPE_VBAD }, 3658 { VBLK, KF_VTYPE_VBLK }, 3659 { VCHR, KF_VTYPE_VCHR }, 3660 { VDIR, KF_VTYPE_VDIR }, 3661 { VFIFO, KF_VTYPE_VFIFO }, 3662 { VLNK, KF_VTYPE_VLNK }, 3663 { VNON, KF_VTYPE_VNON }, 3664 { VREG, KF_VTYPE_VREG }, 3665 { VSOCK, KF_VTYPE_VSOCK } 3666 }; 3667 unsigned int i; 3668 3669 /* 3670 * Perform vtype translation. 3671 */ 3672 for (i = 0; i < nitems(vtypes_table); i++) 3673 if (vtypes_table[i].vtype == vtype) 3674 return (vtypes_table[i].kf_vtype); 3675 3676 return (KF_VTYPE_UNKNOWN); 3677 } 3678 3679 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3680 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3681 "Process filedesc entries"); 3682 3683 /* 3684 * Store a process current working directory information to sbuf. 3685 * 3686 * Takes a locked proc as argument, and returns with the proc unlocked. 3687 */ 3688 int 3689 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3690 { 3691 struct filedesc *fdp; 3692 struct export_fd_buf *efbuf; 3693 int error; 3694 3695 PROC_LOCK_ASSERT(p, MA_OWNED); 3696 3697 fdp = fdhold(p); 3698 PROC_UNLOCK(p); 3699 if (fdp == NULL) 3700 return (EINVAL); 3701 3702 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3703 efbuf->fdp = fdp; 3704 efbuf->sb = sb; 3705 efbuf->remainder = maxlen; 3706 3707 FILEDESC_SLOCK(fdp); 3708 if (fdp->fd_cdir == NULL) 3709 error = EINVAL; 3710 else { 3711 vref(fdp->fd_cdir); 3712 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3713 FREAD, efbuf); 3714 } 3715 FILEDESC_SUNLOCK(fdp); 3716 fddrop(fdp); 3717 free(efbuf, M_TEMP); 3718 return (error); 3719 } 3720 3721 /* 3722 * Get per-process current working directory. 3723 */ 3724 static int 3725 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3726 { 3727 struct sbuf sb; 3728 struct proc *p; 3729 ssize_t maxlen; 3730 int error, error2, *name; 3731 3732 name = (int *)arg1; 3733 3734 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3735 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3736 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3737 if (error != 0) { 3738 sbuf_delete(&sb); 3739 return (error); 3740 } 3741 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3742 error = kern_proc_cwd_out(p, &sb, maxlen); 3743 error2 = sbuf_finish(&sb); 3744 sbuf_delete(&sb); 3745 return (error != 0 ? error : error2); 3746 } 3747 3748 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3749 sysctl_kern_proc_cwd, "Process current working directory"); 3750 3751 #ifdef DDB 3752 /* 3753 * For the purposes of debugging, generate a human-readable string for the 3754 * file type. 3755 */ 3756 static const char * 3757 file_type_to_name(short type) 3758 { 3759 3760 switch (type) { 3761 case 0: 3762 return ("zero"); 3763 case DTYPE_VNODE: 3764 return ("vnod"); 3765 case DTYPE_SOCKET: 3766 return ("sock"); 3767 case DTYPE_PIPE: 3768 return ("pipe"); 3769 case DTYPE_FIFO: 3770 return ("fifo"); 3771 case DTYPE_KQUEUE: 3772 return ("kque"); 3773 case DTYPE_CRYPTO: 3774 return ("crpt"); 3775 case DTYPE_MQUEUE: 3776 return ("mque"); 3777 case DTYPE_SHM: 3778 return ("shm"); 3779 case DTYPE_SEM: 3780 return ("ksem"); 3781 default: 3782 return ("unkn"); 3783 } 3784 } 3785 3786 /* 3787 * For the purposes of debugging, identify a process (if any, perhaps one of 3788 * many) that references the passed file in its file descriptor array. Return 3789 * NULL if none. 3790 */ 3791 static struct proc * 3792 file_to_first_proc(struct file *fp) 3793 { 3794 struct filedesc *fdp; 3795 struct proc *p; 3796 int n; 3797 3798 FOREACH_PROC_IN_SYSTEM(p) { 3799 if (p->p_state == PRS_NEW) 3800 continue; 3801 fdp = p->p_fd; 3802 if (fdp == NULL) 3803 continue; 3804 for (n = 0; n <= fdp->fd_lastfile; n++) { 3805 if (fp == fdp->fd_ofiles[n].fde_file) 3806 return (p); 3807 } 3808 } 3809 return (NULL); 3810 } 3811 3812 static void 3813 db_print_file(struct file *fp, int header) 3814 { 3815 struct proc *p; 3816 3817 if (header) 3818 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3819 "File", "Type", "Data", "Flag", "GCFl", "Count", 3820 "MCount", "Vnode", "FPID", "FCmd"); 3821 p = file_to_first_proc(fp); 3822 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3823 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3824 0, fp->f_count, 0, fp->f_vnode, 3825 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3826 } 3827 3828 DB_SHOW_COMMAND(file, db_show_file) 3829 { 3830 struct file *fp; 3831 3832 if (!have_addr) { 3833 db_printf("usage: show file <addr>\n"); 3834 return; 3835 } 3836 fp = (struct file *)addr; 3837 db_print_file(fp, 1); 3838 } 3839 3840 DB_SHOW_COMMAND(files, db_show_files) 3841 { 3842 struct filedesc *fdp; 3843 struct file *fp; 3844 struct proc *p; 3845 int header; 3846 int n; 3847 3848 header = 1; 3849 FOREACH_PROC_IN_SYSTEM(p) { 3850 if (p->p_state == PRS_NEW) 3851 continue; 3852 if ((fdp = p->p_fd) == NULL) 3853 continue; 3854 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3855 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3856 continue; 3857 db_print_file(fp, header); 3858 header = 0; 3859 } 3860 } 3861 } 3862 #endif 3863 3864 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3865 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3866 3867 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3868 &maxfiles, 0, "Maximum number of files"); 3869 3870 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3871 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3872 3873 /* ARGSUSED*/ 3874 static void 3875 filelistinit(void *dummy) 3876 { 3877 3878 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3879 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3880 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3881 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3882 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3883 } 3884 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3885 3886 /*-------------------------------------------------------------------*/ 3887 3888 static int 3889 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3890 int flags, struct thread *td) 3891 { 3892 3893 return (EBADF); 3894 } 3895 3896 static int 3897 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3898 struct thread *td) 3899 { 3900 3901 return (EINVAL); 3902 } 3903 3904 static int 3905 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3906 struct thread *td) 3907 { 3908 3909 return (EBADF); 3910 } 3911 3912 static int 3913 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3914 struct thread *td) 3915 { 3916 3917 return (0); 3918 } 3919 3920 static int 3921 badfo_kqfilter(struct file *fp, struct knote *kn) 3922 { 3923 3924 return (EBADF); 3925 } 3926 3927 static int 3928 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3929 struct thread *td) 3930 { 3931 3932 return (EBADF); 3933 } 3934 3935 static int 3936 badfo_close(struct file *fp, struct thread *td) 3937 { 3938 3939 return (0); 3940 } 3941 3942 static int 3943 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3944 struct thread *td) 3945 { 3946 3947 return (EBADF); 3948 } 3949 3950 static int 3951 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3952 struct thread *td) 3953 { 3954 3955 return (EBADF); 3956 } 3957 3958 static int 3959 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3960 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3961 struct thread *td) 3962 { 3963 3964 return (EBADF); 3965 } 3966 3967 static int 3968 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3969 { 3970 3971 return (0); 3972 } 3973 3974 struct fileops badfileops = { 3975 .fo_read = badfo_readwrite, 3976 .fo_write = badfo_readwrite, 3977 .fo_truncate = badfo_truncate, 3978 .fo_ioctl = badfo_ioctl, 3979 .fo_poll = badfo_poll, 3980 .fo_kqfilter = badfo_kqfilter, 3981 .fo_stat = badfo_stat, 3982 .fo_close = badfo_close, 3983 .fo_chmod = badfo_chmod, 3984 .fo_chown = badfo_chown, 3985 .fo_sendfile = badfo_sendfile, 3986 .fo_fill_kinfo = badfo_fill_kinfo, 3987 }; 3988 3989 int 3990 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3991 int flags, struct thread *td) 3992 { 3993 3994 return (EOPNOTSUPP); 3995 } 3996 3997 int 3998 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3999 struct thread *td) 4000 { 4001 4002 return (EINVAL); 4003 } 4004 4005 int 4006 invfo_ioctl(struct file *fp, u_long com, void *data, 4007 struct ucred *active_cred, struct thread *td) 4008 { 4009 4010 return (ENOTTY); 4011 } 4012 4013 int 4014 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 4015 struct thread *td) 4016 { 4017 4018 return (poll_no_poll(events)); 4019 } 4020 4021 int 4022 invfo_kqfilter(struct file *fp, struct knote *kn) 4023 { 4024 4025 return (EINVAL); 4026 } 4027 4028 int 4029 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4030 struct thread *td) 4031 { 4032 4033 return (EINVAL); 4034 } 4035 4036 int 4037 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4038 struct thread *td) 4039 { 4040 4041 return (EINVAL); 4042 } 4043 4044 int 4045 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4046 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4047 struct thread *td) 4048 { 4049 4050 return (EINVAL); 4051 } 4052 4053 /*-------------------------------------------------------------------*/ 4054 4055 /* 4056 * File Descriptor pseudo-device driver (/dev/fd/). 4057 * 4058 * Opening minor device N dup()s the file (if any) connected to file 4059 * descriptor N belonging to the calling process. Note that this driver 4060 * consists of only the ``open()'' routine, because all subsequent 4061 * references to this file will be direct to the other driver. 4062 * 4063 * XXX: we could give this one a cloning event handler if necessary. 4064 */ 4065 4066 /* ARGSUSED */ 4067 static int 4068 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4069 { 4070 4071 /* 4072 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4073 * the file descriptor being sought for duplication. The error 4074 * return ensures that the vnode for this device will be released 4075 * by vn_open. Open will detect this special error and take the 4076 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4077 * will simply report the error. 4078 */ 4079 td->td_dupfd = dev2unit(dev); 4080 return (ENODEV); 4081 } 4082 4083 static struct cdevsw fildesc_cdevsw = { 4084 .d_version = D_VERSION, 4085 .d_open = fdopen, 4086 .d_name = "FD", 4087 }; 4088 4089 static void 4090 fildesc_drvinit(void *unused) 4091 { 4092 struct cdev *dev; 4093 4094 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4095 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4096 make_dev_alias(dev, "stdin"); 4097 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4098 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4099 make_dev_alias(dev, "stdout"); 4100 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4101 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4102 make_dev_alias(dev, "stderr"); 4103 } 4104 4105 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4106