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/stat.h> 72 #include <sys/sx.h> 73 #include <sys/syscallsubr.h> 74 #include <sys/sysctl.h> 75 #include <sys/sysproto.h> 76 #include <sys/unistd.h> 77 #include <sys/user.h> 78 #include <sys/vnode.h> 79 #ifdef KTRACE 80 #include <sys/ktrace.h> 81 #endif 82 83 #include <net/vnet.h> 84 85 #include <security/audit/audit.h> 86 87 #include <vm/uma.h> 88 #include <vm/vm.h> 89 90 #include <ddb/ddb.h> 91 92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 94 "file desc to leader structures"); 95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 97 98 MALLOC_DECLARE(M_FADVISE); 99 100 static uma_zone_t file_zone; 101 static uma_zone_t filedesc0_zone; 102 103 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 104 struct thread *td, int holdleaders); 105 static int do_dup(struct thread *td, int flags, int old, int new); 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 /* Flags for do_dup() */ 115 #define DUP_FIXED 0x1 /* Force fixed allocation. */ 116 #define DUP_FCNTL 0x2 /* fcntl()-style errors. */ 117 #define DUP_CLOEXEC 0x4 /* Atomically set FD_CLOEXEC. */ 118 119 /* 120 * Each process has: 121 * 122 * - An array of open file descriptors (fd_ofiles) 123 * - An array of file flags (fd_ofileflags) 124 * - A bitmap recording which descriptors are in use (fd_map) 125 * 126 * A process starts out with NDFILE descriptors. The value of NDFILE has 127 * been selected based the historical limit of 20 open files, and an 128 * assumption that the majority of processes, especially short-lived 129 * processes like shells, will never need more. 130 * 131 * If this initial allocation is exhausted, a larger descriptor table and 132 * map are allocated dynamically, and the pointers in the process's struct 133 * filedesc are updated to point to those. This is repeated every time 134 * the process runs out of file descriptors (provided it hasn't hit its 135 * resource limit). 136 * 137 * Since threads may hold references to individual descriptor table 138 * entries, the tables are never freed. Instead, they are placed on a 139 * linked list and freed only when the struct filedesc is released. 140 */ 141 #define NDFILE 20 142 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 144 #define NDSLOT(x) ((x) / NDENTRIES) 145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 147 148 /* 149 * SLIST entry used to keep track of ofiles which must be reclaimed when 150 * the process exits. 151 */ 152 struct freetable { 153 struct fdescenttbl *ft_table; 154 SLIST_ENTRY(freetable) ft_next; 155 }; 156 157 /* 158 * Initial allocation: a filedesc structure + the head of SLIST used to 159 * keep track of old ofiles + enough space for NDFILE descriptors. 160 */ 161 162 struct fdescenttbl0 { 163 int fdt_nfiles; 164 struct filedescent fdt_ofiles[NDFILE]; 165 }; 166 167 struct filedesc0 { 168 struct filedesc fd_fd; 169 SLIST_HEAD(, freetable) fd_free; 170 struct fdescenttbl0 fd_dfiles; 171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 172 }; 173 174 /* 175 * Descriptor management. 176 */ 177 volatile int openfiles; /* actual number of open files */ 178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 179 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 180 181 /* 182 * If low >= size, just return low. Otherwise find the first zero bit in the 183 * given bitmap, starting at low and not exceeding size - 1. Return size if 184 * not found. 185 */ 186 static int 187 fd_first_free(struct filedesc *fdp, int low, int size) 188 { 189 NDSLOTTYPE *map = fdp->fd_map; 190 NDSLOTTYPE mask; 191 int off, maxoff; 192 193 if (low >= size) 194 return (low); 195 196 off = NDSLOT(low); 197 if (low % NDENTRIES) { 198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 199 if ((mask &= ~map[off]) != 0UL) 200 return (off * NDENTRIES + ffsl(mask) - 1); 201 ++off; 202 } 203 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 204 if (map[off] != ~0UL) 205 return (off * NDENTRIES + ffsl(~map[off]) - 1); 206 return (size); 207 } 208 209 /* 210 * Find the highest non-zero bit in the given bitmap, starting at 0 and 211 * not exceeding size - 1. Return -1 if not found. 212 */ 213 static int 214 fd_last_used(struct filedesc *fdp, int size) 215 { 216 NDSLOTTYPE *map = fdp->fd_map; 217 NDSLOTTYPE mask; 218 int off, minoff; 219 220 off = NDSLOT(size); 221 if (size % NDENTRIES) { 222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 223 if ((mask &= map[off]) != 0) 224 return (off * NDENTRIES + flsl(mask) - 1); 225 --off; 226 } 227 for (minoff = NDSLOT(0); off >= minoff; --off) 228 if (map[off] != 0) 229 return (off * NDENTRIES + flsl(map[off]) - 1); 230 return (-1); 231 } 232 233 #ifdef INVARIANTS 234 static int 235 fdisused(struct filedesc *fdp, int fd) 236 { 237 238 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 239 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 240 241 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 242 } 243 #endif 244 245 /* 246 * Mark a file descriptor as used. 247 */ 248 static void 249 fdused_init(struct filedesc *fdp, int fd) 250 { 251 252 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 253 254 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 255 } 256 257 static void 258 fdused(struct filedesc *fdp, int fd) 259 { 260 261 FILEDESC_XLOCK_ASSERT(fdp); 262 263 fdused_init(fdp, fd); 264 if (fd > fdp->fd_lastfile) 265 fdp->fd_lastfile = fd; 266 if (fd == fdp->fd_freefile) 267 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 268 } 269 270 /* 271 * Mark a file descriptor as unused. 272 */ 273 static void 274 fdunused(struct filedesc *fdp, int fd) 275 { 276 277 FILEDESC_XLOCK_ASSERT(fdp); 278 279 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 280 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 281 ("fd=%d is still in use", fd)); 282 283 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 284 if (fd < fdp->fd_freefile) 285 fdp->fd_freefile = fd; 286 if (fd == fdp->fd_lastfile) 287 fdp->fd_lastfile = fd_last_used(fdp, fd); 288 } 289 290 /* 291 * Free a file descriptor. 292 * 293 * Avoid some work if fdp is about to be destroyed. 294 */ 295 static inline void 296 fdefree_last(struct filedescent *fde) 297 { 298 299 filecaps_free(&fde->fde_caps); 300 } 301 302 static inline void 303 fdfree(struct filedesc *fdp, int fd) 304 { 305 struct filedescent *fde; 306 307 fde = &fdp->fd_ofiles[fd]; 308 #ifdef CAPABILITIES 309 seq_write_begin(&fde->fde_seq); 310 #endif 311 fdefree_last(fde); 312 fde->fde_file = NULL; 313 fdunused(fdp, fd); 314 #ifdef CAPABILITIES 315 seq_write_end(&fde->fde_seq); 316 #endif 317 } 318 319 /* 320 * System calls on descriptors. 321 */ 322 #ifndef _SYS_SYSPROTO_H_ 323 struct getdtablesize_args { 324 int dummy; 325 }; 326 #endif 327 /* ARGSUSED */ 328 int 329 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 330 { 331 #ifdef RACCT 332 uint64_t lim; 333 #endif 334 335 td->td_retval[0] = 336 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); 337 #ifdef RACCT 338 PROC_LOCK(td->td_proc); 339 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 340 PROC_UNLOCK(td->td_proc); 341 if (lim < td->td_retval[0]) 342 td->td_retval[0] = lim; 343 #endif 344 return (0); 345 } 346 347 /* 348 * Duplicate a file descriptor to a particular value. 349 * 350 * Note: keep in mind that a potential race condition exists when closing 351 * descriptors from a shared descriptor table (via rfork). 352 */ 353 #ifndef _SYS_SYSPROTO_H_ 354 struct dup2_args { 355 u_int from; 356 u_int to; 357 }; 358 #endif 359 /* ARGSUSED */ 360 int 361 sys_dup2(struct thread *td, struct dup2_args *uap) 362 { 363 364 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to)); 365 } 366 367 /* 368 * Duplicate a file descriptor. 369 */ 370 #ifndef _SYS_SYSPROTO_H_ 371 struct dup_args { 372 u_int fd; 373 }; 374 #endif 375 /* ARGSUSED */ 376 int 377 sys_dup(struct thread *td, struct dup_args *uap) 378 { 379 380 return (do_dup(td, 0, (int)uap->fd, 0)); 381 } 382 383 /* 384 * The file control system call. 385 */ 386 #ifndef _SYS_SYSPROTO_H_ 387 struct fcntl_args { 388 int fd; 389 int cmd; 390 long arg; 391 }; 392 #endif 393 /* ARGSUSED */ 394 int 395 sys_fcntl(struct thread *td, struct fcntl_args *uap) 396 { 397 398 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 399 } 400 401 int 402 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 403 { 404 struct flock fl; 405 struct __oflock ofl; 406 intptr_t arg1; 407 int error; 408 409 error = 0; 410 switch (cmd) { 411 case F_OGETLK: 412 case F_OSETLK: 413 case F_OSETLKW: 414 /* 415 * Convert old flock structure to new. 416 */ 417 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 418 fl.l_start = ofl.l_start; 419 fl.l_len = ofl.l_len; 420 fl.l_pid = ofl.l_pid; 421 fl.l_type = ofl.l_type; 422 fl.l_whence = ofl.l_whence; 423 fl.l_sysid = 0; 424 425 switch (cmd) { 426 case F_OGETLK: 427 cmd = F_GETLK; 428 break; 429 case F_OSETLK: 430 cmd = F_SETLK; 431 break; 432 case F_OSETLKW: 433 cmd = F_SETLKW; 434 break; 435 } 436 arg1 = (intptr_t)&fl; 437 break; 438 case F_GETLK: 439 case F_SETLK: 440 case F_SETLKW: 441 case F_SETLK_REMOTE: 442 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 443 arg1 = (intptr_t)&fl; 444 break; 445 default: 446 arg1 = arg; 447 break; 448 } 449 if (error) 450 return (error); 451 error = kern_fcntl(td, fd, cmd, arg1); 452 if (error) 453 return (error); 454 if (cmd == F_OGETLK) { 455 ofl.l_start = fl.l_start; 456 ofl.l_len = fl.l_len; 457 ofl.l_pid = fl.l_pid; 458 ofl.l_type = fl.l_type; 459 ofl.l_whence = fl.l_whence; 460 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 461 } else if (cmd == F_GETLK) { 462 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 463 } 464 return (error); 465 } 466 467 int 468 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 469 { 470 struct filedesc *fdp; 471 struct flock *flp; 472 struct file *fp, *fp2; 473 struct filedescent *fde; 474 struct proc *p; 475 struct vnode *vp; 476 cap_rights_t rights; 477 int error, flg, tmp; 478 uint64_t bsize; 479 off_t foffset; 480 481 error = 0; 482 flg = F_POSIX; 483 p = td->td_proc; 484 fdp = p->p_fd; 485 486 switch (cmd) { 487 case F_DUPFD: 488 tmp = arg; 489 error = do_dup(td, DUP_FCNTL, fd, tmp); 490 break; 491 492 case F_DUPFD_CLOEXEC: 493 tmp = arg; 494 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp); 495 break; 496 497 case F_DUP2FD: 498 tmp = arg; 499 error = do_dup(td, DUP_FIXED, fd, tmp); 500 break; 501 502 case F_DUP2FD_CLOEXEC: 503 tmp = arg; 504 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp); 505 break; 506 507 case F_GETFD: 508 FILEDESC_SLOCK(fdp); 509 if (fget_locked(fdp, fd) == NULL) { 510 FILEDESC_SUNLOCK(fdp); 511 error = EBADF; 512 break; 513 } 514 fde = &fdp->fd_ofiles[fd]; 515 td->td_retval[0] = 516 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 517 FILEDESC_SUNLOCK(fdp); 518 break; 519 520 case F_SETFD: 521 FILEDESC_XLOCK(fdp); 522 if (fget_locked(fdp, fd) == NULL) { 523 FILEDESC_XUNLOCK(fdp); 524 error = EBADF; 525 break; 526 } 527 fde = &fdp->fd_ofiles[fd]; 528 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 529 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 530 FILEDESC_XUNLOCK(fdp); 531 break; 532 533 case F_GETFL: 534 error = fget_fcntl(td, fd, 535 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 536 if (error != 0) 537 break; 538 td->td_retval[0] = OFLAGS(fp->f_flag); 539 fdrop(fp, td); 540 break; 541 542 case F_SETFL: 543 error = fget_fcntl(td, fd, 544 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 545 if (error != 0) 546 break; 547 do { 548 tmp = flg = fp->f_flag; 549 tmp &= ~FCNTLFLAGS; 550 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 551 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 552 tmp = fp->f_flag & FNONBLOCK; 553 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 554 if (error != 0) { 555 fdrop(fp, td); 556 break; 557 } 558 tmp = fp->f_flag & FASYNC; 559 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 560 if (error == 0) { 561 fdrop(fp, td); 562 break; 563 } 564 atomic_clear_int(&fp->f_flag, FNONBLOCK); 565 tmp = 0; 566 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 567 fdrop(fp, td); 568 break; 569 570 case F_GETOWN: 571 error = fget_fcntl(td, fd, 572 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 573 if (error != 0) 574 break; 575 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 576 if (error == 0) 577 td->td_retval[0] = tmp; 578 fdrop(fp, td); 579 break; 580 581 case F_SETOWN: 582 error = fget_fcntl(td, fd, 583 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 584 if (error != 0) 585 break; 586 tmp = arg; 587 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 588 fdrop(fp, td); 589 break; 590 591 case F_SETLK_REMOTE: 592 error = priv_check(td, PRIV_NFS_LOCKD); 593 if (error) 594 return (error); 595 flg = F_REMOTE; 596 goto do_setlk; 597 598 case F_SETLKW: 599 flg |= F_WAIT; 600 /* FALLTHROUGH F_SETLK */ 601 602 case F_SETLK: 603 do_setlk: 604 cap_rights_init(&rights, CAP_FLOCK); 605 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 606 if (error != 0) 607 break; 608 if (fp->f_type != DTYPE_VNODE) { 609 error = EBADF; 610 fdrop(fp, td); 611 break; 612 } 613 614 flp = (struct flock *)arg; 615 if (flp->l_whence == SEEK_CUR) { 616 foffset = foffset_get(fp); 617 if (foffset < 0 || 618 (flp->l_start > 0 && 619 foffset > OFF_MAX - flp->l_start)) { 620 error = EOVERFLOW; 621 fdrop(fp, td); 622 break; 623 } 624 flp->l_start += foffset; 625 } 626 627 vp = fp->f_vnode; 628 switch (flp->l_type) { 629 case F_RDLCK: 630 if ((fp->f_flag & FREAD) == 0) { 631 error = EBADF; 632 break; 633 } 634 PROC_LOCK(p->p_leader); 635 p->p_leader->p_flag |= P_ADVLOCK; 636 PROC_UNLOCK(p->p_leader); 637 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 638 flp, flg); 639 break; 640 case F_WRLCK: 641 if ((fp->f_flag & FWRITE) == 0) { 642 error = EBADF; 643 break; 644 } 645 PROC_LOCK(p->p_leader); 646 p->p_leader->p_flag |= P_ADVLOCK; 647 PROC_UNLOCK(p->p_leader); 648 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 649 flp, flg); 650 break; 651 case F_UNLCK: 652 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 653 flp, flg); 654 break; 655 case F_UNLCKSYS: 656 /* 657 * Temporary api for testing remote lock 658 * infrastructure. 659 */ 660 if (flg != F_REMOTE) { 661 error = EINVAL; 662 break; 663 } 664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 665 F_UNLCKSYS, flp, flg); 666 break; 667 default: 668 error = EINVAL; 669 break; 670 } 671 if (error != 0 || flp->l_type == F_UNLCK || 672 flp->l_type == F_UNLCKSYS) { 673 fdrop(fp, td); 674 break; 675 } 676 677 /* 678 * Check for a race with close. 679 * 680 * The vnode is now advisory locked (or unlocked, but this case 681 * is not really important) as the caller requested. 682 * We had to drop the filedesc lock, so we need to recheck if 683 * the descriptor is still valid, because if it was closed 684 * in the meantime we need to remove advisory lock from the 685 * vnode - close on any descriptor leading to an advisory 686 * locked vnode, removes that lock. 687 * We will return 0 on purpose in that case, as the result of 688 * successful advisory lock might have been externally visible 689 * already. This is fine - effectively we pretend to the caller 690 * that the closing thread was a bit slower and that the 691 * advisory lock succeeded before the close. 692 */ 693 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 694 if (error != 0) { 695 fdrop(fp, td); 696 break; 697 } 698 if (fp != fp2) { 699 flp->l_whence = SEEK_SET; 700 flp->l_start = 0; 701 flp->l_len = 0; 702 flp->l_type = F_UNLCK; 703 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 704 F_UNLCK, flp, F_POSIX); 705 } 706 fdrop(fp, td); 707 fdrop(fp2, td); 708 break; 709 710 case F_GETLK: 711 error = fget_unlocked(fdp, fd, 712 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 713 if (error != 0) 714 break; 715 if (fp->f_type != DTYPE_VNODE) { 716 error = EBADF; 717 fdrop(fp, td); 718 break; 719 } 720 flp = (struct flock *)arg; 721 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 722 flp->l_type != F_UNLCK) { 723 error = EINVAL; 724 fdrop(fp, td); 725 break; 726 } 727 if (flp->l_whence == SEEK_CUR) { 728 foffset = foffset_get(fp); 729 if ((flp->l_start > 0 && 730 foffset > OFF_MAX - flp->l_start) || 731 (flp->l_start < 0 && 732 foffset < OFF_MIN - flp->l_start)) { 733 error = EOVERFLOW; 734 fdrop(fp, td); 735 break; 736 } 737 flp->l_start += foffset; 738 } 739 vp = fp->f_vnode; 740 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 741 F_POSIX); 742 fdrop(fp, td); 743 break; 744 745 case F_RDAHEAD: 746 arg = arg ? 128 * 1024: 0; 747 /* FALLTHROUGH */ 748 case F_READAHEAD: 749 error = fget_unlocked(fdp, fd, 750 cap_rights_init(&rights), &fp, NULL); 751 if (error != 0) 752 break; 753 if (fp->f_type != DTYPE_VNODE) { 754 fdrop(fp, td); 755 error = EBADF; 756 break; 757 } 758 vp = fp->f_vnode; 759 /* 760 * Exclusive lock synchronizes against f_seqcount reads and 761 * writes in sequential_heuristic(). 762 */ 763 error = vn_lock(vp, LK_EXCLUSIVE); 764 if (error != 0) { 765 fdrop(fp, td); 766 break; 767 } 768 if (arg >= 0) { 769 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 770 fp->f_seqcount = (arg + bsize - 1) / bsize; 771 atomic_set_int(&fp->f_flag, FRDAHEAD); 772 } else { 773 atomic_clear_int(&fp->f_flag, FRDAHEAD); 774 } 775 VOP_UNLOCK(vp, 0); 776 fdrop(fp, td); 777 break; 778 779 default: 780 error = EINVAL; 781 break; 782 } 783 return (error); 784 } 785 786 static int 787 getmaxfd(struct thread *td) 788 { 789 790 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 791 } 792 793 /* 794 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 795 */ 796 static int 797 do_dup(struct thread *td, int flags, int old, int new) 798 { 799 struct filedesc *fdp; 800 struct filedescent *oldfde, *newfde; 801 struct proc *p; 802 struct file *fp; 803 struct file *delfp; 804 int error, maxfd; 805 806 p = td->td_proc; 807 fdp = p->p_fd; 808 809 /* 810 * Verify we have a valid descriptor to dup from and possibly to 811 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 812 * return EINVAL when the new descriptor is out of bounds. 813 */ 814 if (old < 0) 815 return (EBADF); 816 if (new < 0) 817 return (flags & DUP_FCNTL ? EINVAL : EBADF); 818 maxfd = getmaxfd(td); 819 if (new >= maxfd) 820 return (flags & DUP_FCNTL ? EINVAL : EBADF); 821 822 FILEDESC_XLOCK(fdp); 823 if (fget_locked(fdp, old) == NULL) { 824 FILEDESC_XUNLOCK(fdp); 825 return (EBADF); 826 } 827 oldfde = &fdp->fd_ofiles[old]; 828 if (flags & DUP_FIXED && old == new) { 829 td->td_retval[0] = new; 830 if (flags & DUP_CLOEXEC) 831 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 832 FILEDESC_XUNLOCK(fdp); 833 return (0); 834 } 835 fp = oldfde->fde_file; 836 fhold(fp); 837 838 /* 839 * If the caller specified a file descriptor, make sure the file 840 * table is large enough to hold it, and grab it. Otherwise, just 841 * allocate a new descriptor the usual way. 842 */ 843 if (flags & DUP_FIXED) { 844 if (new >= fdp->fd_nfiles) { 845 /* 846 * The resource limits are here instead of e.g. 847 * fdalloc(), because the file descriptor table may be 848 * shared between processes, so we can't really use 849 * racct_add()/racct_sub(). Instead of counting the 850 * number of actually allocated descriptors, just put 851 * the limit on the size of the file descriptor table. 852 */ 853 #ifdef RACCT 854 if (racct_enable) { 855 PROC_LOCK(p); 856 error = racct_set(p, RACCT_NOFILE, new + 1); 857 PROC_UNLOCK(p); 858 if (error != 0) { 859 FILEDESC_XUNLOCK(fdp); 860 fdrop(fp, td); 861 return (EMFILE); 862 } 863 } 864 #endif 865 fdgrowtable_exp(fdp, new + 1); 866 oldfde = &fdp->fd_ofiles[old]; 867 } 868 newfde = &fdp->fd_ofiles[new]; 869 if (newfde->fde_file == NULL) 870 fdused(fdp, new); 871 } else { 872 if ((error = fdalloc(td, new, &new)) != 0) { 873 FILEDESC_XUNLOCK(fdp); 874 fdrop(fp, td); 875 return (error); 876 } 877 newfde = &fdp->fd_ofiles[new]; 878 } 879 880 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 881 KASSERT(old != new, ("new fd is same as old")); 882 883 delfp = newfde->fde_file; 884 885 /* 886 * Duplicate the source descriptor. 887 */ 888 #ifdef CAPABILITIES 889 seq_write_begin(&newfde->fde_seq); 890 #endif 891 filecaps_free(&newfde->fde_caps); 892 memcpy(newfde, oldfde, fde_change_size); 893 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 894 if ((flags & DUP_CLOEXEC) != 0) 895 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 896 else 897 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 898 #ifdef CAPABILITIES 899 seq_write_end(&newfde->fde_seq); 900 #endif 901 td->td_retval[0] = new; 902 903 if (delfp != NULL) { 904 (void) closefp(fdp, new, delfp, td, 1); 905 /* closefp() drops the FILEDESC lock for us. */ 906 } else { 907 FILEDESC_XUNLOCK(fdp); 908 } 909 910 return (0); 911 } 912 913 /* 914 * If sigio is on the list associated with a process or process group, 915 * disable signalling from the device, remove sigio from the list and 916 * free sigio. 917 */ 918 void 919 funsetown(struct sigio **sigiop) 920 { 921 struct sigio *sigio; 922 923 SIGIO_LOCK(); 924 sigio = *sigiop; 925 if (sigio == NULL) { 926 SIGIO_UNLOCK(); 927 return; 928 } 929 *(sigio->sio_myref) = NULL; 930 if ((sigio)->sio_pgid < 0) { 931 struct pgrp *pg = (sigio)->sio_pgrp; 932 PGRP_LOCK(pg); 933 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 934 sigio, sio_pgsigio); 935 PGRP_UNLOCK(pg); 936 } else { 937 struct proc *p = (sigio)->sio_proc; 938 PROC_LOCK(p); 939 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 940 sigio, sio_pgsigio); 941 PROC_UNLOCK(p); 942 } 943 SIGIO_UNLOCK(); 944 crfree(sigio->sio_ucred); 945 free(sigio, M_SIGIO); 946 } 947 948 /* 949 * Free a list of sigio structures. 950 * We only need to lock the SIGIO_LOCK because we have made ourselves 951 * inaccessible to callers of fsetown and therefore do not need to lock 952 * the proc or pgrp struct for the list manipulation. 953 */ 954 void 955 funsetownlst(struct sigiolst *sigiolst) 956 { 957 struct proc *p; 958 struct pgrp *pg; 959 struct sigio *sigio; 960 961 sigio = SLIST_FIRST(sigiolst); 962 if (sigio == NULL) 963 return; 964 p = NULL; 965 pg = NULL; 966 967 /* 968 * Every entry of the list should belong 969 * to a single proc or pgrp. 970 */ 971 if (sigio->sio_pgid < 0) { 972 pg = sigio->sio_pgrp; 973 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 974 } else /* if (sigio->sio_pgid > 0) */ { 975 p = sigio->sio_proc; 976 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 977 } 978 979 SIGIO_LOCK(); 980 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 981 *(sigio->sio_myref) = NULL; 982 if (pg != NULL) { 983 KASSERT(sigio->sio_pgid < 0, 984 ("Proc sigio in pgrp sigio list")); 985 KASSERT(sigio->sio_pgrp == pg, 986 ("Bogus pgrp in sigio list")); 987 PGRP_LOCK(pg); 988 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 989 sio_pgsigio); 990 PGRP_UNLOCK(pg); 991 } else /* if (p != NULL) */ { 992 KASSERT(sigio->sio_pgid > 0, 993 ("Pgrp sigio in proc sigio list")); 994 KASSERT(sigio->sio_proc == p, 995 ("Bogus proc in sigio list")); 996 PROC_LOCK(p); 997 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 998 sio_pgsigio); 999 PROC_UNLOCK(p); 1000 } 1001 SIGIO_UNLOCK(); 1002 crfree(sigio->sio_ucred); 1003 free(sigio, M_SIGIO); 1004 SIGIO_LOCK(); 1005 } 1006 SIGIO_UNLOCK(); 1007 } 1008 1009 /* 1010 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1011 * 1012 * After permission checking, add a sigio structure to the sigio list for 1013 * the process or process group. 1014 */ 1015 int 1016 fsetown(pid_t pgid, struct sigio **sigiop) 1017 { 1018 struct proc *proc; 1019 struct pgrp *pgrp; 1020 struct sigio *sigio; 1021 int ret; 1022 1023 if (pgid == 0) { 1024 funsetown(sigiop); 1025 return (0); 1026 } 1027 1028 ret = 0; 1029 1030 /* Allocate and fill in the new sigio out of locks. */ 1031 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1032 sigio->sio_pgid = pgid; 1033 sigio->sio_ucred = crhold(curthread->td_ucred); 1034 sigio->sio_myref = sigiop; 1035 1036 sx_slock(&proctree_lock); 1037 if (pgid > 0) { 1038 proc = pfind(pgid); 1039 if (proc == NULL) { 1040 ret = ESRCH; 1041 goto fail; 1042 } 1043 1044 /* 1045 * Policy - Don't allow a process to FSETOWN a process 1046 * in another session. 1047 * 1048 * Remove this test to allow maximum flexibility or 1049 * restrict FSETOWN to the current process or process 1050 * group for maximum safety. 1051 */ 1052 PROC_UNLOCK(proc); 1053 if (proc->p_session != curthread->td_proc->p_session) { 1054 ret = EPERM; 1055 goto fail; 1056 } 1057 1058 pgrp = NULL; 1059 } else /* if (pgid < 0) */ { 1060 pgrp = pgfind(-pgid); 1061 if (pgrp == NULL) { 1062 ret = ESRCH; 1063 goto fail; 1064 } 1065 PGRP_UNLOCK(pgrp); 1066 1067 /* 1068 * Policy - Don't allow a process to FSETOWN a process 1069 * in another session. 1070 * 1071 * Remove this test to allow maximum flexibility or 1072 * restrict FSETOWN to the current process or process 1073 * group for maximum safety. 1074 */ 1075 if (pgrp->pg_session != curthread->td_proc->p_session) { 1076 ret = EPERM; 1077 goto fail; 1078 } 1079 1080 proc = NULL; 1081 } 1082 funsetown(sigiop); 1083 if (pgid > 0) { 1084 PROC_LOCK(proc); 1085 /* 1086 * Since funsetownlst() is called without the proctree 1087 * locked, we need to check for P_WEXIT. 1088 * XXX: is ESRCH correct? 1089 */ 1090 if ((proc->p_flag & P_WEXIT) != 0) { 1091 PROC_UNLOCK(proc); 1092 ret = ESRCH; 1093 goto fail; 1094 } 1095 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1096 sigio->sio_proc = proc; 1097 PROC_UNLOCK(proc); 1098 } else { 1099 PGRP_LOCK(pgrp); 1100 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1101 sigio->sio_pgrp = pgrp; 1102 PGRP_UNLOCK(pgrp); 1103 } 1104 sx_sunlock(&proctree_lock); 1105 SIGIO_LOCK(); 1106 *sigiop = sigio; 1107 SIGIO_UNLOCK(); 1108 return (0); 1109 1110 fail: 1111 sx_sunlock(&proctree_lock); 1112 crfree(sigio->sio_ucred); 1113 free(sigio, M_SIGIO); 1114 return (ret); 1115 } 1116 1117 /* 1118 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1119 */ 1120 pid_t 1121 fgetown(sigiop) 1122 struct sigio **sigiop; 1123 { 1124 pid_t pgid; 1125 1126 SIGIO_LOCK(); 1127 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1128 SIGIO_UNLOCK(); 1129 return (pgid); 1130 } 1131 1132 /* 1133 * Function drops the filedesc lock on return. 1134 */ 1135 static int 1136 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1137 int holdleaders) 1138 { 1139 int error; 1140 1141 FILEDESC_XLOCK_ASSERT(fdp); 1142 1143 if (holdleaders) { 1144 if (td->td_proc->p_fdtol != NULL) { 1145 /* 1146 * Ask fdfree() to sleep to ensure that all relevant 1147 * process leaders can be traversed in closef(). 1148 */ 1149 fdp->fd_holdleaderscount++; 1150 } else { 1151 holdleaders = 0; 1152 } 1153 } 1154 1155 /* 1156 * We now hold the fp reference that used to be owned by the 1157 * descriptor array. We have to unlock the FILEDESC *AFTER* 1158 * knote_fdclose to prevent a race of the fd getting opened, a knote 1159 * added, and deleteing a knote for the new fd. 1160 */ 1161 knote_fdclose(td, fd); 1162 1163 /* 1164 * We need to notify mqueue if the object is of type mqueue. 1165 */ 1166 if (fp->f_type == DTYPE_MQUEUE) 1167 mq_fdclose(td, fd, fp); 1168 FILEDESC_XUNLOCK(fdp); 1169 1170 error = closef(fp, td); 1171 if (holdleaders) { 1172 FILEDESC_XLOCK(fdp); 1173 fdp->fd_holdleaderscount--; 1174 if (fdp->fd_holdleaderscount == 0 && 1175 fdp->fd_holdleaderswakeup != 0) { 1176 fdp->fd_holdleaderswakeup = 0; 1177 wakeup(&fdp->fd_holdleaderscount); 1178 } 1179 FILEDESC_XUNLOCK(fdp); 1180 } 1181 return (error); 1182 } 1183 1184 /* 1185 * Close a file descriptor. 1186 */ 1187 #ifndef _SYS_SYSPROTO_H_ 1188 struct close_args { 1189 int fd; 1190 }; 1191 #endif 1192 /* ARGSUSED */ 1193 int 1194 sys_close(struct thread *td, struct close_args *uap) 1195 { 1196 1197 return (kern_close(td, uap->fd)); 1198 } 1199 1200 int 1201 kern_close(struct thread *td, int fd) 1202 { 1203 struct filedesc *fdp; 1204 struct file *fp; 1205 1206 fdp = td->td_proc->p_fd; 1207 1208 AUDIT_SYSCLOSE(td, fd); 1209 1210 FILEDESC_XLOCK(fdp); 1211 if ((fp = fget_locked(fdp, fd)) == NULL) { 1212 FILEDESC_XUNLOCK(fdp); 1213 return (EBADF); 1214 } 1215 fdfree(fdp, fd); 1216 1217 /* closefp() drops the FILEDESC lock for us. */ 1218 return (closefp(fdp, fd, fp, td, 1)); 1219 } 1220 1221 /* 1222 * Close open file descriptors. 1223 */ 1224 #ifndef _SYS_SYSPROTO_H_ 1225 struct closefrom_args { 1226 int lowfd; 1227 }; 1228 #endif 1229 /* ARGSUSED */ 1230 int 1231 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1232 { 1233 struct filedesc *fdp; 1234 int fd; 1235 1236 fdp = td->td_proc->p_fd; 1237 AUDIT_ARG_FD(uap->lowfd); 1238 1239 /* 1240 * Treat negative starting file descriptor values identical to 1241 * closefrom(0) which closes all files. 1242 */ 1243 if (uap->lowfd < 0) 1244 uap->lowfd = 0; 1245 FILEDESC_SLOCK(fdp); 1246 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1247 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1248 FILEDESC_SUNLOCK(fdp); 1249 (void)kern_close(td, fd); 1250 FILEDESC_SLOCK(fdp); 1251 } 1252 } 1253 FILEDESC_SUNLOCK(fdp); 1254 return (0); 1255 } 1256 1257 #if defined(COMPAT_43) 1258 /* 1259 * Return status information about a file descriptor. 1260 */ 1261 #ifndef _SYS_SYSPROTO_H_ 1262 struct ofstat_args { 1263 int fd; 1264 struct ostat *sb; 1265 }; 1266 #endif 1267 /* ARGSUSED */ 1268 int 1269 ofstat(struct thread *td, struct ofstat_args *uap) 1270 { 1271 struct ostat oub; 1272 struct stat ub; 1273 int error; 1274 1275 error = kern_fstat(td, uap->fd, &ub); 1276 if (error == 0) { 1277 cvtstat(&ub, &oub); 1278 error = copyout(&oub, uap->sb, sizeof(oub)); 1279 } 1280 return (error); 1281 } 1282 #endif /* COMPAT_43 */ 1283 1284 /* 1285 * Return status information about a file descriptor. 1286 */ 1287 #ifndef _SYS_SYSPROTO_H_ 1288 struct fstat_args { 1289 int fd; 1290 struct stat *sb; 1291 }; 1292 #endif 1293 /* ARGSUSED */ 1294 int 1295 sys_fstat(struct thread *td, struct fstat_args *uap) 1296 { 1297 struct stat ub; 1298 int error; 1299 1300 error = kern_fstat(td, uap->fd, &ub); 1301 if (error == 0) 1302 error = copyout(&ub, uap->sb, sizeof(ub)); 1303 return (error); 1304 } 1305 1306 int 1307 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1308 { 1309 struct file *fp; 1310 cap_rights_t rights; 1311 int error; 1312 1313 AUDIT_ARG_FD(fd); 1314 1315 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1316 if (error != 0) 1317 return (error); 1318 1319 AUDIT_ARG_FILE(td->td_proc, fp); 1320 1321 error = fo_stat(fp, sbp, td->td_ucred, td); 1322 fdrop(fp, td); 1323 #ifdef KTRACE 1324 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1325 ktrstat(sbp); 1326 #endif 1327 return (error); 1328 } 1329 1330 /* 1331 * Return status information about a file descriptor. 1332 */ 1333 #ifndef _SYS_SYSPROTO_H_ 1334 struct nfstat_args { 1335 int fd; 1336 struct nstat *sb; 1337 }; 1338 #endif 1339 /* ARGSUSED */ 1340 int 1341 sys_nfstat(struct thread *td, struct nfstat_args *uap) 1342 { 1343 struct nstat nub; 1344 struct stat ub; 1345 int error; 1346 1347 error = kern_fstat(td, uap->fd, &ub); 1348 if (error == 0) { 1349 cvtnstat(&ub, &nub); 1350 error = copyout(&nub, uap->sb, sizeof(nub)); 1351 } 1352 return (error); 1353 } 1354 1355 /* 1356 * Return pathconf information about a file descriptor. 1357 */ 1358 #ifndef _SYS_SYSPROTO_H_ 1359 struct fpathconf_args { 1360 int fd; 1361 int name; 1362 }; 1363 #endif 1364 /* ARGSUSED */ 1365 int 1366 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1367 { 1368 struct file *fp; 1369 struct vnode *vp; 1370 cap_rights_t rights; 1371 int error; 1372 1373 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1374 if (error != 0) 1375 return (error); 1376 1377 /* If asynchronous I/O is available, it works for all descriptors. */ 1378 if (uap->name == _PC_ASYNC_IO) { 1379 td->td_retval[0] = async_io_version; 1380 goto out; 1381 } 1382 vp = fp->f_vnode; 1383 if (vp != NULL) { 1384 vn_lock(vp, LK_SHARED | LK_RETRY); 1385 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1386 VOP_UNLOCK(vp, 0); 1387 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1388 if (uap->name != _PC_PIPE_BUF) { 1389 error = EINVAL; 1390 } else { 1391 td->td_retval[0] = PIPE_BUF; 1392 error = 0; 1393 } 1394 } else { 1395 error = EOPNOTSUPP; 1396 } 1397 out: 1398 fdrop(fp, td); 1399 return (error); 1400 } 1401 1402 /* 1403 * Initialize filecaps structure. 1404 */ 1405 void 1406 filecaps_init(struct filecaps *fcaps) 1407 { 1408 1409 bzero(fcaps, sizeof(*fcaps)); 1410 fcaps->fc_nioctls = -1; 1411 } 1412 1413 /* 1414 * Copy filecaps structure allocating memory for ioctls array if needed. 1415 */ 1416 void 1417 filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1418 { 1419 size_t size; 1420 1421 *dst = *src; 1422 if (src->fc_ioctls != NULL) { 1423 KASSERT(src->fc_nioctls > 0, 1424 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1425 1426 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1427 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1428 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1429 } 1430 } 1431 1432 /* 1433 * Move filecaps structure to the new place and clear the old place. 1434 */ 1435 void 1436 filecaps_move(struct filecaps *src, struct filecaps *dst) 1437 { 1438 1439 *dst = *src; 1440 bzero(src, sizeof(*src)); 1441 } 1442 1443 /* 1444 * Fill the given filecaps structure with full rights. 1445 */ 1446 static void 1447 filecaps_fill(struct filecaps *fcaps) 1448 { 1449 1450 CAP_ALL(&fcaps->fc_rights); 1451 fcaps->fc_ioctls = NULL; 1452 fcaps->fc_nioctls = -1; 1453 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1454 } 1455 1456 /* 1457 * Free memory allocated within filecaps structure. 1458 */ 1459 void 1460 filecaps_free(struct filecaps *fcaps) 1461 { 1462 1463 free(fcaps->fc_ioctls, M_FILECAPS); 1464 bzero(fcaps, sizeof(*fcaps)); 1465 } 1466 1467 /* 1468 * Validate the given filecaps structure. 1469 */ 1470 static void 1471 filecaps_validate(const struct filecaps *fcaps, const char *func) 1472 { 1473 1474 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1475 ("%s: invalid rights", func)); 1476 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1477 ("%s: invalid fcntls", func)); 1478 KASSERT(fcaps->fc_fcntls == 0 || 1479 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1480 ("%s: fcntls without CAP_FCNTL", func)); 1481 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1482 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1483 ("%s: invalid ioctls", func)); 1484 KASSERT(fcaps->fc_nioctls == 0 || 1485 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1486 ("%s: ioctls without CAP_IOCTL", func)); 1487 } 1488 1489 static void 1490 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1491 { 1492 int nfd1; 1493 1494 FILEDESC_XLOCK_ASSERT(fdp); 1495 1496 nfd1 = fdp->fd_nfiles * 2; 1497 if (nfd1 < nfd) 1498 nfd1 = nfd; 1499 fdgrowtable(fdp, nfd1); 1500 } 1501 1502 /* 1503 * Grow the file table to accomodate (at least) nfd descriptors. 1504 */ 1505 static void 1506 fdgrowtable(struct filedesc *fdp, int nfd) 1507 { 1508 struct filedesc0 *fdp0; 1509 struct freetable *ft; 1510 struct fdescenttbl *ntable; 1511 struct fdescenttbl *otable; 1512 int nnfiles, onfiles; 1513 NDSLOTTYPE *nmap, *omap; 1514 1515 /* 1516 * If lastfile is -1 this struct filedesc was just allocated and we are 1517 * growing it to accomodate for the one we are going to copy from. There 1518 * is no need to have a lock on this one as it's not visible to anyone. 1519 */ 1520 if (fdp->fd_lastfile != -1) 1521 FILEDESC_XLOCK_ASSERT(fdp); 1522 1523 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1524 1525 /* save old values */ 1526 onfiles = fdp->fd_nfiles; 1527 otable = fdp->fd_files; 1528 omap = fdp->fd_map; 1529 1530 /* compute the size of the new table */ 1531 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1532 if (nnfiles <= onfiles) 1533 /* the table is already large enough */ 1534 return; 1535 1536 /* 1537 * Allocate a new table. We need enough space for the number of 1538 * entries, file entries themselves and the struct freetable we will use 1539 * when we decommission the table and place it on the freelist. 1540 * We place the struct freetable in the middle so we don't have 1541 * to worry about padding. 1542 */ 1543 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1544 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1545 sizeof(struct freetable), 1546 M_FILEDESC, M_ZERO | M_WAITOK); 1547 /* copy the old data */ 1548 ntable->fdt_nfiles = nnfiles; 1549 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1550 onfiles * sizeof(ntable->fdt_ofiles[0])); 1551 1552 /* 1553 * Allocate a new map only if the old is not large enough. It will 1554 * grow at a slower rate than the table as it can map more 1555 * entries than the table can hold. 1556 */ 1557 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1558 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1559 M_ZERO | M_WAITOK); 1560 /* copy over the old data and update the pointer */ 1561 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1562 fdp->fd_map = nmap; 1563 } 1564 1565 /* 1566 * Make sure that ntable is correctly initialized before we replace 1567 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1568 * data. 1569 */ 1570 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1571 1572 /* 1573 * Do not free the old file table, as some threads may still 1574 * reference entries within it. Instead, place it on a freelist 1575 * which will be processed when the struct filedesc is released. 1576 * 1577 * Note that if onfiles == NDFILE, we're dealing with the original 1578 * static allocation contained within (struct filedesc0 *)fdp, 1579 * which must not be freed. 1580 */ 1581 if (onfiles > NDFILE) { 1582 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1583 fdp0 = (struct filedesc0 *)fdp; 1584 ft->ft_table = otable; 1585 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1586 } 1587 /* 1588 * The map does not have the same possibility of threads still 1589 * holding references to it. So always free it as long as it 1590 * does not reference the original static allocation. 1591 */ 1592 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1593 free(omap, M_FILEDESC); 1594 } 1595 1596 /* 1597 * Allocate a file descriptor for the process. 1598 */ 1599 int 1600 fdalloc(struct thread *td, int minfd, int *result) 1601 { 1602 struct proc *p = td->td_proc; 1603 struct filedesc *fdp = p->p_fd; 1604 int fd, maxfd, allocfd; 1605 #ifdef RACCT 1606 int error; 1607 #endif 1608 1609 FILEDESC_XLOCK_ASSERT(fdp); 1610 1611 if (fdp->fd_freefile > minfd) 1612 minfd = fdp->fd_freefile; 1613 1614 maxfd = getmaxfd(td); 1615 1616 /* 1617 * Search the bitmap for a free descriptor starting at minfd. 1618 * If none is found, grow the file table. 1619 */ 1620 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1621 if (fd >= maxfd) 1622 return (EMFILE); 1623 if (fd >= fdp->fd_nfiles) { 1624 allocfd = min(fd * 2, maxfd); 1625 #ifdef RACCT 1626 if (racct_enable) { 1627 PROC_LOCK(p); 1628 error = racct_set(p, RACCT_NOFILE, allocfd); 1629 PROC_UNLOCK(p); 1630 if (error != 0) 1631 return (EMFILE); 1632 } 1633 #endif 1634 /* 1635 * fd is already equal to first free descriptor >= minfd, so 1636 * we only need to grow the table and we are done. 1637 */ 1638 fdgrowtable_exp(fdp, allocfd); 1639 } 1640 1641 /* 1642 * Perform some sanity checks, then mark the file descriptor as 1643 * used and return it to the caller. 1644 */ 1645 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1646 ("invalid descriptor %d", fd)); 1647 KASSERT(!fdisused(fdp, fd), 1648 ("fd_first_free() returned non-free descriptor")); 1649 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1650 ("file descriptor isn't free")); 1651 fdused(fdp, fd); 1652 *result = fd; 1653 return (0); 1654 } 1655 1656 /* 1657 * Allocate n file descriptors for the process. 1658 */ 1659 int 1660 fdallocn(struct thread *td, int minfd, int *fds, int n) 1661 { 1662 struct proc *p = td->td_proc; 1663 struct filedesc *fdp = p->p_fd; 1664 int i; 1665 1666 FILEDESC_XLOCK_ASSERT(fdp); 1667 1668 for (i = 0; i < n; i++) 1669 if (fdalloc(td, 0, &fds[i]) != 0) 1670 break; 1671 1672 if (i < n) { 1673 for (i--; i >= 0; i--) 1674 fdunused(fdp, fds[i]); 1675 return (EMFILE); 1676 } 1677 1678 return (0); 1679 } 1680 1681 /* 1682 * Create a new open file structure and allocate a file decriptor for the 1683 * process that refers to it. We add one reference to the file for the 1684 * descriptor table and one reference for resultfp. This is to prevent us 1685 * being preempted and the entry in the descriptor table closed after we 1686 * release the FILEDESC lock. 1687 */ 1688 int 1689 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1690 { 1691 struct file *fp; 1692 int error, fd; 1693 1694 error = falloc_noinstall(td, &fp); 1695 if (error) 1696 return (error); /* no reference held on error */ 1697 1698 error = finstall(td, fp, &fd, flags, NULL); 1699 if (error) { 1700 fdrop(fp, td); /* one reference (fp only) */ 1701 return (error); 1702 } 1703 1704 if (resultfp != NULL) 1705 *resultfp = fp; /* copy out result */ 1706 else 1707 fdrop(fp, td); /* release local reference */ 1708 1709 if (resultfd != NULL) 1710 *resultfd = fd; 1711 1712 return (0); 1713 } 1714 1715 /* 1716 * Create a new open file structure without allocating a file descriptor. 1717 */ 1718 int 1719 falloc_noinstall(struct thread *td, struct file **resultfp) 1720 { 1721 struct file *fp; 1722 int maxuserfiles = maxfiles - (maxfiles / 20); 1723 static struct timeval lastfail; 1724 static int curfail; 1725 1726 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1727 1728 if ((openfiles >= maxuserfiles && 1729 priv_check(td, PRIV_MAXFILES) != 0) || 1730 openfiles >= maxfiles) { 1731 if (ppsratecheck(&lastfail, &curfail, 1)) { 1732 printf("kern.maxfiles limit exceeded by uid %i, " 1733 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1734 } 1735 return (ENFILE); 1736 } 1737 atomic_add_int(&openfiles, 1); 1738 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1739 refcount_init(&fp->f_count, 1); 1740 fp->f_cred = crhold(td->td_ucred); 1741 fp->f_ops = &badfileops; 1742 *resultfp = fp; 1743 return (0); 1744 } 1745 1746 /* 1747 * Install a file in a file descriptor table. 1748 */ 1749 void 1750 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1751 struct filecaps *fcaps) 1752 { 1753 struct filedescent *fde; 1754 1755 MPASS(fp != NULL); 1756 if (fcaps != NULL) 1757 filecaps_validate(fcaps, __func__); 1758 FILEDESC_XLOCK_ASSERT(fdp); 1759 1760 fde = &fdp->fd_ofiles[fd]; 1761 #ifdef CAPABILITIES 1762 seq_write_begin(&fde->fde_seq); 1763 #endif 1764 fde->fde_file = fp; 1765 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1766 if (fcaps != NULL) 1767 filecaps_move(fcaps, &fde->fde_caps); 1768 else 1769 filecaps_fill(&fde->fde_caps); 1770 #ifdef CAPABILITIES 1771 seq_write_end(&fde->fde_seq); 1772 #endif 1773 } 1774 1775 int 1776 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1777 struct filecaps *fcaps) 1778 { 1779 struct filedesc *fdp = td->td_proc->p_fd; 1780 int error; 1781 1782 MPASS(fd != NULL); 1783 1784 FILEDESC_XLOCK(fdp); 1785 if ((error = fdalloc(td, 0, fd))) { 1786 FILEDESC_XUNLOCK(fdp); 1787 return (error); 1788 } 1789 fhold(fp); 1790 _finstall(fdp, fp, *fd, flags, fcaps); 1791 FILEDESC_XUNLOCK(fdp); 1792 return (0); 1793 } 1794 1795 /* 1796 * Build a new filedesc structure from another. 1797 * Copy the current, root, and jail root vnode references. 1798 * 1799 * If fdp is not NULL, return with it shared locked. 1800 */ 1801 struct filedesc * 1802 fdinit(struct filedesc *fdp, bool prepfiles) 1803 { 1804 struct filedesc0 *newfdp0; 1805 struct filedesc *newfdp; 1806 1807 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1808 newfdp = &newfdp0->fd_fd; 1809 1810 /* Create the file descriptor table. */ 1811 FILEDESC_LOCK_INIT(newfdp); 1812 refcount_init(&newfdp->fd_refcnt, 1); 1813 refcount_init(&newfdp->fd_holdcnt, 1); 1814 newfdp->fd_cmask = CMASK; 1815 newfdp->fd_map = newfdp0->fd_dmap; 1816 newfdp->fd_lastfile = -1; 1817 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1818 newfdp->fd_files->fdt_nfiles = NDFILE; 1819 1820 if (fdp == NULL) 1821 return (newfdp); 1822 1823 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1824 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1825 1826 FILEDESC_SLOCK(fdp); 1827 newfdp->fd_cdir = fdp->fd_cdir; 1828 if (newfdp->fd_cdir) 1829 VREF(newfdp->fd_cdir); 1830 newfdp->fd_rdir = fdp->fd_rdir; 1831 if (newfdp->fd_rdir) 1832 VREF(newfdp->fd_rdir); 1833 newfdp->fd_jdir = fdp->fd_jdir; 1834 if (newfdp->fd_jdir) 1835 VREF(newfdp->fd_jdir); 1836 1837 if (!prepfiles) { 1838 FILEDESC_SUNLOCK(fdp); 1839 } else { 1840 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1841 FILEDESC_SUNLOCK(fdp); 1842 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1843 FILEDESC_SLOCK(fdp); 1844 } 1845 } 1846 1847 return (newfdp); 1848 } 1849 1850 static struct filedesc * 1851 fdhold(struct proc *p) 1852 { 1853 struct filedesc *fdp; 1854 1855 PROC_LOCK_ASSERT(p, MA_OWNED); 1856 fdp = p->p_fd; 1857 if (fdp != NULL) 1858 refcount_acquire(&fdp->fd_holdcnt); 1859 return (fdp); 1860 } 1861 1862 static void 1863 fddrop(struct filedesc *fdp) 1864 { 1865 1866 if (fdp->fd_holdcnt > 1) { 1867 if (refcount_release(&fdp->fd_holdcnt) == 0) 1868 return; 1869 } 1870 1871 FILEDESC_LOCK_DESTROY(fdp); 1872 uma_zfree(filedesc0_zone, fdp); 1873 } 1874 1875 /* 1876 * Share a filedesc structure. 1877 */ 1878 struct filedesc * 1879 fdshare(struct filedesc *fdp) 1880 { 1881 1882 refcount_acquire(&fdp->fd_refcnt); 1883 return (fdp); 1884 } 1885 1886 /* 1887 * Unshare a filedesc structure, if necessary by making a copy 1888 */ 1889 void 1890 fdunshare(struct thread *td) 1891 { 1892 struct filedesc *tmp; 1893 struct proc *p = td->td_proc; 1894 1895 if (p->p_fd->fd_refcnt == 1) 1896 return; 1897 1898 tmp = fdcopy(p->p_fd); 1899 fdescfree(td); 1900 p->p_fd = tmp; 1901 } 1902 1903 /* 1904 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1905 * this is to ease callers, not catch errors. 1906 */ 1907 struct filedesc * 1908 fdcopy(struct filedesc *fdp) 1909 { 1910 struct filedesc *newfdp; 1911 struct filedescent *nfde, *ofde; 1912 int i; 1913 1914 MPASS(fdp != NULL); 1915 1916 newfdp = fdinit(fdp, true); 1917 /* copy all passable descriptors (i.e. not kqueue) */ 1918 newfdp->fd_freefile = -1; 1919 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1920 ofde = &fdp->fd_ofiles[i]; 1921 if (ofde->fde_file == NULL || 1922 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1923 if (newfdp->fd_freefile == -1) 1924 newfdp->fd_freefile = i; 1925 continue; 1926 } 1927 nfde = &newfdp->fd_ofiles[i]; 1928 *nfde = *ofde; 1929 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1930 fhold(nfde->fde_file); 1931 fdused_init(newfdp, i); 1932 newfdp->fd_lastfile = i; 1933 } 1934 if (newfdp->fd_freefile == -1) 1935 newfdp->fd_freefile = i; 1936 newfdp->fd_cmask = fdp->fd_cmask; 1937 FILEDESC_SUNLOCK(fdp); 1938 return (newfdp); 1939 } 1940 1941 /* 1942 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 1943 * one of processes using it exits) and the table used to be shared. 1944 */ 1945 static void 1946 fdclearlocks(struct thread *td) 1947 { 1948 struct filedesc *fdp; 1949 struct filedesc_to_leader *fdtol; 1950 struct flock lf; 1951 struct file *fp; 1952 struct proc *p; 1953 struct vnode *vp; 1954 int i; 1955 1956 p = td->td_proc; 1957 fdp = p->p_fd; 1958 fdtol = p->p_fdtol; 1959 MPASS(fdtol != NULL); 1960 1961 FILEDESC_XLOCK(fdp); 1962 KASSERT(fdtol->fdl_refcount > 0, 1963 ("filedesc_to_refcount botch: fdl_refcount=%d", 1964 fdtol->fdl_refcount)); 1965 if (fdtol->fdl_refcount == 1 && 1966 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1967 for (i = 0; i <= fdp->fd_lastfile; i++) { 1968 fp = fdp->fd_ofiles[i].fde_file; 1969 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1970 continue; 1971 fhold(fp); 1972 FILEDESC_XUNLOCK(fdp); 1973 lf.l_whence = SEEK_SET; 1974 lf.l_start = 0; 1975 lf.l_len = 0; 1976 lf.l_type = F_UNLCK; 1977 vp = fp->f_vnode; 1978 (void) VOP_ADVLOCK(vp, 1979 (caddr_t)p->p_leader, F_UNLCK, 1980 &lf, F_POSIX); 1981 FILEDESC_XLOCK(fdp); 1982 fdrop(fp, td); 1983 } 1984 } 1985 retry: 1986 if (fdtol->fdl_refcount == 1) { 1987 if (fdp->fd_holdleaderscount > 0 && 1988 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1989 /* 1990 * close() or do_dup() has cleared a reference 1991 * in a shared file descriptor table. 1992 */ 1993 fdp->fd_holdleaderswakeup = 1; 1994 sx_sleep(&fdp->fd_holdleaderscount, 1995 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1996 goto retry; 1997 } 1998 if (fdtol->fdl_holdcount > 0) { 1999 /* 2000 * Ensure that fdtol->fdl_leader remains 2001 * valid in closef(). 2002 */ 2003 fdtol->fdl_wakeup = 1; 2004 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2005 "fdlhold", 0); 2006 goto retry; 2007 } 2008 } 2009 fdtol->fdl_refcount--; 2010 if (fdtol->fdl_refcount == 0 && 2011 fdtol->fdl_holdcount == 0) { 2012 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2013 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2014 } else 2015 fdtol = NULL; 2016 p->p_fdtol = NULL; 2017 FILEDESC_XUNLOCK(fdp); 2018 if (fdtol != NULL) 2019 free(fdtol, M_FILEDESC_TO_LEADER); 2020 } 2021 2022 /* 2023 * Release a filedesc structure. 2024 */ 2025 void 2026 fdescfree(struct thread *td) 2027 { 2028 struct proc *p; 2029 struct filedesc0 *fdp0; 2030 struct filedesc *fdp; 2031 struct freetable *ft, *tft; 2032 struct filedescent *fde; 2033 struct file *fp; 2034 struct vnode *cdir, *jdir, *rdir; 2035 int i; 2036 2037 p = td->td_proc; 2038 fdp = p->p_fd; 2039 MPASS(fdp != NULL); 2040 2041 #ifdef RACCT 2042 if (racct_enable) { 2043 PROC_LOCK(p); 2044 racct_set(p, RACCT_NOFILE, 0); 2045 PROC_UNLOCK(p); 2046 } 2047 #endif 2048 2049 if (td->td_proc->p_fdtol != NULL) 2050 fdclearlocks(td); 2051 2052 PROC_LOCK(p); 2053 p->p_fd = NULL; 2054 PROC_UNLOCK(p); 2055 2056 if (refcount_release(&fdp->fd_refcnt) == 0) 2057 return; 2058 2059 FILEDESC_XLOCK(fdp); 2060 cdir = fdp->fd_cdir; 2061 fdp->fd_cdir = NULL; 2062 rdir = fdp->fd_rdir; 2063 fdp->fd_rdir = NULL; 2064 jdir = fdp->fd_jdir; 2065 fdp->fd_jdir = NULL; 2066 FILEDESC_XUNLOCK(fdp); 2067 2068 for (i = 0; i <= fdp->fd_lastfile; i++) { 2069 fde = &fdp->fd_ofiles[i]; 2070 fp = fde->fde_file; 2071 if (fp != NULL) { 2072 fdefree_last(fde); 2073 (void) closef(fp, td); 2074 } 2075 } 2076 2077 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2078 free(fdp->fd_map, M_FILEDESC); 2079 if (fdp->fd_nfiles > NDFILE) 2080 free(fdp->fd_files, M_FILEDESC); 2081 2082 fdp0 = (struct filedesc0 *)fdp; 2083 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2084 free(ft->ft_table, M_FILEDESC); 2085 2086 if (cdir != NULL) 2087 vrele(cdir); 2088 if (rdir != NULL) 2089 vrele(rdir); 2090 if (jdir != NULL) 2091 vrele(jdir); 2092 2093 fddrop(fdp); 2094 } 2095 2096 /* 2097 * For setugid programs, we don't want to people to use that setugidness 2098 * to generate error messages which write to a file which otherwise would 2099 * otherwise be off-limits to the process. We check for filesystems where 2100 * the vnode can change out from under us after execve (like [lin]procfs). 2101 * 2102 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2103 * sufficient. We also don't check for setugidness since we know we are. 2104 */ 2105 static bool 2106 is_unsafe(struct file *fp) 2107 { 2108 struct vnode *vp; 2109 2110 if (fp->f_type != DTYPE_VNODE) 2111 return (false); 2112 2113 vp = fp->f_vnode; 2114 return ((vp->v_vflag & VV_PROCDEP) != 0); 2115 } 2116 2117 /* 2118 * Make this setguid thing safe, if at all possible. 2119 */ 2120 void 2121 fdsetugidsafety(struct thread *td) 2122 { 2123 struct filedesc *fdp; 2124 struct file *fp; 2125 int i; 2126 2127 fdp = td->td_proc->p_fd; 2128 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2129 MPASS(fdp->fd_nfiles >= 3); 2130 for (i = 0; i <= 2; i++) { 2131 fp = fdp->fd_ofiles[i].fde_file; 2132 if (fp != NULL && is_unsafe(fp)) { 2133 FILEDESC_XLOCK(fdp); 2134 knote_fdclose(td, i); 2135 /* 2136 * NULL-out descriptor prior to close to avoid 2137 * a race while close blocks. 2138 */ 2139 fdfree(fdp, i); 2140 FILEDESC_XUNLOCK(fdp); 2141 (void) closef(fp, td); 2142 } 2143 } 2144 } 2145 2146 /* 2147 * If a specific file object occupies a specific file descriptor, close the 2148 * file descriptor entry and drop a reference on the file object. This is a 2149 * convenience function to handle a subsequent error in a function that calls 2150 * falloc() that handles the race that another thread might have closed the 2151 * file descriptor out from under the thread creating the file object. 2152 */ 2153 void 2154 fdclose(struct thread *td, struct file *fp, int idx) 2155 { 2156 struct filedesc *fdp = td->td_proc->p_fd; 2157 2158 FILEDESC_XLOCK(fdp); 2159 if (fdp->fd_ofiles[idx].fde_file == fp) { 2160 fdfree(fdp, idx); 2161 FILEDESC_XUNLOCK(fdp); 2162 fdrop(fp, td); 2163 } else 2164 FILEDESC_XUNLOCK(fdp); 2165 } 2166 2167 /* 2168 * Close any files on exec? 2169 */ 2170 void 2171 fdcloseexec(struct thread *td) 2172 { 2173 struct filedesc *fdp; 2174 struct filedescent *fde; 2175 struct file *fp; 2176 int i; 2177 2178 fdp = td->td_proc->p_fd; 2179 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2180 for (i = 0; i <= fdp->fd_lastfile; i++) { 2181 fde = &fdp->fd_ofiles[i]; 2182 fp = fde->fde_file; 2183 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2184 (fde->fde_flags & UF_EXCLOSE))) { 2185 FILEDESC_XLOCK(fdp); 2186 fdfree(fdp, i); 2187 (void) closefp(fdp, i, fp, td, 0); 2188 /* closefp() drops the FILEDESC lock. */ 2189 } 2190 } 2191 } 2192 2193 /* 2194 * It is unsafe for set[ug]id processes to be started with file 2195 * descriptors 0..2 closed, as these descriptors are given implicit 2196 * significance in the Standard C library. fdcheckstd() will create a 2197 * descriptor referencing /dev/null for each of stdin, stdout, and 2198 * stderr that is not already open. 2199 */ 2200 int 2201 fdcheckstd(struct thread *td) 2202 { 2203 struct filedesc *fdp; 2204 register_t save; 2205 int i, error, devnull; 2206 2207 fdp = td->td_proc->p_fd; 2208 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2209 MPASS(fdp->fd_nfiles >= 3); 2210 devnull = -1; 2211 for (i = 0; i <= 2; i++) { 2212 if (fdp->fd_ofiles[i].fde_file != NULL) 2213 continue; 2214 2215 save = td->td_retval[0]; 2216 if (devnull != -1) { 2217 error = do_dup(td, DUP_FIXED, devnull, i); 2218 } else { 2219 error = kern_openat(td, AT_FDCWD, "/dev/null", 2220 UIO_SYSSPACE, O_RDWR, 0); 2221 if (error == 0) { 2222 devnull = td->td_retval[0]; 2223 KASSERT(devnull == i, ("we didn't get our fd")); 2224 } 2225 } 2226 td->td_retval[0] = save; 2227 if (error != 0) 2228 return (error); 2229 } 2230 return (0); 2231 } 2232 2233 /* 2234 * Internal form of close. Decrement reference count on file structure. 2235 * Note: td may be NULL when closing a file that was being passed in a 2236 * message. 2237 * 2238 * XXXRW: Giant is not required for the caller, but often will be held; this 2239 * makes it moderately likely the Giant will be recursed in the VFS case. 2240 */ 2241 int 2242 closef(struct file *fp, struct thread *td) 2243 { 2244 struct vnode *vp; 2245 struct flock lf; 2246 struct filedesc_to_leader *fdtol; 2247 struct filedesc *fdp; 2248 2249 /* 2250 * POSIX record locking dictates that any close releases ALL 2251 * locks owned by this process. This is handled by setting 2252 * a flag in the unlock to free ONLY locks obeying POSIX 2253 * semantics, and not to free BSD-style file locks. 2254 * If the descriptor was in a message, POSIX-style locks 2255 * aren't passed with the descriptor, and the thread pointer 2256 * will be NULL. Callers should be careful only to pass a 2257 * NULL thread pointer when there really is no owning 2258 * context that might have locks, or the locks will be 2259 * leaked. 2260 */ 2261 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2262 vp = fp->f_vnode; 2263 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2264 lf.l_whence = SEEK_SET; 2265 lf.l_start = 0; 2266 lf.l_len = 0; 2267 lf.l_type = F_UNLCK; 2268 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2269 F_UNLCK, &lf, F_POSIX); 2270 } 2271 fdtol = td->td_proc->p_fdtol; 2272 if (fdtol != NULL) { 2273 /* 2274 * Handle special case where file descriptor table is 2275 * shared between multiple process leaders. 2276 */ 2277 fdp = td->td_proc->p_fd; 2278 FILEDESC_XLOCK(fdp); 2279 for (fdtol = fdtol->fdl_next; 2280 fdtol != td->td_proc->p_fdtol; 2281 fdtol = fdtol->fdl_next) { 2282 if ((fdtol->fdl_leader->p_flag & 2283 P_ADVLOCK) == 0) 2284 continue; 2285 fdtol->fdl_holdcount++; 2286 FILEDESC_XUNLOCK(fdp); 2287 lf.l_whence = SEEK_SET; 2288 lf.l_start = 0; 2289 lf.l_len = 0; 2290 lf.l_type = F_UNLCK; 2291 vp = fp->f_vnode; 2292 (void) VOP_ADVLOCK(vp, 2293 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2294 F_POSIX); 2295 FILEDESC_XLOCK(fdp); 2296 fdtol->fdl_holdcount--; 2297 if (fdtol->fdl_holdcount == 0 && 2298 fdtol->fdl_wakeup != 0) { 2299 fdtol->fdl_wakeup = 0; 2300 wakeup(fdtol); 2301 } 2302 } 2303 FILEDESC_XUNLOCK(fdp); 2304 } 2305 } 2306 return (fdrop(fp, td)); 2307 } 2308 2309 /* 2310 * Initialize the file pointer with the specified properties. 2311 * 2312 * The ops are set with release semantics to be certain that the flags, type, 2313 * and data are visible when ops is. This is to prevent ops methods from being 2314 * called with bad data. 2315 */ 2316 void 2317 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2318 { 2319 fp->f_data = data; 2320 fp->f_flag = flag; 2321 fp->f_type = type; 2322 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2323 } 2324 2325 int 2326 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2327 struct file **fpp, seq_t *seqp) 2328 { 2329 #ifdef CAPABILITIES 2330 struct filedescent *fde; 2331 #endif 2332 struct fdescenttbl *fdt; 2333 struct file *fp; 2334 u_int count; 2335 #ifdef CAPABILITIES 2336 seq_t seq; 2337 cap_rights_t haverights; 2338 int error; 2339 #endif 2340 2341 fdt = fdp->fd_files; 2342 if ((u_int)fd >= fdt->fdt_nfiles) 2343 return (EBADF); 2344 /* 2345 * Fetch the descriptor locklessly. We avoid fdrop() races by 2346 * never raising a refcount above 0. To accomplish this we have 2347 * to use a cmpset loop rather than an atomic_add. The descriptor 2348 * must be re-verified once we acquire a reference to be certain 2349 * that the identity is still correct and we did not lose a race 2350 * due to preemption. 2351 */ 2352 for (;;) { 2353 #ifdef CAPABILITIES 2354 seq = seq_read(fd_seq(fdt, fd)); 2355 fde = &fdt->fdt_ofiles[fd]; 2356 haverights = *cap_rights_fde(fde); 2357 fp = fde->fde_file; 2358 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2359 cpu_spinwait(); 2360 continue; 2361 } 2362 #else 2363 fp = fdt->fdt_ofiles[fd].fde_file; 2364 #endif 2365 if (fp == NULL) 2366 return (EBADF); 2367 #ifdef CAPABILITIES 2368 error = cap_check(&haverights, needrightsp); 2369 if (error != 0) 2370 return (error); 2371 #endif 2372 retry: 2373 count = fp->f_count; 2374 if (count == 0) { 2375 /* 2376 * Force a reload. Other thread could reallocate the 2377 * table before this fd was closed, so it possible that 2378 * there is a stale fp pointer in cached version. 2379 */ 2380 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2381 continue; 2382 } 2383 /* 2384 * Use an acquire barrier to force re-reading of fdt so it is 2385 * refreshed for verification. 2386 */ 2387 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2388 goto retry; 2389 fdt = fdp->fd_files; 2390 #ifdef CAPABILITIES 2391 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2392 #else 2393 if (fp == fdt->fdt_ofiles[fd].fde_file) 2394 #endif 2395 break; 2396 fdrop(fp, curthread); 2397 } 2398 *fpp = fp; 2399 if (seqp != NULL) { 2400 #ifdef CAPABILITIES 2401 *seqp = seq; 2402 #endif 2403 } 2404 return (0); 2405 } 2406 2407 /* 2408 * Extract the file pointer associated with the specified descriptor for the 2409 * current user process. 2410 * 2411 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2412 * returned. 2413 * 2414 * File's rights will be checked against the capability rights mask. 2415 * 2416 * If an error occured the non-zero error is returned and *fpp is set to 2417 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2418 * responsible for fdrop(). 2419 */ 2420 static __inline int 2421 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2422 cap_rights_t *needrightsp, seq_t *seqp) 2423 { 2424 struct filedesc *fdp; 2425 struct file *fp; 2426 int error; 2427 2428 *fpp = NULL; 2429 fdp = td->td_proc->p_fd; 2430 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2431 if (error != 0) 2432 return (error); 2433 if (fp->f_ops == &badfileops) { 2434 fdrop(fp, td); 2435 return (EBADF); 2436 } 2437 2438 /* 2439 * FREAD and FWRITE failure return EBADF as per POSIX. 2440 */ 2441 error = 0; 2442 switch (flags) { 2443 case FREAD: 2444 case FWRITE: 2445 if ((fp->f_flag & flags) == 0) 2446 error = EBADF; 2447 break; 2448 case FEXEC: 2449 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2450 ((fp->f_flag & FWRITE) != 0)) 2451 error = EBADF; 2452 break; 2453 case 0: 2454 break; 2455 default: 2456 KASSERT(0, ("wrong flags")); 2457 } 2458 2459 if (error != 0) { 2460 fdrop(fp, td); 2461 return (error); 2462 } 2463 2464 *fpp = fp; 2465 return (0); 2466 } 2467 2468 int 2469 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2470 { 2471 2472 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2473 } 2474 2475 int 2476 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2477 struct file **fpp) 2478 { 2479 int error; 2480 #ifndef CAPABILITIES 2481 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2482 if (maxprotp != NULL) 2483 *maxprotp = VM_PROT_ALL; 2484 #else 2485 struct filedesc *fdp = td->td_proc->p_fd; 2486 seq_t seq; 2487 2488 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2489 for (;;) { 2490 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2491 if (error != 0) 2492 return (error); 2493 /* 2494 * If requested, convert capability rights to access flags. 2495 */ 2496 if (maxprotp != NULL) 2497 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2498 if (!fd_modified(fdp, fd, seq)) 2499 break; 2500 fdrop(*fpp, td); 2501 } 2502 #endif 2503 return (error); 2504 } 2505 2506 int 2507 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2508 { 2509 2510 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2511 } 2512 2513 int 2514 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2515 { 2516 2517 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2518 } 2519 2520 int 2521 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2522 struct file **fpp) 2523 { 2524 struct filedesc *fdp = td->td_proc->p_fd; 2525 #ifndef CAPABILITIES 2526 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2527 #else 2528 int error; 2529 seq_t seq; 2530 2531 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2532 for (;;) { 2533 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2534 if (error != 0) 2535 return (error); 2536 error = cap_fcntl_check(fdp, fd, needfcntl); 2537 if (!fd_modified(fdp, fd, seq)) 2538 break; 2539 fdrop(*fpp, td); 2540 } 2541 if (error != 0) { 2542 fdrop(*fpp, td); 2543 *fpp = NULL; 2544 } 2545 return (error); 2546 #endif 2547 } 2548 2549 /* 2550 * Like fget() but loads the underlying vnode, or returns an error if the 2551 * descriptor does not represent a vnode. Note that pipes use vnodes but 2552 * never have VM objects. The returned vnode will be vref()'d. 2553 * 2554 * XXX: what about the unused flags ? 2555 */ 2556 static __inline int 2557 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2558 struct vnode **vpp) 2559 { 2560 struct file *fp; 2561 int error; 2562 2563 *vpp = NULL; 2564 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2565 if (error != 0) 2566 return (error); 2567 if (fp->f_vnode == NULL) { 2568 error = EINVAL; 2569 } else { 2570 *vpp = fp->f_vnode; 2571 vref(*vpp); 2572 } 2573 fdrop(fp, td); 2574 2575 return (error); 2576 } 2577 2578 int 2579 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2580 { 2581 2582 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2583 } 2584 2585 int 2586 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2587 struct filecaps *havecaps, struct vnode **vpp) 2588 { 2589 struct filedesc *fdp; 2590 struct file *fp; 2591 #ifdef CAPABILITIES 2592 int error; 2593 #endif 2594 2595 fdp = td->td_proc->p_fd; 2596 fp = fget_locked(fdp, fd); 2597 if (fp == NULL || fp->f_ops == &badfileops) 2598 return (EBADF); 2599 2600 #ifdef CAPABILITIES 2601 if (needrightsp != NULL) { 2602 error = cap_check(cap_rights(fdp, fd), needrightsp); 2603 if (error != 0) 2604 return (error); 2605 } 2606 #endif 2607 2608 if (fp->f_vnode == NULL) 2609 return (EINVAL); 2610 2611 *vpp = fp->f_vnode; 2612 vref(*vpp); 2613 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2614 2615 return (0); 2616 } 2617 2618 int 2619 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2620 { 2621 2622 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2623 } 2624 2625 int 2626 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2627 { 2628 2629 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2630 } 2631 2632 #ifdef notyet 2633 int 2634 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2635 struct vnode **vpp) 2636 { 2637 2638 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2639 } 2640 #endif 2641 2642 /* 2643 * Like fget() but loads the underlying socket, or returns an error if the 2644 * descriptor does not represent a socket. 2645 * 2646 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2647 * in the future. 2648 * 2649 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2650 * on their file descriptor reference to prevent the socket from being free'd 2651 * during use. 2652 */ 2653 int 2654 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2655 u_int *fflagp) 2656 { 2657 struct file *fp; 2658 int error; 2659 2660 *spp = NULL; 2661 if (fflagp != NULL) 2662 *fflagp = 0; 2663 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2664 return (error); 2665 if (fp->f_type != DTYPE_SOCKET) { 2666 error = ENOTSOCK; 2667 } else { 2668 *spp = fp->f_data; 2669 if (fflagp) 2670 *fflagp = fp->f_flag; 2671 SOCK_LOCK(*spp); 2672 soref(*spp); 2673 SOCK_UNLOCK(*spp); 2674 } 2675 fdrop(fp, td); 2676 2677 return (error); 2678 } 2679 2680 /* 2681 * Drop the reference count on the socket and XXX release the SX lock in the 2682 * future. The last reference closes the socket. 2683 * 2684 * Note: fputsock() is deprecated, see comment for fgetsock(). 2685 */ 2686 void 2687 fputsock(struct socket *so) 2688 { 2689 2690 ACCEPT_LOCK(); 2691 SOCK_LOCK(so); 2692 CURVNET_SET(so->so_vnet); 2693 sorele(so); 2694 CURVNET_RESTORE(); 2695 } 2696 2697 /* 2698 * Handle the last reference to a file being closed. 2699 */ 2700 int 2701 _fdrop(struct file *fp, struct thread *td) 2702 { 2703 int error; 2704 2705 if (fp->f_count != 0) 2706 panic("fdrop: count %d", fp->f_count); 2707 error = fo_close(fp, td); 2708 atomic_subtract_int(&openfiles, 1); 2709 crfree(fp->f_cred); 2710 free(fp->f_advice, M_FADVISE); 2711 uma_zfree(file_zone, fp); 2712 2713 return (error); 2714 } 2715 2716 /* 2717 * Apply an advisory lock on a file descriptor. 2718 * 2719 * Just attempt to get a record lock of the requested type on the entire file 2720 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2721 */ 2722 #ifndef _SYS_SYSPROTO_H_ 2723 struct flock_args { 2724 int fd; 2725 int how; 2726 }; 2727 #endif 2728 /* ARGSUSED */ 2729 int 2730 sys_flock(struct thread *td, struct flock_args *uap) 2731 { 2732 struct file *fp; 2733 struct vnode *vp; 2734 struct flock lf; 2735 cap_rights_t rights; 2736 int error; 2737 2738 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2739 if (error != 0) 2740 return (error); 2741 if (fp->f_type != DTYPE_VNODE) { 2742 fdrop(fp, td); 2743 return (EOPNOTSUPP); 2744 } 2745 2746 vp = fp->f_vnode; 2747 lf.l_whence = SEEK_SET; 2748 lf.l_start = 0; 2749 lf.l_len = 0; 2750 if (uap->how & LOCK_UN) { 2751 lf.l_type = F_UNLCK; 2752 atomic_clear_int(&fp->f_flag, FHASLOCK); 2753 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2754 goto done2; 2755 } 2756 if (uap->how & LOCK_EX) 2757 lf.l_type = F_WRLCK; 2758 else if (uap->how & LOCK_SH) 2759 lf.l_type = F_RDLCK; 2760 else { 2761 error = EBADF; 2762 goto done2; 2763 } 2764 atomic_set_int(&fp->f_flag, FHASLOCK); 2765 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2766 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2767 done2: 2768 fdrop(fp, td); 2769 return (error); 2770 } 2771 /* 2772 * Duplicate the specified descriptor to a free descriptor. 2773 */ 2774 int 2775 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2776 int openerror, int *indxp) 2777 { 2778 struct filedescent *newfde, *oldfde; 2779 struct file *fp; 2780 int error, indx; 2781 2782 KASSERT(openerror == ENODEV || openerror == ENXIO, 2783 ("unexpected error %d in %s", openerror, __func__)); 2784 2785 /* 2786 * If the to-be-dup'd fd number is greater than the allowed number 2787 * of file descriptors, or the fd to be dup'd has already been 2788 * closed, then reject. 2789 */ 2790 FILEDESC_XLOCK(fdp); 2791 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2792 FILEDESC_XUNLOCK(fdp); 2793 return (EBADF); 2794 } 2795 2796 error = fdalloc(td, 0, &indx); 2797 if (error != 0) { 2798 FILEDESC_XUNLOCK(fdp); 2799 return (error); 2800 } 2801 2802 /* 2803 * There are two cases of interest here. 2804 * 2805 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2806 * 2807 * For ENXIO steal away the file structure from (dfd) and store it in 2808 * (indx). (dfd) is effectively closed by this operation. 2809 */ 2810 switch (openerror) { 2811 case ENODEV: 2812 /* 2813 * Check that the mode the file is being opened for is a 2814 * subset of the mode of the existing descriptor. 2815 */ 2816 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2817 fdunused(fdp, indx); 2818 FILEDESC_XUNLOCK(fdp); 2819 return (EACCES); 2820 } 2821 fhold(fp); 2822 newfde = &fdp->fd_ofiles[indx]; 2823 oldfde = &fdp->fd_ofiles[dfd]; 2824 #ifdef CAPABILITIES 2825 seq_write_begin(&newfde->fde_seq); 2826 #endif 2827 memcpy(newfde, oldfde, fde_change_size); 2828 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 2829 #ifdef CAPABILITIES 2830 seq_write_end(&newfde->fde_seq); 2831 #endif 2832 break; 2833 case ENXIO: 2834 /* 2835 * Steal away the file pointer from dfd and stuff it into indx. 2836 */ 2837 newfde = &fdp->fd_ofiles[indx]; 2838 oldfde = &fdp->fd_ofiles[dfd]; 2839 #ifdef CAPABILITIES 2840 seq_write_begin(&newfde->fde_seq); 2841 #endif 2842 memcpy(newfde, oldfde, fde_change_size); 2843 oldfde->fde_file = NULL; 2844 fdunused(fdp, dfd); 2845 #ifdef CAPABILITIES 2846 seq_write_end(&newfde->fde_seq); 2847 #endif 2848 break; 2849 } 2850 FILEDESC_XUNLOCK(fdp); 2851 *indxp = indx; 2852 return (0); 2853 } 2854 2855 /* 2856 * Scan all active processes and prisons to see if any of them have a current 2857 * or root directory of `olddp'. If so, replace them with the new mount point. 2858 */ 2859 void 2860 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2861 { 2862 struct filedesc *fdp; 2863 struct prison *pr; 2864 struct proc *p; 2865 int nrele; 2866 2867 if (vrefcnt(olddp) == 1) 2868 return; 2869 nrele = 0; 2870 sx_slock(&allproc_lock); 2871 FOREACH_PROC_IN_SYSTEM(p) { 2872 PROC_LOCK(p); 2873 fdp = fdhold(p); 2874 PROC_UNLOCK(p); 2875 if (fdp == NULL) 2876 continue; 2877 FILEDESC_XLOCK(fdp); 2878 if (fdp->fd_cdir == olddp) { 2879 vref(newdp); 2880 fdp->fd_cdir = newdp; 2881 nrele++; 2882 } 2883 if (fdp->fd_rdir == olddp) { 2884 vref(newdp); 2885 fdp->fd_rdir = newdp; 2886 nrele++; 2887 } 2888 if (fdp->fd_jdir == olddp) { 2889 vref(newdp); 2890 fdp->fd_jdir = newdp; 2891 nrele++; 2892 } 2893 FILEDESC_XUNLOCK(fdp); 2894 fddrop(fdp); 2895 } 2896 sx_sunlock(&allproc_lock); 2897 if (rootvnode == olddp) { 2898 vref(newdp); 2899 rootvnode = newdp; 2900 nrele++; 2901 } 2902 mtx_lock(&prison0.pr_mtx); 2903 if (prison0.pr_root == olddp) { 2904 vref(newdp); 2905 prison0.pr_root = newdp; 2906 nrele++; 2907 } 2908 mtx_unlock(&prison0.pr_mtx); 2909 sx_slock(&allprison_lock); 2910 TAILQ_FOREACH(pr, &allprison, pr_list) { 2911 mtx_lock(&pr->pr_mtx); 2912 if (pr->pr_root == olddp) { 2913 vref(newdp); 2914 pr->pr_root = newdp; 2915 nrele++; 2916 } 2917 mtx_unlock(&pr->pr_mtx); 2918 } 2919 sx_sunlock(&allprison_lock); 2920 while (nrele--) 2921 vrele(olddp); 2922 } 2923 2924 struct filedesc_to_leader * 2925 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2926 { 2927 struct filedesc_to_leader *fdtol; 2928 2929 fdtol = malloc(sizeof(struct filedesc_to_leader), 2930 M_FILEDESC_TO_LEADER, M_WAITOK); 2931 fdtol->fdl_refcount = 1; 2932 fdtol->fdl_holdcount = 0; 2933 fdtol->fdl_wakeup = 0; 2934 fdtol->fdl_leader = leader; 2935 if (old != NULL) { 2936 FILEDESC_XLOCK(fdp); 2937 fdtol->fdl_next = old->fdl_next; 2938 fdtol->fdl_prev = old; 2939 old->fdl_next = fdtol; 2940 fdtol->fdl_next->fdl_prev = fdtol; 2941 FILEDESC_XUNLOCK(fdp); 2942 } else { 2943 fdtol->fdl_next = fdtol; 2944 fdtol->fdl_prev = fdtol; 2945 } 2946 return (fdtol); 2947 } 2948 2949 /* 2950 * Get file structures globally. 2951 */ 2952 static int 2953 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2954 { 2955 struct xfile xf; 2956 struct filedesc *fdp; 2957 struct file *fp; 2958 struct proc *p; 2959 int error, n; 2960 2961 error = sysctl_wire_old_buffer(req, 0); 2962 if (error != 0) 2963 return (error); 2964 if (req->oldptr == NULL) { 2965 n = 0; 2966 sx_slock(&allproc_lock); 2967 FOREACH_PROC_IN_SYSTEM(p) { 2968 PROC_LOCK(p); 2969 if (p->p_state == PRS_NEW) { 2970 PROC_UNLOCK(p); 2971 continue; 2972 } 2973 fdp = fdhold(p); 2974 PROC_UNLOCK(p); 2975 if (fdp == NULL) 2976 continue; 2977 /* overestimates sparse tables. */ 2978 if (fdp->fd_lastfile > 0) 2979 n += fdp->fd_lastfile; 2980 fddrop(fdp); 2981 } 2982 sx_sunlock(&allproc_lock); 2983 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2984 } 2985 error = 0; 2986 bzero(&xf, sizeof(xf)); 2987 xf.xf_size = sizeof(xf); 2988 sx_slock(&allproc_lock); 2989 FOREACH_PROC_IN_SYSTEM(p) { 2990 PROC_LOCK(p); 2991 if (p->p_state == PRS_NEW) { 2992 PROC_UNLOCK(p); 2993 continue; 2994 } 2995 if (p_cansee(req->td, p) != 0) { 2996 PROC_UNLOCK(p); 2997 continue; 2998 } 2999 xf.xf_pid = p->p_pid; 3000 xf.xf_uid = p->p_ucred->cr_uid; 3001 fdp = fdhold(p); 3002 PROC_UNLOCK(p); 3003 if (fdp == NULL) 3004 continue; 3005 FILEDESC_SLOCK(fdp); 3006 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3007 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3008 continue; 3009 xf.xf_fd = n; 3010 xf.xf_file = fp; 3011 xf.xf_data = fp->f_data; 3012 xf.xf_vnode = fp->f_vnode; 3013 xf.xf_type = fp->f_type; 3014 xf.xf_count = fp->f_count; 3015 xf.xf_msgcount = 0; 3016 xf.xf_offset = foffset_get(fp); 3017 xf.xf_flag = fp->f_flag; 3018 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3019 if (error) 3020 break; 3021 } 3022 FILEDESC_SUNLOCK(fdp); 3023 fddrop(fdp); 3024 if (error) 3025 break; 3026 } 3027 sx_sunlock(&allproc_lock); 3028 return (error); 3029 } 3030 3031 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3032 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3033 3034 #ifdef KINFO_FILE_SIZE 3035 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3036 #endif 3037 3038 static int 3039 xlate_fflags(int fflags) 3040 { 3041 static const struct { 3042 int fflag; 3043 int kf_fflag; 3044 } fflags_table[] = { 3045 { FAPPEND, KF_FLAG_APPEND }, 3046 { FASYNC, KF_FLAG_ASYNC }, 3047 { FFSYNC, KF_FLAG_FSYNC }, 3048 { FHASLOCK, KF_FLAG_HASLOCK }, 3049 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3050 { FREAD, KF_FLAG_READ }, 3051 { FWRITE, KF_FLAG_WRITE }, 3052 { O_CREAT, KF_FLAG_CREAT }, 3053 { O_DIRECT, KF_FLAG_DIRECT }, 3054 { O_EXCL, KF_FLAG_EXCL }, 3055 { O_EXEC, KF_FLAG_EXEC }, 3056 { O_EXLOCK, KF_FLAG_EXLOCK }, 3057 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3058 { O_SHLOCK, KF_FLAG_SHLOCK }, 3059 { O_TRUNC, KF_FLAG_TRUNC } 3060 }; 3061 unsigned int i; 3062 int kflags; 3063 3064 kflags = 0; 3065 for (i = 0; i < nitems(fflags_table); i++) 3066 if (fflags & fflags_table[i].fflag) 3067 kflags |= fflags_table[i].kf_fflag; 3068 return (kflags); 3069 } 3070 3071 /* Trim unused data from kf_path by truncating the structure size. */ 3072 static void 3073 pack_kinfo(struct kinfo_file *kif) 3074 { 3075 3076 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3077 strlen(kif->kf_path) + 1; 3078 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3079 } 3080 3081 static void 3082 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3083 struct kinfo_file *kif, struct filedesc *fdp) 3084 { 3085 int error; 3086 3087 bzero(kif, sizeof(*kif)); 3088 3089 /* Set a default type to allow for empty fill_kinfo() methods. */ 3090 kif->kf_type = KF_TYPE_UNKNOWN; 3091 kif->kf_flags = xlate_fflags(fp->f_flag); 3092 if (rightsp != NULL) 3093 kif->kf_cap_rights = *rightsp; 3094 else 3095 cap_rights_init(&kif->kf_cap_rights); 3096 kif->kf_fd = fd; 3097 kif->kf_ref_count = fp->f_count; 3098 kif->kf_offset = foffset_get(fp); 3099 3100 /* 3101 * This may drop the filedesc lock, so the 'fp' cannot be 3102 * accessed after this call. 3103 */ 3104 error = fo_fill_kinfo(fp, kif, fdp); 3105 if (error == 0) 3106 kif->kf_status |= KF_ATTR_VALID; 3107 pack_kinfo(kif); 3108 } 3109 3110 static void 3111 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3112 struct kinfo_file *kif) 3113 { 3114 int error; 3115 3116 bzero(kif, sizeof(*kif)); 3117 3118 kif->kf_type = KF_TYPE_VNODE; 3119 error = vn_fill_kinfo_vnode(vp, kif); 3120 if (error == 0) 3121 kif->kf_status |= KF_ATTR_VALID; 3122 kif->kf_flags = xlate_fflags(fflags); 3123 cap_rights_init(&kif->kf_cap_rights); 3124 kif->kf_fd = fd; 3125 kif->kf_ref_count = -1; 3126 kif->kf_offset = -1; 3127 pack_kinfo(kif); 3128 vrele(vp); 3129 } 3130 3131 struct export_fd_buf { 3132 struct filedesc *fdp; 3133 struct sbuf *sb; 3134 ssize_t remainder; 3135 struct kinfo_file kif; 3136 }; 3137 3138 static int 3139 export_kinfo_to_sb(struct export_fd_buf *efbuf) 3140 { 3141 struct kinfo_file *kif; 3142 3143 kif = &efbuf->kif; 3144 if (efbuf->remainder != -1) { 3145 if (efbuf->remainder < kif->kf_structsize) { 3146 /* Terminate export. */ 3147 efbuf->remainder = 0; 3148 return (0); 3149 } 3150 efbuf->remainder -= kif->kf_structsize; 3151 } 3152 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3153 } 3154 3155 static int 3156 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3157 struct export_fd_buf *efbuf) 3158 { 3159 int error; 3160 3161 if (efbuf->remainder == 0) 3162 return (0); 3163 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); 3164 FILEDESC_SUNLOCK(efbuf->fdp); 3165 error = export_kinfo_to_sb(efbuf); 3166 FILEDESC_SLOCK(efbuf->fdp); 3167 return (error); 3168 } 3169 3170 static int 3171 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3172 struct export_fd_buf *efbuf) 3173 { 3174 int error; 3175 3176 if (efbuf->remainder == 0) 3177 return (0); 3178 if (efbuf->fdp != NULL) 3179 FILEDESC_SUNLOCK(efbuf->fdp); 3180 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); 3181 error = export_kinfo_to_sb(efbuf); 3182 if (efbuf->fdp != NULL) 3183 FILEDESC_SLOCK(efbuf->fdp); 3184 return (error); 3185 } 3186 3187 /* 3188 * Store a process file descriptor information to sbuf. 3189 * 3190 * Takes a locked proc as argument, and returns with the proc unlocked. 3191 */ 3192 int 3193 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3194 { 3195 struct file *fp; 3196 struct filedesc *fdp; 3197 struct export_fd_buf *efbuf; 3198 struct vnode *cttyvp, *textvp, *tracevp; 3199 int error, i; 3200 cap_rights_t rights; 3201 3202 PROC_LOCK_ASSERT(p, MA_OWNED); 3203 3204 /* ktrace vnode */ 3205 tracevp = p->p_tracevp; 3206 if (tracevp != NULL) 3207 vref(tracevp); 3208 /* text vnode */ 3209 textvp = p->p_textvp; 3210 if (textvp != NULL) 3211 vref(textvp); 3212 /* Controlling tty. */ 3213 cttyvp = NULL; 3214 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3215 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3216 if (cttyvp != NULL) 3217 vref(cttyvp); 3218 } 3219 fdp = fdhold(p); 3220 PROC_UNLOCK(p); 3221 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3222 efbuf->fdp = NULL; 3223 efbuf->sb = sb; 3224 efbuf->remainder = maxlen; 3225 if (tracevp != NULL) 3226 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3227 efbuf); 3228 if (textvp != NULL) 3229 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3230 if (cttyvp != NULL) 3231 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3232 efbuf); 3233 error = 0; 3234 if (fdp == NULL) 3235 goto fail; 3236 efbuf->fdp = fdp; 3237 FILEDESC_SLOCK(fdp); 3238 /* working directory */ 3239 if (fdp->fd_cdir != NULL) { 3240 vref(fdp->fd_cdir); 3241 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3242 } 3243 /* root directory */ 3244 if (fdp->fd_rdir != NULL) { 3245 vref(fdp->fd_rdir); 3246 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3247 } 3248 /* jail directory */ 3249 if (fdp->fd_jdir != NULL) { 3250 vref(fdp->fd_jdir); 3251 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3252 } 3253 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3254 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3255 continue; 3256 #ifdef CAPABILITIES 3257 rights = *cap_rights(fdp, i); 3258 #else /* !CAPABILITIES */ 3259 cap_rights_init(&rights); 3260 #endif 3261 /* 3262 * Create sysctl entry. It is OK to drop the filedesc 3263 * lock inside of export_file_to_sb() as we will 3264 * re-validate and re-evaluate its properties when the 3265 * loop continues. 3266 */ 3267 error = export_file_to_sb(fp, i, &rights, efbuf); 3268 if (error != 0 || efbuf->remainder == 0) 3269 break; 3270 } 3271 FILEDESC_SUNLOCK(fdp); 3272 fddrop(fdp); 3273 fail: 3274 free(efbuf, M_TEMP); 3275 return (error); 3276 } 3277 3278 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3279 3280 /* 3281 * Get per-process file descriptors for use by procstat(1), et al. 3282 */ 3283 static int 3284 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3285 { 3286 struct sbuf sb; 3287 struct proc *p; 3288 ssize_t maxlen; 3289 int error, error2, *name; 3290 3291 name = (int *)arg1; 3292 3293 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3294 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3295 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3296 if (error != 0) { 3297 sbuf_delete(&sb); 3298 return (error); 3299 } 3300 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3301 error = kern_proc_filedesc_out(p, &sb, maxlen); 3302 error2 = sbuf_finish(&sb); 3303 sbuf_delete(&sb); 3304 return (error != 0 ? error : error2); 3305 } 3306 3307 #ifdef KINFO_OFILE_SIZE 3308 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3309 #endif 3310 3311 #ifdef COMPAT_FREEBSD7 3312 static void 3313 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3314 { 3315 3316 okif->kf_structsize = sizeof(*okif); 3317 okif->kf_type = kif->kf_type; 3318 okif->kf_fd = kif->kf_fd; 3319 okif->kf_ref_count = kif->kf_ref_count; 3320 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3321 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3322 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3323 okif->kf_offset = kif->kf_offset; 3324 okif->kf_vnode_type = kif->kf_vnode_type; 3325 okif->kf_sock_domain = kif->kf_sock_domain; 3326 okif->kf_sock_type = kif->kf_sock_type; 3327 okif->kf_sock_protocol = kif->kf_sock_protocol; 3328 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3329 okif->kf_sa_local = kif->kf_sa_local; 3330 okif->kf_sa_peer = kif->kf_sa_peer; 3331 } 3332 3333 static int 3334 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3335 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3336 { 3337 int error; 3338 3339 vref(vp); 3340 FILEDESC_SUNLOCK(fdp); 3341 export_vnode_to_kinfo(vp, type, 0, kif); 3342 kinfo_to_okinfo(kif, okif); 3343 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3344 FILEDESC_SLOCK(fdp); 3345 return (error); 3346 } 3347 3348 /* 3349 * Get per-process file descriptors for use by procstat(1), et al. 3350 */ 3351 static int 3352 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3353 { 3354 struct kinfo_ofile *okif; 3355 struct kinfo_file *kif; 3356 struct filedesc *fdp; 3357 int error, i, *name; 3358 struct file *fp; 3359 struct proc *p; 3360 3361 name = (int *)arg1; 3362 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3363 if (error != 0) 3364 return (error); 3365 fdp = fdhold(p); 3366 PROC_UNLOCK(p); 3367 if (fdp == NULL) 3368 return (ENOENT); 3369 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3370 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3371 FILEDESC_SLOCK(fdp); 3372 if (fdp->fd_cdir != NULL) 3373 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3374 okif, fdp, req); 3375 if (fdp->fd_rdir != NULL) 3376 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3377 okif, fdp, req); 3378 if (fdp->fd_jdir != NULL) 3379 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3380 okif, fdp, req); 3381 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3382 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3383 continue; 3384 export_file_to_kinfo(fp, i, NULL, kif, fdp); 3385 FILEDESC_SUNLOCK(fdp); 3386 kinfo_to_okinfo(kif, okif); 3387 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3388 FILEDESC_SLOCK(fdp); 3389 if (error) 3390 break; 3391 } 3392 FILEDESC_SUNLOCK(fdp); 3393 fddrop(fdp); 3394 free(kif, M_TEMP); 3395 free(okif, M_TEMP); 3396 return (0); 3397 } 3398 3399 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3400 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3401 "Process ofiledesc entries"); 3402 #endif /* COMPAT_FREEBSD7 */ 3403 3404 int 3405 vntype_to_kinfo(int vtype) 3406 { 3407 struct { 3408 int vtype; 3409 int kf_vtype; 3410 } vtypes_table[] = { 3411 { VBAD, KF_VTYPE_VBAD }, 3412 { VBLK, KF_VTYPE_VBLK }, 3413 { VCHR, KF_VTYPE_VCHR }, 3414 { VDIR, KF_VTYPE_VDIR }, 3415 { VFIFO, KF_VTYPE_VFIFO }, 3416 { VLNK, KF_VTYPE_VLNK }, 3417 { VNON, KF_VTYPE_VNON }, 3418 { VREG, KF_VTYPE_VREG }, 3419 { VSOCK, KF_VTYPE_VSOCK } 3420 }; 3421 unsigned int i; 3422 3423 /* 3424 * Perform vtype translation. 3425 */ 3426 for (i = 0; i < nitems(vtypes_table); i++) 3427 if (vtypes_table[i].vtype == vtype) 3428 return (vtypes_table[i].kf_vtype); 3429 3430 return (KF_VTYPE_UNKNOWN); 3431 } 3432 3433 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3434 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3435 "Process filedesc entries"); 3436 3437 /* 3438 * Store a process current working directory information to sbuf. 3439 * 3440 * Takes a locked proc as argument, and returns with the proc unlocked. 3441 */ 3442 int 3443 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3444 { 3445 struct filedesc *fdp; 3446 struct export_fd_buf *efbuf; 3447 int error; 3448 3449 PROC_LOCK_ASSERT(p, MA_OWNED); 3450 3451 fdp = fdhold(p); 3452 PROC_UNLOCK(p); 3453 if (fdp == NULL) 3454 return (EINVAL); 3455 3456 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3457 efbuf->fdp = fdp; 3458 efbuf->sb = sb; 3459 efbuf->remainder = maxlen; 3460 3461 FILEDESC_SLOCK(fdp); 3462 if (fdp->fd_cdir == NULL) 3463 error = EINVAL; 3464 else { 3465 vref(fdp->fd_cdir); 3466 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3467 FREAD, efbuf); 3468 } 3469 FILEDESC_SUNLOCK(fdp); 3470 fddrop(fdp); 3471 free(efbuf, M_TEMP); 3472 return (error); 3473 } 3474 3475 /* 3476 * Get per-process current working directory. 3477 */ 3478 static int 3479 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3480 { 3481 struct sbuf sb; 3482 struct proc *p; 3483 ssize_t maxlen; 3484 int error, error2, *name; 3485 3486 name = (int *)arg1; 3487 3488 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3489 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3490 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3491 if (error != 0) { 3492 sbuf_delete(&sb); 3493 return (error); 3494 } 3495 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3496 error = kern_proc_cwd_out(p, &sb, maxlen); 3497 error2 = sbuf_finish(&sb); 3498 sbuf_delete(&sb); 3499 return (error != 0 ? error : error2); 3500 } 3501 3502 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3503 sysctl_kern_proc_cwd, "Process current working directory"); 3504 3505 #ifdef DDB 3506 /* 3507 * For the purposes of debugging, generate a human-readable string for the 3508 * file type. 3509 */ 3510 static const char * 3511 file_type_to_name(short type) 3512 { 3513 3514 switch (type) { 3515 case 0: 3516 return ("zero"); 3517 case DTYPE_VNODE: 3518 return ("vnod"); 3519 case DTYPE_SOCKET: 3520 return ("sock"); 3521 case DTYPE_PIPE: 3522 return ("pipe"); 3523 case DTYPE_FIFO: 3524 return ("fifo"); 3525 case DTYPE_KQUEUE: 3526 return ("kque"); 3527 case DTYPE_CRYPTO: 3528 return ("crpt"); 3529 case DTYPE_MQUEUE: 3530 return ("mque"); 3531 case DTYPE_SHM: 3532 return ("shm"); 3533 case DTYPE_SEM: 3534 return ("ksem"); 3535 default: 3536 return ("unkn"); 3537 } 3538 } 3539 3540 /* 3541 * For the purposes of debugging, identify a process (if any, perhaps one of 3542 * many) that references the passed file in its file descriptor array. Return 3543 * NULL if none. 3544 */ 3545 static struct proc * 3546 file_to_first_proc(struct file *fp) 3547 { 3548 struct filedesc *fdp; 3549 struct proc *p; 3550 int n; 3551 3552 FOREACH_PROC_IN_SYSTEM(p) { 3553 if (p->p_state == PRS_NEW) 3554 continue; 3555 fdp = p->p_fd; 3556 if (fdp == NULL) 3557 continue; 3558 for (n = 0; n <= fdp->fd_lastfile; n++) { 3559 if (fp == fdp->fd_ofiles[n].fde_file) 3560 return (p); 3561 } 3562 } 3563 return (NULL); 3564 } 3565 3566 static void 3567 db_print_file(struct file *fp, int header) 3568 { 3569 struct proc *p; 3570 3571 if (header) 3572 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3573 "File", "Type", "Data", "Flag", "GCFl", "Count", 3574 "MCount", "Vnode", "FPID", "FCmd"); 3575 p = file_to_first_proc(fp); 3576 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3577 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3578 0, fp->f_count, 0, fp->f_vnode, 3579 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3580 } 3581 3582 DB_SHOW_COMMAND(file, db_show_file) 3583 { 3584 struct file *fp; 3585 3586 if (!have_addr) { 3587 db_printf("usage: show file <addr>\n"); 3588 return; 3589 } 3590 fp = (struct file *)addr; 3591 db_print_file(fp, 1); 3592 } 3593 3594 DB_SHOW_COMMAND(files, db_show_files) 3595 { 3596 struct filedesc *fdp; 3597 struct file *fp; 3598 struct proc *p; 3599 int header; 3600 int n; 3601 3602 header = 1; 3603 FOREACH_PROC_IN_SYSTEM(p) { 3604 if (p->p_state == PRS_NEW) 3605 continue; 3606 if ((fdp = p->p_fd) == NULL) 3607 continue; 3608 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3609 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3610 continue; 3611 db_print_file(fp, header); 3612 header = 0; 3613 } 3614 } 3615 } 3616 #endif 3617 3618 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3619 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3620 3621 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3622 &maxfiles, 0, "Maximum number of files"); 3623 3624 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3625 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3626 3627 /* ARGSUSED*/ 3628 static void 3629 filelistinit(void *dummy) 3630 { 3631 3632 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3633 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3634 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3635 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3636 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3637 } 3638 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3639 3640 /*-------------------------------------------------------------------*/ 3641 3642 static int 3643 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3644 int flags, struct thread *td) 3645 { 3646 3647 return (EBADF); 3648 } 3649 3650 static int 3651 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3652 struct thread *td) 3653 { 3654 3655 return (EINVAL); 3656 } 3657 3658 static int 3659 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3660 struct thread *td) 3661 { 3662 3663 return (EBADF); 3664 } 3665 3666 static int 3667 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3668 struct thread *td) 3669 { 3670 3671 return (0); 3672 } 3673 3674 static int 3675 badfo_kqfilter(struct file *fp, struct knote *kn) 3676 { 3677 3678 return (EBADF); 3679 } 3680 3681 static int 3682 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3683 struct thread *td) 3684 { 3685 3686 return (EBADF); 3687 } 3688 3689 static int 3690 badfo_close(struct file *fp, struct thread *td) 3691 { 3692 3693 return (0); 3694 } 3695 3696 static int 3697 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3698 struct thread *td) 3699 { 3700 3701 return (EBADF); 3702 } 3703 3704 static int 3705 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3706 struct thread *td) 3707 { 3708 3709 return (EBADF); 3710 } 3711 3712 static int 3713 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3714 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3715 int kflags, struct thread *td) 3716 { 3717 3718 return (EBADF); 3719 } 3720 3721 static int 3722 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3723 { 3724 3725 return (0); 3726 } 3727 3728 struct fileops badfileops = { 3729 .fo_read = badfo_readwrite, 3730 .fo_write = badfo_readwrite, 3731 .fo_truncate = badfo_truncate, 3732 .fo_ioctl = badfo_ioctl, 3733 .fo_poll = badfo_poll, 3734 .fo_kqfilter = badfo_kqfilter, 3735 .fo_stat = badfo_stat, 3736 .fo_close = badfo_close, 3737 .fo_chmod = badfo_chmod, 3738 .fo_chown = badfo_chown, 3739 .fo_sendfile = badfo_sendfile, 3740 .fo_fill_kinfo = badfo_fill_kinfo, 3741 }; 3742 3743 int 3744 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3745 int flags, struct thread *td) 3746 { 3747 3748 return (EOPNOTSUPP); 3749 } 3750 3751 int 3752 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3753 struct thread *td) 3754 { 3755 3756 return (EINVAL); 3757 } 3758 3759 int 3760 invfo_ioctl(struct file *fp, u_long com, void *data, 3761 struct ucred *active_cred, struct thread *td) 3762 { 3763 3764 return (ENOTTY); 3765 } 3766 3767 int 3768 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 3769 struct thread *td) 3770 { 3771 3772 return (poll_no_poll(events)); 3773 } 3774 3775 int 3776 invfo_kqfilter(struct file *fp, struct knote *kn) 3777 { 3778 3779 return (EINVAL); 3780 } 3781 3782 int 3783 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3784 struct thread *td) 3785 { 3786 3787 return (EINVAL); 3788 } 3789 3790 int 3791 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3792 struct thread *td) 3793 { 3794 3795 return (EINVAL); 3796 } 3797 3798 int 3799 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3800 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3801 int kflags, struct thread *td) 3802 { 3803 3804 return (EINVAL); 3805 } 3806 3807 /*-------------------------------------------------------------------*/ 3808 3809 /* 3810 * File Descriptor pseudo-device driver (/dev/fd/). 3811 * 3812 * Opening minor device N dup()s the file (if any) connected to file 3813 * descriptor N belonging to the calling process. Note that this driver 3814 * consists of only the ``open()'' routine, because all subsequent 3815 * references to this file will be direct to the other driver. 3816 * 3817 * XXX: we could give this one a cloning event handler if necessary. 3818 */ 3819 3820 /* ARGSUSED */ 3821 static int 3822 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3823 { 3824 3825 /* 3826 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3827 * the file descriptor being sought for duplication. The error 3828 * return ensures that the vnode for this device will be released 3829 * by vn_open. Open will detect this special error and take the 3830 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3831 * will simply report the error. 3832 */ 3833 td->td_dupfd = dev2unit(dev); 3834 return (ENODEV); 3835 } 3836 3837 static struct cdevsw fildesc_cdevsw = { 3838 .d_version = D_VERSION, 3839 .d_open = fdopen, 3840 .d_name = "FD", 3841 }; 3842 3843 static void 3844 fildesc_drvinit(void *unused) 3845 { 3846 struct cdev *dev; 3847 3848 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3849 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3850 make_dev_alias(dev, "stdin"); 3851 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3852 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3853 make_dev_alias(dev, "stdout"); 3854 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3855 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3856 make_dev_alias(dev, "stderr"); 3857 } 3858 3859 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3860