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 AUDIT_ARG_FD(old); 824 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 825 826 /* 827 * Verify we have a valid descriptor to dup from and possibly to 828 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 829 * return EINVAL when the new descriptor is out of bounds. 830 */ 831 if (old < 0) 832 return (EBADF); 833 if (new < 0) 834 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 835 maxfd = getmaxfd(td); 836 if (new >= maxfd) 837 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 838 839 error = EBADF; 840 FILEDESC_XLOCK(fdp); 841 if (fget_locked(fdp, old) == NULL) 842 goto unlock; 843 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { 844 td->td_retval[0] = new; 845 if (flags & FDDUP_FLAG_CLOEXEC) 846 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 847 error = 0; 848 goto unlock; 849 } 850 851 /* 852 * If the caller specified a file descriptor, make sure the file 853 * table is large enough to hold it, and grab it. Otherwise, just 854 * allocate a new descriptor the usual way. 855 */ 856 switch (mode) { 857 case FDDUP_NORMAL: 858 case FDDUP_FCNTL: 859 if ((error = fdalloc(td, new, &new)) != 0) 860 goto unlock; 861 break; 862 case FDDUP_MUSTREPLACE: 863 /* Target file descriptor must exist. */ 864 if (fget_locked(fdp, new) == NULL) 865 goto unlock; 866 break; 867 case FDDUP_FIXED: 868 if (new >= fdp->fd_nfiles) { 869 /* 870 * The resource limits are here instead of e.g. 871 * fdalloc(), because the file descriptor table may be 872 * shared between processes, so we can't really use 873 * racct_add()/racct_sub(). Instead of counting the 874 * number of actually allocated descriptors, just put 875 * the limit on the size of the file descriptor table. 876 */ 877 #ifdef RACCT 878 if (racct_enable) { 879 PROC_LOCK(p); 880 error = racct_set(p, RACCT_NOFILE, new + 1); 881 PROC_UNLOCK(p); 882 if (error != 0) { 883 error = EMFILE; 884 goto unlock; 885 } 886 } 887 #endif 888 fdgrowtable_exp(fdp, new + 1); 889 } 890 if (!fdisused(fdp, new)) 891 fdused(fdp, new); 892 break; 893 default: 894 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 895 } 896 897 KASSERT(old != new, ("new fd is same as old")); 898 899 oldfde = &fdp->fd_ofiles[old]; 900 fhold(oldfde->fde_file); 901 newfde = &fdp->fd_ofiles[new]; 902 delfp = newfde->fde_file; 903 904 /* 905 * Duplicate the source descriptor. 906 */ 907 #ifdef CAPABILITIES 908 seq_write_begin(&newfde->fde_seq); 909 #endif 910 filecaps_free(&newfde->fde_caps); 911 memcpy(newfde, oldfde, fde_change_size); 912 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); 913 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 914 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 915 else 916 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 917 #ifdef CAPABILITIES 918 seq_write_end(&newfde->fde_seq); 919 #endif 920 td->td_retval[0] = new; 921 922 error = 0; 923 924 if (delfp != NULL) { 925 (void) closefp(fdp, new, delfp, td, 1); 926 FILEDESC_UNLOCK_ASSERT(fdp); 927 } else { 928 unlock: 929 FILEDESC_XUNLOCK(fdp); 930 } 931 932 return (error); 933 } 934 935 /* 936 * If sigio is on the list associated with a process or process group, 937 * disable signalling from the device, remove sigio from the list and 938 * free sigio. 939 */ 940 void 941 funsetown(struct sigio **sigiop) 942 { 943 struct sigio *sigio; 944 945 if (*sigiop == NULL) 946 return; 947 SIGIO_LOCK(); 948 sigio = *sigiop; 949 if (sigio == NULL) { 950 SIGIO_UNLOCK(); 951 return; 952 } 953 *(sigio->sio_myref) = NULL; 954 if ((sigio)->sio_pgid < 0) { 955 struct pgrp *pg = (sigio)->sio_pgrp; 956 PGRP_LOCK(pg); 957 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 958 sigio, sio_pgsigio); 959 PGRP_UNLOCK(pg); 960 } else { 961 struct proc *p = (sigio)->sio_proc; 962 PROC_LOCK(p); 963 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 964 sigio, sio_pgsigio); 965 PROC_UNLOCK(p); 966 } 967 SIGIO_UNLOCK(); 968 crfree(sigio->sio_ucred); 969 free(sigio, M_SIGIO); 970 } 971 972 /* 973 * Free a list of sigio structures. 974 * We only need to lock the SIGIO_LOCK because we have made ourselves 975 * inaccessible to callers of fsetown and therefore do not need to lock 976 * the proc or pgrp struct for the list manipulation. 977 */ 978 void 979 funsetownlst(struct sigiolst *sigiolst) 980 { 981 struct proc *p; 982 struct pgrp *pg; 983 struct sigio *sigio; 984 985 sigio = SLIST_FIRST(sigiolst); 986 if (sigio == NULL) 987 return; 988 p = NULL; 989 pg = NULL; 990 991 /* 992 * Every entry of the list should belong 993 * to a single proc or pgrp. 994 */ 995 if (sigio->sio_pgid < 0) { 996 pg = sigio->sio_pgrp; 997 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 998 } else /* if (sigio->sio_pgid > 0) */ { 999 p = sigio->sio_proc; 1000 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1001 } 1002 1003 SIGIO_LOCK(); 1004 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1005 *(sigio->sio_myref) = NULL; 1006 if (pg != NULL) { 1007 KASSERT(sigio->sio_pgid < 0, 1008 ("Proc sigio in pgrp sigio list")); 1009 KASSERT(sigio->sio_pgrp == pg, 1010 ("Bogus pgrp in sigio list")); 1011 PGRP_LOCK(pg); 1012 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1013 sio_pgsigio); 1014 PGRP_UNLOCK(pg); 1015 } else /* if (p != NULL) */ { 1016 KASSERT(sigio->sio_pgid > 0, 1017 ("Pgrp sigio in proc sigio list")); 1018 KASSERT(sigio->sio_proc == p, 1019 ("Bogus proc in sigio list")); 1020 PROC_LOCK(p); 1021 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1022 sio_pgsigio); 1023 PROC_UNLOCK(p); 1024 } 1025 SIGIO_UNLOCK(); 1026 crfree(sigio->sio_ucred); 1027 free(sigio, M_SIGIO); 1028 SIGIO_LOCK(); 1029 } 1030 SIGIO_UNLOCK(); 1031 } 1032 1033 /* 1034 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1035 * 1036 * After permission checking, add a sigio structure to the sigio list for 1037 * the process or process group. 1038 */ 1039 int 1040 fsetown(pid_t pgid, struct sigio **sigiop) 1041 { 1042 struct proc *proc; 1043 struct pgrp *pgrp; 1044 struct sigio *sigio; 1045 int ret; 1046 1047 if (pgid == 0) { 1048 funsetown(sigiop); 1049 return (0); 1050 } 1051 1052 ret = 0; 1053 1054 /* Allocate and fill in the new sigio out of locks. */ 1055 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1056 sigio->sio_pgid = pgid; 1057 sigio->sio_ucred = crhold(curthread->td_ucred); 1058 sigio->sio_myref = sigiop; 1059 1060 sx_slock(&proctree_lock); 1061 if (pgid > 0) { 1062 proc = pfind(pgid); 1063 if (proc == NULL) { 1064 ret = ESRCH; 1065 goto fail; 1066 } 1067 1068 /* 1069 * Policy - Don't allow a process to FSETOWN a process 1070 * in another session. 1071 * 1072 * Remove this test to allow maximum flexibility or 1073 * restrict FSETOWN to the current process or process 1074 * group for maximum safety. 1075 */ 1076 PROC_UNLOCK(proc); 1077 if (proc->p_session != curthread->td_proc->p_session) { 1078 ret = EPERM; 1079 goto fail; 1080 } 1081 1082 pgrp = NULL; 1083 } else /* if (pgid < 0) */ { 1084 pgrp = pgfind(-pgid); 1085 if (pgrp == NULL) { 1086 ret = ESRCH; 1087 goto fail; 1088 } 1089 PGRP_UNLOCK(pgrp); 1090 1091 /* 1092 * Policy - Don't allow a process to FSETOWN a process 1093 * in another session. 1094 * 1095 * Remove this test to allow maximum flexibility or 1096 * restrict FSETOWN to the current process or process 1097 * group for maximum safety. 1098 */ 1099 if (pgrp->pg_session != curthread->td_proc->p_session) { 1100 ret = EPERM; 1101 goto fail; 1102 } 1103 1104 proc = NULL; 1105 } 1106 funsetown(sigiop); 1107 if (pgid > 0) { 1108 PROC_LOCK(proc); 1109 /* 1110 * Since funsetownlst() is called without the proctree 1111 * locked, we need to check for P_WEXIT. 1112 * XXX: is ESRCH correct? 1113 */ 1114 if ((proc->p_flag & P_WEXIT) != 0) { 1115 PROC_UNLOCK(proc); 1116 ret = ESRCH; 1117 goto fail; 1118 } 1119 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1120 sigio->sio_proc = proc; 1121 PROC_UNLOCK(proc); 1122 } else { 1123 PGRP_LOCK(pgrp); 1124 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1125 sigio->sio_pgrp = pgrp; 1126 PGRP_UNLOCK(pgrp); 1127 } 1128 sx_sunlock(&proctree_lock); 1129 SIGIO_LOCK(); 1130 *sigiop = sigio; 1131 SIGIO_UNLOCK(); 1132 return (0); 1133 1134 fail: 1135 sx_sunlock(&proctree_lock); 1136 crfree(sigio->sio_ucred); 1137 free(sigio, M_SIGIO); 1138 return (ret); 1139 } 1140 1141 /* 1142 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1143 */ 1144 pid_t 1145 fgetown(sigiop) 1146 struct sigio **sigiop; 1147 { 1148 pid_t pgid; 1149 1150 SIGIO_LOCK(); 1151 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1152 SIGIO_UNLOCK(); 1153 return (pgid); 1154 } 1155 1156 /* 1157 * Function drops the filedesc lock on return. 1158 */ 1159 static int 1160 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1161 int holdleaders) 1162 { 1163 int error; 1164 1165 FILEDESC_XLOCK_ASSERT(fdp); 1166 1167 if (holdleaders) { 1168 if (td->td_proc->p_fdtol != NULL) { 1169 /* 1170 * Ask fdfree() to sleep to ensure that all relevant 1171 * process leaders can be traversed in closef(). 1172 */ 1173 fdp->fd_holdleaderscount++; 1174 } else { 1175 holdleaders = 0; 1176 } 1177 } 1178 1179 /* 1180 * We now hold the fp reference that used to be owned by the 1181 * descriptor array. We have to unlock the FILEDESC *AFTER* 1182 * knote_fdclose to prevent a race of the fd getting opened, a knote 1183 * added, and deleteing a knote for the new fd. 1184 */ 1185 knote_fdclose(td, fd); 1186 1187 /* 1188 * We need to notify mqueue if the object is of type mqueue. 1189 */ 1190 if (fp->f_type == DTYPE_MQUEUE) 1191 mq_fdclose(td, fd, fp); 1192 FILEDESC_XUNLOCK(fdp); 1193 1194 error = closef(fp, td); 1195 if (holdleaders) { 1196 FILEDESC_XLOCK(fdp); 1197 fdp->fd_holdleaderscount--; 1198 if (fdp->fd_holdleaderscount == 0 && 1199 fdp->fd_holdleaderswakeup != 0) { 1200 fdp->fd_holdleaderswakeup = 0; 1201 wakeup(&fdp->fd_holdleaderscount); 1202 } 1203 FILEDESC_XUNLOCK(fdp); 1204 } 1205 return (error); 1206 } 1207 1208 /* 1209 * Close a file descriptor. 1210 */ 1211 #ifndef _SYS_SYSPROTO_H_ 1212 struct close_args { 1213 int fd; 1214 }; 1215 #endif 1216 /* ARGSUSED */ 1217 int 1218 sys_close(struct thread *td, struct close_args *uap) 1219 { 1220 1221 return (kern_close(td, uap->fd)); 1222 } 1223 1224 int 1225 kern_close(struct thread *td, int fd) 1226 { 1227 struct filedesc *fdp; 1228 struct file *fp; 1229 1230 fdp = td->td_proc->p_fd; 1231 1232 AUDIT_SYSCLOSE(td, fd); 1233 1234 FILEDESC_XLOCK(fdp); 1235 if ((fp = fget_locked(fdp, fd)) == NULL) { 1236 FILEDESC_XUNLOCK(fdp); 1237 return (EBADF); 1238 } 1239 fdfree(fdp, fd); 1240 1241 /* closefp() drops the FILEDESC lock for us. */ 1242 return (closefp(fdp, fd, fp, td, 1)); 1243 } 1244 1245 /* 1246 * Close open file descriptors. 1247 */ 1248 #ifndef _SYS_SYSPROTO_H_ 1249 struct closefrom_args { 1250 int lowfd; 1251 }; 1252 #endif 1253 /* ARGSUSED */ 1254 int 1255 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1256 { 1257 struct filedesc *fdp; 1258 int fd; 1259 1260 fdp = td->td_proc->p_fd; 1261 AUDIT_ARG_FD(uap->lowfd); 1262 1263 /* 1264 * Treat negative starting file descriptor values identical to 1265 * closefrom(0) which closes all files. 1266 */ 1267 if (uap->lowfd < 0) 1268 uap->lowfd = 0; 1269 FILEDESC_SLOCK(fdp); 1270 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1271 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1272 FILEDESC_SUNLOCK(fdp); 1273 (void)kern_close(td, fd); 1274 FILEDESC_SLOCK(fdp); 1275 } 1276 } 1277 FILEDESC_SUNLOCK(fdp); 1278 return (0); 1279 } 1280 1281 #if defined(COMPAT_43) 1282 /* 1283 * Return status information about a file descriptor. 1284 */ 1285 #ifndef _SYS_SYSPROTO_H_ 1286 struct ofstat_args { 1287 int fd; 1288 struct ostat *sb; 1289 }; 1290 #endif 1291 /* ARGSUSED */ 1292 int 1293 ofstat(struct thread *td, struct ofstat_args *uap) 1294 { 1295 struct ostat oub; 1296 struct stat ub; 1297 int error; 1298 1299 error = kern_fstat(td, uap->fd, &ub); 1300 if (error == 0) { 1301 cvtstat(&ub, &oub); 1302 error = copyout(&oub, uap->sb, sizeof(oub)); 1303 } 1304 return (error); 1305 } 1306 #endif /* COMPAT_43 */ 1307 1308 /* 1309 * Return status information about a file descriptor. 1310 */ 1311 #ifndef _SYS_SYSPROTO_H_ 1312 struct fstat_args { 1313 int fd; 1314 struct stat *sb; 1315 }; 1316 #endif 1317 /* ARGSUSED */ 1318 int 1319 sys_fstat(struct thread *td, struct fstat_args *uap) 1320 { 1321 struct stat ub; 1322 int error; 1323 1324 error = kern_fstat(td, uap->fd, &ub); 1325 if (error == 0) 1326 error = copyout(&ub, uap->sb, sizeof(ub)); 1327 return (error); 1328 } 1329 1330 int 1331 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1332 { 1333 struct file *fp; 1334 cap_rights_t rights; 1335 int error; 1336 1337 AUDIT_ARG_FD(fd); 1338 1339 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1340 if (error != 0) 1341 return (error); 1342 1343 AUDIT_ARG_FILE(td->td_proc, fp); 1344 1345 error = fo_stat(fp, sbp, td->td_ucred, td); 1346 fdrop(fp, td); 1347 #ifdef KTRACE 1348 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1349 ktrstat(sbp); 1350 #endif 1351 return (error); 1352 } 1353 1354 /* 1355 * Return status information about a file descriptor. 1356 */ 1357 #ifndef _SYS_SYSPROTO_H_ 1358 struct nfstat_args { 1359 int fd; 1360 struct nstat *sb; 1361 }; 1362 #endif 1363 /* ARGSUSED */ 1364 int 1365 sys_nfstat(struct thread *td, struct nfstat_args *uap) 1366 { 1367 struct nstat nub; 1368 struct stat ub; 1369 int error; 1370 1371 error = kern_fstat(td, uap->fd, &ub); 1372 if (error == 0) { 1373 cvtnstat(&ub, &nub); 1374 error = copyout(&nub, uap->sb, sizeof(nub)); 1375 } 1376 return (error); 1377 } 1378 1379 /* 1380 * Return pathconf information about a file descriptor. 1381 */ 1382 #ifndef _SYS_SYSPROTO_H_ 1383 struct fpathconf_args { 1384 int fd; 1385 int name; 1386 }; 1387 #endif 1388 /* ARGSUSED */ 1389 int 1390 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1391 { 1392 struct file *fp; 1393 struct vnode *vp; 1394 cap_rights_t rights; 1395 int error; 1396 1397 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1398 if (error != 0) 1399 return (error); 1400 1401 if (uap->name == _PC_ASYNC_IO) { 1402 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO; 1403 goto out; 1404 } 1405 vp = fp->f_vnode; 1406 if (vp != NULL) { 1407 vn_lock(vp, LK_SHARED | LK_RETRY); 1408 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1409 VOP_UNLOCK(vp, 0); 1410 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1411 if (uap->name != _PC_PIPE_BUF) { 1412 error = EINVAL; 1413 } else { 1414 td->td_retval[0] = PIPE_BUF; 1415 error = 0; 1416 } 1417 } else { 1418 error = EOPNOTSUPP; 1419 } 1420 out: 1421 fdrop(fp, td); 1422 return (error); 1423 } 1424 1425 /* 1426 * Initialize filecaps structure. 1427 */ 1428 void 1429 filecaps_init(struct filecaps *fcaps) 1430 { 1431 1432 bzero(fcaps, sizeof(*fcaps)); 1433 fcaps->fc_nioctls = -1; 1434 } 1435 1436 /* 1437 * Copy filecaps structure allocating memory for ioctls array if needed. 1438 * 1439 * The last parameter indicates whether the fdtable is locked. If it is not and 1440 * ioctls are encountered, copying fails and the caller must lock the table. 1441 * 1442 * Note that if the table was not locked, the caller has to check the relevant 1443 * sequence counter to determine whether the operation was successful. 1444 */ 1445 int 1446 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1447 { 1448 size_t size; 1449 1450 *dst = *src; 1451 if (src->fc_ioctls == NULL) 1452 return (0); 1453 if (!locked) 1454 return (1); 1455 1456 KASSERT(src->fc_nioctls > 0, 1457 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1458 1459 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1460 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1461 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1462 return (0); 1463 } 1464 1465 /* 1466 * Move filecaps structure to the new place and clear the old place. 1467 */ 1468 void 1469 filecaps_move(struct filecaps *src, struct filecaps *dst) 1470 { 1471 1472 *dst = *src; 1473 bzero(src, sizeof(*src)); 1474 } 1475 1476 /* 1477 * Fill the given filecaps structure with full rights. 1478 */ 1479 static void 1480 filecaps_fill(struct filecaps *fcaps) 1481 { 1482 1483 CAP_ALL(&fcaps->fc_rights); 1484 fcaps->fc_ioctls = NULL; 1485 fcaps->fc_nioctls = -1; 1486 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1487 } 1488 1489 /* 1490 * Free memory allocated within filecaps structure. 1491 */ 1492 void 1493 filecaps_free(struct filecaps *fcaps) 1494 { 1495 1496 free(fcaps->fc_ioctls, M_FILECAPS); 1497 bzero(fcaps, sizeof(*fcaps)); 1498 } 1499 1500 /* 1501 * Validate the given filecaps structure. 1502 */ 1503 static void 1504 filecaps_validate(const struct filecaps *fcaps, const char *func) 1505 { 1506 1507 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1508 ("%s: invalid rights", func)); 1509 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1510 ("%s: invalid fcntls", func)); 1511 KASSERT(fcaps->fc_fcntls == 0 || 1512 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1513 ("%s: fcntls without CAP_FCNTL", func)); 1514 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1515 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1516 ("%s: invalid ioctls", func)); 1517 KASSERT(fcaps->fc_nioctls == 0 || 1518 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1519 ("%s: ioctls without CAP_IOCTL", func)); 1520 } 1521 1522 static void 1523 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1524 { 1525 int nfd1; 1526 1527 FILEDESC_XLOCK_ASSERT(fdp); 1528 1529 nfd1 = fdp->fd_nfiles * 2; 1530 if (nfd1 < nfd) 1531 nfd1 = nfd; 1532 fdgrowtable(fdp, nfd1); 1533 } 1534 1535 /* 1536 * Grow the file table to accommodate (at least) nfd descriptors. 1537 */ 1538 static void 1539 fdgrowtable(struct filedesc *fdp, int nfd) 1540 { 1541 struct filedesc0 *fdp0; 1542 struct freetable *ft; 1543 struct fdescenttbl *ntable; 1544 struct fdescenttbl *otable; 1545 int nnfiles, onfiles; 1546 NDSLOTTYPE *nmap, *omap; 1547 1548 /* 1549 * If lastfile is -1 this struct filedesc was just allocated and we are 1550 * growing it to accommodate for the one we are going to copy from. There 1551 * is no need to have a lock on this one as it's not visible to anyone. 1552 */ 1553 if (fdp->fd_lastfile != -1) 1554 FILEDESC_XLOCK_ASSERT(fdp); 1555 1556 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1557 1558 /* save old values */ 1559 onfiles = fdp->fd_nfiles; 1560 otable = fdp->fd_files; 1561 omap = fdp->fd_map; 1562 1563 /* compute the size of the new table */ 1564 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1565 if (nnfiles <= onfiles) 1566 /* the table is already large enough */ 1567 return; 1568 1569 /* 1570 * Allocate a new table. We need enough space for the number of 1571 * entries, file entries themselves and the struct freetable we will use 1572 * when we decommission the table and place it on the freelist. 1573 * We place the struct freetable in the middle so we don't have 1574 * to worry about padding. 1575 */ 1576 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1577 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1578 sizeof(struct freetable), 1579 M_FILEDESC, M_ZERO | M_WAITOK); 1580 /* copy the old data */ 1581 ntable->fdt_nfiles = nnfiles; 1582 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1583 onfiles * sizeof(ntable->fdt_ofiles[0])); 1584 1585 /* 1586 * Allocate a new map only if the old is not large enough. It will 1587 * grow at a slower rate than the table as it can map more 1588 * entries than the table can hold. 1589 */ 1590 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1591 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1592 M_ZERO | M_WAITOK); 1593 /* copy over the old data and update the pointer */ 1594 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1595 fdp->fd_map = nmap; 1596 } 1597 1598 /* 1599 * Make sure that ntable is correctly initialized before we replace 1600 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1601 * data. 1602 */ 1603 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1604 1605 /* 1606 * Do not free the old file table, as some threads may still 1607 * reference entries within it. Instead, place it on a freelist 1608 * which will be processed when the struct filedesc is released. 1609 * 1610 * Note that if onfiles == NDFILE, we're dealing with the original 1611 * static allocation contained within (struct filedesc0 *)fdp, 1612 * which must not be freed. 1613 */ 1614 if (onfiles > NDFILE) { 1615 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1616 fdp0 = (struct filedesc0 *)fdp; 1617 ft->ft_table = otable; 1618 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1619 } 1620 /* 1621 * The map does not have the same possibility of threads still 1622 * holding references to it. So always free it as long as it 1623 * does not reference the original static allocation. 1624 */ 1625 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1626 free(omap, M_FILEDESC); 1627 } 1628 1629 /* 1630 * Allocate a file descriptor for the process. 1631 */ 1632 int 1633 fdalloc(struct thread *td, int minfd, int *result) 1634 { 1635 struct proc *p = td->td_proc; 1636 struct filedesc *fdp = p->p_fd; 1637 int fd, maxfd, allocfd; 1638 #ifdef RACCT 1639 int error; 1640 #endif 1641 1642 FILEDESC_XLOCK_ASSERT(fdp); 1643 1644 if (fdp->fd_freefile > minfd) 1645 minfd = fdp->fd_freefile; 1646 1647 maxfd = getmaxfd(td); 1648 1649 /* 1650 * Search the bitmap for a free descriptor starting at minfd. 1651 * If none is found, grow the file table. 1652 */ 1653 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1654 if (fd >= maxfd) 1655 return (EMFILE); 1656 if (fd >= fdp->fd_nfiles) { 1657 allocfd = min(fd * 2, maxfd); 1658 #ifdef RACCT 1659 if (racct_enable) { 1660 PROC_LOCK(p); 1661 error = racct_set(p, RACCT_NOFILE, allocfd); 1662 PROC_UNLOCK(p); 1663 if (error != 0) 1664 return (EMFILE); 1665 } 1666 #endif 1667 /* 1668 * fd is already equal to first free descriptor >= minfd, so 1669 * we only need to grow the table and we are done. 1670 */ 1671 fdgrowtable_exp(fdp, allocfd); 1672 } 1673 1674 /* 1675 * Perform some sanity checks, then mark the file descriptor as 1676 * used and return it to the caller. 1677 */ 1678 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1679 ("invalid descriptor %d", fd)); 1680 KASSERT(!fdisused(fdp, fd), 1681 ("fd_first_free() returned non-free descriptor")); 1682 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1683 ("file descriptor isn't free")); 1684 fdused(fdp, fd); 1685 *result = fd; 1686 return (0); 1687 } 1688 1689 /* 1690 * Allocate n file descriptors for the process. 1691 */ 1692 int 1693 fdallocn(struct thread *td, int minfd, int *fds, int n) 1694 { 1695 struct proc *p = td->td_proc; 1696 struct filedesc *fdp = p->p_fd; 1697 int i; 1698 1699 FILEDESC_XLOCK_ASSERT(fdp); 1700 1701 for (i = 0; i < n; i++) 1702 if (fdalloc(td, 0, &fds[i]) != 0) 1703 break; 1704 1705 if (i < n) { 1706 for (i--; i >= 0; i--) 1707 fdunused(fdp, fds[i]); 1708 return (EMFILE); 1709 } 1710 1711 return (0); 1712 } 1713 1714 /* 1715 * Create a new open file structure and allocate a file descriptor for the 1716 * process that refers to it. We add one reference to the file for the 1717 * descriptor table and one reference for resultfp. This is to prevent us 1718 * being preempted and the entry in the descriptor table closed after we 1719 * release the FILEDESC lock. 1720 */ 1721 int 1722 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 1723 struct filecaps *fcaps) 1724 { 1725 struct file *fp; 1726 int error, fd; 1727 1728 error = falloc_noinstall(td, &fp); 1729 if (error) 1730 return (error); /* no reference held on error */ 1731 1732 error = finstall(td, fp, &fd, flags, fcaps); 1733 if (error) { 1734 fdrop(fp, td); /* one reference (fp only) */ 1735 return (error); 1736 } 1737 1738 if (resultfp != NULL) 1739 *resultfp = fp; /* copy out result */ 1740 else 1741 fdrop(fp, td); /* release local reference */ 1742 1743 if (resultfd != NULL) 1744 *resultfd = fd; 1745 1746 return (0); 1747 } 1748 1749 /* 1750 * Create a new open file structure without allocating a file descriptor. 1751 */ 1752 int 1753 falloc_noinstall(struct thread *td, struct file **resultfp) 1754 { 1755 struct file *fp; 1756 int maxuserfiles = maxfiles - (maxfiles / 20); 1757 static struct timeval lastfail; 1758 static int curfail; 1759 1760 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1761 1762 if ((openfiles >= maxuserfiles && 1763 priv_check(td, PRIV_MAXFILES) != 0) || 1764 openfiles >= maxfiles) { 1765 if (ppsratecheck(&lastfail, &curfail, 1)) { 1766 printf("kern.maxfiles limit exceeded by uid %i, " 1767 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1768 } 1769 return (ENFILE); 1770 } 1771 atomic_add_int(&openfiles, 1); 1772 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1773 refcount_init(&fp->f_count, 1); 1774 fp->f_cred = crhold(td->td_ucred); 1775 fp->f_ops = &badfileops; 1776 *resultfp = fp; 1777 return (0); 1778 } 1779 1780 /* 1781 * Install a file in a file descriptor table. 1782 */ 1783 void 1784 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1785 struct filecaps *fcaps) 1786 { 1787 struct filedescent *fde; 1788 1789 MPASS(fp != NULL); 1790 if (fcaps != NULL) 1791 filecaps_validate(fcaps, __func__); 1792 FILEDESC_XLOCK_ASSERT(fdp); 1793 1794 fde = &fdp->fd_ofiles[fd]; 1795 #ifdef CAPABILITIES 1796 seq_write_begin(&fde->fde_seq); 1797 #endif 1798 fde->fde_file = fp; 1799 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1800 if (fcaps != NULL) 1801 filecaps_move(fcaps, &fde->fde_caps); 1802 else 1803 filecaps_fill(&fde->fde_caps); 1804 #ifdef CAPABILITIES 1805 seq_write_end(&fde->fde_seq); 1806 #endif 1807 } 1808 1809 int 1810 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1811 struct filecaps *fcaps) 1812 { 1813 struct filedesc *fdp = td->td_proc->p_fd; 1814 int error; 1815 1816 MPASS(fd != NULL); 1817 1818 FILEDESC_XLOCK(fdp); 1819 if ((error = fdalloc(td, 0, fd))) { 1820 FILEDESC_XUNLOCK(fdp); 1821 return (error); 1822 } 1823 fhold(fp); 1824 _finstall(fdp, fp, *fd, flags, fcaps); 1825 FILEDESC_XUNLOCK(fdp); 1826 return (0); 1827 } 1828 1829 /* 1830 * Build a new filedesc structure from another. 1831 * Copy the current, root, and jail root vnode references. 1832 * 1833 * If fdp is not NULL, return with it shared locked. 1834 */ 1835 struct filedesc * 1836 fdinit(struct filedesc *fdp, bool prepfiles) 1837 { 1838 struct filedesc0 *newfdp0; 1839 struct filedesc *newfdp; 1840 1841 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1842 newfdp = &newfdp0->fd_fd; 1843 1844 /* Create the file descriptor table. */ 1845 FILEDESC_LOCK_INIT(newfdp); 1846 refcount_init(&newfdp->fd_refcnt, 1); 1847 refcount_init(&newfdp->fd_holdcnt, 1); 1848 newfdp->fd_cmask = CMASK; 1849 newfdp->fd_map = newfdp0->fd_dmap; 1850 newfdp->fd_lastfile = -1; 1851 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1852 newfdp->fd_files->fdt_nfiles = NDFILE; 1853 1854 if (fdp == NULL) 1855 return (newfdp); 1856 1857 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1858 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1859 1860 FILEDESC_SLOCK(fdp); 1861 newfdp->fd_cdir = fdp->fd_cdir; 1862 if (newfdp->fd_cdir) 1863 VREF(newfdp->fd_cdir); 1864 newfdp->fd_rdir = fdp->fd_rdir; 1865 if (newfdp->fd_rdir) 1866 VREF(newfdp->fd_rdir); 1867 newfdp->fd_jdir = fdp->fd_jdir; 1868 if (newfdp->fd_jdir) 1869 VREF(newfdp->fd_jdir); 1870 1871 if (!prepfiles) { 1872 FILEDESC_SUNLOCK(fdp); 1873 } else { 1874 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1875 FILEDESC_SUNLOCK(fdp); 1876 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1877 FILEDESC_SLOCK(fdp); 1878 } 1879 } 1880 1881 return (newfdp); 1882 } 1883 1884 static struct filedesc * 1885 fdhold(struct proc *p) 1886 { 1887 struct filedesc *fdp; 1888 1889 PROC_LOCK_ASSERT(p, MA_OWNED); 1890 fdp = p->p_fd; 1891 if (fdp != NULL) 1892 refcount_acquire(&fdp->fd_holdcnt); 1893 return (fdp); 1894 } 1895 1896 static void 1897 fddrop(struct filedesc *fdp) 1898 { 1899 1900 if (fdp->fd_holdcnt > 1) { 1901 if (refcount_release(&fdp->fd_holdcnt) == 0) 1902 return; 1903 } 1904 1905 FILEDESC_LOCK_DESTROY(fdp); 1906 uma_zfree(filedesc0_zone, fdp); 1907 } 1908 1909 /* 1910 * Share a filedesc structure. 1911 */ 1912 struct filedesc * 1913 fdshare(struct filedesc *fdp) 1914 { 1915 1916 refcount_acquire(&fdp->fd_refcnt); 1917 return (fdp); 1918 } 1919 1920 /* 1921 * Unshare a filedesc structure, if necessary by making a copy 1922 */ 1923 void 1924 fdunshare(struct thread *td) 1925 { 1926 struct filedesc *tmp; 1927 struct proc *p = td->td_proc; 1928 1929 if (p->p_fd->fd_refcnt == 1) 1930 return; 1931 1932 tmp = fdcopy(p->p_fd); 1933 fdescfree(td); 1934 p->p_fd = tmp; 1935 } 1936 1937 void 1938 fdinstall_remapped(struct thread *td, struct filedesc *fdp) 1939 { 1940 1941 fdescfree(td); 1942 td->td_proc->p_fd = fdp; 1943 } 1944 1945 /* 1946 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1947 * this is to ease callers, not catch errors. 1948 */ 1949 struct filedesc * 1950 fdcopy(struct filedesc *fdp) 1951 { 1952 struct filedesc *newfdp; 1953 struct filedescent *nfde, *ofde; 1954 int i; 1955 1956 MPASS(fdp != NULL); 1957 1958 newfdp = fdinit(fdp, true); 1959 /* copy all passable descriptors (i.e. not kqueue) */ 1960 newfdp->fd_freefile = -1; 1961 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1962 ofde = &fdp->fd_ofiles[i]; 1963 if (ofde->fde_file == NULL || 1964 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1965 if (newfdp->fd_freefile == -1) 1966 newfdp->fd_freefile = i; 1967 continue; 1968 } 1969 nfde = &newfdp->fd_ofiles[i]; 1970 *nfde = *ofde; 1971 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 1972 fhold(nfde->fde_file); 1973 fdused_init(newfdp, i); 1974 newfdp->fd_lastfile = i; 1975 } 1976 if (newfdp->fd_freefile == -1) 1977 newfdp->fd_freefile = i; 1978 newfdp->fd_cmask = fdp->fd_cmask; 1979 FILEDESC_SUNLOCK(fdp); 1980 return (newfdp); 1981 } 1982 1983 /* 1984 * Copies a filedesc structure, while remapping all file descriptors 1985 * stored inside using a translation table. 1986 * 1987 * File descriptors are copied over to the new file descriptor table, 1988 * regardless of whether the close-on-exec flag is set. 1989 */ 1990 int 1991 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, 1992 struct filedesc **ret) 1993 { 1994 struct filedesc *newfdp; 1995 struct filedescent *nfde, *ofde; 1996 int error, i; 1997 1998 MPASS(fdp != NULL); 1999 2000 newfdp = fdinit(fdp, true); 2001 if (nfds > fdp->fd_lastfile + 1) { 2002 /* New table cannot be larger than the old one. */ 2003 error = E2BIG; 2004 goto bad; 2005 } 2006 /* Copy all passable descriptors (i.e. not kqueue). */ 2007 newfdp->fd_freefile = nfds; 2008 for (i = 0; i < nfds; ++i) { 2009 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { 2010 /* File descriptor out of bounds. */ 2011 error = EBADF; 2012 goto bad; 2013 } 2014 ofde = &fdp->fd_ofiles[fds[i]]; 2015 if (ofde->fde_file == NULL) { 2016 /* Unused file descriptor. */ 2017 error = EBADF; 2018 goto bad; 2019 } 2020 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 2021 /* File descriptor cannot be passed. */ 2022 error = EINVAL; 2023 goto bad; 2024 } 2025 nfde = &newfdp->fd_ofiles[i]; 2026 *nfde = *ofde; 2027 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2028 fhold(nfde->fde_file); 2029 fdused_init(newfdp, i); 2030 newfdp->fd_lastfile = i; 2031 } 2032 newfdp->fd_cmask = fdp->fd_cmask; 2033 FILEDESC_SUNLOCK(fdp); 2034 *ret = newfdp; 2035 return (0); 2036 bad: 2037 FILEDESC_SUNLOCK(fdp); 2038 fdescfree_remapped(newfdp); 2039 return (error); 2040 } 2041 2042 /* 2043 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2044 * one of processes using it exits) and the table used to be shared. 2045 */ 2046 static void 2047 fdclearlocks(struct thread *td) 2048 { 2049 struct filedesc *fdp; 2050 struct filedesc_to_leader *fdtol; 2051 struct flock lf; 2052 struct file *fp; 2053 struct proc *p; 2054 struct vnode *vp; 2055 int i; 2056 2057 p = td->td_proc; 2058 fdp = p->p_fd; 2059 fdtol = p->p_fdtol; 2060 MPASS(fdtol != NULL); 2061 2062 FILEDESC_XLOCK(fdp); 2063 KASSERT(fdtol->fdl_refcount > 0, 2064 ("filedesc_to_refcount botch: fdl_refcount=%d", 2065 fdtol->fdl_refcount)); 2066 if (fdtol->fdl_refcount == 1 && 2067 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2068 for (i = 0; i <= fdp->fd_lastfile; i++) { 2069 fp = fdp->fd_ofiles[i].fde_file; 2070 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2071 continue; 2072 fhold(fp); 2073 FILEDESC_XUNLOCK(fdp); 2074 lf.l_whence = SEEK_SET; 2075 lf.l_start = 0; 2076 lf.l_len = 0; 2077 lf.l_type = F_UNLCK; 2078 vp = fp->f_vnode; 2079 (void) VOP_ADVLOCK(vp, 2080 (caddr_t)p->p_leader, F_UNLCK, 2081 &lf, F_POSIX); 2082 FILEDESC_XLOCK(fdp); 2083 fdrop(fp, td); 2084 } 2085 } 2086 retry: 2087 if (fdtol->fdl_refcount == 1) { 2088 if (fdp->fd_holdleaderscount > 0 && 2089 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2090 /* 2091 * close() or kern_dup() has cleared a reference 2092 * in a shared file descriptor table. 2093 */ 2094 fdp->fd_holdleaderswakeup = 1; 2095 sx_sleep(&fdp->fd_holdleaderscount, 2096 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2097 goto retry; 2098 } 2099 if (fdtol->fdl_holdcount > 0) { 2100 /* 2101 * Ensure that fdtol->fdl_leader remains 2102 * valid in closef(). 2103 */ 2104 fdtol->fdl_wakeup = 1; 2105 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2106 "fdlhold", 0); 2107 goto retry; 2108 } 2109 } 2110 fdtol->fdl_refcount--; 2111 if (fdtol->fdl_refcount == 0 && 2112 fdtol->fdl_holdcount == 0) { 2113 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2114 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2115 } else 2116 fdtol = NULL; 2117 p->p_fdtol = NULL; 2118 FILEDESC_XUNLOCK(fdp); 2119 if (fdtol != NULL) 2120 free(fdtol, M_FILEDESC_TO_LEADER); 2121 } 2122 2123 /* 2124 * Release a filedesc structure. 2125 */ 2126 static void 2127 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) 2128 { 2129 struct filedesc0 *fdp0; 2130 struct freetable *ft, *tft; 2131 struct filedescent *fde; 2132 struct file *fp; 2133 int i; 2134 2135 for (i = 0; i <= fdp->fd_lastfile; i++) { 2136 fde = &fdp->fd_ofiles[i]; 2137 fp = fde->fde_file; 2138 if (fp != NULL) { 2139 fdefree_last(fde); 2140 if (needclose) 2141 (void) closef(fp, td); 2142 else 2143 fdrop(fp, td); 2144 } 2145 } 2146 2147 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2148 free(fdp->fd_map, M_FILEDESC); 2149 if (fdp->fd_nfiles > NDFILE) 2150 free(fdp->fd_files, M_FILEDESC); 2151 2152 fdp0 = (struct filedesc0 *)fdp; 2153 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2154 free(ft->ft_table, M_FILEDESC); 2155 2156 fddrop(fdp); 2157 } 2158 2159 void 2160 fdescfree(struct thread *td) 2161 { 2162 struct proc *p; 2163 struct filedesc *fdp; 2164 struct vnode *cdir, *jdir, *rdir; 2165 2166 p = td->td_proc; 2167 fdp = p->p_fd; 2168 MPASS(fdp != NULL); 2169 2170 #ifdef RACCT 2171 if (racct_enable) { 2172 PROC_LOCK(p); 2173 racct_set(p, RACCT_NOFILE, 0); 2174 PROC_UNLOCK(p); 2175 } 2176 #endif 2177 2178 if (p->p_fdtol != NULL) 2179 fdclearlocks(td); 2180 2181 PROC_LOCK(p); 2182 p->p_fd = NULL; 2183 PROC_UNLOCK(p); 2184 2185 if (refcount_release(&fdp->fd_refcnt) == 0) 2186 return; 2187 2188 FILEDESC_XLOCK(fdp); 2189 cdir = fdp->fd_cdir; 2190 fdp->fd_cdir = NULL; 2191 rdir = fdp->fd_rdir; 2192 fdp->fd_rdir = NULL; 2193 jdir = fdp->fd_jdir; 2194 fdp->fd_jdir = NULL; 2195 FILEDESC_XUNLOCK(fdp); 2196 2197 if (cdir != NULL) 2198 vrele(cdir); 2199 if (rdir != NULL) 2200 vrele(rdir); 2201 if (jdir != NULL) 2202 vrele(jdir); 2203 2204 fdescfree_fds(td, fdp, 1); 2205 } 2206 2207 void 2208 fdescfree_remapped(struct filedesc *fdp) 2209 { 2210 2211 if (fdp->fd_cdir != NULL) 2212 vrele(fdp->fd_cdir); 2213 if (fdp->fd_rdir != NULL) 2214 vrele(fdp->fd_rdir); 2215 if (fdp->fd_jdir != NULL) 2216 vrele(fdp->fd_jdir); 2217 2218 fdescfree_fds(curthread, fdp, 0); 2219 } 2220 2221 /* 2222 * For setugid programs, we don't want to people to use that setugidness 2223 * to generate error messages which write to a file which otherwise would 2224 * otherwise be off-limits to the process. We check for filesystems where 2225 * the vnode can change out from under us after execve (like [lin]procfs). 2226 * 2227 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2228 * sufficient. We also don't check for setugidness since we know we are. 2229 */ 2230 static bool 2231 is_unsafe(struct file *fp) 2232 { 2233 struct vnode *vp; 2234 2235 if (fp->f_type != DTYPE_VNODE) 2236 return (false); 2237 2238 vp = fp->f_vnode; 2239 return ((vp->v_vflag & VV_PROCDEP) != 0); 2240 } 2241 2242 /* 2243 * Make this setguid thing safe, if at all possible. 2244 */ 2245 void 2246 fdsetugidsafety(struct thread *td) 2247 { 2248 struct filedesc *fdp; 2249 struct file *fp; 2250 int i; 2251 2252 fdp = td->td_proc->p_fd; 2253 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2254 MPASS(fdp->fd_nfiles >= 3); 2255 for (i = 0; i <= 2; i++) { 2256 fp = fdp->fd_ofiles[i].fde_file; 2257 if (fp != NULL && is_unsafe(fp)) { 2258 FILEDESC_XLOCK(fdp); 2259 knote_fdclose(td, i); 2260 /* 2261 * NULL-out descriptor prior to close to avoid 2262 * a race while close blocks. 2263 */ 2264 fdfree(fdp, i); 2265 FILEDESC_XUNLOCK(fdp); 2266 (void) closef(fp, td); 2267 } 2268 } 2269 } 2270 2271 /* 2272 * If a specific file object occupies a specific file descriptor, close the 2273 * file descriptor entry and drop a reference on the file object. This is a 2274 * convenience function to handle a subsequent error in a function that calls 2275 * falloc() that handles the race that another thread might have closed the 2276 * file descriptor out from under the thread creating the file object. 2277 */ 2278 void 2279 fdclose(struct thread *td, struct file *fp, int idx) 2280 { 2281 struct filedesc *fdp = td->td_proc->p_fd; 2282 2283 FILEDESC_XLOCK(fdp); 2284 if (fdp->fd_ofiles[idx].fde_file == fp) { 2285 fdfree(fdp, idx); 2286 FILEDESC_XUNLOCK(fdp); 2287 fdrop(fp, td); 2288 } else 2289 FILEDESC_XUNLOCK(fdp); 2290 } 2291 2292 /* 2293 * Close any files on exec? 2294 */ 2295 void 2296 fdcloseexec(struct thread *td) 2297 { 2298 struct filedesc *fdp; 2299 struct filedescent *fde; 2300 struct file *fp; 2301 int i; 2302 2303 fdp = td->td_proc->p_fd; 2304 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2305 for (i = 0; i <= fdp->fd_lastfile; i++) { 2306 fde = &fdp->fd_ofiles[i]; 2307 fp = fde->fde_file; 2308 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2309 (fde->fde_flags & UF_EXCLOSE))) { 2310 FILEDESC_XLOCK(fdp); 2311 fdfree(fdp, i); 2312 (void) closefp(fdp, i, fp, td, 0); 2313 FILEDESC_UNLOCK_ASSERT(fdp); 2314 } 2315 } 2316 } 2317 2318 /* 2319 * It is unsafe for set[ug]id processes to be started with file 2320 * descriptors 0..2 closed, as these descriptors are given implicit 2321 * significance in the Standard C library. fdcheckstd() will create a 2322 * descriptor referencing /dev/null for each of stdin, stdout, and 2323 * stderr that is not already open. 2324 */ 2325 int 2326 fdcheckstd(struct thread *td) 2327 { 2328 struct filedesc *fdp; 2329 register_t save; 2330 int i, error, devnull; 2331 2332 fdp = td->td_proc->p_fd; 2333 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2334 MPASS(fdp->fd_nfiles >= 3); 2335 devnull = -1; 2336 for (i = 0; i <= 2; i++) { 2337 if (fdp->fd_ofiles[i].fde_file != NULL) 2338 continue; 2339 2340 save = td->td_retval[0]; 2341 if (devnull != -1) { 2342 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2343 } else { 2344 error = kern_openat(td, AT_FDCWD, "/dev/null", 2345 UIO_SYSSPACE, O_RDWR, 0); 2346 if (error == 0) { 2347 devnull = td->td_retval[0]; 2348 KASSERT(devnull == i, ("we didn't get our fd")); 2349 } 2350 } 2351 td->td_retval[0] = save; 2352 if (error != 0) 2353 return (error); 2354 } 2355 return (0); 2356 } 2357 2358 /* 2359 * Internal form of close. Decrement reference count on file structure. 2360 * Note: td may be NULL when closing a file that was being passed in a 2361 * message. 2362 * 2363 * XXXRW: Giant is not required for the caller, but often will be held; this 2364 * makes it moderately likely the Giant will be recursed in the VFS case. 2365 */ 2366 int 2367 closef(struct file *fp, struct thread *td) 2368 { 2369 struct vnode *vp; 2370 struct flock lf; 2371 struct filedesc_to_leader *fdtol; 2372 struct filedesc *fdp; 2373 2374 /* 2375 * POSIX record locking dictates that any close releases ALL 2376 * locks owned by this process. This is handled by setting 2377 * a flag in the unlock to free ONLY locks obeying POSIX 2378 * semantics, and not to free BSD-style file locks. 2379 * If the descriptor was in a message, POSIX-style locks 2380 * aren't passed with the descriptor, and the thread pointer 2381 * will be NULL. Callers should be careful only to pass a 2382 * NULL thread pointer when there really is no owning 2383 * context that might have locks, or the locks will be 2384 * leaked. 2385 */ 2386 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2387 vp = fp->f_vnode; 2388 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2389 lf.l_whence = SEEK_SET; 2390 lf.l_start = 0; 2391 lf.l_len = 0; 2392 lf.l_type = F_UNLCK; 2393 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2394 F_UNLCK, &lf, F_POSIX); 2395 } 2396 fdtol = td->td_proc->p_fdtol; 2397 if (fdtol != NULL) { 2398 /* 2399 * Handle special case where file descriptor table is 2400 * shared between multiple process leaders. 2401 */ 2402 fdp = td->td_proc->p_fd; 2403 FILEDESC_XLOCK(fdp); 2404 for (fdtol = fdtol->fdl_next; 2405 fdtol != td->td_proc->p_fdtol; 2406 fdtol = fdtol->fdl_next) { 2407 if ((fdtol->fdl_leader->p_flag & 2408 P_ADVLOCK) == 0) 2409 continue; 2410 fdtol->fdl_holdcount++; 2411 FILEDESC_XUNLOCK(fdp); 2412 lf.l_whence = SEEK_SET; 2413 lf.l_start = 0; 2414 lf.l_len = 0; 2415 lf.l_type = F_UNLCK; 2416 vp = fp->f_vnode; 2417 (void) VOP_ADVLOCK(vp, 2418 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2419 F_POSIX); 2420 FILEDESC_XLOCK(fdp); 2421 fdtol->fdl_holdcount--; 2422 if (fdtol->fdl_holdcount == 0 && 2423 fdtol->fdl_wakeup != 0) { 2424 fdtol->fdl_wakeup = 0; 2425 wakeup(fdtol); 2426 } 2427 } 2428 FILEDESC_XUNLOCK(fdp); 2429 } 2430 } 2431 return (fdrop(fp, td)); 2432 } 2433 2434 /* 2435 * Initialize the file pointer with the specified properties. 2436 * 2437 * The ops are set with release semantics to be certain that the flags, type, 2438 * and data are visible when ops is. This is to prevent ops methods from being 2439 * called with bad data. 2440 */ 2441 void 2442 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2443 { 2444 fp->f_data = data; 2445 fp->f_flag = flag; 2446 fp->f_type = type; 2447 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2448 } 2449 2450 int 2451 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2452 struct file **fpp, seq_t *seqp) 2453 { 2454 #ifdef CAPABILITIES 2455 struct filedescent *fde; 2456 #endif 2457 struct fdescenttbl *fdt; 2458 struct file *fp; 2459 u_int count; 2460 #ifdef CAPABILITIES 2461 seq_t seq; 2462 cap_rights_t haverights; 2463 int error; 2464 #endif 2465 2466 fdt = fdp->fd_files; 2467 if ((u_int)fd >= fdt->fdt_nfiles) 2468 return (EBADF); 2469 /* 2470 * Fetch the descriptor locklessly. We avoid fdrop() races by 2471 * never raising a refcount above 0. To accomplish this we have 2472 * to use a cmpset loop rather than an atomic_add. The descriptor 2473 * must be re-verified once we acquire a reference to be certain 2474 * that the identity is still correct and we did not lose a race 2475 * due to preemption. 2476 */ 2477 for (;;) { 2478 #ifdef CAPABILITIES 2479 seq = seq_read(fd_seq(fdt, fd)); 2480 fde = &fdt->fdt_ofiles[fd]; 2481 haverights = *cap_rights_fde(fde); 2482 fp = fde->fde_file; 2483 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2484 cpu_spinwait(); 2485 continue; 2486 } 2487 #else 2488 fp = fdt->fdt_ofiles[fd].fde_file; 2489 #endif 2490 if (fp == NULL) 2491 return (EBADF); 2492 #ifdef CAPABILITIES 2493 error = cap_check(&haverights, needrightsp); 2494 if (error != 0) 2495 return (error); 2496 #endif 2497 retry: 2498 count = fp->f_count; 2499 if (count == 0) { 2500 /* 2501 * Force a reload. Other thread could reallocate the 2502 * table before this fd was closed, so it possible that 2503 * there is a stale fp pointer in cached version. 2504 */ 2505 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2506 continue; 2507 } 2508 /* 2509 * Use an acquire barrier to force re-reading of fdt so it is 2510 * refreshed for verification. 2511 */ 2512 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2513 goto retry; 2514 fdt = fdp->fd_files; 2515 #ifdef CAPABILITIES 2516 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2517 #else 2518 if (fp == fdt->fdt_ofiles[fd].fde_file) 2519 #endif 2520 break; 2521 fdrop(fp, curthread); 2522 } 2523 *fpp = fp; 2524 if (seqp != NULL) { 2525 #ifdef CAPABILITIES 2526 *seqp = seq; 2527 #endif 2528 } 2529 return (0); 2530 } 2531 2532 /* 2533 * Extract the file pointer associated with the specified descriptor for the 2534 * current user process. 2535 * 2536 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2537 * returned. 2538 * 2539 * File's rights will be checked against the capability rights mask. 2540 * 2541 * If an error occurred the non-zero error is returned and *fpp is set to 2542 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2543 * responsible for fdrop(). 2544 */ 2545 static __inline int 2546 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2547 cap_rights_t *needrightsp, seq_t *seqp) 2548 { 2549 struct filedesc *fdp; 2550 struct file *fp; 2551 int error; 2552 2553 *fpp = NULL; 2554 fdp = td->td_proc->p_fd; 2555 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2556 if (error != 0) 2557 return (error); 2558 if (fp->f_ops == &badfileops) { 2559 fdrop(fp, td); 2560 return (EBADF); 2561 } 2562 2563 /* 2564 * FREAD and FWRITE failure return EBADF as per POSIX. 2565 */ 2566 error = 0; 2567 switch (flags) { 2568 case FREAD: 2569 case FWRITE: 2570 if ((fp->f_flag & flags) == 0) 2571 error = EBADF; 2572 break; 2573 case FEXEC: 2574 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2575 ((fp->f_flag & FWRITE) != 0)) 2576 error = EBADF; 2577 break; 2578 case 0: 2579 break; 2580 default: 2581 KASSERT(0, ("wrong flags")); 2582 } 2583 2584 if (error != 0) { 2585 fdrop(fp, td); 2586 return (error); 2587 } 2588 2589 *fpp = fp; 2590 return (0); 2591 } 2592 2593 int 2594 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2595 { 2596 2597 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2598 } 2599 2600 int 2601 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2602 struct file **fpp) 2603 { 2604 int error; 2605 #ifndef CAPABILITIES 2606 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2607 if (maxprotp != NULL) 2608 *maxprotp = VM_PROT_ALL; 2609 #else 2610 struct filedesc *fdp = td->td_proc->p_fd; 2611 seq_t seq; 2612 2613 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2614 for (;;) { 2615 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2616 if (error != 0) 2617 return (error); 2618 /* 2619 * If requested, convert capability rights to access flags. 2620 */ 2621 if (maxprotp != NULL) 2622 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2623 if (!fd_modified(fdp, fd, seq)) 2624 break; 2625 fdrop(*fpp, td); 2626 } 2627 #endif 2628 return (error); 2629 } 2630 2631 int 2632 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2633 { 2634 2635 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2636 } 2637 2638 int 2639 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2640 { 2641 2642 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2643 } 2644 2645 int 2646 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2647 struct file **fpp) 2648 { 2649 struct filedesc *fdp = td->td_proc->p_fd; 2650 #ifndef CAPABILITIES 2651 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2652 #else 2653 int error; 2654 seq_t seq; 2655 2656 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2657 for (;;) { 2658 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2659 if (error != 0) 2660 return (error); 2661 error = cap_fcntl_check(fdp, fd, needfcntl); 2662 if (!fd_modified(fdp, fd, seq)) 2663 break; 2664 fdrop(*fpp, td); 2665 } 2666 if (error != 0) { 2667 fdrop(*fpp, td); 2668 *fpp = NULL; 2669 } 2670 return (error); 2671 #endif 2672 } 2673 2674 /* 2675 * Like fget() but loads the underlying vnode, or returns an error if the 2676 * descriptor does not represent a vnode. Note that pipes use vnodes but 2677 * never have VM objects. The returned vnode will be vref()'d. 2678 * 2679 * XXX: what about the unused flags ? 2680 */ 2681 static __inline int 2682 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2683 struct vnode **vpp) 2684 { 2685 struct file *fp; 2686 int error; 2687 2688 *vpp = NULL; 2689 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2690 if (error != 0) 2691 return (error); 2692 if (fp->f_vnode == NULL) { 2693 error = EINVAL; 2694 } else { 2695 *vpp = fp->f_vnode; 2696 vref(*vpp); 2697 } 2698 fdrop(fp, td); 2699 2700 return (error); 2701 } 2702 2703 int 2704 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2705 { 2706 2707 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2708 } 2709 2710 int 2711 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2712 struct filecaps *havecaps, struct vnode **vpp) 2713 { 2714 struct filedesc *fdp; 2715 struct file *fp; 2716 #ifdef CAPABILITIES 2717 int error; 2718 #endif 2719 2720 fdp = td->td_proc->p_fd; 2721 fp = fget_locked(fdp, fd); 2722 if (fp == NULL || fp->f_ops == &badfileops) 2723 return (EBADF); 2724 2725 #ifdef CAPABILITIES 2726 error = cap_check(cap_rights(fdp, fd), needrightsp); 2727 if (error != 0) 2728 return (error); 2729 #endif 2730 2731 if (fp->f_vnode == NULL) 2732 return (EINVAL); 2733 2734 *vpp = fp->f_vnode; 2735 vref(*vpp); 2736 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps, true); 2737 2738 return (0); 2739 } 2740 2741 int 2742 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2743 { 2744 2745 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2746 } 2747 2748 int 2749 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2750 { 2751 2752 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2753 } 2754 2755 #ifdef notyet 2756 int 2757 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2758 struct vnode **vpp) 2759 { 2760 2761 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2762 } 2763 #endif 2764 2765 /* 2766 * Like fget() but loads the underlying socket, or returns an error if the 2767 * descriptor does not represent a socket. 2768 * 2769 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2770 * in the future. 2771 * 2772 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2773 * on their file descriptor reference to prevent the socket from being free'd 2774 * during use. 2775 */ 2776 int 2777 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2778 u_int *fflagp) 2779 { 2780 struct file *fp; 2781 int error; 2782 2783 *spp = NULL; 2784 if (fflagp != NULL) 2785 *fflagp = 0; 2786 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2787 return (error); 2788 if (fp->f_type != DTYPE_SOCKET) { 2789 error = ENOTSOCK; 2790 } else { 2791 *spp = fp->f_data; 2792 if (fflagp) 2793 *fflagp = fp->f_flag; 2794 SOCK_LOCK(*spp); 2795 soref(*spp); 2796 SOCK_UNLOCK(*spp); 2797 } 2798 fdrop(fp, td); 2799 2800 return (error); 2801 } 2802 2803 /* 2804 * Drop the reference count on the socket and XXX release the SX lock in the 2805 * future. The last reference closes the socket. 2806 * 2807 * Note: fputsock() is deprecated, see comment for fgetsock(). 2808 */ 2809 void 2810 fputsock(struct socket *so) 2811 { 2812 2813 ACCEPT_LOCK(); 2814 SOCK_LOCK(so); 2815 CURVNET_SET(so->so_vnet); 2816 sorele(so); 2817 CURVNET_RESTORE(); 2818 } 2819 2820 /* 2821 * Handle the last reference to a file being closed. 2822 */ 2823 int 2824 _fdrop(struct file *fp, struct thread *td) 2825 { 2826 int error; 2827 2828 if (fp->f_count != 0) 2829 panic("fdrop: count %d", fp->f_count); 2830 error = fo_close(fp, td); 2831 atomic_subtract_int(&openfiles, 1); 2832 crfree(fp->f_cred); 2833 free(fp->f_advice, M_FADVISE); 2834 uma_zfree(file_zone, fp); 2835 2836 return (error); 2837 } 2838 2839 /* 2840 * Apply an advisory lock on a file descriptor. 2841 * 2842 * Just attempt to get a record lock of the requested type on the entire file 2843 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2844 */ 2845 #ifndef _SYS_SYSPROTO_H_ 2846 struct flock_args { 2847 int fd; 2848 int how; 2849 }; 2850 #endif 2851 /* ARGSUSED */ 2852 int 2853 sys_flock(struct thread *td, struct flock_args *uap) 2854 { 2855 struct file *fp; 2856 struct vnode *vp; 2857 struct flock lf; 2858 cap_rights_t rights; 2859 int error; 2860 2861 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2862 if (error != 0) 2863 return (error); 2864 if (fp->f_type != DTYPE_VNODE) { 2865 fdrop(fp, td); 2866 return (EOPNOTSUPP); 2867 } 2868 2869 vp = fp->f_vnode; 2870 lf.l_whence = SEEK_SET; 2871 lf.l_start = 0; 2872 lf.l_len = 0; 2873 if (uap->how & LOCK_UN) { 2874 lf.l_type = F_UNLCK; 2875 atomic_clear_int(&fp->f_flag, FHASLOCK); 2876 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2877 goto done2; 2878 } 2879 if (uap->how & LOCK_EX) 2880 lf.l_type = F_WRLCK; 2881 else if (uap->how & LOCK_SH) 2882 lf.l_type = F_RDLCK; 2883 else { 2884 error = EBADF; 2885 goto done2; 2886 } 2887 atomic_set_int(&fp->f_flag, FHASLOCK); 2888 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2889 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2890 done2: 2891 fdrop(fp, td); 2892 return (error); 2893 } 2894 /* 2895 * Duplicate the specified descriptor to a free descriptor. 2896 */ 2897 int 2898 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2899 int openerror, int *indxp) 2900 { 2901 struct filedescent *newfde, *oldfde; 2902 struct file *fp; 2903 int error, indx; 2904 2905 KASSERT(openerror == ENODEV || openerror == ENXIO, 2906 ("unexpected error %d in %s", openerror, __func__)); 2907 2908 /* 2909 * If the to-be-dup'd fd number is greater than the allowed number 2910 * of file descriptors, or the fd to be dup'd has already been 2911 * closed, then reject. 2912 */ 2913 FILEDESC_XLOCK(fdp); 2914 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2915 FILEDESC_XUNLOCK(fdp); 2916 return (EBADF); 2917 } 2918 2919 error = fdalloc(td, 0, &indx); 2920 if (error != 0) { 2921 FILEDESC_XUNLOCK(fdp); 2922 return (error); 2923 } 2924 2925 /* 2926 * There are two cases of interest here. 2927 * 2928 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2929 * 2930 * For ENXIO steal away the file structure from (dfd) and store it in 2931 * (indx). (dfd) is effectively closed by this operation. 2932 */ 2933 switch (openerror) { 2934 case ENODEV: 2935 /* 2936 * Check that the mode the file is being opened for is a 2937 * subset of the mode of the existing descriptor. 2938 */ 2939 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2940 fdunused(fdp, indx); 2941 FILEDESC_XUNLOCK(fdp); 2942 return (EACCES); 2943 } 2944 fhold(fp); 2945 newfde = &fdp->fd_ofiles[indx]; 2946 oldfde = &fdp->fd_ofiles[dfd]; 2947 #ifdef CAPABILITIES 2948 seq_write_begin(&newfde->fde_seq); 2949 #endif 2950 memcpy(newfde, oldfde, fde_change_size); 2951 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); 2952 #ifdef CAPABILITIES 2953 seq_write_end(&newfde->fde_seq); 2954 #endif 2955 break; 2956 case ENXIO: 2957 /* 2958 * Steal away the file pointer from dfd and stuff it into indx. 2959 */ 2960 newfde = &fdp->fd_ofiles[indx]; 2961 oldfde = &fdp->fd_ofiles[dfd]; 2962 #ifdef CAPABILITIES 2963 seq_write_begin(&newfde->fde_seq); 2964 #endif 2965 memcpy(newfde, oldfde, fde_change_size); 2966 oldfde->fde_file = NULL; 2967 fdunused(fdp, dfd); 2968 #ifdef CAPABILITIES 2969 seq_write_end(&newfde->fde_seq); 2970 #endif 2971 break; 2972 } 2973 FILEDESC_XUNLOCK(fdp); 2974 *indxp = indx; 2975 return (0); 2976 } 2977 2978 /* 2979 * This sysctl determines if we will allow a process to chroot(2) if it 2980 * has a directory open: 2981 * 0: disallowed for all processes. 2982 * 1: allowed for processes that were not already chroot(2)'ed. 2983 * 2: allowed for all processes. 2984 */ 2985 2986 static int chroot_allow_open_directories = 1; 2987 2988 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 2989 &chroot_allow_open_directories, 0, 2990 "Allow a process to chroot(2) if it has a directory open"); 2991 2992 /* 2993 * Helper function for raised chroot(2) security function: Refuse if 2994 * any filedescriptors are open directories. 2995 */ 2996 static int 2997 chroot_refuse_vdir_fds(struct filedesc *fdp) 2998 { 2999 struct vnode *vp; 3000 struct file *fp; 3001 int fd; 3002 3003 FILEDESC_LOCK_ASSERT(fdp); 3004 3005 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 3006 fp = fget_locked(fdp, fd); 3007 if (fp == NULL) 3008 continue; 3009 if (fp->f_type == DTYPE_VNODE) { 3010 vp = fp->f_vnode; 3011 if (vp->v_type == VDIR) 3012 return (EPERM); 3013 } 3014 } 3015 return (0); 3016 } 3017 3018 /* 3019 * Common routine for kern_chroot() and jail_attach(). The caller is 3020 * responsible for invoking priv_check() and mac_vnode_check_chroot() to 3021 * authorize this operation. 3022 */ 3023 int 3024 pwd_chroot(struct thread *td, struct vnode *vp) 3025 { 3026 struct filedesc *fdp; 3027 struct vnode *oldvp; 3028 int error; 3029 3030 fdp = td->td_proc->p_fd; 3031 FILEDESC_XLOCK(fdp); 3032 if (chroot_allow_open_directories == 0 || 3033 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 3034 error = chroot_refuse_vdir_fds(fdp); 3035 if (error != 0) { 3036 FILEDESC_XUNLOCK(fdp); 3037 return (error); 3038 } 3039 } 3040 oldvp = fdp->fd_rdir; 3041 VREF(vp); 3042 fdp->fd_rdir = vp; 3043 if (fdp->fd_jdir == NULL) { 3044 VREF(vp); 3045 fdp->fd_jdir = vp; 3046 } 3047 FILEDESC_XUNLOCK(fdp); 3048 vrele(oldvp); 3049 return (0); 3050 } 3051 3052 void 3053 pwd_chdir(struct thread *td, struct vnode *vp) 3054 { 3055 struct filedesc *fdp; 3056 struct vnode *oldvp; 3057 3058 fdp = td->td_proc->p_fd; 3059 FILEDESC_XLOCK(fdp); 3060 VNASSERT(vp->v_usecount > 0, vp, 3061 ("chdir to a vnode with zero usecount")); 3062 oldvp = fdp->fd_cdir; 3063 fdp->fd_cdir = vp; 3064 FILEDESC_XUNLOCK(fdp); 3065 vrele(oldvp); 3066 } 3067 3068 /* 3069 * Scan all active processes and prisons to see if any of them have a current 3070 * or root directory of `olddp'. If so, replace them with the new mount point. 3071 */ 3072 void 3073 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 3074 { 3075 struct filedesc *fdp; 3076 struct prison *pr; 3077 struct proc *p; 3078 int nrele; 3079 3080 if (vrefcnt(olddp) == 1) 3081 return; 3082 nrele = 0; 3083 sx_slock(&allproc_lock); 3084 FOREACH_PROC_IN_SYSTEM(p) { 3085 PROC_LOCK(p); 3086 fdp = fdhold(p); 3087 PROC_UNLOCK(p); 3088 if (fdp == NULL) 3089 continue; 3090 FILEDESC_XLOCK(fdp); 3091 if (fdp->fd_cdir == olddp) { 3092 vref(newdp); 3093 fdp->fd_cdir = newdp; 3094 nrele++; 3095 } 3096 if (fdp->fd_rdir == olddp) { 3097 vref(newdp); 3098 fdp->fd_rdir = newdp; 3099 nrele++; 3100 } 3101 if (fdp->fd_jdir == olddp) { 3102 vref(newdp); 3103 fdp->fd_jdir = newdp; 3104 nrele++; 3105 } 3106 FILEDESC_XUNLOCK(fdp); 3107 fddrop(fdp); 3108 } 3109 sx_sunlock(&allproc_lock); 3110 if (rootvnode == olddp) { 3111 vref(newdp); 3112 rootvnode = newdp; 3113 nrele++; 3114 } 3115 mtx_lock(&prison0.pr_mtx); 3116 if (prison0.pr_root == olddp) { 3117 vref(newdp); 3118 prison0.pr_root = newdp; 3119 nrele++; 3120 } 3121 mtx_unlock(&prison0.pr_mtx); 3122 sx_slock(&allprison_lock); 3123 TAILQ_FOREACH(pr, &allprison, pr_list) { 3124 mtx_lock(&pr->pr_mtx); 3125 if (pr->pr_root == olddp) { 3126 vref(newdp); 3127 pr->pr_root = newdp; 3128 nrele++; 3129 } 3130 mtx_unlock(&pr->pr_mtx); 3131 } 3132 sx_sunlock(&allprison_lock); 3133 while (nrele--) 3134 vrele(olddp); 3135 } 3136 3137 struct filedesc_to_leader * 3138 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 3139 { 3140 struct filedesc_to_leader *fdtol; 3141 3142 fdtol = malloc(sizeof(struct filedesc_to_leader), 3143 M_FILEDESC_TO_LEADER, M_WAITOK); 3144 fdtol->fdl_refcount = 1; 3145 fdtol->fdl_holdcount = 0; 3146 fdtol->fdl_wakeup = 0; 3147 fdtol->fdl_leader = leader; 3148 if (old != NULL) { 3149 FILEDESC_XLOCK(fdp); 3150 fdtol->fdl_next = old->fdl_next; 3151 fdtol->fdl_prev = old; 3152 old->fdl_next = fdtol; 3153 fdtol->fdl_next->fdl_prev = fdtol; 3154 FILEDESC_XUNLOCK(fdp); 3155 } else { 3156 fdtol->fdl_next = fdtol; 3157 fdtol->fdl_prev = fdtol; 3158 } 3159 return (fdtol); 3160 } 3161 3162 static int 3163 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 3164 { 3165 struct filedesc *fdp; 3166 int i, count, slots; 3167 3168 if (*(int *)arg1 != 0) 3169 return (EINVAL); 3170 3171 fdp = curproc->p_fd; 3172 count = 0; 3173 FILEDESC_SLOCK(fdp); 3174 slots = NDSLOTS(fdp->fd_lastfile + 1); 3175 for (i = 0; i < slots; i++) 3176 count += bitcountl(fdp->fd_map[i]); 3177 FILEDESC_SUNLOCK(fdp); 3178 3179 return (SYSCTL_OUT(req, &count, sizeof(count))); 3180 } 3181 3182 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 3183 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 3184 "Number of open file descriptors"); 3185 3186 /* 3187 * Get file structures globally. 3188 */ 3189 static int 3190 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 3191 { 3192 struct xfile xf; 3193 struct filedesc *fdp; 3194 struct file *fp; 3195 struct proc *p; 3196 int error, n; 3197 3198 error = sysctl_wire_old_buffer(req, 0); 3199 if (error != 0) 3200 return (error); 3201 if (req->oldptr == NULL) { 3202 n = 0; 3203 sx_slock(&allproc_lock); 3204 FOREACH_PROC_IN_SYSTEM(p) { 3205 PROC_LOCK(p); 3206 if (p->p_state == PRS_NEW) { 3207 PROC_UNLOCK(p); 3208 continue; 3209 } 3210 fdp = fdhold(p); 3211 PROC_UNLOCK(p); 3212 if (fdp == NULL) 3213 continue; 3214 /* overestimates sparse tables. */ 3215 if (fdp->fd_lastfile > 0) 3216 n += fdp->fd_lastfile; 3217 fddrop(fdp); 3218 } 3219 sx_sunlock(&allproc_lock); 3220 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 3221 } 3222 error = 0; 3223 bzero(&xf, sizeof(xf)); 3224 xf.xf_size = sizeof(xf); 3225 sx_slock(&allproc_lock); 3226 FOREACH_PROC_IN_SYSTEM(p) { 3227 PROC_LOCK(p); 3228 if (p->p_state == PRS_NEW) { 3229 PROC_UNLOCK(p); 3230 continue; 3231 } 3232 if (p_cansee(req->td, p) != 0) { 3233 PROC_UNLOCK(p); 3234 continue; 3235 } 3236 xf.xf_pid = p->p_pid; 3237 xf.xf_uid = p->p_ucred->cr_uid; 3238 fdp = fdhold(p); 3239 PROC_UNLOCK(p); 3240 if (fdp == NULL) 3241 continue; 3242 FILEDESC_SLOCK(fdp); 3243 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3244 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3245 continue; 3246 xf.xf_fd = n; 3247 xf.xf_file = fp; 3248 xf.xf_data = fp->f_data; 3249 xf.xf_vnode = fp->f_vnode; 3250 xf.xf_type = fp->f_type; 3251 xf.xf_count = fp->f_count; 3252 xf.xf_msgcount = 0; 3253 xf.xf_offset = foffset_get(fp); 3254 xf.xf_flag = fp->f_flag; 3255 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3256 if (error) 3257 break; 3258 } 3259 FILEDESC_SUNLOCK(fdp); 3260 fddrop(fdp); 3261 if (error) 3262 break; 3263 } 3264 sx_sunlock(&allproc_lock); 3265 return (error); 3266 } 3267 3268 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3269 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3270 3271 #ifdef KINFO_FILE_SIZE 3272 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3273 #endif 3274 3275 static int 3276 xlate_fflags(int fflags) 3277 { 3278 static const struct { 3279 int fflag; 3280 int kf_fflag; 3281 } fflags_table[] = { 3282 { FAPPEND, KF_FLAG_APPEND }, 3283 { FASYNC, KF_FLAG_ASYNC }, 3284 { FFSYNC, KF_FLAG_FSYNC }, 3285 { FHASLOCK, KF_FLAG_HASLOCK }, 3286 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3287 { FREAD, KF_FLAG_READ }, 3288 { FWRITE, KF_FLAG_WRITE }, 3289 { O_CREAT, KF_FLAG_CREAT }, 3290 { O_DIRECT, KF_FLAG_DIRECT }, 3291 { O_EXCL, KF_FLAG_EXCL }, 3292 { O_EXEC, KF_FLAG_EXEC }, 3293 { O_EXLOCK, KF_FLAG_EXLOCK }, 3294 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3295 { O_SHLOCK, KF_FLAG_SHLOCK }, 3296 { O_TRUNC, KF_FLAG_TRUNC } 3297 }; 3298 unsigned int i; 3299 int kflags; 3300 3301 kflags = 0; 3302 for (i = 0; i < nitems(fflags_table); i++) 3303 if (fflags & fflags_table[i].fflag) 3304 kflags |= fflags_table[i].kf_fflag; 3305 return (kflags); 3306 } 3307 3308 /* Trim unused data from kf_path by truncating the structure size. */ 3309 static void 3310 pack_kinfo(struct kinfo_file *kif) 3311 { 3312 3313 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3314 strlen(kif->kf_path) + 1; 3315 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3316 } 3317 3318 static void 3319 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3320 struct kinfo_file *kif, struct filedesc *fdp, int flags) 3321 { 3322 int error; 3323 3324 bzero(kif, sizeof(*kif)); 3325 3326 /* Set a default type to allow for empty fill_kinfo() methods. */ 3327 kif->kf_type = KF_TYPE_UNKNOWN; 3328 kif->kf_flags = xlate_fflags(fp->f_flag); 3329 if (rightsp != NULL) 3330 kif->kf_cap_rights = *rightsp; 3331 else 3332 cap_rights_init(&kif->kf_cap_rights); 3333 kif->kf_fd = fd; 3334 kif->kf_ref_count = fp->f_count; 3335 kif->kf_offset = foffset_get(fp); 3336 3337 /* 3338 * This may drop the filedesc lock, so the 'fp' cannot be 3339 * accessed after this call. 3340 */ 3341 error = fo_fill_kinfo(fp, kif, fdp); 3342 if (error == 0) 3343 kif->kf_status |= KF_ATTR_VALID; 3344 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3345 pack_kinfo(kif); 3346 else 3347 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3348 } 3349 3350 static void 3351 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3352 struct kinfo_file *kif, int flags) 3353 { 3354 int error; 3355 3356 bzero(kif, sizeof(*kif)); 3357 3358 kif->kf_type = KF_TYPE_VNODE; 3359 error = vn_fill_kinfo_vnode(vp, kif); 3360 if (error == 0) 3361 kif->kf_status |= KF_ATTR_VALID; 3362 kif->kf_flags = xlate_fflags(fflags); 3363 cap_rights_init(&kif->kf_cap_rights); 3364 kif->kf_fd = fd; 3365 kif->kf_ref_count = -1; 3366 kif->kf_offset = -1; 3367 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 3368 pack_kinfo(kif); 3369 else 3370 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 3371 vrele(vp); 3372 } 3373 3374 struct export_fd_buf { 3375 struct filedesc *fdp; 3376 struct sbuf *sb; 3377 ssize_t remainder; 3378 struct kinfo_file kif; 3379 int flags; 3380 }; 3381 3382 static int 3383 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3384 { 3385 struct kinfo_file *kif; 3386 3387 kif = &efbuf->kif; 3388 if (efbuf->remainder != -1) { 3389 if (efbuf->remainder < kif->kf_structsize) { 3390 /* Terminate export. */ 3391 efbuf->remainder = 0; 3392 return (0); 3393 } 3394 efbuf->remainder -= kif->kf_structsize; 3395 } 3396 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3397 } 3398 3399 static int 3400 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3401 struct export_fd_buf *efbuf) 3402 { 3403 int error; 3404 3405 if (efbuf->remainder == 0) 3406 return (0); 3407 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 3408 efbuf->flags); 3409 FILEDESC_SUNLOCK(efbuf->fdp); 3410 error = export_kinfo_to_sb(efbuf); 3411 FILEDESC_SLOCK(efbuf->fdp); 3412 return (error); 3413 } 3414 3415 static int 3416 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3417 struct export_fd_buf *efbuf) 3418 { 3419 int error; 3420 3421 if (efbuf->remainder == 0) 3422 return (0); 3423 if (efbuf->fdp != NULL) 3424 FILEDESC_SUNLOCK(efbuf->fdp); 3425 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 3426 error = export_kinfo_to_sb(efbuf); 3427 if (efbuf->fdp != NULL) 3428 FILEDESC_SLOCK(efbuf->fdp); 3429 return (error); 3430 } 3431 3432 /* 3433 * Store a process file descriptor information to sbuf. 3434 * 3435 * Takes a locked proc as argument, and returns with the proc unlocked. 3436 */ 3437 int 3438 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 3439 int flags) 3440 { 3441 struct file *fp; 3442 struct filedesc *fdp; 3443 struct export_fd_buf *efbuf; 3444 struct vnode *cttyvp, *textvp, *tracevp; 3445 int error, i; 3446 cap_rights_t rights; 3447 3448 PROC_LOCK_ASSERT(p, MA_OWNED); 3449 3450 /* ktrace vnode */ 3451 tracevp = p->p_tracevp; 3452 if (tracevp != NULL) 3453 vref(tracevp); 3454 /* text vnode */ 3455 textvp = p->p_textvp; 3456 if (textvp != NULL) 3457 vref(textvp); 3458 /* Controlling tty. */ 3459 cttyvp = NULL; 3460 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3461 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3462 if (cttyvp != NULL) 3463 vref(cttyvp); 3464 } 3465 fdp = fdhold(p); 3466 PROC_UNLOCK(p); 3467 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3468 efbuf->fdp = NULL; 3469 efbuf->sb = sb; 3470 efbuf->remainder = maxlen; 3471 efbuf->flags = flags; 3472 if (tracevp != NULL) 3473 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3474 efbuf); 3475 if (textvp != NULL) 3476 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3477 if (cttyvp != NULL) 3478 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3479 efbuf); 3480 error = 0; 3481 if (fdp == NULL) 3482 goto fail; 3483 efbuf->fdp = fdp; 3484 FILEDESC_SLOCK(fdp); 3485 /* working directory */ 3486 if (fdp->fd_cdir != NULL) { 3487 vref(fdp->fd_cdir); 3488 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3489 } 3490 /* root directory */ 3491 if (fdp->fd_rdir != NULL) { 3492 vref(fdp->fd_rdir); 3493 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3494 } 3495 /* jail directory */ 3496 if (fdp->fd_jdir != NULL) { 3497 vref(fdp->fd_jdir); 3498 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3499 } 3500 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3501 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3502 continue; 3503 #ifdef CAPABILITIES 3504 rights = *cap_rights(fdp, i); 3505 #else /* !CAPABILITIES */ 3506 cap_rights_init(&rights); 3507 #endif 3508 /* 3509 * Create sysctl entry. It is OK to drop the filedesc 3510 * lock inside of export_file_to_sb() as we will 3511 * re-validate and re-evaluate its properties when the 3512 * loop continues. 3513 */ 3514 error = export_file_to_sb(fp, i, &rights, efbuf); 3515 if (error != 0 || efbuf->remainder == 0) 3516 break; 3517 } 3518 FILEDESC_SUNLOCK(fdp); 3519 fddrop(fdp); 3520 fail: 3521 free(efbuf, M_TEMP); 3522 return (error); 3523 } 3524 3525 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3526 3527 /* 3528 * Get per-process file descriptors for use by procstat(1), et al. 3529 */ 3530 static int 3531 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3532 { 3533 struct sbuf sb; 3534 struct proc *p; 3535 ssize_t maxlen; 3536 int error, error2, *name; 3537 3538 name = (int *)arg1; 3539 3540 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3541 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3542 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3543 if (error != 0) { 3544 sbuf_delete(&sb); 3545 return (error); 3546 } 3547 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3548 error = kern_proc_filedesc_out(p, &sb, maxlen, 3549 KERN_FILEDESC_PACK_KINFO); 3550 error2 = sbuf_finish(&sb); 3551 sbuf_delete(&sb); 3552 return (error != 0 ? error : error2); 3553 } 3554 3555 #ifdef KINFO_OFILE_SIZE 3556 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3557 #endif 3558 3559 #ifdef COMPAT_FREEBSD7 3560 static void 3561 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3562 { 3563 3564 okif->kf_structsize = sizeof(*okif); 3565 okif->kf_type = kif->kf_type; 3566 okif->kf_fd = kif->kf_fd; 3567 okif->kf_ref_count = kif->kf_ref_count; 3568 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3569 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3570 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3571 okif->kf_offset = kif->kf_offset; 3572 okif->kf_vnode_type = kif->kf_vnode_type; 3573 okif->kf_sock_domain = kif->kf_sock_domain; 3574 okif->kf_sock_type = kif->kf_sock_type; 3575 okif->kf_sock_protocol = kif->kf_sock_protocol; 3576 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3577 okif->kf_sa_local = kif->kf_sa_local; 3578 okif->kf_sa_peer = kif->kf_sa_peer; 3579 } 3580 3581 static int 3582 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3583 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3584 { 3585 int error; 3586 3587 vref(vp); 3588 FILEDESC_SUNLOCK(fdp); 3589 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 3590 kinfo_to_okinfo(kif, okif); 3591 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3592 FILEDESC_SLOCK(fdp); 3593 return (error); 3594 } 3595 3596 /* 3597 * Get per-process file descriptors for use by procstat(1), et al. 3598 */ 3599 static int 3600 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3601 { 3602 struct kinfo_ofile *okif; 3603 struct kinfo_file *kif; 3604 struct filedesc *fdp; 3605 int error, i, *name; 3606 struct file *fp; 3607 struct proc *p; 3608 3609 name = (int *)arg1; 3610 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3611 if (error != 0) 3612 return (error); 3613 fdp = fdhold(p); 3614 PROC_UNLOCK(p); 3615 if (fdp == NULL) 3616 return (ENOENT); 3617 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3618 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3619 FILEDESC_SLOCK(fdp); 3620 if (fdp->fd_cdir != NULL) 3621 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3622 okif, fdp, req); 3623 if (fdp->fd_rdir != NULL) 3624 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3625 okif, fdp, req); 3626 if (fdp->fd_jdir != NULL) 3627 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3628 okif, fdp, req); 3629 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3630 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3631 continue; 3632 export_file_to_kinfo(fp, i, NULL, kif, fdp, 3633 KERN_FILEDESC_PACK_KINFO); 3634 FILEDESC_SUNLOCK(fdp); 3635 kinfo_to_okinfo(kif, okif); 3636 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3637 FILEDESC_SLOCK(fdp); 3638 if (error) 3639 break; 3640 } 3641 FILEDESC_SUNLOCK(fdp); 3642 fddrop(fdp); 3643 free(kif, M_TEMP); 3644 free(okif, M_TEMP); 3645 return (0); 3646 } 3647 3648 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3649 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3650 "Process ofiledesc entries"); 3651 #endif /* COMPAT_FREEBSD7 */ 3652 3653 int 3654 vntype_to_kinfo(int vtype) 3655 { 3656 struct { 3657 int vtype; 3658 int kf_vtype; 3659 } vtypes_table[] = { 3660 { VBAD, KF_VTYPE_VBAD }, 3661 { VBLK, KF_VTYPE_VBLK }, 3662 { VCHR, KF_VTYPE_VCHR }, 3663 { VDIR, KF_VTYPE_VDIR }, 3664 { VFIFO, KF_VTYPE_VFIFO }, 3665 { VLNK, KF_VTYPE_VLNK }, 3666 { VNON, KF_VTYPE_VNON }, 3667 { VREG, KF_VTYPE_VREG }, 3668 { VSOCK, KF_VTYPE_VSOCK } 3669 }; 3670 unsigned int i; 3671 3672 /* 3673 * Perform vtype translation. 3674 */ 3675 for (i = 0; i < nitems(vtypes_table); i++) 3676 if (vtypes_table[i].vtype == vtype) 3677 return (vtypes_table[i].kf_vtype); 3678 3679 return (KF_VTYPE_UNKNOWN); 3680 } 3681 3682 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3683 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3684 "Process filedesc entries"); 3685 3686 /* 3687 * Store a process current working directory information to sbuf. 3688 * 3689 * Takes a locked proc as argument, and returns with the proc unlocked. 3690 */ 3691 int 3692 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3693 { 3694 struct filedesc *fdp; 3695 struct export_fd_buf *efbuf; 3696 int error; 3697 3698 PROC_LOCK_ASSERT(p, MA_OWNED); 3699 3700 fdp = fdhold(p); 3701 PROC_UNLOCK(p); 3702 if (fdp == NULL) 3703 return (EINVAL); 3704 3705 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3706 efbuf->fdp = fdp; 3707 efbuf->sb = sb; 3708 efbuf->remainder = maxlen; 3709 3710 FILEDESC_SLOCK(fdp); 3711 if (fdp->fd_cdir == NULL) 3712 error = EINVAL; 3713 else { 3714 vref(fdp->fd_cdir); 3715 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3716 FREAD, efbuf); 3717 } 3718 FILEDESC_SUNLOCK(fdp); 3719 fddrop(fdp); 3720 free(efbuf, M_TEMP); 3721 return (error); 3722 } 3723 3724 /* 3725 * Get per-process current working directory. 3726 */ 3727 static int 3728 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3729 { 3730 struct sbuf sb; 3731 struct proc *p; 3732 ssize_t maxlen; 3733 int error, error2, *name; 3734 3735 name = (int *)arg1; 3736 3737 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3738 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3739 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3740 if (error != 0) { 3741 sbuf_delete(&sb); 3742 return (error); 3743 } 3744 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3745 error = kern_proc_cwd_out(p, &sb, maxlen); 3746 error2 = sbuf_finish(&sb); 3747 sbuf_delete(&sb); 3748 return (error != 0 ? error : error2); 3749 } 3750 3751 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3752 sysctl_kern_proc_cwd, "Process current working directory"); 3753 3754 #ifdef DDB 3755 /* 3756 * For the purposes of debugging, generate a human-readable string for the 3757 * file type. 3758 */ 3759 static const char * 3760 file_type_to_name(short type) 3761 { 3762 3763 switch (type) { 3764 case 0: 3765 return ("zero"); 3766 case DTYPE_VNODE: 3767 return ("vnod"); 3768 case DTYPE_SOCKET: 3769 return ("sock"); 3770 case DTYPE_PIPE: 3771 return ("pipe"); 3772 case DTYPE_FIFO: 3773 return ("fifo"); 3774 case DTYPE_KQUEUE: 3775 return ("kque"); 3776 case DTYPE_CRYPTO: 3777 return ("crpt"); 3778 case DTYPE_MQUEUE: 3779 return ("mque"); 3780 case DTYPE_SHM: 3781 return ("shm"); 3782 case DTYPE_SEM: 3783 return ("ksem"); 3784 default: 3785 return ("unkn"); 3786 } 3787 } 3788 3789 /* 3790 * For the purposes of debugging, identify a process (if any, perhaps one of 3791 * many) that references the passed file in its file descriptor array. Return 3792 * NULL if none. 3793 */ 3794 static struct proc * 3795 file_to_first_proc(struct file *fp) 3796 { 3797 struct filedesc *fdp; 3798 struct proc *p; 3799 int n; 3800 3801 FOREACH_PROC_IN_SYSTEM(p) { 3802 if (p->p_state == PRS_NEW) 3803 continue; 3804 fdp = p->p_fd; 3805 if (fdp == NULL) 3806 continue; 3807 for (n = 0; n <= fdp->fd_lastfile; n++) { 3808 if (fp == fdp->fd_ofiles[n].fde_file) 3809 return (p); 3810 } 3811 } 3812 return (NULL); 3813 } 3814 3815 static void 3816 db_print_file(struct file *fp, int header) 3817 { 3818 struct proc *p; 3819 3820 if (header) 3821 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3822 "File", "Type", "Data", "Flag", "GCFl", "Count", 3823 "MCount", "Vnode", "FPID", "FCmd"); 3824 p = file_to_first_proc(fp); 3825 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3826 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3827 0, fp->f_count, 0, fp->f_vnode, 3828 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3829 } 3830 3831 DB_SHOW_COMMAND(file, db_show_file) 3832 { 3833 struct file *fp; 3834 3835 if (!have_addr) { 3836 db_printf("usage: show file <addr>\n"); 3837 return; 3838 } 3839 fp = (struct file *)addr; 3840 db_print_file(fp, 1); 3841 } 3842 3843 DB_SHOW_COMMAND(files, db_show_files) 3844 { 3845 struct filedesc *fdp; 3846 struct file *fp; 3847 struct proc *p; 3848 int header; 3849 int n; 3850 3851 header = 1; 3852 FOREACH_PROC_IN_SYSTEM(p) { 3853 if (p->p_state == PRS_NEW) 3854 continue; 3855 if ((fdp = p->p_fd) == NULL) 3856 continue; 3857 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3858 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3859 continue; 3860 db_print_file(fp, header); 3861 header = 0; 3862 } 3863 } 3864 } 3865 #endif 3866 3867 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3868 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3869 3870 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3871 &maxfiles, 0, "Maximum number of files"); 3872 3873 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3874 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3875 3876 /* ARGSUSED*/ 3877 static void 3878 filelistinit(void *dummy) 3879 { 3880 3881 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3882 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3883 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3884 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3885 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3886 } 3887 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3888 3889 /*-------------------------------------------------------------------*/ 3890 3891 static int 3892 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3893 int flags, struct thread *td) 3894 { 3895 3896 return (EBADF); 3897 } 3898 3899 static int 3900 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3901 struct thread *td) 3902 { 3903 3904 return (EINVAL); 3905 } 3906 3907 static int 3908 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3909 struct thread *td) 3910 { 3911 3912 return (EBADF); 3913 } 3914 3915 static int 3916 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3917 struct thread *td) 3918 { 3919 3920 return (0); 3921 } 3922 3923 static int 3924 badfo_kqfilter(struct file *fp, struct knote *kn) 3925 { 3926 3927 return (EBADF); 3928 } 3929 3930 static int 3931 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3932 struct thread *td) 3933 { 3934 3935 return (EBADF); 3936 } 3937 3938 static int 3939 badfo_close(struct file *fp, struct thread *td) 3940 { 3941 3942 return (0); 3943 } 3944 3945 static int 3946 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3947 struct thread *td) 3948 { 3949 3950 return (EBADF); 3951 } 3952 3953 static int 3954 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3955 struct thread *td) 3956 { 3957 3958 return (EBADF); 3959 } 3960 3961 static int 3962 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3963 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3964 struct thread *td) 3965 { 3966 3967 return (EBADF); 3968 } 3969 3970 static int 3971 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3972 { 3973 3974 return (0); 3975 } 3976 3977 struct fileops badfileops = { 3978 .fo_read = badfo_readwrite, 3979 .fo_write = badfo_readwrite, 3980 .fo_truncate = badfo_truncate, 3981 .fo_ioctl = badfo_ioctl, 3982 .fo_poll = badfo_poll, 3983 .fo_kqfilter = badfo_kqfilter, 3984 .fo_stat = badfo_stat, 3985 .fo_close = badfo_close, 3986 .fo_chmod = badfo_chmod, 3987 .fo_chown = badfo_chown, 3988 .fo_sendfile = badfo_sendfile, 3989 .fo_fill_kinfo = badfo_fill_kinfo, 3990 }; 3991 3992 int 3993 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3994 int flags, struct thread *td) 3995 { 3996 3997 return (EOPNOTSUPP); 3998 } 3999 4000 int 4001 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4002 struct thread *td) 4003 { 4004 4005 return (EINVAL); 4006 } 4007 4008 int 4009 invfo_ioctl(struct file *fp, u_long com, void *data, 4010 struct ucred *active_cred, struct thread *td) 4011 { 4012 4013 return (ENOTTY); 4014 } 4015 4016 int 4017 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 4018 struct thread *td) 4019 { 4020 4021 return (poll_no_poll(events)); 4022 } 4023 4024 int 4025 invfo_kqfilter(struct file *fp, struct knote *kn) 4026 { 4027 4028 return (EINVAL); 4029 } 4030 4031 int 4032 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 4033 struct thread *td) 4034 { 4035 4036 return (EINVAL); 4037 } 4038 4039 int 4040 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 4041 struct thread *td) 4042 { 4043 4044 return (EINVAL); 4045 } 4046 4047 int 4048 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4049 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4050 struct thread *td) 4051 { 4052 4053 return (EINVAL); 4054 } 4055 4056 /*-------------------------------------------------------------------*/ 4057 4058 /* 4059 * File Descriptor pseudo-device driver (/dev/fd/). 4060 * 4061 * Opening minor device N dup()s the file (if any) connected to file 4062 * descriptor N belonging to the calling process. Note that this driver 4063 * consists of only the ``open()'' routine, because all subsequent 4064 * references to this file will be direct to the other driver. 4065 * 4066 * XXX: we could give this one a cloning event handler if necessary. 4067 */ 4068 4069 /* ARGSUSED */ 4070 static int 4071 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4072 { 4073 4074 /* 4075 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4076 * the file descriptor being sought for duplication. The error 4077 * return ensures that the vnode for this device will be released 4078 * by vn_open. Open will detect this special error and take the 4079 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4080 * will simply report the error. 4081 */ 4082 td->td_dupfd = dev2unit(dev); 4083 return (ENODEV); 4084 } 4085 4086 static struct cdevsw fildesc_cdevsw = { 4087 .d_version = D_VERSION, 4088 .d_open = fdopen, 4089 .d_name = "FD", 4090 }; 4091 4092 static void 4093 fildesc_drvinit(void *unused) 4094 { 4095 struct cdev *dev; 4096 4097 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4098 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4099 make_dev_alias(dev, "stdin"); 4100 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4101 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4102 make_dev_alias(dev, "stdout"); 4103 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4104 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4105 make_dev_alias(dev, "stderr"); 4106 } 4107 4108 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4109