1 /* $NetBSD: kern_descrip.c,v 1.227 2014/09/05 09:20:59 matt Exp $ */ 2 3 /*- 4 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 1982, 1986, 1989, 1991, 1993 34 * The Regents of the University of California. All rights reserved. 35 * (c) UNIX System Laboratories, Inc. 36 * All or some portions of this file are derived from material licensed 37 * to the University of California by American Telephone and Telegraph 38 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 39 * the permission of UNIX System Laboratories, Inc. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)kern_descrip.c 8.8 (Berkeley) 2/14/95 66 */ 67 68 /* 69 * File descriptor management. 70 */ 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: kern_descrip.c,v 1.227 2014/09/05 09:20:59 matt Exp $"); 74 75 #include <sys/param.h> 76 #include <sys/systm.h> 77 #include <sys/filedesc.h> 78 #include <sys/kernel.h> 79 #include <sys/proc.h> 80 #include <sys/file.h> 81 #include <sys/socket.h> 82 #include <sys/socketvar.h> 83 #include <sys/stat.h> 84 #include <sys/ioctl.h> 85 #include <sys/fcntl.h> 86 #include <sys/pool.h> 87 #include <sys/unistd.h> 88 #include <sys/resourcevar.h> 89 #include <sys/conf.h> 90 #include <sys/event.h> 91 #include <sys/kauth.h> 92 #include <sys/atomic.h> 93 #include <sys/syscallargs.h> 94 #include <sys/cpu.h> 95 #include <sys/kmem.h> 96 #include <sys/vnode.h> 97 #include <sys/sysctl.h> 98 #include <sys/ktrace.h> 99 100 /* 101 * A list (head) of open files, counter, and lock protecting them. 102 */ 103 struct filelist filehead __cacheline_aligned; 104 static u_int nfiles __cacheline_aligned; 105 kmutex_t filelist_lock __cacheline_aligned; 106 107 static pool_cache_t filedesc_cache __read_mostly; 108 static pool_cache_t file_cache __read_mostly; 109 static pool_cache_t fdfile_cache __read_mostly; 110 111 static int file_ctor(void *, void *, int); 112 static void file_dtor(void *, void *); 113 static int fdfile_ctor(void *, void *, int); 114 static void fdfile_dtor(void *, void *); 115 static int filedesc_ctor(void *, void *, int); 116 static void filedesc_dtor(void *, void *); 117 static int filedescopen(dev_t, int, int, lwp_t *); 118 119 static int sysctl_kern_file(SYSCTLFN_PROTO); 120 static int sysctl_kern_file2(SYSCTLFN_PROTO); 121 static void fill_file(struct kinfo_file *, const file_t *, const fdfile_t *, 122 int, pid_t); 123 124 const struct cdevsw filedesc_cdevsw = { 125 .d_open = filedescopen, 126 .d_close = noclose, 127 .d_read = noread, 128 .d_write = nowrite, 129 .d_ioctl = noioctl, 130 .d_stop = nostop, 131 .d_tty = notty, 132 .d_poll = nopoll, 133 .d_mmap = nommap, 134 .d_kqfilter = nokqfilter, 135 .d_discard = nodiscard, 136 .d_flag = D_OTHER | D_MPSAFE 137 }; 138 139 /* For ease of reading. */ 140 __strong_alias(fd_putvnode,fd_putfile) 141 __strong_alias(fd_putsock,fd_putfile) 142 143 /* 144 * Initialize the descriptor system. 145 */ 146 void 147 fd_sys_init(void) 148 { 149 static struct sysctllog *clog; 150 151 mutex_init(&filelist_lock, MUTEX_DEFAULT, IPL_NONE); 152 153 file_cache = pool_cache_init(sizeof(file_t), coherency_unit, 0, 154 0, "file", NULL, IPL_NONE, file_ctor, file_dtor, NULL); 155 KASSERT(file_cache != NULL); 156 157 fdfile_cache = pool_cache_init(sizeof(fdfile_t), coherency_unit, 0, 158 PR_LARGECACHE, "fdfile", NULL, IPL_NONE, fdfile_ctor, fdfile_dtor, 159 NULL); 160 KASSERT(fdfile_cache != NULL); 161 162 filedesc_cache = pool_cache_init(sizeof(filedesc_t), coherency_unit, 163 0, 0, "filedesc", NULL, IPL_NONE, filedesc_ctor, filedesc_dtor, 164 NULL); 165 KASSERT(filedesc_cache != NULL); 166 167 sysctl_createv(&clog, 0, NULL, NULL, 168 CTLFLAG_PERMANENT, 169 CTLTYPE_STRUCT, "file", 170 SYSCTL_DESCR("System open file table"), 171 sysctl_kern_file, 0, NULL, 0, 172 CTL_KERN, KERN_FILE, CTL_EOL); 173 sysctl_createv(&clog, 0, NULL, NULL, 174 CTLFLAG_PERMANENT, 175 CTLTYPE_STRUCT, "file2", 176 SYSCTL_DESCR("System open file table"), 177 sysctl_kern_file2, 0, NULL, 0, 178 CTL_KERN, KERN_FILE2, CTL_EOL); 179 } 180 181 static bool 182 fd_isused(filedesc_t *fdp, unsigned fd) 183 { 184 u_int off = fd >> NDENTRYSHIFT; 185 186 KASSERT(fd < fdp->fd_dt->dt_nfiles); 187 188 return (fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0; 189 } 190 191 /* 192 * Verify that the bitmaps match the descriptor table. 193 */ 194 static inline void 195 fd_checkmaps(filedesc_t *fdp) 196 { 197 #ifdef DEBUG 198 fdtab_t *dt; 199 u_int fd; 200 201 dt = fdp->fd_dt; 202 if (fdp->fd_refcnt == -1) { 203 /* 204 * fd_free tears down the table without maintaining its bitmap. 205 */ 206 return; 207 } 208 for (fd = 0; fd < dt->dt_nfiles; fd++) { 209 if (fd < NDFDFILE) { 210 KASSERT(dt->dt_ff[fd] == 211 (fdfile_t *)fdp->fd_dfdfile[fd]); 212 } 213 if (dt->dt_ff[fd] == NULL) { 214 KASSERT(!fd_isused(fdp, fd)); 215 } else if (dt->dt_ff[fd]->ff_file != NULL) { 216 KASSERT(fd_isused(fdp, fd)); 217 } 218 } 219 #endif 220 } 221 222 static int 223 fd_next_zero(filedesc_t *fdp, uint32_t *bitmap, int want, u_int bits) 224 { 225 int i, off, maxoff; 226 uint32_t sub; 227 228 KASSERT(mutex_owned(&fdp->fd_lock)); 229 230 fd_checkmaps(fdp); 231 232 if (want > bits) 233 return -1; 234 235 off = want >> NDENTRYSHIFT; 236 i = want & NDENTRYMASK; 237 if (i) { 238 sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i)); 239 if (sub != ~0) 240 goto found; 241 off++; 242 } 243 244 maxoff = NDLOSLOTS(bits); 245 while (off < maxoff) { 246 if ((sub = bitmap[off]) != ~0) 247 goto found; 248 off++; 249 } 250 251 return -1; 252 253 found: 254 return (off << NDENTRYSHIFT) + ffs(~sub) - 1; 255 } 256 257 static int 258 fd_last_set(filedesc_t *fd, int last) 259 { 260 int off, i; 261 fdfile_t **ff = fd->fd_dt->dt_ff; 262 uint32_t *bitmap = fd->fd_lomap; 263 264 KASSERT(mutex_owned(&fd->fd_lock)); 265 266 fd_checkmaps(fd); 267 268 off = (last - 1) >> NDENTRYSHIFT; 269 270 while (off >= 0 && !bitmap[off]) 271 off--; 272 273 if (off < 0) 274 return -1; 275 276 i = ((off + 1) << NDENTRYSHIFT) - 1; 277 if (i >= last) 278 i = last - 1; 279 280 /* XXX should use bitmap */ 281 while (i > 0 && (ff[i] == NULL || !ff[i]->ff_allocated)) 282 i--; 283 284 return i; 285 } 286 287 static inline void 288 fd_used(filedesc_t *fdp, unsigned fd) 289 { 290 u_int off = fd >> NDENTRYSHIFT; 291 fdfile_t *ff; 292 293 ff = fdp->fd_dt->dt_ff[fd]; 294 295 KASSERT(mutex_owned(&fdp->fd_lock)); 296 KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0); 297 KASSERT(ff != NULL); 298 KASSERT(ff->ff_file == NULL); 299 KASSERT(!ff->ff_allocated); 300 301 ff->ff_allocated = true; 302 fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK); 303 if (__predict_false(fdp->fd_lomap[off] == ~0)) { 304 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] & 305 (1 << (off & NDENTRYMASK))) == 0); 306 fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK); 307 } 308 309 if ((int)fd > fdp->fd_lastfile) { 310 fdp->fd_lastfile = fd; 311 } 312 313 fd_checkmaps(fdp); 314 } 315 316 static inline void 317 fd_unused(filedesc_t *fdp, unsigned fd) 318 { 319 u_int off = fd >> NDENTRYSHIFT; 320 fdfile_t *ff; 321 322 ff = fdp->fd_dt->dt_ff[fd]; 323 324 /* 325 * Don't assert the lock is held here, as we may be copying 326 * the table during exec() and it is not needed there. 327 * procfs and sysctl are locked out by proc::p_reflock. 328 * 329 * KASSERT(mutex_owned(&fdp->fd_lock)); 330 */ 331 KASSERT(ff != NULL); 332 KASSERT(ff->ff_file == NULL); 333 KASSERT(ff->ff_allocated); 334 335 if (fd < fdp->fd_freefile) { 336 fdp->fd_freefile = fd; 337 } 338 339 if (fdp->fd_lomap[off] == ~0) { 340 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] & 341 (1 << (off & NDENTRYMASK))) != 0); 342 fdp->fd_himap[off >> NDENTRYSHIFT] &= 343 ~(1 << (off & NDENTRYMASK)); 344 } 345 KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0); 346 fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK)); 347 ff->ff_allocated = false; 348 349 KASSERT(fd <= fdp->fd_lastfile); 350 if (fd == fdp->fd_lastfile) { 351 fdp->fd_lastfile = fd_last_set(fdp, fd); 352 } 353 fd_checkmaps(fdp); 354 } 355 356 /* 357 * Look up the file structure corresponding to a file descriptor 358 * and return the file, holding a reference on the descriptor. 359 */ 360 file_t * 361 fd_getfile(unsigned fd) 362 { 363 filedesc_t *fdp; 364 fdfile_t *ff; 365 file_t *fp; 366 fdtab_t *dt; 367 368 /* 369 * Look up the fdfile structure representing this descriptor. 370 * We are doing this unlocked. See fd_tryexpand(). 371 */ 372 fdp = curlwp->l_fd; 373 dt = fdp->fd_dt; 374 if (__predict_false(fd >= dt->dt_nfiles)) { 375 return NULL; 376 } 377 ff = dt->dt_ff[fd]; 378 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 379 if (__predict_false(ff == NULL)) { 380 return NULL; 381 } 382 383 /* Now get a reference to the descriptor. */ 384 if (fdp->fd_refcnt == 1) { 385 /* 386 * Single threaded: don't need to worry about concurrent 387 * access (other than earlier calls to kqueue, which may 388 * hold a reference to the descriptor). 389 */ 390 ff->ff_refcnt++; 391 } else { 392 /* 393 * Multi threaded: issue a memory barrier to ensure that we 394 * acquire the file pointer _after_ adding a reference. If 395 * no memory barrier, we could fetch a stale pointer. 396 */ 397 atomic_inc_uint(&ff->ff_refcnt); 398 #ifndef __HAVE_ATOMIC_AS_MEMBAR 399 membar_enter(); 400 #endif 401 } 402 403 /* 404 * If the file is not open or is being closed then put the 405 * reference back. 406 */ 407 fp = ff->ff_file; 408 if (__predict_true(fp != NULL)) { 409 return fp; 410 } 411 fd_putfile(fd); 412 return NULL; 413 } 414 415 /* 416 * Release a reference to a file descriptor acquired with fd_getfile(). 417 */ 418 void 419 fd_putfile(unsigned fd) 420 { 421 filedesc_t *fdp; 422 fdfile_t *ff; 423 u_int u, v; 424 425 fdp = curlwp->l_fd; 426 ff = fdp->fd_dt->dt_ff[fd]; 427 428 KASSERT(fd < fdp->fd_dt->dt_nfiles); 429 KASSERT(ff != NULL); 430 KASSERT((ff->ff_refcnt & FR_MASK) > 0); 431 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 432 433 if (fdp->fd_refcnt == 1) { 434 /* 435 * Single threaded: don't need to worry about concurrent 436 * access (other than earlier calls to kqueue, which may 437 * hold a reference to the descriptor). 438 */ 439 if (__predict_false((ff->ff_refcnt & FR_CLOSING) != 0)) { 440 fd_close(fd); 441 return; 442 } 443 ff->ff_refcnt--; 444 return; 445 } 446 447 /* 448 * Ensure that any use of the file is complete and globally 449 * visible before dropping the final reference. If no membar, 450 * the current CPU could still access memory associated with 451 * the file after it has been freed or recycled by another 452 * CPU. 453 */ 454 #ifndef __HAVE_ATOMIC_AS_MEMBAR 455 membar_exit(); 456 #endif 457 458 /* 459 * Be optimistic and start out with the assumption that no other 460 * threads are trying to close the descriptor. If the CAS fails, 461 * we lost a race and/or it's being closed. 462 */ 463 for (u = ff->ff_refcnt & FR_MASK;; u = v) { 464 v = atomic_cas_uint(&ff->ff_refcnt, u, u - 1); 465 if (__predict_true(u == v)) { 466 return; 467 } 468 if (__predict_false((v & FR_CLOSING) != 0)) { 469 break; 470 } 471 } 472 473 /* Another thread is waiting to close the file: join it. */ 474 (void)fd_close(fd); 475 } 476 477 /* 478 * Convenience wrapper around fd_getfile() that returns reference 479 * to a vnode. 480 */ 481 int 482 fd_getvnode(unsigned fd, file_t **fpp) 483 { 484 vnode_t *vp; 485 file_t *fp; 486 487 fp = fd_getfile(fd); 488 if (__predict_false(fp == NULL)) { 489 return EBADF; 490 } 491 if (__predict_false(fp->f_type != DTYPE_VNODE)) { 492 fd_putfile(fd); 493 return EINVAL; 494 } 495 vp = fp->f_vnode; 496 if (__predict_false(vp->v_type == VBAD)) { 497 /* XXX Is this case really necessary? */ 498 fd_putfile(fd); 499 return EBADF; 500 } 501 *fpp = fp; 502 return 0; 503 } 504 505 /* 506 * Convenience wrapper around fd_getfile() that returns reference 507 * to a socket. 508 */ 509 int 510 fd_getsock1(unsigned fd, struct socket **sop, file_t **fp) 511 { 512 *fp = fd_getfile(fd); 513 if (__predict_false(*fp == NULL)) { 514 return EBADF; 515 } 516 if (__predict_false((*fp)->f_type != DTYPE_SOCKET)) { 517 fd_putfile(fd); 518 return ENOTSOCK; 519 } 520 *sop = (*fp)->f_socket; 521 return 0; 522 } 523 524 int 525 fd_getsock(unsigned fd, struct socket **sop) 526 { 527 file_t *fp; 528 return fd_getsock1(fd, sop, &fp); 529 } 530 531 /* 532 * Look up the file structure corresponding to a file descriptor 533 * and return it with a reference held on the file, not the 534 * descriptor. 535 * 536 * This is heavyweight and only used when accessing descriptors 537 * from a foreign process. The caller must ensure that `p' does 538 * not exit or fork across this call. 539 * 540 * To release the file (not descriptor) reference, use closef(). 541 */ 542 file_t * 543 fd_getfile2(proc_t *p, unsigned fd) 544 { 545 filedesc_t *fdp; 546 fdfile_t *ff; 547 file_t *fp; 548 fdtab_t *dt; 549 550 fdp = p->p_fd; 551 mutex_enter(&fdp->fd_lock); 552 dt = fdp->fd_dt; 553 if (fd >= dt->dt_nfiles) { 554 mutex_exit(&fdp->fd_lock); 555 return NULL; 556 } 557 if ((ff = dt->dt_ff[fd]) == NULL) { 558 mutex_exit(&fdp->fd_lock); 559 return NULL; 560 } 561 if ((fp = ff->ff_file) == NULL) { 562 mutex_exit(&fdp->fd_lock); 563 return NULL; 564 } 565 mutex_enter(&fp->f_lock); 566 fp->f_count++; 567 mutex_exit(&fp->f_lock); 568 mutex_exit(&fdp->fd_lock); 569 570 return fp; 571 } 572 573 /* 574 * Internal form of close. Must be called with a reference to the 575 * descriptor, and will drop the reference. When all descriptor 576 * references are dropped, releases the descriptor slot and a single 577 * reference to the file structure. 578 */ 579 int 580 fd_close(unsigned fd) 581 { 582 struct flock lf; 583 filedesc_t *fdp; 584 fdfile_t *ff; 585 file_t *fp; 586 proc_t *p; 587 lwp_t *l; 588 u_int refcnt; 589 590 l = curlwp; 591 p = l->l_proc; 592 fdp = l->l_fd; 593 ff = fdp->fd_dt->dt_ff[fd]; 594 595 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 596 597 mutex_enter(&fdp->fd_lock); 598 KASSERT((ff->ff_refcnt & FR_MASK) > 0); 599 if (__predict_false(ff->ff_file == NULL)) { 600 /* 601 * Another user of the file is already closing, and is 602 * waiting for other users of the file to drain. Release 603 * our reference, and wake up the closer. 604 */ 605 atomic_dec_uint(&ff->ff_refcnt); 606 cv_broadcast(&ff->ff_closing); 607 mutex_exit(&fdp->fd_lock); 608 609 /* 610 * An application error, so pretend that the descriptor 611 * was already closed. We can't safely wait for it to 612 * be closed without potentially deadlocking. 613 */ 614 return (EBADF); 615 } 616 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0); 617 618 /* 619 * There may be multiple users of this file within the process. 620 * Notify existing and new users that the file is closing. This 621 * will prevent them from adding additional uses to this file 622 * while we are closing it. 623 */ 624 fp = ff->ff_file; 625 ff->ff_file = NULL; 626 ff->ff_exclose = false; 627 628 /* 629 * We expect the caller to hold a descriptor reference - drop it. 630 * The reference count may increase beyond zero at this point due 631 * to an erroneous descriptor reference by an application, but 632 * fd_getfile() will notice that the file is being closed and drop 633 * the reference again. 634 */ 635 if (fdp->fd_refcnt == 1) { 636 /* Single threaded. */ 637 refcnt = --(ff->ff_refcnt); 638 } else { 639 /* Multi threaded. */ 640 #ifndef __HAVE_ATOMIC_AS_MEMBAR 641 membar_producer(); 642 #endif 643 refcnt = atomic_dec_uint_nv(&ff->ff_refcnt); 644 } 645 if (__predict_false(refcnt != 0)) { 646 /* 647 * Wait for other references to drain. This is typically 648 * an application error - the descriptor is being closed 649 * while still in use. 650 * (Or just a threaded application trying to unblock its 651 * thread that sleeps in (say) accept()). 652 */ 653 atomic_or_uint(&ff->ff_refcnt, FR_CLOSING); 654 655 /* 656 * Remove any knotes attached to the file. A knote 657 * attached to the descriptor can hold references on it. 658 */ 659 mutex_exit(&fdp->fd_lock); 660 if (!SLIST_EMPTY(&ff->ff_knlist)) { 661 knote_fdclose(fd); 662 } 663 664 /* 665 * Since the file system code doesn't know which fd 666 * each request came from (think dup()), we have to 667 * ask it to return ERESTART for any long-term blocks. 668 * The re-entry through read/write/etc will detect the 669 * closed fd and return EBAFD. 670 * Blocked partial writes may return a short length. 671 */ 672 (*fp->f_ops->fo_restart)(fp); 673 mutex_enter(&fdp->fd_lock); 674 675 /* 676 * We need to see the count drop to zero at least once, 677 * in order to ensure that all pre-existing references 678 * have been drained. New references past this point are 679 * of no interest. 680 * XXX (dsl) this may need to call fo_restart() after a 681 * timeout to guarantee that all the system calls exit. 682 */ 683 while ((ff->ff_refcnt & FR_MASK) != 0) { 684 cv_wait(&ff->ff_closing, &fdp->fd_lock); 685 } 686 atomic_and_uint(&ff->ff_refcnt, ~FR_CLOSING); 687 } else { 688 /* If no references, there must be no knotes. */ 689 KASSERT(SLIST_EMPTY(&ff->ff_knlist)); 690 } 691 692 /* 693 * POSIX record locking dictates that any close releases ALL 694 * locks owned by this process. This is handled by setting 695 * a flag in the unlock to free ONLY locks obeying POSIX 696 * semantics, and not to free BSD-style file locks. 697 * If the descriptor was in a message, POSIX-style locks 698 * aren't passed with the descriptor. 699 */ 700 if (__predict_false((p->p_flag & PK_ADVLOCK) != 0 && 701 fp->f_type == DTYPE_VNODE)) { 702 lf.l_whence = SEEK_SET; 703 lf.l_start = 0; 704 lf.l_len = 0; 705 lf.l_type = F_UNLCK; 706 mutex_exit(&fdp->fd_lock); 707 (void)VOP_ADVLOCK(fp->f_vnode, p, F_UNLCK, &lf, F_POSIX); 708 mutex_enter(&fdp->fd_lock); 709 } 710 711 /* Free descriptor slot. */ 712 fd_unused(fdp, fd); 713 mutex_exit(&fdp->fd_lock); 714 715 /* Now drop reference to the file itself. */ 716 return closef(fp); 717 } 718 719 /* 720 * Duplicate a file descriptor. 721 */ 722 int 723 fd_dup(file_t *fp, int minfd, int *newp, bool exclose) 724 { 725 proc_t *p = curproc; 726 int error; 727 728 while ((error = fd_alloc(p, minfd, newp)) != 0) { 729 if (error != ENOSPC) { 730 return error; 731 } 732 fd_tryexpand(p); 733 } 734 735 curlwp->l_fd->fd_dt->dt_ff[*newp]->ff_exclose = exclose; 736 fd_affix(p, fp, *newp); 737 return 0; 738 } 739 740 /* 741 * dup2 operation. 742 */ 743 int 744 fd_dup2(file_t *fp, unsigned newfd, int flags) 745 { 746 filedesc_t *fdp = curlwp->l_fd; 747 fdfile_t *ff; 748 fdtab_t *dt; 749 750 if (flags & ~(O_CLOEXEC|O_NONBLOCK)) 751 return EINVAL; 752 /* 753 * Ensure there are enough slots in the descriptor table, 754 * and allocate an fdfile_t up front in case we need it. 755 */ 756 while (newfd >= fdp->fd_dt->dt_nfiles) { 757 fd_tryexpand(curproc); 758 } 759 ff = pool_cache_get(fdfile_cache, PR_WAITOK); 760 761 /* 762 * If there is already a file open, close it. If the file is 763 * half open, wait for it to be constructed before closing it. 764 * XXX Potential for deadlock here? 765 */ 766 mutex_enter(&fdp->fd_lock); 767 while (fd_isused(fdp, newfd)) { 768 mutex_exit(&fdp->fd_lock); 769 if (fd_getfile(newfd) != NULL) { 770 (void)fd_close(newfd); 771 } else { 772 /* 773 * Crummy, but unlikely to happen. 774 * Can occur if we interrupt another 775 * thread while it is opening a file. 776 */ 777 kpause("dup2", false, 1, NULL); 778 } 779 mutex_enter(&fdp->fd_lock); 780 } 781 dt = fdp->fd_dt; 782 if (dt->dt_ff[newfd] == NULL) { 783 KASSERT(newfd >= NDFDFILE); 784 dt->dt_ff[newfd] = ff; 785 ff = NULL; 786 } 787 fd_used(fdp, newfd); 788 mutex_exit(&fdp->fd_lock); 789 790 dt->dt_ff[newfd]->ff_exclose = (flags & O_CLOEXEC) != 0; 791 fp->f_flag |= flags & FNONBLOCK; 792 /* Slot is now allocated. Insert copy of the file. */ 793 fd_affix(curproc, fp, newfd); 794 if (ff != NULL) { 795 pool_cache_put(fdfile_cache, ff); 796 } 797 return 0; 798 } 799 800 /* 801 * Drop reference to a file structure. 802 */ 803 int 804 closef(file_t *fp) 805 { 806 struct flock lf; 807 int error; 808 809 /* 810 * Drop reference. If referenced elsewhere it's still open 811 * and we have nothing more to do. 812 */ 813 mutex_enter(&fp->f_lock); 814 KASSERT(fp->f_count > 0); 815 if (--fp->f_count > 0) { 816 mutex_exit(&fp->f_lock); 817 return 0; 818 } 819 KASSERT(fp->f_count == 0); 820 mutex_exit(&fp->f_lock); 821 822 /* We held the last reference - release locks, close and free. */ 823 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) { 824 lf.l_whence = SEEK_SET; 825 lf.l_start = 0; 826 lf.l_len = 0; 827 lf.l_type = F_UNLCK; 828 (void)VOP_ADVLOCK(fp->f_vnode, fp, F_UNLCK, &lf, F_FLOCK); 829 } 830 if (fp->f_ops != NULL) { 831 error = (*fp->f_ops->fo_close)(fp); 832 } else { 833 error = 0; 834 } 835 KASSERT(fp->f_count == 0); 836 KASSERT(fp->f_cred != NULL); 837 pool_cache_put(file_cache, fp); 838 839 return error; 840 } 841 842 /* 843 * Allocate a file descriptor for the process. 844 */ 845 int 846 fd_alloc(proc_t *p, int want, int *result) 847 { 848 filedesc_t *fdp = p->p_fd; 849 int i, lim, last, error, hi; 850 u_int off; 851 fdtab_t *dt; 852 853 KASSERT(p == curproc || p == &proc0); 854 855 /* 856 * Search for a free descriptor starting at the higher 857 * of want or fd_freefile. 858 */ 859 mutex_enter(&fdp->fd_lock); 860 fd_checkmaps(fdp); 861 dt = fdp->fd_dt; 862 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 863 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles); 864 last = min(dt->dt_nfiles, lim); 865 for (;;) { 866 if ((i = want) < fdp->fd_freefile) 867 i = fdp->fd_freefile; 868 off = i >> NDENTRYSHIFT; 869 hi = fd_next_zero(fdp, fdp->fd_himap, off, 870 (last + NDENTRIES - 1) >> NDENTRYSHIFT); 871 if (hi == -1) 872 break; 873 i = fd_next_zero(fdp, &fdp->fd_lomap[hi], 874 hi > off ? 0 : i & NDENTRYMASK, NDENTRIES); 875 if (i == -1) { 876 /* 877 * Free file descriptor in this block was 878 * below want, try again with higher want. 879 */ 880 want = (hi + 1) << NDENTRYSHIFT; 881 continue; 882 } 883 i += (hi << NDENTRYSHIFT); 884 if (i >= last) { 885 break; 886 } 887 if (dt->dt_ff[i] == NULL) { 888 KASSERT(i >= NDFDFILE); 889 dt->dt_ff[i] = pool_cache_get(fdfile_cache, PR_WAITOK); 890 } 891 KASSERT(dt->dt_ff[i]->ff_file == NULL); 892 fd_used(fdp, i); 893 if (want <= fdp->fd_freefile) { 894 fdp->fd_freefile = i; 895 } 896 *result = i; 897 KASSERT(i >= NDFDFILE || 898 dt->dt_ff[i] == (fdfile_t *)fdp->fd_dfdfile[i]); 899 fd_checkmaps(fdp); 900 mutex_exit(&fdp->fd_lock); 901 return 0; 902 } 903 904 /* No space in current array. Let the caller expand and retry. */ 905 error = (dt->dt_nfiles >= lim) ? EMFILE : ENOSPC; 906 mutex_exit(&fdp->fd_lock); 907 return error; 908 } 909 910 /* 911 * Allocate memory for a descriptor table. 912 */ 913 static fdtab_t * 914 fd_dtab_alloc(int n) 915 { 916 fdtab_t *dt; 917 size_t sz; 918 919 KASSERT(n > NDFILE); 920 921 sz = sizeof(*dt) + (n - NDFILE) * sizeof(dt->dt_ff[0]); 922 dt = kmem_alloc(sz, KM_SLEEP); 923 #ifdef DIAGNOSTIC 924 memset(dt, 0xff, sz); 925 #endif 926 dt->dt_nfiles = n; 927 dt->dt_link = NULL; 928 return dt; 929 } 930 931 /* 932 * Free a descriptor table, and all tables linked for deferred free. 933 */ 934 static void 935 fd_dtab_free(fdtab_t *dt) 936 { 937 fdtab_t *next; 938 size_t sz; 939 940 do { 941 next = dt->dt_link; 942 KASSERT(dt->dt_nfiles > NDFILE); 943 sz = sizeof(*dt) + 944 (dt->dt_nfiles - NDFILE) * sizeof(dt->dt_ff[0]); 945 #ifdef DIAGNOSTIC 946 memset(dt, 0xff, sz); 947 #endif 948 kmem_free(dt, sz); 949 dt = next; 950 } while (dt != NULL); 951 } 952 953 /* 954 * Allocate descriptor bitmap. 955 */ 956 static void 957 fd_map_alloc(int n, uint32_t **lo, uint32_t **hi) 958 { 959 uint8_t *ptr; 960 size_t szlo, szhi; 961 962 KASSERT(n > NDENTRIES); 963 964 szlo = NDLOSLOTS(n) * sizeof(uint32_t); 965 szhi = NDHISLOTS(n) * sizeof(uint32_t); 966 ptr = kmem_alloc(szlo + szhi, KM_SLEEP); 967 *lo = (uint32_t *)ptr; 968 *hi = (uint32_t *)(ptr + szlo); 969 } 970 971 /* 972 * Free descriptor bitmap. 973 */ 974 static void 975 fd_map_free(int n, uint32_t *lo, uint32_t *hi) 976 { 977 size_t szlo, szhi; 978 979 KASSERT(n > NDENTRIES); 980 981 szlo = NDLOSLOTS(n) * sizeof(uint32_t); 982 szhi = NDHISLOTS(n) * sizeof(uint32_t); 983 KASSERT(hi == (uint32_t *)((uint8_t *)lo + szlo)); 984 kmem_free(lo, szlo + szhi); 985 } 986 987 /* 988 * Expand a process' descriptor table. 989 */ 990 void 991 fd_tryexpand(proc_t *p) 992 { 993 filedesc_t *fdp; 994 int i, numfiles, oldnfiles; 995 fdtab_t *newdt, *dt; 996 uint32_t *newhimap, *newlomap; 997 998 KASSERT(p == curproc || p == &proc0); 999 1000 fdp = p->p_fd; 1001 newhimap = NULL; 1002 newlomap = NULL; 1003 oldnfiles = fdp->fd_dt->dt_nfiles; 1004 1005 if (oldnfiles < NDEXTENT) 1006 numfiles = NDEXTENT; 1007 else 1008 numfiles = 2 * oldnfiles; 1009 1010 newdt = fd_dtab_alloc(numfiles); 1011 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1012 fd_map_alloc(numfiles, &newlomap, &newhimap); 1013 } 1014 1015 mutex_enter(&fdp->fd_lock); 1016 dt = fdp->fd_dt; 1017 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 1018 if (dt->dt_nfiles != oldnfiles) { 1019 /* fdp changed; caller must retry */ 1020 mutex_exit(&fdp->fd_lock); 1021 fd_dtab_free(newdt); 1022 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1023 fd_map_free(numfiles, newlomap, newhimap); 1024 } 1025 return; 1026 } 1027 1028 /* Copy the existing descriptor table and zero the new portion. */ 1029 i = sizeof(fdfile_t *) * oldnfiles; 1030 memcpy(newdt->dt_ff, dt->dt_ff, i); 1031 memset((uint8_t *)newdt->dt_ff + i, 0, 1032 numfiles * sizeof(fdfile_t *) - i); 1033 1034 /* 1035 * Link old descriptor array into list to be discarded. We defer 1036 * freeing until the last reference to the descriptor table goes 1037 * away (usually process exit). This allows us to do lockless 1038 * lookups in fd_getfile(). 1039 */ 1040 if (oldnfiles > NDFILE) { 1041 if (fdp->fd_refcnt > 1) { 1042 newdt->dt_link = dt; 1043 } else { 1044 fd_dtab_free(dt); 1045 } 1046 } 1047 1048 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1049 i = NDHISLOTS(oldnfiles) * sizeof(uint32_t); 1050 memcpy(newhimap, fdp->fd_himap, i); 1051 memset((uint8_t *)newhimap + i, 0, 1052 NDHISLOTS(numfiles) * sizeof(uint32_t) - i); 1053 1054 i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t); 1055 memcpy(newlomap, fdp->fd_lomap, i); 1056 memset((uint8_t *)newlomap + i, 0, 1057 NDLOSLOTS(numfiles) * sizeof(uint32_t) - i); 1058 1059 if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) { 1060 fd_map_free(oldnfiles, fdp->fd_lomap, fdp->fd_himap); 1061 } 1062 fdp->fd_himap = newhimap; 1063 fdp->fd_lomap = newlomap; 1064 } 1065 1066 /* 1067 * All other modifications must become globally visible before 1068 * the change to fd_dt. See fd_getfile(). 1069 */ 1070 membar_producer(); 1071 fdp->fd_dt = newdt; 1072 KASSERT(newdt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 1073 fd_checkmaps(fdp); 1074 mutex_exit(&fdp->fd_lock); 1075 } 1076 1077 /* 1078 * Create a new open file structure and allocate a file descriptor 1079 * for the current process. 1080 */ 1081 int 1082 fd_allocfile(file_t **resultfp, int *resultfd) 1083 { 1084 proc_t *p = curproc; 1085 kauth_cred_t cred; 1086 file_t *fp; 1087 int error; 1088 1089 while ((error = fd_alloc(p, 0, resultfd)) != 0) { 1090 if (error != ENOSPC) { 1091 return error; 1092 } 1093 fd_tryexpand(p); 1094 } 1095 1096 fp = pool_cache_get(file_cache, PR_WAITOK); 1097 if (fp == NULL) { 1098 fd_abort(p, NULL, *resultfd); 1099 return ENFILE; 1100 } 1101 KASSERT(fp->f_count == 0); 1102 KASSERT(fp->f_msgcount == 0); 1103 KASSERT(fp->f_unpcount == 0); 1104 1105 /* Replace cached credentials if not what we need. */ 1106 cred = curlwp->l_cred; 1107 if (__predict_false(cred != fp->f_cred)) { 1108 kauth_cred_free(fp->f_cred); 1109 kauth_cred_hold(cred); 1110 fp->f_cred = cred; 1111 } 1112 1113 /* 1114 * Don't allow recycled files to be scanned. 1115 * See uipc_usrreq.c. 1116 */ 1117 if (__predict_false((fp->f_flag & FSCAN) != 0)) { 1118 mutex_enter(&fp->f_lock); 1119 atomic_and_uint(&fp->f_flag, ~FSCAN); 1120 mutex_exit(&fp->f_lock); 1121 } 1122 1123 fp->f_advice = 0; 1124 fp->f_offset = 0; 1125 *resultfp = fp; 1126 1127 return 0; 1128 } 1129 1130 /* 1131 * Successful creation of a new descriptor: make visible to the process. 1132 */ 1133 void 1134 fd_affix(proc_t *p, file_t *fp, unsigned fd) 1135 { 1136 fdfile_t *ff; 1137 filedesc_t *fdp; 1138 1139 KASSERT(p == curproc || p == &proc0); 1140 1141 /* Add a reference to the file structure. */ 1142 mutex_enter(&fp->f_lock); 1143 fp->f_count++; 1144 mutex_exit(&fp->f_lock); 1145 1146 /* 1147 * Insert the new file into the descriptor slot. 1148 * 1149 * The memory barriers provided by lock activity in this routine 1150 * ensure that any updates to the file structure become globally 1151 * visible before the file becomes visible to other LWPs in the 1152 * current process. 1153 */ 1154 fdp = p->p_fd; 1155 ff = fdp->fd_dt->dt_ff[fd]; 1156 1157 KASSERT(ff != NULL); 1158 KASSERT(ff->ff_file == NULL); 1159 KASSERT(ff->ff_allocated); 1160 KASSERT(fd_isused(fdp, fd)); 1161 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1162 1163 /* No need to lock in order to make file initially visible. */ 1164 ff->ff_file = fp; 1165 } 1166 1167 /* 1168 * Abort creation of a new descriptor: free descriptor slot and file. 1169 */ 1170 void 1171 fd_abort(proc_t *p, file_t *fp, unsigned fd) 1172 { 1173 filedesc_t *fdp; 1174 fdfile_t *ff; 1175 1176 KASSERT(p == curproc || p == &proc0); 1177 1178 fdp = p->p_fd; 1179 ff = fdp->fd_dt->dt_ff[fd]; 1180 ff->ff_exclose = false; 1181 1182 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1183 1184 mutex_enter(&fdp->fd_lock); 1185 KASSERT(fd_isused(fdp, fd)); 1186 fd_unused(fdp, fd); 1187 mutex_exit(&fdp->fd_lock); 1188 1189 if (fp != NULL) { 1190 KASSERT(fp->f_count == 0); 1191 KASSERT(fp->f_cred != NULL); 1192 pool_cache_put(file_cache, fp); 1193 } 1194 } 1195 1196 static int 1197 file_ctor(void *arg, void *obj, int flags) 1198 { 1199 file_t *fp = obj; 1200 1201 memset(fp, 0, sizeof(*fp)); 1202 1203 mutex_enter(&filelist_lock); 1204 if (__predict_false(nfiles >= maxfiles)) { 1205 mutex_exit(&filelist_lock); 1206 tablefull("file", "increase kern.maxfiles or MAXFILES"); 1207 return ENFILE; 1208 } 1209 nfiles++; 1210 LIST_INSERT_HEAD(&filehead, fp, f_list); 1211 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE); 1212 fp->f_cred = curlwp->l_cred; 1213 kauth_cred_hold(fp->f_cred); 1214 mutex_exit(&filelist_lock); 1215 1216 return 0; 1217 } 1218 1219 static void 1220 file_dtor(void *arg, void *obj) 1221 { 1222 file_t *fp = obj; 1223 1224 mutex_enter(&filelist_lock); 1225 nfiles--; 1226 LIST_REMOVE(fp, f_list); 1227 mutex_exit(&filelist_lock); 1228 1229 kauth_cred_free(fp->f_cred); 1230 mutex_destroy(&fp->f_lock); 1231 } 1232 1233 static int 1234 fdfile_ctor(void *arg, void *obj, int flags) 1235 { 1236 fdfile_t *ff = obj; 1237 1238 memset(ff, 0, sizeof(*ff)); 1239 cv_init(&ff->ff_closing, "fdclose"); 1240 1241 return 0; 1242 } 1243 1244 static void 1245 fdfile_dtor(void *arg, void *obj) 1246 { 1247 fdfile_t *ff = obj; 1248 1249 cv_destroy(&ff->ff_closing); 1250 } 1251 1252 file_t * 1253 fgetdummy(void) 1254 { 1255 file_t *fp; 1256 1257 fp = kmem_zalloc(sizeof(*fp), KM_SLEEP); 1258 if (fp != NULL) { 1259 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE); 1260 } 1261 return fp; 1262 } 1263 1264 void 1265 fputdummy(file_t *fp) 1266 { 1267 1268 mutex_destroy(&fp->f_lock); 1269 kmem_free(fp, sizeof(*fp)); 1270 } 1271 1272 /* 1273 * Create an initial filedesc structure. 1274 */ 1275 filedesc_t * 1276 fd_init(filedesc_t *fdp) 1277 { 1278 #ifdef DIAGNOSTIC 1279 unsigned fd; 1280 #endif 1281 1282 if (__predict_true(fdp == NULL)) { 1283 fdp = pool_cache_get(filedesc_cache, PR_WAITOK); 1284 } else { 1285 KASSERT(fdp == &filedesc0); 1286 filedesc_ctor(NULL, fdp, PR_WAITOK); 1287 } 1288 1289 #ifdef DIAGNOSTIC 1290 KASSERT(fdp->fd_lastfile == -1); 1291 KASSERT(fdp->fd_lastkqfile == -1); 1292 KASSERT(fdp->fd_knhash == NULL); 1293 KASSERT(fdp->fd_freefile == 0); 1294 KASSERT(fdp->fd_exclose == false); 1295 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin); 1296 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1297 for (fd = 0; fd < NDFDFILE; fd++) { 1298 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == 1299 (fdfile_t *)fdp->fd_dfdfile[fd]); 1300 } 1301 for (fd = NDFDFILE; fd < NDFILE; fd++) { 1302 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == NULL); 1303 } 1304 KASSERT(fdp->fd_himap == fdp->fd_dhimap); 1305 KASSERT(fdp->fd_lomap == fdp->fd_dlomap); 1306 #endif /* DIAGNOSTIC */ 1307 1308 fdp->fd_refcnt = 1; 1309 fd_checkmaps(fdp); 1310 1311 return fdp; 1312 } 1313 1314 /* 1315 * Initialize a file descriptor table. 1316 */ 1317 static int 1318 filedesc_ctor(void *arg, void *obj, int flag) 1319 { 1320 filedesc_t *fdp = obj; 1321 fdfile_t **ffp; 1322 int i; 1323 1324 memset(fdp, 0, sizeof(*fdp)); 1325 mutex_init(&fdp->fd_lock, MUTEX_DEFAULT, IPL_NONE); 1326 fdp->fd_lastfile = -1; 1327 fdp->fd_lastkqfile = -1; 1328 fdp->fd_dt = &fdp->fd_dtbuiltin; 1329 fdp->fd_dtbuiltin.dt_nfiles = NDFILE; 1330 fdp->fd_himap = fdp->fd_dhimap; 1331 fdp->fd_lomap = fdp->fd_dlomap; 1332 1333 CTASSERT(sizeof(fdp->fd_dfdfile[0]) >= sizeof(fdfile_t)); 1334 for (i = 0, ffp = fdp->fd_dt->dt_ff; i < NDFDFILE; i++, ffp++) { 1335 *ffp = (fdfile_t *)fdp->fd_dfdfile[i]; 1336 (void)fdfile_ctor(NULL, fdp->fd_dfdfile[i], PR_WAITOK); 1337 } 1338 1339 return 0; 1340 } 1341 1342 static void 1343 filedesc_dtor(void *arg, void *obj) 1344 { 1345 filedesc_t *fdp = obj; 1346 int i; 1347 1348 for (i = 0; i < NDFDFILE; i++) { 1349 fdfile_dtor(NULL, fdp->fd_dfdfile[i]); 1350 } 1351 1352 mutex_destroy(&fdp->fd_lock); 1353 } 1354 1355 /* 1356 * Make p share curproc's filedesc structure. 1357 */ 1358 void 1359 fd_share(struct proc *p) 1360 { 1361 filedesc_t *fdp; 1362 1363 fdp = curlwp->l_fd; 1364 p->p_fd = fdp; 1365 atomic_inc_uint(&fdp->fd_refcnt); 1366 } 1367 1368 /* 1369 * Acquire a hold on a filedesc structure. 1370 */ 1371 void 1372 fd_hold(lwp_t *l) 1373 { 1374 filedesc_t *fdp = l->l_fd; 1375 1376 atomic_inc_uint(&fdp->fd_refcnt); 1377 } 1378 1379 /* 1380 * Copy a filedesc structure. 1381 */ 1382 filedesc_t * 1383 fd_copy(void) 1384 { 1385 filedesc_t *newfdp, *fdp; 1386 fdfile_t *ff, **ffp, **nffp, *ff2; 1387 int i, j, numfiles, lastfile, newlast; 1388 file_t *fp; 1389 fdtab_t *newdt; 1390 1391 fdp = curproc->p_fd; 1392 newfdp = pool_cache_get(filedesc_cache, PR_WAITOK); 1393 newfdp->fd_refcnt = 1; 1394 1395 #ifdef DIAGNOSTIC 1396 KASSERT(newfdp->fd_lastfile == -1); 1397 KASSERT(newfdp->fd_lastkqfile == -1); 1398 KASSERT(newfdp->fd_knhash == NULL); 1399 KASSERT(newfdp->fd_freefile == 0); 1400 KASSERT(newfdp->fd_exclose == false); 1401 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin); 1402 KASSERT(newfdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1403 for (i = 0; i < NDFDFILE; i++) { 1404 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == 1405 (fdfile_t *)&newfdp->fd_dfdfile[i]); 1406 } 1407 for (i = NDFDFILE; i < NDFILE; i++) { 1408 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == NULL); 1409 } 1410 #endif /* DIAGNOSTIC */ 1411 1412 mutex_enter(&fdp->fd_lock); 1413 fd_checkmaps(fdp); 1414 numfiles = fdp->fd_dt->dt_nfiles; 1415 lastfile = fdp->fd_lastfile; 1416 1417 /* 1418 * If the number of open files fits in the internal arrays 1419 * of the open file structure, use them, otherwise allocate 1420 * additional memory for the number of descriptors currently 1421 * in use. 1422 */ 1423 if (lastfile < NDFILE) { 1424 i = NDFILE; 1425 newdt = newfdp->fd_dt; 1426 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin); 1427 } else { 1428 /* 1429 * Compute the smallest multiple of NDEXTENT needed 1430 * for the file descriptors currently in use, 1431 * allowing the table to shrink. 1432 */ 1433 i = numfiles; 1434 while (i >= 2 * NDEXTENT && i > lastfile * 2) { 1435 i /= 2; 1436 } 1437 KASSERT(i > NDFILE); 1438 newdt = fd_dtab_alloc(i); 1439 newfdp->fd_dt = newdt; 1440 memcpy(newdt->dt_ff, newfdp->fd_dtbuiltin.dt_ff, 1441 NDFDFILE * sizeof(fdfile_t **)); 1442 memset(newdt->dt_ff + NDFDFILE, 0, 1443 (i - NDFDFILE) * sizeof(fdfile_t **)); 1444 } 1445 if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) { 1446 newfdp->fd_himap = newfdp->fd_dhimap; 1447 newfdp->fd_lomap = newfdp->fd_dlomap; 1448 } else { 1449 fd_map_alloc(i, &newfdp->fd_lomap, &newfdp->fd_himap); 1450 KASSERT(i >= NDENTRIES * NDENTRIES); 1451 memset(newfdp->fd_himap, 0, NDHISLOTS(i)*sizeof(uint32_t)); 1452 memset(newfdp->fd_lomap, 0, NDLOSLOTS(i)*sizeof(uint32_t)); 1453 } 1454 newfdp->fd_freefile = fdp->fd_freefile; 1455 newfdp->fd_exclose = fdp->fd_exclose; 1456 1457 ffp = fdp->fd_dt->dt_ff; 1458 nffp = newdt->dt_ff; 1459 newlast = -1; 1460 for (i = 0; i <= (int)lastfile; i++, ffp++, nffp++) { 1461 KASSERT(i >= NDFDFILE || 1462 *nffp == (fdfile_t *)newfdp->fd_dfdfile[i]); 1463 ff = *ffp; 1464 if (ff == NULL || (fp = ff->ff_file) == NULL) { 1465 /* Descriptor unused, or descriptor half open. */ 1466 KASSERT(!fd_isused(newfdp, i)); 1467 continue; 1468 } 1469 if (__predict_false(fp->f_type == DTYPE_KQUEUE)) { 1470 /* kqueue descriptors cannot be copied. */ 1471 if (i < newfdp->fd_freefile) { 1472 newfdp->fd_freefile = i; 1473 } 1474 continue; 1475 } 1476 /* It's active: add a reference to the file. */ 1477 mutex_enter(&fp->f_lock); 1478 fp->f_count++; 1479 mutex_exit(&fp->f_lock); 1480 1481 /* Allocate an fdfile_t to represent it. */ 1482 if (i >= NDFDFILE) { 1483 ff2 = pool_cache_get(fdfile_cache, PR_WAITOK); 1484 *nffp = ff2; 1485 } else { 1486 ff2 = newdt->dt_ff[i]; 1487 } 1488 ff2->ff_file = fp; 1489 ff2->ff_exclose = ff->ff_exclose; 1490 ff2->ff_allocated = true; 1491 1492 /* Fix up bitmaps. */ 1493 j = i >> NDENTRYSHIFT; 1494 KASSERT((newfdp->fd_lomap[j] & (1 << (i & NDENTRYMASK))) == 0); 1495 newfdp->fd_lomap[j] |= 1 << (i & NDENTRYMASK); 1496 if (__predict_false(newfdp->fd_lomap[j] == ~0)) { 1497 KASSERT((newfdp->fd_himap[j >> NDENTRYSHIFT] & 1498 (1 << (j & NDENTRYMASK))) == 0); 1499 newfdp->fd_himap[j >> NDENTRYSHIFT] |= 1500 1 << (j & NDENTRYMASK); 1501 } 1502 newlast = i; 1503 } 1504 KASSERT(newdt->dt_ff[0] == (fdfile_t *)newfdp->fd_dfdfile[0]); 1505 newfdp->fd_lastfile = newlast; 1506 fd_checkmaps(newfdp); 1507 mutex_exit(&fdp->fd_lock); 1508 1509 return newfdp; 1510 } 1511 1512 /* 1513 * Release a filedesc structure. 1514 */ 1515 void 1516 fd_free(void) 1517 { 1518 fdfile_t *ff; 1519 file_t *fp; 1520 int fd, nf; 1521 fdtab_t *dt; 1522 lwp_t * const l = curlwp; 1523 filedesc_t * const fdp = l->l_fd; 1524 const bool noadvlock = (l->l_proc->p_flag & PK_ADVLOCK) == 0; 1525 1526 KASSERT(fdp->fd_dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 1527 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1528 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL); 1529 1530 #ifndef __HAVE_ATOMIC_AS_MEMBAR 1531 membar_exit(); 1532 #endif 1533 if (atomic_dec_uint_nv(&fdp->fd_refcnt) > 0) 1534 return; 1535 1536 /* 1537 * Close any files that the process holds open. 1538 */ 1539 dt = fdp->fd_dt; 1540 fd_checkmaps(fdp); 1541 #ifdef DEBUG 1542 fdp->fd_refcnt = -1; /* see fd_checkmaps */ 1543 #endif 1544 for (fd = 0, nf = dt->dt_nfiles; fd < nf; fd++) { 1545 ff = dt->dt_ff[fd]; 1546 KASSERT(fd >= NDFDFILE || 1547 ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1548 if (ff == NULL) 1549 continue; 1550 if ((fp = ff->ff_file) != NULL) { 1551 /* 1552 * Must use fd_close() here if there is 1553 * a reference from kqueue or we might have posix 1554 * advisory locks. 1555 */ 1556 if (__predict_true(ff->ff_refcnt == 0) && 1557 (noadvlock || fp->f_type != DTYPE_VNODE)) { 1558 ff->ff_file = NULL; 1559 ff->ff_exclose = false; 1560 ff->ff_allocated = false; 1561 closef(fp); 1562 } else { 1563 ff->ff_refcnt++; 1564 fd_close(fd); 1565 } 1566 } 1567 KASSERT(ff->ff_refcnt == 0); 1568 KASSERT(ff->ff_file == NULL); 1569 KASSERT(!ff->ff_exclose); 1570 KASSERT(!ff->ff_allocated); 1571 if (fd >= NDFDFILE) { 1572 pool_cache_put(fdfile_cache, ff); 1573 dt->dt_ff[fd] = NULL; 1574 } 1575 } 1576 1577 /* 1578 * Clean out the descriptor table for the next user and return 1579 * to the cache. 1580 */ 1581 if (__predict_false(dt != &fdp->fd_dtbuiltin)) { 1582 fd_dtab_free(fdp->fd_dt); 1583 /* Otherwise, done above. */ 1584 memset(&fdp->fd_dtbuiltin.dt_ff[NDFDFILE], 0, 1585 (NDFILE - NDFDFILE) * sizeof(fdp->fd_dtbuiltin.dt_ff[0])); 1586 fdp->fd_dt = &fdp->fd_dtbuiltin; 1587 } 1588 if (__predict_false(NDHISLOTS(nf) > NDHISLOTS(NDFILE))) { 1589 KASSERT(fdp->fd_himap != fdp->fd_dhimap); 1590 KASSERT(fdp->fd_lomap != fdp->fd_dlomap); 1591 fd_map_free(nf, fdp->fd_lomap, fdp->fd_himap); 1592 } 1593 if (__predict_false(fdp->fd_knhash != NULL)) { 1594 hashdone(fdp->fd_knhash, HASH_LIST, fdp->fd_knhashmask); 1595 fdp->fd_knhash = NULL; 1596 fdp->fd_knhashmask = 0; 1597 } else { 1598 KASSERT(fdp->fd_knhashmask == 0); 1599 } 1600 fdp->fd_dt = &fdp->fd_dtbuiltin; 1601 fdp->fd_lastkqfile = -1; 1602 fdp->fd_lastfile = -1; 1603 fdp->fd_freefile = 0; 1604 fdp->fd_exclose = false; 1605 memset(&fdp->fd_startzero, 0, sizeof(*fdp) - 1606 offsetof(filedesc_t, fd_startzero)); 1607 fdp->fd_himap = fdp->fd_dhimap; 1608 fdp->fd_lomap = fdp->fd_dlomap; 1609 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1610 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL); 1611 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin); 1612 #ifdef DEBUG 1613 fdp->fd_refcnt = 0; /* see fd_checkmaps */ 1614 #endif 1615 fd_checkmaps(fdp); 1616 pool_cache_put(filedesc_cache, fdp); 1617 } 1618 1619 /* 1620 * File Descriptor pseudo-device driver (/dev/fd/). 1621 * 1622 * Opening minor device N dup()s the file (if any) connected to file 1623 * descriptor N belonging to the calling process. Note that this driver 1624 * consists of only the ``open()'' routine, because all subsequent 1625 * references to this file will be direct to the other driver. 1626 */ 1627 static int 1628 filedescopen(dev_t dev, int mode, int type, lwp_t *l) 1629 { 1630 1631 /* 1632 * XXX Kludge: set dupfd to contain the value of the 1633 * the file descriptor being sought for duplication. The error 1634 * return ensures that the vnode for this device will be released 1635 * by vn_open. Open will detect this special error and take the 1636 * actions in fd_dupopen below. Other callers of vn_open or VOP_OPEN 1637 * will simply report the error. 1638 */ 1639 l->l_dupfd = minor(dev); /* XXX */ 1640 return EDUPFD; 1641 } 1642 1643 /* 1644 * Duplicate the specified descriptor to a free descriptor. 1645 */ 1646 int 1647 fd_dupopen(int old, int *newp, int mode, int error) 1648 { 1649 filedesc_t *fdp; 1650 fdfile_t *ff; 1651 file_t *fp; 1652 fdtab_t *dt; 1653 1654 if ((fp = fd_getfile(old)) == NULL) { 1655 return EBADF; 1656 } 1657 fdp = curlwp->l_fd; 1658 dt = fdp->fd_dt; 1659 ff = dt->dt_ff[old]; 1660 1661 /* 1662 * There are two cases of interest here. 1663 * 1664 * For EDUPFD simply dup (old) to file descriptor 1665 * (new) and return. 1666 * 1667 * For EMOVEFD steal away the file structure from (old) and 1668 * store it in (new). (old) is effectively closed by 1669 * this operation. 1670 * 1671 * Any other error code is just returned. 1672 */ 1673 switch (error) { 1674 case EDUPFD: 1675 /* 1676 * Check that the mode the file is being opened for is a 1677 * subset of the mode of the existing descriptor. 1678 */ 1679 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 1680 error = EACCES; 1681 break; 1682 } 1683 1684 /* Copy it. */ 1685 error = fd_dup(fp, 0, newp, ff->ff_exclose); 1686 break; 1687 1688 case EMOVEFD: 1689 /* Copy it. */ 1690 error = fd_dup(fp, 0, newp, ff->ff_exclose); 1691 if (error != 0) { 1692 break; 1693 } 1694 1695 /* Steal away the file pointer from 'old'. */ 1696 (void)fd_close(old); 1697 return 0; 1698 } 1699 1700 fd_putfile(old); 1701 return error; 1702 } 1703 1704 /* 1705 * Close open files on exec. 1706 */ 1707 void 1708 fd_closeexec(void) 1709 { 1710 proc_t *p; 1711 filedesc_t *fdp; 1712 fdfile_t *ff; 1713 lwp_t *l; 1714 fdtab_t *dt; 1715 int fd; 1716 1717 l = curlwp; 1718 p = l->l_proc; 1719 fdp = p->p_fd; 1720 1721 if (fdp->fd_refcnt > 1) { 1722 fdp = fd_copy(); 1723 fd_free(); 1724 p->p_fd = fdp; 1725 l->l_fd = fdp; 1726 } 1727 if (!fdp->fd_exclose) { 1728 return; 1729 } 1730 fdp->fd_exclose = false; 1731 dt = fdp->fd_dt; 1732 1733 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 1734 if ((ff = dt->dt_ff[fd]) == NULL) { 1735 KASSERT(fd >= NDFDFILE); 1736 continue; 1737 } 1738 KASSERT(fd >= NDFDFILE || 1739 ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1740 if (ff->ff_file == NULL) 1741 continue; 1742 if (ff->ff_exclose) { 1743 /* 1744 * We need a reference to close the file. 1745 * No other threads can see the fdfile_t at 1746 * this point, so don't bother locking. 1747 */ 1748 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0); 1749 ff->ff_refcnt++; 1750 fd_close(fd); 1751 } 1752 } 1753 } 1754 1755 /* 1756 * Sets descriptor owner. If the owner is a process, 'pgid' 1757 * is set to positive value, process ID. If the owner is process group, 1758 * 'pgid' is set to -pg_id. 1759 */ 1760 int 1761 fsetown(pid_t *pgid, u_long cmd, const void *data) 1762 { 1763 pid_t id = *(const pid_t *)data; 1764 int error; 1765 1766 switch (cmd) { 1767 case TIOCSPGRP: 1768 if (id < 0) 1769 return EINVAL; 1770 id = -id; 1771 break; 1772 default: 1773 break; 1774 } 1775 if (id > 0) { 1776 mutex_enter(proc_lock); 1777 error = proc_find(id) ? 0 : ESRCH; 1778 mutex_exit(proc_lock); 1779 } else if (id < 0) { 1780 error = pgid_in_session(curproc, -id); 1781 } else { 1782 error = 0; 1783 } 1784 if (!error) { 1785 *pgid = id; 1786 } 1787 return error; 1788 } 1789 1790 void 1791 fd_set_exclose(struct lwp *l, int fd, bool exclose) 1792 { 1793 filedesc_t *fdp = l->l_fd; 1794 fdfile_t *ff = fdp->fd_dt->dt_ff[fd]; 1795 1796 ff->ff_exclose = exclose; 1797 if (exclose) 1798 fdp->fd_exclose = true; 1799 } 1800 1801 /* 1802 * Return descriptor owner information. If the value is positive, 1803 * it's process ID. If it's negative, it's process group ID and 1804 * needs the sign removed before use. 1805 */ 1806 int 1807 fgetown(pid_t pgid, u_long cmd, void *data) 1808 { 1809 1810 switch (cmd) { 1811 case TIOCGPGRP: 1812 *(int *)data = -pgid; 1813 break; 1814 default: 1815 *(int *)data = pgid; 1816 break; 1817 } 1818 return 0; 1819 } 1820 1821 /* 1822 * Send signal to descriptor owner, either process or process group. 1823 */ 1824 void 1825 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata) 1826 { 1827 ksiginfo_t ksi; 1828 1829 KASSERT(!cpu_intr_p()); 1830 1831 if (pgid == 0) { 1832 return; 1833 } 1834 1835 KSI_INIT(&ksi); 1836 ksi.ksi_signo = signo; 1837 ksi.ksi_code = code; 1838 ksi.ksi_band = band; 1839 1840 mutex_enter(proc_lock); 1841 if (pgid > 0) { 1842 struct proc *p1; 1843 1844 p1 = proc_find(pgid); 1845 if (p1 != NULL) { 1846 kpsignal(p1, &ksi, fdescdata); 1847 } 1848 } else { 1849 struct pgrp *pgrp; 1850 1851 KASSERT(pgid < 0); 1852 pgrp = pgrp_find(-pgid); 1853 if (pgrp != NULL) { 1854 kpgsignal(pgrp, &ksi, fdescdata, 0); 1855 } 1856 } 1857 mutex_exit(proc_lock); 1858 } 1859 1860 int 1861 fd_clone(file_t *fp, unsigned fd, int flag, const struct fileops *fops, 1862 void *data) 1863 { 1864 1865 fp->f_flag = flag; 1866 fp->f_type = DTYPE_MISC; 1867 fp->f_ops = fops; 1868 fp->f_data = data; 1869 curlwp->l_dupfd = fd; 1870 fd_affix(curproc, fp, fd); 1871 1872 return EMOVEFD; 1873 } 1874 1875 int 1876 fnullop_fcntl(file_t *fp, u_int cmd, void *data) 1877 { 1878 1879 if (cmd == F_SETFL) 1880 return 0; 1881 1882 return EOPNOTSUPP; 1883 } 1884 1885 int 1886 fnullop_poll(file_t *fp, int which) 1887 { 1888 1889 return 0; 1890 } 1891 1892 int 1893 fnullop_kqfilter(file_t *fp, struct knote *kn) 1894 { 1895 1896 return EOPNOTSUPP; 1897 } 1898 1899 void 1900 fnullop_restart(file_t *fp) 1901 { 1902 1903 } 1904 1905 int 1906 fbadop_read(file_t *fp, off_t *offset, struct uio *uio, 1907 kauth_cred_t cred, int flags) 1908 { 1909 1910 return EOPNOTSUPP; 1911 } 1912 1913 int 1914 fbadop_write(file_t *fp, off_t *offset, struct uio *uio, 1915 kauth_cred_t cred, int flags) 1916 { 1917 1918 return EOPNOTSUPP; 1919 } 1920 1921 int 1922 fbadop_ioctl(file_t *fp, u_long com, void *data) 1923 { 1924 1925 return EOPNOTSUPP; 1926 } 1927 1928 int 1929 fbadop_stat(file_t *fp, struct stat *sb) 1930 { 1931 1932 return EOPNOTSUPP; 1933 } 1934 1935 int 1936 fbadop_close(file_t *fp) 1937 { 1938 1939 return EOPNOTSUPP; 1940 } 1941 1942 /* 1943 * sysctl routines pertaining to file descriptors 1944 */ 1945 1946 /* Initialized in sysctl_init() for now... */ 1947 extern kmutex_t sysctl_file_marker_lock; 1948 static u_int sysctl_file_marker = 1; 1949 1950 /* 1951 * Expects to be called with proc_lock and sysctl_file_marker_lock locked. 1952 */ 1953 static void 1954 sysctl_file_marker_reset(void) 1955 { 1956 struct proc *p; 1957 1958 PROCLIST_FOREACH(p, &allproc) { 1959 struct filedesc *fd = p->p_fd; 1960 fdtab_t *dt; 1961 u_int i; 1962 1963 mutex_enter(&fd->fd_lock); 1964 dt = fd->fd_dt; 1965 for (i = 0; i < dt->dt_nfiles; i++) { 1966 struct file *fp; 1967 fdfile_t *ff; 1968 1969 if ((ff = dt->dt_ff[i]) == NULL) { 1970 continue; 1971 } 1972 if ((fp = ff->ff_file) == NULL) { 1973 continue; 1974 } 1975 fp->f_marker = 0; 1976 } 1977 mutex_exit(&fd->fd_lock); 1978 } 1979 } 1980 1981 /* 1982 * sysctl helper routine for kern.file pseudo-subtree. 1983 */ 1984 static int 1985 sysctl_kern_file(SYSCTLFN_ARGS) 1986 { 1987 int error; 1988 size_t buflen; 1989 struct file *fp, fbuf; 1990 char *start, *where; 1991 struct proc *p; 1992 1993 start = where = oldp; 1994 buflen = *oldlenp; 1995 1996 if (where == NULL) { 1997 /* 1998 * overestimate by 10 files 1999 */ 2000 *oldlenp = sizeof(filehead) + (nfiles + 10) * 2001 sizeof(struct file); 2002 return 0; 2003 } 2004 2005 /* 2006 * first sysctl_copyout filehead 2007 */ 2008 if (buflen < sizeof(filehead)) { 2009 *oldlenp = 0; 2010 return 0; 2011 } 2012 sysctl_unlock(); 2013 error = sysctl_copyout(l, &filehead, where, sizeof(filehead)); 2014 if (error) { 2015 sysctl_relock(); 2016 return error; 2017 } 2018 buflen -= sizeof(filehead); 2019 where += sizeof(filehead); 2020 2021 /* 2022 * followed by an array of file structures 2023 */ 2024 mutex_enter(&sysctl_file_marker_lock); 2025 mutex_enter(proc_lock); 2026 PROCLIST_FOREACH(p, &allproc) { 2027 struct filedesc *fd; 2028 fdtab_t *dt; 2029 u_int i; 2030 2031 if (p->p_stat == SIDL) { 2032 /* skip embryonic processes */ 2033 continue; 2034 } 2035 mutex_enter(p->p_lock); 2036 error = kauth_authorize_process(l->l_cred, 2037 KAUTH_PROCESS_CANSEE, p, 2038 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), 2039 NULL, NULL); 2040 mutex_exit(p->p_lock); 2041 if (error != 0) { 2042 /* 2043 * Don't leak kauth retval if we're silently 2044 * skipping this entry. 2045 */ 2046 error = 0; 2047 continue; 2048 } 2049 2050 /* 2051 * Grab a hold on the process. 2052 */ 2053 if (!rw_tryenter(&p->p_reflock, RW_READER)) { 2054 continue; 2055 } 2056 mutex_exit(proc_lock); 2057 2058 fd = p->p_fd; 2059 mutex_enter(&fd->fd_lock); 2060 dt = fd->fd_dt; 2061 for (i = 0; i < dt->dt_nfiles; i++) { 2062 fdfile_t *ff; 2063 2064 if ((ff = dt->dt_ff[i]) == NULL) { 2065 continue; 2066 } 2067 if ((fp = ff->ff_file) == NULL) { 2068 continue; 2069 } 2070 2071 mutex_enter(&fp->f_lock); 2072 2073 if ((fp->f_count == 0) || 2074 (fp->f_marker == sysctl_file_marker)) { 2075 mutex_exit(&fp->f_lock); 2076 continue; 2077 } 2078 2079 /* Check that we have enough space. */ 2080 if (buflen < sizeof(struct file)) { 2081 *oldlenp = where - start; 2082 mutex_exit(&fp->f_lock); 2083 error = ENOMEM; 2084 break; 2085 } 2086 2087 memcpy(&fbuf, fp, sizeof(fbuf)); 2088 mutex_exit(&fp->f_lock); 2089 error = sysctl_copyout(l, &fbuf, where, sizeof(fbuf)); 2090 if (error) { 2091 break; 2092 } 2093 buflen -= sizeof(struct file); 2094 where += sizeof(struct file); 2095 2096 fp->f_marker = sysctl_file_marker; 2097 } 2098 mutex_exit(&fd->fd_lock); 2099 2100 /* 2101 * Release reference to process. 2102 */ 2103 mutex_enter(proc_lock); 2104 rw_exit(&p->p_reflock); 2105 2106 if (error) 2107 break; 2108 } 2109 2110 sysctl_file_marker++; 2111 /* Reset all markers if wrapped. */ 2112 if (sysctl_file_marker == 0) { 2113 sysctl_file_marker_reset(); 2114 sysctl_file_marker++; 2115 } 2116 2117 mutex_exit(proc_lock); 2118 mutex_exit(&sysctl_file_marker_lock); 2119 2120 *oldlenp = where - start; 2121 sysctl_relock(); 2122 return error; 2123 } 2124 2125 /* 2126 * sysctl helper function for kern.file2 2127 */ 2128 static int 2129 sysctl_kern_file2(SYSCTLFN_ARGS) 2130 { 2131 struct proc *p; 2132 struct file *fp; 2133 struct filedesc *fd; 2134 struct kinfo_file kf; 2135 char *dp; 2136 u_int i, op; 2137 size_t len, needed, elem_size, out_size; 2138 int error, arg, elem_count; 2139 fdfile_t *ff; 2140 fdtab_t *dt; 2141 2142 if (namelen == 1 && name[0] == CTL_QUERY) 2143 return sysctl_query(SYSCTLFN_CALL(rnode)); 2144 2145 if (namelen != 4) 2146 return EINVAL; 2147 2148 error = 0; 2149 dp = oldp; 2150 len = (oldp != NULL) ? *oldlenp : 0; 2151 op = name[0]; 2152 arg = name[1]; 2153 elem_size = name[2]; 2154 elem_count = name[3]; 2155 out_size = MIN(sizeof(kf), elem_size); 2156 needed = 0; 2157 2158 if (elem_size < 1 || elem_count < 0) 2159 return EINVAL; 2160 2161 switch (op) { 2162 case KERN_FILE_BYFILE: 2163 case KERN_FILE_BYPID: 2164 /* 2165 * We're traversing the process list in both cases; the BYFILE 2166 * case does additional work of keeping track of files already 2167 * looked at. 2168 */ 2169 2170 /* doesn't use arg so it must be zero */ 2171 if ((op == KERN_FILE_BYFILE) && (arg != 0)) 2172 return EINVAL; 2173 2174 if ((op == KERN_FILE_BYPID) && (arg < -1)) 2175 /* -1 means all processes */ 2176 return EINVAL; 2177 2178 sysctl_unlock(); 2179 if (op == KERN_FILE_BYFILE) 2180 mutex_enter(&sysctl_file_marker_lock); 2181 mutex_enter(proc_lock); 2182 PROCLIST_FOREACH(p, &allproc) { 2183 if (p->p_stat == SIDL) { 2184 /* skip embryonic processes */ 2185 continue; 2186 } 2187 if (arg > 0 && p->p_pid != arg) { 2188 /* pick only the one we want */ 2189 /* XXX want 0 to mean "kernel files" */ 2190 continue; 2191 } 2192 mutex_enter(p->p_lock); 2193 error = kauth_authorize_process(l->l_cred, 2194 KAUTH_PROCESS_CANSEE, p, 2195 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), 2196 NULL, NULL); 2197 mutex_exit(p->p_lock); 2198 if (error != 0) { 2199 /* 2200 * Don't leak kauth retval if we're silently 2201 * skipping this entry. 2202 */ 2203 error = 0; 2204 continue; 2205 } 2206 2207 /* 2208 * Grab a hold on the process. 2209 */ 2210 if (!rw_tryenter(&p->p_reflock, RW_READER)) { 2211 continue; 2212 } 2213 mutex_exit(proc_lock); 2214 2215 fd = p->p_fd; 2216 mutex_enter(&fd->fd_lock); 2217 dt = fd->fd_dt; 2218 for (i = 0; i < dt->dt_nfiles; i++) { 2219 if ((ff = dt->dt_ff[i]) == NULL) { 2220 continue; 2221 } 2222 if ((fp = ff->ff_file) == NULL) { 2223 continue; 2224 } 2225 2226 if ((op == KERN_FILE_BYFILE) && 2227 (fp->f_marker == sysctl_file_marker)) { 2228 continue; 2229 } 2230 if (len >= elem_size && elem_count > 0) { 2231 mutex_enter(&fp->f_lock); 2232 fill_file(&kf, fp, ff, i, p->p_pid); 2233 mutex_exit(&fp->f_lock); 2234 mutex_exit(&fd->fd_lock); 2235 error = sysctl_copyout(l, 2236 &kf, dp, out_size); 2237 mutex_enter(&fd->fd_lock); 2238 if (error) 2239 break; 2240 dp += elem_size; 2241 len -= elem_size; 2242 } 2243 if (op == KERN_FILE_BYFILE) 2244 fp->f_marker = sysctl_file_marker; 2245 needed += elem_size; 2246 if (elem_count > 0 && elem_count != INT_MAX) 2247 elem_count--; 2248 } 2249 mutex_exit(&fd->fd_lock); 2250 2251 /* 2252 * Release reference to process. 2253 */ 2254 mutex_enter(proc_lock); 2255 rw_exit(&p->p_reflock); 2256 } 2257 if (op == KERN_FILE_BYFILE) { 2258 sysctl_file_marker++; 2259 2260 /* Reset all markers if wrapped. */ 2261 if (sysctl_file_marker == 0) { 2262 sysctl_file_marker_reset(); 2263 sysctl_file_marker++; 2264 } 2265 } 2266 mutex_exit(proc_lock); 2267 if (op == KERN_FILE_BYFILE) 2268 mutex_exit(&sysctl_file_marker_lock); 2269 sysctl_relock(); 2270 break; 2271 default: 2272 return EINVAL; 2273 } 2274 2275 if (oldp == NULL) 2276 needed += KERN_FILESLOP * elem_size; 2277 *oldlenp = needed; 2278 2279 return error; 2280 } 2281 2282 static void 2283 fill_file(struct kinfo_file *kp, const file_t *fp, const fdfile_t *ff, 2284 int i, pid_t pid) 2285 { 2286 2287 memset(kp, 0, sizeof(*kp)); 2288 2289 kp->ki_fileaddr = PTRTOUINT64(fp); 2290 kp->ki_flag = fp->f_flag; 2291 kp->ki_iflags = 0; 2292 kp->ki_ftype = fp->f_type; 2293 kp->ki_count = fp->f_count; 2294 kp->ki_msgcount = fp->f_msgcount; 2295 kp->ki_fucred = PTRTOUINT64(fp->f_cred); 2296 kp->ki_fuid = kauth_cred_geteuid(fp->f_cred); 2297 kp->ki_fgid = kauth_cred_getegid(fp->f_cred); 2298 kp->ki_fops = PTRTOUINT64(fp->f_ops); 2299 kp->ki_foffset = fp->f_offset; 2300 kp->ki_fdata = PTRTOUINT64(fp->f_data); 2301 2302 /* vnode information to glue this file to something */ 2303 if (fp->f_type == DTYPE_VNODE) { 2304 struct vnode *vp = fp->f_vnode; 2305 2306 kp->ki_vun = PTRTOUINT64(vp->v_un.vu_socket); 2307 kp->ki_vsize = vp->v_size; 2308 kp->ki_vtype = vp->v_type; 2309 kp->ki_vtag = vp->v_tag; 2310 kp->ki_vdata = PTRTOUINT64(vp->v_data); 2311 } 2312 2313 /* process information when retrieved via KERN_FILE_BYPID */ 2314 if (ff != NULL) { 2315 kp->ki_pid = pid; 2316 kp->ki_fd = i; 2317 kp->ki_ofileflags = ff->ff_exclose; 2318 kp->ki_usecount = ff->ff_refcnt; 2319 } 2320 } 2321