1 /* $NetBSD: kern_descrip.c,v 1.265 2024/12/21 19:02:31 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 2008, 2009, 2023 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 1.265 2024/12/21 19:02:31 riastradh 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; 109 110 static int file_ctor(void *, void *, int); 111 static void file_dtor(void *, void *); 112 static void fdfile_ctor(fdfile_t *); 113 static void fdfile_dtor(fdfile_t *); 114 static int filedesc_ctor(void *, void *, int); 115 static void filedesc_dtor(void *, void *); 116 static int filedescopen(dev_t, int, int, lwp_t *); 117 118 static int sysctl_kern_file(SYSCTLFN_PROTO); 119 static int sysctl_kern_file2(SYSCTLFN_PROTO); 120 static void fill_file(struct file *, const struct file *); 121 static void fill_file2(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 LIST_INIT(&filehead); 154 155 file_cache = pool_cache_init(sizeof(file_t), coherency_unit, 0, 156 0, "file", NULL, IPL_NONE, file_ctor, file_dtor, NULL); 157 KASSERT(file_cache != NULL); 158 159 filedesc_cache = pool_cache_init(sizeof(filedesc_t), coherency_unit, 160 0, 0, "filedesc", NULL, IPL_NONE, filedesc_ctor, filedesc_dtor, 161 NULL); 162 KASSERT(filedesc_cache != NULL); 163 164 sysctl_createv(&clog, 0, NULL, NULL, 165 CTLFLAG_PERMANENT, 166 CTLTYPE_STRUCT, "file", 167 SYSCTL_DESCR("System open file table"), 168 sysctl_kern_file, 0, NULL, 0, 169 CTL_KERN, KERN_FILE, CTL_EOL); 170 sysctl_createv(&clog, 0, NULL, NULL, 171 CTLFLAG_PERMANENT, 172 CTLTYPE_STRUCT, "file2", 173 SYSCTL_DESCR("System open file table"), 174 sysctl_kern_file2, 0, NULL, 0, 175 CTL_KERN, KERN_FILE2, CTL_EOL); 176 } 177 178 static bool 179 fd_isused(filedesc_t *fdp, unsigned fd) 180 { 181 u_int off = fd >> NDENTRYSHIFT; 182 183 KASSERT(fd < atomic_load_consume(&fdp->fd_dt)->dt_nfiles); 184 185 return (fdp->fd_lomap[off] & (1U << (fd & NDENTRYMASK))) != 0; 186 } 187 188 /* 189 * Verify that the bitmaps match the descriptor table. 190 */ 191 static inline void 192 fd_checkmaps(filedesc_t *fdp) 193 { 194 #ifdef DEBUG 195 fdtab_t *dt; 196 u_int fd; 197 198 KASSERT(fdp->fd_refcnt <= 1 || mutex_owned(&fdp->fd_lock)); 199 200 dt = fdp->fd_dt; 201 if (fdp->fd_refcnt == -1) { 202 /* 203 * fd_free tears down the table without maintaining its bitmap. 204 */ 205 return; 206 } 207 for (fd = 0; fd < dt->dt_nfiles; fd++) { 208 if (fd < NDFDFILE) { 209 KASSERT(dt->dt_ff[fd] == 210 (fdfile_t *)fdp->fd_dfdfile[fd]); 211 } 212 if (dt->dt_ff[fd] == NULL) { 213 KASSERT(!fd_isused(fdp, fd)); 214 } else if (dt->dt_ff[fd]->ff_file != NULL) { 215 KASSERT(fd_isused(fdp, fd)); 216 } 217 } 218 #endif 219 } 220 221 static int 222 fd_next_zero(filedesc_t *fdp, uint32_t *bitmap, int want, u_int bits) 223 { 224 int i, off, maxoff; 225 uint32_t sub; 226 227 KASSERT(mutex_owned(&fdp->fd_lock)); 228 229 fd_checkmaps(fdp); 230 231 if (want > bits) 232 return -1; 233 234 off = want >> NDENTRYSHIFT; 235 i = want & NDENTRYMASK; 236 if (i) { 237 sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i)); 238 if (sub != ~0) 239 goto found; 240 off++; 241 } 242 243 maxoff = NDLOSLOTS(bits); 244 while (off < maxoff) { 245 if ((sub = bitmap[off]) != ~0) 246 goto found; 247 off++; 248 } 249 250 return -1; 251 252 found: 253 return (off << NDENTRYSHIFT) + ffs(~sub) - 1; 254 } 255 256 static int 257 fd_last_set(filedesc_t *fd, int last) 258 { 259 int off, i; 260 fdfile_t **ff = fd->fd_dt->dt_ff; 261 uint32_t *bitmap = fd->fd_lomap; 262 263 KASSERT(mutex_owned(&fd->fd_lock)); 264 265 fd_checkmaps(fd); 266 267 off = (last - 1) >> NDENTRYSHIFT; 268 269 while (off >= 0 && !bitmap[off]) 270 off--; 271 272 if (off < 0) 273 return -1; 274 275 i = ((off + 1) << NDENTRYSHIFT) - 1; 276 if (i >= last) 277 i = last - 1; 278 279 /* XXX should use bitmap */ 280 while (i > 0 && (ff[i] == NULL || !ff[i]->ff_allocated)) 281 i--; 282 283 return i; 284 } 285 286 static inline void 287 fd_used(filedesc_t *fdp, unsigned fd) 288 { 289 u_int off = fd >> NDENTRYSHIFT; 290 fdfile_t *ff; 291 292 ff = fdp->fd_dt->dt_ff[fd]; 293 294 KASSERT(mutex_owned(&fdp->fd_lock)); 295 KASSERT((fdp->fd_lomap[off] & (1U << (fd & NDENTRYMASK))) == 0); 296 KASSERT(ff != NULL); 297 KASSERT(ff->ff_file == NULL); 298 KASSERT(!ff->ff_allocated); 299 300 ff->ff_allocated = true; 301 fdp->fd_lomap[off] |= 1U << (fd & NDENTRYMASK); 302 if (__predict_false(fdp->fd_lomap[off] == ~0)) { 303 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] & 304 (1U << (off & NDENTRYMASK))) == 0); 305 fdp->fd_himap[off >> NDENTRYSHIFT] |= 1U << (off & NDENTRYMASK); 306 } 307 308 if ((int)fd > fdp->fd_lastfile) { 309 fdp->fd_lastfile = fd; 310 } 311 312 fd_checkmaps(fdp); 313 } 314 315 static inline void 316 fd_unused(filedesc_t *fdp, unsigned fd) 317 { 318 u_int off = fd >> NDENTRYSHIFT; 319 fdfile_t *ff; 320 321 ff = fdp->fd_dt->dt_ff[fd]; 322 323 KASSERT(mutex_owned(&fdp->fd_lock)); 324 KASSERT(ff != NULL); 325 KASSERT(ff->ff_file == NULL); 326 KASSERT(ff->ff_allocated); 327 328 if (fd < fdp->fd_freefile) { 329 fdp->fd_freefile = fd; 330 } 331 332 if (fdp->fd_lomap[off] == ~0) { 333 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] & 334 (1U << (off & NDENTRYMASK))) != 0); 335 fdp->fd_himap[off >> NDENTRYSHIFT] &= 336 ~(1U << (off & NDENTRYMASK)); 337 } 338 KASSERT((fdp->fd_lomap[off] & (1U << (fd & NDENTRYMASK))) != 0); 339 fdp->fd_lomap[off] &= ~(1U << (fd & NDENTRYMASK)); 340 ff->ff_allocated = false; 341 342 KASSERT(fd <= fdp->fd_lastfile); 343 if (fd == fdp->fd_lastfile) { 344 fdp->fd_lastfile = fd_last_set(fdp, fd); 345 } 346 fd_checkmaps(fdp); 347 } 348 349 /* 350 * Look up the file structure corresponding to a file descriptor 351 * and return the file, holding a reference on the descriptor. 352 */ 353 file_t * 354 fd_getfile(unsigned fd) 355 { 356 filedesc_t *fdp; 357 fdfile_t *ff; 358 file_t *fp; 359 fdtab_t *dt; 360 361 /* 362 * Look up the fdfile structure representing this descriptor. 363 * We are doing this unlocked. See fd_tryexpand(). 364 */ 365 fdp = curlwp->l_fd; 366 dt = atomic_load_consume(&fdp->fd_dt); 367 if (__predict_false(fd >= dt->dt_nfiles)) { 368 return NULL; 369 } 370 ff = dt->dt_ff[fd]; 371 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 372 if (__predict_false(ff == NULL)) { 373 return NULL; 374 } 375 376 /* Now get a reference to the descriptor. */ 377 if (fdp->fd_refcnt == 1) { 378 /* 379 * Single threaded: don't need to worry about concurrent 380 * access (other than earlier calls to kqueue, which may 381 * hold a reference to the descriptor). 382 */ 383 ff->ff_refcnt++; 384 } else { 385 /* 386 * Multi threaded: issue a memory barrier to ensure that we 387 * acquire the file pointer _after_ adding a reference. If 388 * no memory barrier, we could fetch a stale pointer. 389 * 390 * In particular, we must coordinate the following four 391 * memory operations: 392 * 393 * A. fd_close store ff->ff_file = NULL 394 * B. fd_close refcnt = atomic_dec_uint_nv(&ff->ff_refcnt) 395 * C. fd_getfile atomic_inc_uint(&ff->ff_refcnt) 396 * D. fd_getfile load fp = ff->ff_file 397 * 398 * If the order is D;A;B;C: 399 * 400 * 1. D: fp = ff->ff_file 401 * 2. A: ff->ff_file = NULL 402 * 3. B: refcnt = atomic_dec_uint_nv(&ff->ff_refcnt) 403 * 4. C: atomic_inc_uint(&ff->ff_refcnt) 404 * 405 * then fd_close determines that there are no more 406 * references and decides to free fp immediately, at 407 * the same that fd_getfile ends up with an fp that's 408 * about to be freed. *boom* 409 * 410 * By making B a release operation in fd_close, and by 411 * making C an acquire operation in fd_getfile, since 412 * they are atomic operations on the same object, which 413 * has a total modification order, we guarantee either: 414 * 415 * - B happens before C. Then since A is 416 * sequenced before B in fd_close, and C is 417 * sequenced before D in fd_getfile, we 418 * guarantee A happens before D, so fd_getfile 419 * reads a null fp and safely fails. 420 * 421 * - C happens before B. Then fd_getfile may read 422 * null or nonnull, but either way, fd_close 423 * will safely wait for references to drain. 424 */ 425 atomic_inc_uint(&ff->ff_refcnt); 426 membar_acquire(); 427 } 428 429 /* 430 * If the file is not open or is being closed then put the 431 * reference back. 432 */ 433 fp = atomic_load_consume(&ff->ff_file); 434 if (__predict_true(fp != NULL)) { 435 return fp; 436 } 437 fd_putfile(fd); 438 return NULL; 439 } 440 441 /* 442 * Release a reference to a file descriptor acquired with fd_getfile(). 443 */ 444 void 445 fd_putfile(unsigned fd) 446 { 447 filedesc_t *fdp; 448 fdfile_t *ff; 449 u_int u, v; 450 451 fdp = curlwp->l_fd; 452 KASSERT(fd < atomic_load_consume(&fdp->fd_dt)->dt_nfiles); 453 ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd]; 454 455 KASSERT(ff != NULL); 456 KASSERT((ff->ff_refcnt & FR_MASK) > 0); 457 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 458 459 if (fdp->fd_refcnt == 1) { 460 /* 461 * Single threaded: don't need to worry about concurrent 462 * access (other than earlier calls to kqueue, which may 463 * hold a reference to the descriptor). 464 */ 465 if (__predict_false((ff->ff_refcnt & FR_CLOSING) != 0)) { 466 fd_close(fd); 467 return; 468 } 469 ff->ff_refcnt--; 470 return; 471 } 472 473 /* 474 * Ensure that any use of the file is complete and globally 475 * visible before dropping the final reference. If no membar, 476 * the current CPU could still access memory associated with 477 * the file after it has been freed or recycled by another 478 * CPU. 479 */ 480 membar_release(); 481 482 /* 483 * Be optimistic and start out with the assumption that no other 484 * threads are trying to close the descriptor. If the CAS fails, 485 * we lost a race and/or it's being closed. 486 */ 487 for (u = ff->ff_refcnt & FR_MASK;; u = v) { 488 v = atomic_cas_uint(&ff->ff_refcnt, u, u - 1); 489 if (__predict_true(u == v)) { 490 return; 491 } 492 if (__predict_false((v & FR_CLOSING) != 0)) { 493 break; 494 } 495 } 496 497 /* Another thread is waiting to close the file: join it. */ 498 (void)fd_close(fd); 499 } 500 501 /* 502 * Convenience wrapper around fd_getfile() that returns reference 503 * to a vnode. 504 */ 505 int 506 fd_getvnode(unsigned fd, file_t **fpp) 507 { 508 vnode_t *vp; 509 file_t *fp; 510 511 fp = fd_getfile(fd); 512 if (__predict_false(fp == NULL)) { 513 return EBADF; 514 } 515 if (__predict_false(fp->f_type != DTYPE_VNODE)) { 516 fd_putfile(fd); 517 return EINVAL; 518 } 519 vp = fp->f_vnode; 520 if (__predict_false(vp->v_type == VBAD)) { 521 /* XXX Is this case really necessary? */ 522 fd_putfile(fd); 523 return EBADF; 524 } 525 *fpp = fp; 526 return 0; 527 } 528 529 /* 530 * Convenience wrapper around fd_getfile() that returns reference 531 * to a socket. 532 */ 533 int 534 fd_getsock1(unsigned fd, struct socket **sop, file_t **fp) 535 { 536 *fp = fd_getfile(fd); 537 if (__predict_false(*fp == NULL)) { 538 return EBADF; 539 } 540 if (__predict_false((*fp)->f_type != DTYPE_SOCKET)) { 541 fd_putfile(fd); 542 return ENOTSOCK; 543 } 544 *sop = (*fp)->f_socket; 545 return 0; 546 } 547 548 int 549 fd_getsock(unsigned fd, struct socket **sop) 550 { 551 file_t *fp; 552 return fd_getsock1(fd, sop, &fp); 553 } 554 555 /* 556 * Look up the file structure corresponding to a file descriptor 557 * and return it with a reference held on the file, not the 558 * descriptor. 559 * 560 * This is heavyweight and only used when accessing descriptors 561 * from a foreign process. The caller must ensure that `p' does 562 * not exit or fork across this call. 563 * 564 * To release the file (not descriptor) reference, use closef(). 565 */ 566 file_t * 567 fd_getfile2(proc_t *p, unsigned fd) 568 { 569 filedesc_t *fdp; 570 fdfile_t *ff; 571 file_t *fp; 572 fdtab_t *dt; 573 574 fdp = p->p_fd; 575 mutex_enter(&fdp->fd_lock); 576 dt = fdp->fd_dt; 577 if (fd >= dt->dt_nfiles) { 578 mutex_exit(&fdp->fd_lock); 579 return NULL; 580 } 581 if ((ff = dt->dt_ff[fd]) == NULL) { 582 mutex_exit(&fdp->fd_lock); 583 return NULL; 584 } 585 if ((fp = atomic_load_consume(&ff->ff_file)) == NULL) { 586 mutex_exit(&fdp->fd_lock); 587 return NULL; 588 } 589 mutex_enter(&fp->f_lock); 590 fp->f_count++; 591 mutex_exit(&fp->f_lock); 592 mutex_exit(&fdp->fd_lock); 593 594 return fp; 595 } 596 597 /* 598 * Internal form of close. Must be called with a reference to the 599 * descriptor, and will drop the reference. When all descriptor 600 * references are dropped, releases the descriptor slot and a single 601 * reference to the file structure. 602 */ 603 int 604 fd_close(unsigned fd) 605 { 606 struct flock lf; 607 filedesc_t *fdp; 608 fdfile_t *ff; 609 file_t *fp; 610 proc_t *p; 611 lwp_t *l; 612 u_int refcnt; 613 614 l = curlwp; 615 p = l->l_proc; 616 fdp = l->l_fd; 617 ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd]; 618 619 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 620 621 mutex_enter(&fdp->fd_lock); 622 KASSERT((ff->ff_refcnt & FR_MASK) > 0); 623 fp = atomic_load_consume(&ff->ff_file); 624 if (__predict_false(fp == NULL)) { 625 /* 626 * Another user of the file is already closing, and is 627 * waiting for other users of the file to drain. Release 628 * our reference, and wake up the closer. 629 */ 630 membar_release(); 631 atomic_dec_uint(&ff->ff_refcnt); 632 cv_broadcast(&ff->ff_closing); 633 mutex_exit(&fdp->fd_lock); 634 635 /* 636 * An application error, so pretend that the descriptor 637 * was already closed. We can't safely wait for it to 638 * be closed without potentially deadlocking. 639 */ 640 return (EBADF); 641 } 642 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0); 643 644 /* 645 * There may be multiple users of this file within the process. 646 * Notify existing and new users that the file is closing. This 647 * will prevent them from adding additional uses to this file 648 * while we are closing it. 649 */ 650 atomic_store_relaxed(&ff->ff_file, NULL); 651 ff->ff_exclose = false; 652 653 /* 654 * We expect the caller to hold a descriptor reference - drop it. 655 * The reference count may increase beyond zero at this point due 656 * to an erroneous descriptor reference by an application, but 657 * fd_getfile() will notice that the file is being closed and drop 658 * the reference again. 659 */ 660 if (fdp->fd_refcnt == 1) { 661 /* Single threaded. */ 662 refcnt = --(ff->ff_refcnt); 663 } else { 664 /* Multi threaded. */ 665 membar_release(); 666 refcnt = atomic_dec_uint_nv(&ff->ff_refcnt); 667 membar_acquire(); 668 } 669 if (__predict_false(refcnt != 0)) { 670 /* 671 * Wait for other references to drain. This is typically 672 * an application error - the descriptor is being closed 673 * while still in use. 674 * (Or just a threaded application trying to unblock its 675 * thread that sleeps in (say) accept()). 676 */ 677 atomic_or_uint(&ff->ff_refcnt, FR_CLOSING); 678 679 /* 680 * Remove any knotes attached to the file. A knote 681 * attached to the descriptor can hold references on it. 682 */ 683 mutex_exit(&fdp->fd_lock); 684 if (!SLIST_EMPTY(&ff->ff_knlist)) { 685 knote_fdclose(fd); 686 } 687 688 /* 689 * Since the file system code doesn't know which fd 690 * each request came from (think dup()), we have to 691 * ask it to return ERESTART for any long-term blocks. 692 * The re-entry through read/write/etc will detect the 693 * closed fd and return EBAFD. 694 * Blocked partial writes may return a short length. 695 */ 696 (*fp->f_ops->fo_restart)(fp); 697 mutex_enter(&fdp->fd_lock); 698 699 /* 700 * We need to see the count drop to zero at least once, 701 * in order to ensure that all pre-existing references 702 * have been drained. New references past this point are 703 * of no interest. 704 * XXX (dsl) this may need to call fo_restart() after a 705 * timeout to guarantee that all the system calls exit. 706 */ 707 while ((ff->ff_refcnt & FR_MASK) != 0) { 708 cv_wait(&ff->ff_closing, &fdp->fd_lock); 709 } 710 atomic_and_uint(&ff->ff_refcnt, ~FR_CLOSING); 711 } else { 712 /* If no references, there must be no knotes. */ 713 KASSERT(SLIST_EMPTY(&ff->ff_knlist)); 714 } 715 716 /* 717 * POSIX record locking dictates that any close releases ALL 718 * locks owned by this process. This is handled by setting 719 * a flag in the unlock to free ONLY locks obeying POSIX 720 * semantics, and not to free BSD-style file locks. 721 * If the descriptor was in a message, POSIX-style locks 722 * aren't passed with the descriptor. 723 */ 724 if (__predict_false((p->p_flag & PK_ADVLOCK) != 0) && 725 fp->f_ops->fo_advlock != NULL) { 726 lf.l_whence = SEEK_SET; 727 lf.l_start = 0; 728 lf.l_len = 0; 729 lf.l_type = F_UNLCK; 730 mutex_exit(&fdp->fd_lock); 731 (void)(*fp->f_ops->fo_advlock)(fp, p, F_UNLCK, &lf, F_POSIX); 732 mutex_enter(&fdp->fd_lock); 733 } 734 735 /* Free descriptor slot. */ 736 fd_unused(fdp, fd); 737 mutex_exit(&fdp->fd_lock); 738 739 /* Now drop reference to the file itself. */ 740 return closef(fp); 741 } 742 743 /* 744 * Duplicate a file descriptor. 745 */ 746 int 747 fd_dup(file_t *fp, int minfd, int *newp, bool exclose) 748 { 749 proc_t *p = curproc; 750 int error; 751 752 while ((error = fd_alloc(p, minfd, newp)) != 0) { 753 if (error != ENOSPC) { 754 return error; 755 } 756 fd_tryexpand(p); 757 } 758 759 fd_set_exclose(curlwp, *newp, exclose); 760 fd_affix(p, fp, *newp); 761 return 0; 762 } 763 764 /* 765 * dup2 operation. 766 */ 767 int 768 fd_dup2(file_t *fp, unsigned newfd, int flags) 769 { 770 filedesc_t *fdp = curlwp->l_fd; 771 fdfile_t *ff; 772 fdtab_t *dt; 773 774 if (flags & ~(O_CLOEXEC|O_NONBLOCK|O_NOSIGPIPE)) 775 return EINVAL; 776 /* 777 * Ensure there are enough slots in the descriptor table, 778 * and allocate an fdfile_t up front in case we need it. 779 */ 780 while (newfd >= atomic_load_consume(&fdp->fd_dt)->dt_nfiles) { 781 fd_tryexpand(curproc); 782 } 783 ff = kmem_alloc(sizeof(*ff), KM_SLEEP); 784 fdfile_ctor(ff); 785 786 /* 787 * If there is already a file open, close it. If the file is 788 * half open, wait for it to be constructed before closing it. 789 * XXX Potential for deadlock here? 790 */ 791 mutex_enter(&fdp->fd_lock); 792 while (fd_isused(fdp, newfd)) { 793 mutex_exit(&fdp->fd_lock); 794 if (fd_getfile(newfd) != NULL) { 795 (void)fd_close(newfd); 796 } else { 797 /* 798 * Crummy, but unlikely to happen. 799 * Can occur if we interrupt another 800 * thread while it is opening a file. 801 */ 802 kpause("dup2", false, 1, NULL); 803 } 804 mutex_enter(&fdp->fd_lock); 805 } 806 dt = fdp->fd_dt; 807 if (dt->dt_ff[newfd] == NULL) { 808 KASSERT(newfd >= NDFDFILE); 809 dt->dt_ff[newfd] = ff; 810 ff = NULL; 811 } 812 fd_used(fdp, newfd); 813 mutex_exit(&fdp->fd_lock); 814 815 fd_set_exclose(curlwp, newfd, (flags & O_CLOEXEC) != 0); 816 fp->f_flag |= flags & (FNONBLOCK|FNOSIGPIPE); 817 /* Slot is now allocated. Insert copy of the file. */ 818 fd_affix(curproc, fp, newfd); 819 if (ff != NULL) { 820 cv_destroy(&ff->ff_closing); 821 kmem_free(ff, sizeof(*ff)); 822 } 823 return 0; 824 } 825 826 /* 827 * Drop reference to a file structure. 828 */ 829 int 830 closef(file_t *fp) 831 { 832 struct flock lf; 833 int error; 834 835 /* 836 * Drop reference. If referenced elsewhere it's still open 837 * and we have nothing more to do. 838 */ 839 mutex_enter(&fp->f_lock); 840 KASSERT(fp->f_count > 0); 841 if (--fp->f_count > 0) { 842 mutex_exit(&fp->f_lock); 843 return 0; 844 } 845 KASSERT(fp->f_count == 0); 846 mutex_exit(&fp->f_lock); 847 848 /* We held the last reference - release locks, close and free. */ 849 if (fp->f_ops->fo_advlock == NULL) { 850 KASSERT((fp->f_flag & FHASLOCK) == 0); 851 } else if (fp->f_flag & FHASLOCK) { 852 lf.l_whence = SEEK_SET; 853 lf.l_start = 0; 854 lf.l_len = 0; 855 lf.l_type = F_UNLCK; 856 (void)(*fp->f_ops->fo_advlock)(fp, fp, F_UNLCK, &lf, F_FLOCK); 857 } 858 if (fp->f_ops != NULL) { 859 error = (*fp->f_ops->fo_close)(fp); 860 861 /* 862 * .fo_close is final, so real errors are frowned on 863 * (but allowed and passed on to close(2)), and 864 * ERESTART is absolutely forbidden because the file 865 * descriptor is gone and there is no chance to retry. 866 */ 867 KASSERTMSG(error != ERESTART, 868 "file %p f_ops %p fo_close %p returned ERESTART", 869 fp, fp->f_ops, fp->f_ops->fo_close); 870 } else { 871 error = 0; 872 } 873 KASSERT(fp->f_count == 0); 874 KASSERT(fp->f_cred != NULL); 875 pool_cache_put(file_cache, fp); 876 877 return error; 878 } 879 880 /* 881 * Allocate a file descriptor for the process. 882 * 883 * Future idea for experimentation: replace all of this with radixtree. 884 */ 885 int 886 fd_alloc(proc_t *p, int want, int *result) 887 { 888 filedesc_t *fdp = p->p_fd; 889 int i, lim, last, error, hi; 890 u_int off; 891 fdtab_t *dt; 892 893 KASSERT(p == curproc || p == &proc0); 894 895 /* 896 * Search for a free descriptor starting at the higher 897 * of want or fd_freefile. 898 */ 899 mutex_enter(&fdp->fd_lock); 900 fd_checkmaps(fdp); 901 dt = fdp->fd_dt; 902 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 903 lim = uimin((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles); 904 last = uimin(dt->dt_nfiles, lim); 905 906 for (;;) { 907 if ((i = want) < fdp->fd_freefile) 908 i = fdp->fd_freefile; 909 off = i >> NDENTRYSHIFT; 910 hi = fd_next_zero(fdp, fdp->fd_himap, off, 911 (last + NDENTRIES - 1) >> NDENTRYSHIFT); 912 if (hi == -1) 913 break; 914 i = fd_next_zero(fdp, &fdp->fd_lomap[hi], 915 hi > off ? 0 : i & NDENTRYMASK, NDENTRIES); 916 if (i == -1) { 917 /* 918 * Free file descriptor in this block was 919 * below want, try again with higher want. 920 */ 921 want = (hi + 1) << NDENTRYSHIFT; 922 continue; 923 } 924 i += (hi << NDENTRYSHIFT); 925 if (i >= last) { 926 break; 927 } 928 if (dt->dt_ff[i] == NULL) { 929 KASSERT(i >= NDFDFILE); 930 dt->dt_ff[i] = kmem_alloc(sizeof(fdfile_t), KM_SLEEP); 931 fdfile_ctor(dt->dt_ff[i]); 932 } 933 KASSERT(dt->dt_ff[i]->ff_file == NULL); 934 fd_used(fdp, i); 935 if (want <= fdp->fd_freefile) { 936 fdp->fd_freefile = i; 937 } 938 *result = i; 939 KASSERT(i >= NDFDFILE || 940 dt->dt_ff[i] == (fdfile_t *)fdp->fd_dfdfile[i]); 941 fd_checkmaps(fdp); 942 mutex_exit(&fdp->fd_lock); 943 return 0; 944 } 945 946 /* No space in current array. Let the caller expand and retry. */ 947 error = (dt->dt_nfiles >= lim) ? EMFILE : ENOSPC; 948 mutex_exit(&fdp->fd_lock); 949 return error; 950 } 951 952 /* 953 * Allocate memory for a descriptor table. 954 */ 955 static fdtab_t * 956 fd_dtab_alloc(int n) 957 { 958 fdtab_t *dt; 959 size_t sz; 960 961 KASSERT(n > NDFILE); 962 963 sz = sizeof(*dt) + (n - NDFILE) * sizeof(dt->dt_ff[0]); 964 dt = kmem_alloc(sz, KM_SLEEP); 965 #ifdef DIAGNOSTIC 966 memset(dt, 0xff, sz); 967 #endif 968 dt->dt_nfiles = n; 969 dt->dt_link = NULL; 970 return dt; 971 } 972 973 /* 974 * Free a descriptor table, and all tables linked for deferred free. 975 */ 976 static void 977 fd_dtab_free(fdtab_t *dt) 978 { 979 fdtab_t *next; 980 size_t sz; 981 982 do { 983 next = dt->dt_link; 984 KASSERT(dt->dt_nfiles > NDFILE); 985 sz = sizeof(*dt) + 986 (dt->dt_nfiles - NDFILE) * sizeof(dt->dt_ff[0]); 987 #ifdef DIAGNOSTIC 988 memset(dt, 0xff, sz); 989 #endif 990 kmem_free(dt, sz); 991 dt = next; 992 } while (dt != NULL); 993 } 994 995 /* 996 * Allocate descriptor bitmap. 997 */ 998 static void 999 fd_map_alloc(int n, uint32_t **lo, uint32_t **hi) 1000 { 1001 uint8_t *ptr; 1002 size_t szlo, szhi; 1003 1004 KASSERT(n > NDENTRIES); 1005 1006 szlo = NDLOSLOTS(n) * sizeof(uint32_t); 1007 szhi = NDHISLOTS(n) * sizeof(uint32_t); 1008 ptr = kmem_alloc(szlo + szhi, KM_SLEEP); 1009 *lo = (uint32_t *)ptr; 1010 *hi = (uint32_t *)(ptr + szlo); 1011 } 1012 1013 /* 1014 * Free descriptor bitmap. 1015 */ 1016 static void 1017 fd_map_free(int n, uint32_t *lo, uint32_t *hi) 1018 { 1019 size_t szlo, szhi; 1020 1021 KASSERT(n > NDENTRIES); 1022 1023 szlo = NDLOSLOTS(n) * sizeof(uint32_t); 1024 szhi = NDHISLOTS(n) * sizeof(uint32_t); 1025 KASSERT(hi == (uint32_t *)((uint8_t *)lo + szlo)); 1026 kmem_free(lo, szlo + szhi); 1027 } 1028 1029 /* 1030 * Expand a process' descriptor table. 1031 */ 1032 void 1033 fd_tryexpand(proc_t *p) 1034 { 1035 filedesc_t *fdp; 1036 int i, numfiles, oldnfiles; 1037 fdtab_t *newdt, *dt; 1038 uint32_t *newhimap, *newlomap; 1039 1040 KASSERT(p == curproc || p == &proc0); 1041 1042 fdp = p->p_fd; 1043 newhimap = NULL; 1044 newlomap = NULL; 1045 oldnfiles = atomic_load_consume(&fdp->fd_dt)->dt_nfiles; 1046 1047 if (oldnfiles < NDEXTENT) 1048 numfiles = NDEXTENT; 1049 else 1050 numfiles = 2 * oldnfiles; 1051 1052 newdt = fd_dtab_alloc(numfiles); 1053 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1054 fd_map_alloc(numfiles, &newlomap, &newhimap); 1055 } 1056 1057 mutex_enter(&fdp->fd_lock); 1058 dt = fdp->fd_dt; 1059 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 1060 if (dt->dt_nfiles != oldnfiles) { 1061 /* fdp changed; caller must retry */ 1062 mutex_exit(&fdp->fd_lock); 1063 fd_dtab_free(newdt); 1064 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1065 fd_map_free(numfiles, newlomap, newhimap); 1066 } 1067 return; 1068 } 1069 1070 /* Copy the existing descriptor table and zero the new portion. */ 1071 i = sizeof(fdfile_t *) * oldnfiles; 1072 memcpy(newdt->dt_ff, dt->dt_ff, i); 1073 memset((uint8_t *)newdt->dt_ff + i, 0, 1074 numfiles * sizeof(fdfile_t *) - i); 1075 1076 /* 1077 * Link old descriptor array into list to be discarded. We defer 1078 * freeing until the last reference to the descriptor table goes 1079 * away (usually process exit). This allows us to do lockless 1080 * lookups in fd_getfile(). 1081 */ 1082 if (oldnfiles > NDFILE) { 1083 if (fdp->fd_refcnt > 1) { 1084 newdt->dt_link = dt; 1085 } else { 1086 fd_dtab_free(dt); 1087 } 1088 } 1089 1090 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) { 1091 i = NDHISLOTS(oldnfiles) * sizeof(uint32_t); 1092 memcpy(newhimap, fdp->fd_himap, i); 1093 memset((uint8_t *)newhimap + i, 0, 1094 NDHISLOTS(numfiles) * sizeof(uint32_t) - i); 1095 1096 i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t); 1097 memcpy(newlomap, fdp->fd_lomap, i); 1098 memset((uint8_t *)newlomap + i, 0, 1099 NDLOSLOTS(numfiles) * sizeof(uint32_t) - i); 1100 1101 if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) { 1102 fd_map_free(oldnfiles, fdp->fd_lomap, fdp->fd_himap); 1103 } 1104 fdp->fd_himap = newhimap; 1105 fdp->fd_lomap = newlomap; 1106 } 1107 1108 /* 1109 * All other modifications must become globally visible before 1110 * the change to fd_dt. See fd_getfile(). 1111 */ 1112 atomic_store_release(&fdp->fd_dt, newdt); 1113 KASSERT(newdt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]); 1114 fd_checkmaps(fdp); 1115 mutex_exit(&fdp->fd_lock); 1116 } 1117 1118 /* 1119 * Create a new open file structure and allocate a file descriptor 1120 * for the current process. 1121 */ 1122 int 1123 fd_allocfile(file_t **resultfp, int *resultfd) 1124 { 1125 proc_t *p = curproc; 1126 kauth_cred_t cred; 1127 file_t *fp; 1128 int error; 1129 1130 while ((error = fd_alloc(p, 0, resultfd)) != 0) { 1131 if (error != ENOSPC) { 1132 return error; 1133 } 1134 fd_tryexpand(p); 1135 } 1136 1137 fp = pool_cache_get(file_cache, PR_WAITOK); 1138 if (fp == NULL) { 1139 fd_abort(p, NULL, *resultfd); 1140 return ENFILE; 1141 } 1142 KASSERT(fp->f_count == 0); 1143 KASSERT(fp->f_msgcount == 0); 1144 KASSERT(fp->f_unpcount == 0); 1145 1146 /* Replace cached credentials if not what we need. */ 1147 cred = curlwp->l_cred; 1148 if (__predict_false(cred != fp->f_cred)) { 1149 kauth_cred_free(fp->f_cred); 1150 fp->f_cred = kauth_cred_hold(cred); 1151 } 1152 1153 /* 1154 * Don't allow recycled files to be scanned. 1155 * See uipc_usrreq.c. 1156 */ 1157 if (__predict_false((fp->f_flag & FSCAN) != 0)) { 1158 mutex_enter(&fp->f_lock); 1159 atomic_and_uint(&fp->f_flag, ~FSCAN); 1160 mutex_exit(&fp->f_lock); 1161 } 1162 1163 fp->f_advice = 0; 1164 fp->f_offset = 0; 1165 *resultfp = fp; 1166 1167 return 0; 1168 } 1169 1170 /* 1171 * Successful creation of a new descriptor: make visible to the process. 1172 */ 1173 void 1174 fd_affix(proc_t *p, file_t *fp, unsigned fd) 1175 { 1176 fdfile_t *ff; 1177 filedesc_t *fdp; 1178 fdtab_t *dt; 1179 1180 KASSERT(p == curproc || p == &proc0); 1181 1182 /* Add a reference to the file structure. */ 1183 mutex_enter(&fp->f_lock); 1184 fp->f_count++; 1185 mutex_exit(&fp->f_lock); 1186 1187 /* 1188 * Insert the new file into the descriptor slot. 1189 */ 1190 fdp = p->p_fd; 1191 dt = atomic_load_consume(&fdp->fd_dt); 1192 ff = dt->dt_ff[fd]; 1193 1194 KASSERT(ff != NULL); 1195 KASSERT(ff->ff_file == NULL); 1196 KASSERT(ff->ff_allocated); 1197 KASSERT(fd_isused(fdp, fd)); 1198 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1199 1200 /* No need to lock in order to make file initially visible. */ 1201 atomic_store_release(&ff->ff_file, fp); 1202 } 1203 1204 /* 1205 * Abort creation of a new descriptor: free descriptor slot and file. 1206 */ 1207 void 1208 fd_abort(proc_t *p, file_t *fp, unsigned fd) 1209 { 1210 filedesc_t *fdp; 1211 fdfile_t *ff; 1212 1213 KASSERT(p == curproc || p == &proc0); 1214 1215 fdp = p->p_fd; 1216 ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd]; 1217 ff->ff_exclose = false; 1218 1219 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1220 1221 mutex_enter(&fdp->fd_lock); 1222 KASSERT(fd_isused(fdp, fd)); 1223 fd_unused(fdp, fd); 1224 mutex_exit(&fdp->fd_lock); 1225 1226 if (fp != NULL) { 1227 KASSERT(fp->f_count == 0); 1228 KASSERT(fp->f_cred != NULL); 1229 pool_cache_put(file_cache, fp); 1230 } 1231 } 1232 1233 static int 1234 file_ctor(void *arg, void *obj, int flags) 1235 { 1236 /* 1237 * It's easy to exhaust the open file limit on a system with many 1238 * CPUs due to caching. Allow a bit of leeway to reduce the element 1239 * of surprise. 1240 */ 1241 u_int slop = PCG_NOBJECTS_NORMAL * (ncpu - 1); 1242 file_t *fp = obj; 1243 1244 memset(fp, 0, sizeof(*fp)); 1245 1246 mutex_enter(&filelist_lock); 1247 if (__predict_false(nfiles >= slop + maxfiles)) { 1248 mutex_exit(&filelist_lock); 1249 tablefull("file", "increase kern.maxfiles or MAXFILES"); 1250 return ENFILE; 1251 } 1252 nfiles++; 1253 LIST_INSERT_HEAD(&filehead, fp, f_list); 1254 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE); 1255 fp->f_cred = kauth_cred_hold(curlwp->l_cred); 1256 mutex_exit(&filelist_lock); 1257 1258 return 0; 1259 } 1260 1261 static void 1262 file_dtor(void *arg, void *obj) 1263 { 1264 file_t *fp = obj; 1265 1266 mutex_enter(&filelist_lock); 1267 nfiles--; 1268 LIST_REMOVE(fp, f_list); 1269 mutex_exit(&filelist_lock); 1270 1271 KASSERT(fp->f_count == 0); 1272 kauth_cred_free(fp->f_cred); 1273 mutex_destroy(&fp->f_lock); 1274 } 1275 1276 static void 1277 fdfile_ctor(fdfile_t *ff) 1278 { 1279 1280 memset(ff, 0, sizeof(*ff)); 1281 cv_init(&ff->ff_closing, "fdclose"); 1282 } 1283 1284 static void 1285 fdfile_dtor(fdfile_t *ff) 1286 { 1287 1288 cv_destroy(&ff->ff_closing); 1289 } 1290 1291 file_t * 1292 fgetdummy(void) 1293 { 1294 file_t *fp; 1295 1296 fp = kmem_zalloc(sizeof(*fp), KM_SLEEP); 1297 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE); 1298 return fp; 1299 } 1300 1301 void 1302 fputdummy(file_t *fp) 1303 { 1304 1305 mutex_destroy(&fp->f_lock); 1306 kmem_free(fp, sizeof(*fp)); 1307 } 1308 1309 /* 1310 * Create an initial filedesc structure. 1311 */ 1312 filedesc_t * 1313 fd_init(filedesc_t *fdp) 1314 { 1315 #ifdef DIAGNOSTIC 1316 unsigned fd; 1317 #endif 1318 1319 if (__predict_true(fdp == NULL)) { 1320 fdp = pool_cache_get(filedesc_cache, PR_WAITOK); 1321 } else { 1322 KASSERT(fdp == &filedesc0); 1323 filedesc_ctor(NULL, fdp, PR_WAITOK); 1324 } 1325 1326 #ifdef DIAGNOSTIC 1327 KASSERT(fdp->fd_lastfile == -1); 1328 KASSERT(fdp->fd_lastkqfile == -1); 1329 KASSERT(fdp->fd_knhash == NULL); 1330 KASSERT(fdp->fd_freefile == 0); 1331 KASSERT(fdp->fd_exclose == false); 1332 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin); 1333 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1334 for (fd = 0; fd < NDFDFILE; fd++) { 1335 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == 1336 (fdfile_t *)fdp->fd_dfdfile[fd]); 1337 } 1338 for (fd = NDFDFILE; fd < NDFILE; fd++) { 1339 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == NULL); 1340 } 1341 KASSERT(fdp->fd_himap == fdp->fd_dhimap); 1342 KASSERT(fdp->fd_lomap == fdp->fd_dlomap); 1343 #endif /* DIAGNOSTIC */ 1344 1345 fdp->fd_refcnt = 1; 1346 fd_checkmaps(fdp); 1347 1348 return fdp; 1349 } 1350 1351 /* 1352 * Initialize a file descriptor table. 1353 */ 1354 static int 1355 filedesc_ctor(void *arg, void *obj, int flag) 1356 { 1357 filedesc_t *fdp = obj; 1358 fdfile_t **ffp; 1359 int i; 1360 1361 memset(fdp, 0, sizeof(*fdp)); 1362 mutex_init(&fdp->fd_lock, MUTEX_DEFAULT, IPL_NONE); 1363 fdp->fd_lastfile = -1; 1364 fdp->fd_lastkqfile = -1; 1365 fdp->fd_dt = &fdp->fd_dtbuiltin; 1366 fdp->fd_dtbuiltin.dt_nfiles = NDFILE; 1367 fdp->fd_himap = fdp->fd_dhimap; 1368 fdp->fd_lomap = fdp->fd_dlomap; 1369 1370 CTASSERT(sizeof(fdp->fd_dfdfile[0]) >= sizeof(fdfile_t)); 1371 for (i = 0, ffp = fdp->fd_dt->dt_ff; i < NDFDFILE; i++, ffp++) { 1372 fdfile_ctor(*ffp = (fdfile_t *)fdp->fd_dfdfile[i]); 1373 } 1374 1375 return 0; 1376 } 1377 1378 static void 1379 filedesc_dtor(void *arg, void *obj) 1380 { 1381 filedesc_t *fdp = obj; 1382 int i; 1383 1384 for (i = 0; i < NDFDFILE; i++) { 1385 fdfile_dtor((fdfile_t *)fdp->fd_dfdfile[i]); 1386 } 1387 1388 mutex_destroy(&fdp->fd_lock); 1389 } 1390 1391 /* 1392 * Make p share curproc's filedesc structure. 1393 */ 1394 void 1395 fd_share(struct proc *p) 1396 { 1397 filedesc_t *fdp; 1398 1399 fdp = curlwp->l_fd; 1400 p->p_fd = fdp; 1401 atomic_inc_uint(&fdp->fd_refcnt); 1402 } 1403 1404 /* 1405 * Acquire a hold on a filedesc structure. 1406 */ 1407 void 1408 fd_hold(lwp_t *l) 1409 { 1410 filedesc_t *fdp = l->l_fd; 1411 1412 atomic_inc_uint(&fdp->fd_refcnt); 1413 } 1414 1415 /* 1416 * Copy a filedesc structure. 1417 */ 1418 filedesc_t * 1419 fd_copy(void) 1420 { 1421 filedesc_t *newfdp, *fdp; 1422 fdfile_t *ff, **ffp, **nffp, *ff2; 1423 int i, j, numfiles, lastfile, newlast; 1424 file_t *fp; 1425 fdtab_t *newdt; 1426 1427 fdp = curproc->p_fd; 1428 newfdp = pool_cache_get(filedesc_cache, PR_WAITOK); 1429 newfdp->fd_refcnt = 1; 1430 1431 #ifdef DIAGNOSTIC 1432 KASSERT(newfdp->fd_lastfile == -1); 1433 KASSERT(newfdp->fd_lastkqfile == -1); 1434 KASSERT(newfdp->fd_knhash == NULL); 1435 KASSERT(newfdp->fd_freefile == 0); 1436 KASSERT(newfdp->fd_exclose == false); 1437 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin); 1438 KASSERT(newfdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1439 for (i = 0; i < NDFDFILE; i++) { 1440 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == 1441 (fdfile_t *)&newfdp->fd_dfdfile[i]); 1442 } 1443 for (i = NDFDFILE; i < NDFILE; i++) { 1444 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == NULL); 1445 } 1446 #endif /* DIAGNOSTIC */ 1447 1448 mutex_enter(&fdp->fd_lock); 1449 fd_checkmaps(fdp); 1450 numfiles = fdp->fd_dt->dt_nfiles; 1451 lastfile = fdp->fd_lastfile; 1452 1453 /* 1454 * If the number of open files fits in the internal arrays 1455 * of the open file structure, use them, otherwise allocate 1456 * additional memory for the number of descriptors currently 1457 * in use. 1458 */ 1459 if (lastfile < NDFILE) { 1460 i = NDFILE; 1461 newdt = newfdp->fd_dt; 1462 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin); 1463 } else { 1464 /* 1465 * Compute the smallest multiple of NDEXTENT needed 1466 * for the file descriptors currently in use, 1467 * allowing the table to shrink. 1468 */ 1469 i = numfiles; 1470 while (i >= 2 * NDEXTENT && i > lastfile * 2) { 1471 i /= 2; 1472 } 1473 KASSERT(i > NDFILE); 1474 newdt = fd_dtab_alloc(i); 1475 newfdp->fd_dt = newdt; 1476 memcpy(newdt->dt_ff, newfdp->fd_dtbuiltin.dt_ff, 1477 NDFDFILE * sizeof(fdfile_t **)); 1478 memset(newdt->dt_ff + NDFDFILE, 0, 1479 (i - NDFDFILE) * sizeof(fdfile_t **)); 1480 } 1481 if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) { 1482 newfdp->fd_himap = newfdp->fd_dhimap; 1483 newfdp->fd_lomap = newfdp->fd_dlomap; 1484 } else { 1485 fd_map_alloc(i, &newfdp->fd_lomap, &newfdp->fd_himap); 1486 KASSERT(i >= NDENTRIES * NDENTRIES); 1487 memset(newfdp->fd_himap, 0, NDHISLOTS(i)*sizeof(uint32_t)); 1488 memset(newfdp->fd_lomap, 0, NDLOSLOTS(i)*sizeof(uint32_t)); 1489 } 1490 newfdp->fd_freefile = fdp->fd_freefile; 1491 newfdp->fd_exclose = fdp->fd_exclose; 1492 1493 ffp = fdp->fd_dt->dt_ff; 1494 nffp = newdt->dt_ff; 1495 newlast = -1; 1496 for (i = 0; i <= lastfile; i++, ffp++, nffp++) { 1497 KASSERT(i >= NDFDFILE || 1498 *nffp == (fdfile_t *)newfdp->fd_dfdfile[i]); 1499 ff = *ffp; 1500 if (ff == NULL || 1501 (fp = atomic_load_consume(&ff->ff_file)) == NULL) { 1502 /* Descriptor unused, or descriptor half open. */ 1503 KASSERT(!fd_isused(newfdp, i)); 1504 continue; 1505 } 1506 if (__predict_false(fp->f_type == DTYPE_KQUEUE)) { 1507 /* kqueue descriptors cannot be copied. */ 1508 if (i < newfdp->fd_freefile) { 1509 newfdp->fd_freefile = i; 1510 } 1511 continue; 1512 } 1513 /* It's active: add a reference to the file. */ 1514 mutex_enter(&fp->f_lock); 1515 fp->f_count++; 1516 mutex_exit(&fp->f_lock); 1517 1518 /* Allocate an fdfile_t to represent it. */ 1519 if (i >= NDFDFILE) { 1520 ff2 = kmem_alloc(sizeof(*ff2), KM_SLEEP); 1521 fdfile_ctor(ff2); 1522 *nffp = ff2; 1523 } else { 1524 ff2 = newdt->dt_ff[i]; 1525 } 1526 ff2->ff_file = fp; 1527 ff2->ff_exclose = ff->ff_exclose; 1528 ff2->ff_allocated = true; 1529 1530 /* Fix up bitmaps. */ 1531 j = i >> NDENTRYSHIFT; 1532 KASSERT((newfdp->fd_lomap[j] & (1U << (i & NDENTRYMASK))) == 0); 1533 newfdp->fd_lomap[j] |= 1U << (i & NDENTRYMASK); 1534 if (__predict_false(newfdp->fd_lomap[j] == ~0)) { 1535 KASSERT((newfdp->fd_himap[j >> NDENTRYSHIFT] & 1536 (1U << (j & NDENTRYMASK))) == 0); 1537 newfdp->fd_himap[j >> NDENTRYSHIFT] |= 1538 1U << (j & NDENTRYMASK); 1539 } 1540 newlast = i; 1541 } 1542 KASSERT(newdt->dt_ff[0] == (fdfile_t *)newfdp->fd_dfdfile[0]); 1543 newfdp->fd_lastfile = newlast; 1544 fd_checkmaps(newfdp); 1545 mutex_exit(&fdp->fd_lock); 1546 1547 return newfdp; 1548 } 1549 1550 /* 1551 * Release a filedesc structure. 1552 */ 1553 void 1554 fd_free(void) 1555 { 1556 fdfile_t *ff; 1557 file_t *fp; 1558 int fd, nf; 1559 fdtab_t *dt; 1560 lwp_t * const l = curlwp; 1561 filedesc_t * const fdp = l->l_fd; 1562 const bool noadvlock = (l->l_proc->p_flag & PK_ADVLOCK) == 0; 1563 1564 KASSERT(atomic_load_consume(&fdp->fd_dt)->dt_ff[0] == 1565 (fdfile_t *)fdp->fd_dfdfile[0]); 1566 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1567 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL); 1568 1569 membar_release(); 1570 if (atomic_dec_uint_nv(&fdp->fd_refcnt) > 0) 1571 return; 1572 membar_acquire(); 1573 1574 /* 1575 * Close any files that the process holds open. 1576 */ 1577 dt = fdp->fd_dt; 1578 fd_checkmaps(fdp); 1579 #ifdef DEBUG 1580 fdp->fd_refcnt = -1; /* see fd_checkmaps */ 1581 #endif 1582 for (fd = 0, nf = dt->dt_nfiles; fd < nf; fd++) { 1583 ff = dt->dt_ff[fd]; 1584 KASSERT(fd >= NDFDFILE || 1585 ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1586 if (ff == NULL) 1587 continue; 1588 if ((fp = atomic_load_consume(&ff->ff_file)) != NULL) { 1589 /* 1590 * Must use fd_close() here if there is 1591 * a reference from kqueue or we might have posix 1592 * advisory locks. 1593 */ 1594 if (__predict_true(ff->ff_refcnt == 0) && 1595 (noadvlock || fp->f_type != DTYPE_VNODE)) { 1596 ff->ff_file = NULL; 1597 ff->ff_exclose = false; 1598 ff->ff_allocated = false; 1599 closef(fp); 1600 } else { 1601 ff->ff_refcnt++; 1602 fd_close(fd); 1603 } 1604 } 1605 KASSERT(ff->ff_refcnt == 0); 1606 KASSERT(ff->ff_file == NULL); 1607 KASSERT(!ff->ff_exclose); 1608 KASSERT(!ff->ff_allocated); 1609 if (fd >= NDFDFILE) { 1610 cv_destroy(&ff->ff_closing); 1611 kmem_free(ff, sizeof(*ff)); 1612 dt->dt_ff[fd] = NULL; 1613 } 1614 } 1615 1616 /* 1617 * Clean out the descriptor table for the next user and return 1618 * to the cache. 1619 */ 1620 if (__predict_false(dt != &fdp->fd_dtbuiltin)) { 1621 fd_dtab_free(fdp->fd_dt); 1622 /* Otherwise, done above. */ 1623 memset(&fdp->fd_dtbuiltin.dt_ff[NDFDFILE], 0, 1624 (NDFILE - NDFDFILE) * sizeof(fdp->fd_dtbuiltin.dt_ff[0])); 1625 fdp->fd_dt = &fdp->fd_dtbuiltin; 1626 } 1627 if (__predict_false(NDHISLOTS(nf) > NDHISLOTS(NDFILE))) { 1628 KASSERT(fdp->fd_himap != fdp->fd_dhimap); 1629 KASSERT(fdp->fd_lomap != fdp->fd_dlomap); 1630 fd_map_free(nf, fdp->fd_lomap, fdp->fd_himap); 1631 } 1632 if (__predict_false(fdp->fd_knhash != NULL)) { 1633 hashdone(fdp->fd_knhash, HASH_LIST, fdp->fd_knhashmask); 1634 fdp->fd_knhash = NULL; 1635 fdp->fd_knhashmask = 0; 1636 } else { 1637 KASSERT(fdp->fd_knhashmask == 0); 1638 } 1639 fdp->fd_dt = &fdp->fd_dtbuiltin; 1640 fdp->fd_lastkqfile = -1; 1641 fdp->fd_lastfile = -1; 1642 fdp->fd_freefile = 0; 1643 fdp->fd_exclose = false; 1644 memset(&fdp->fd_startzero, 0, sizeof(*fdp) - 1645 offsetof(filedesc_t, fd_startzero)); 1646 fdp->fd_himap = fdp->fd_dhimap; 1647 fdp->fd_lomap = fdp->fd_dlomap; 1648 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE); 1649 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL); 1650 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin); 1651 #ifdef DEBUG 1652 fdp->fd_refcnt = 0; /* see fd_checkmaps */ 1653 #endif 1654 fd_checkmaps(fdp); 1655 pool_cache_put(filedesc_cache, fdp); 1656 } 1657 1658 /* 1659 * File Descriptor pseudo-device driver (/dev/fd/). 1660 * 1661 * Opening minor device N dup()s the file (if any) connected to file 1662 * descriptor N belonging to the calling process. Note that this driver 1663 * consists of only the ``open()'' routine, because all subsequent 1664 * references to this file will be direct to the other driver. 1665 */ 1666 static int 1667 filedescopen(dev_t dev, int mode, int type, lwp_t *l) 1668 { 1669 1670 /* 1671 * XXX Kludge: set dupfd to contain the value of the 1672 * the file descriptor being sought for duplication. The error 1673 * return ensures that the vnode for this device will be released 1674 * by vn_open. Open will detect this special error and take the 1675 * actions in fd_dupopen below. Other callers of vn_open or VOP_OPEN 1676 * will simply report the error. 1677 */ 1678 l->l_dupfd = minor(dev); /* XXX */ 1679 return EDUPFD; 1680 } 1681 1682 /* 1683 * Duplicate the specified descriptor to a free descriptor. 1684 * 1685 * old is the original fd. 1686 * moveit is true if we should move rather than duplicate. 1687 * flags are the open flags (converted from O_* to F*). 1688 * newp returns the new fd on success. 1689 * 1690 * These two cases are produced by the EDUPFD and EMOVEFD magic 1691 * errnos, but in the interest of removing that regrettable interface, 1692 * vn_open has been changed to intercept them. Now vn_open returns 1693 * either a vnode or a filehandle, and the filehandle is accompanied 1694 * by a boolean that says whether we should dup (moveit == false) or 1695 * move (moveit == true) the fd. 1696 * 1697 * The dup case is used by /dev/stderr, /proc/self/fd, and such. The 1698 * move case is used by cloner devices that allocate a fd of their 1699 * own (a layering violation that should go away eventually) that 1700 * then needs to be put in the place open() expects it. 1701 */ 1702 int 1703 fd_dupopen(int old, bool moveit, int flags, int *newp) 1704 { 1705 filedesc_t *fdp; 1706 fdfile_t *ff; 1707 file_t *fp; 1708 fdtab_t *dt; 1709 int error; 1710 1711 if ((fp = fd_getfile(old)) == NULL) { 1712 return EBADF; 1713 } 1714 fdp = curlwp->l_fd; 1715 dt = atomic_load_consume(&fdp->fd_dt); 1716 ff = dt->dt_ff[old]; 1717 1718 /* 1719 * There are two cases of interest here. 1720 * 1721 * 1. moveit == false (used to be the EDUPFD magic errno): 1722 * simply dup (old) to file descriptor (new) and return. 1723 * 1724 * 2. moveit == true (used to be the EMOVEFD magic errno): 1725 * steal away the file structure from (old) and store it in 1726 * (new). (old) is effectively closed by this operation. 1727 */ 1728 if (moveit == false) { 1729 /* 1730 * Check that the mode the file is being opened for is a 1731 * subset of the mode of the existing descriptor. 1732 */ 1733 if (((flags & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 1734 error = EACCES; 1735 goto out; 1736 } 1737 1738 /* Copy it. */ 1739 error = fd_dup(fp, 0, newp, ff->ff_exclose); 1740 } else { 1741 /* Copy it. */ 1742 error = fd_dup(fp, 0, newp, ff->ff_exclose); 1743 if (error != 0) { 1744 goto out; 1745 } 1746 1747 /* Steal away the file pointer from 'old'. */ 1748 (void)fd_close(old); 1749 return 0; 1750 } 1751 1752 out: 1753 fd_putfile(old); 1754 return error; 1755 } 1756 1757 /* 1758 * Close open files on exec. 1759 */ 1760 void 1761 fd_closeexec(void) 1762 { 1763 proc_t *p; 1764 filedesc_t *fdp; 1765 fdfile_t *ff; 1766 lwp_t *l; 1767 fdtab_t *dt; 1768 int fd; 1769 1770 l = curlwp; 1771 p = l->l_proc; 1772 fdp = p->p_fd; 1773 1774 if (fdp->fd_refcnt > 1) { 1775 fdp = fd_copy(); 1776 fd_free(); 1777 p->p_fd = fdp; 1778 l->l_fd = fdp; 1779 } 1780 if (!fdp->fd_exclose) { 1781 return; 1782 } 1783 fdp->fd_exclose = false; 1784 dt = atomic_load_consume(&fdp->fd_dt); 1785 1786 for (fd = 0; fd <= fdp->fd_lastfile; fd++) { 1787 if ((ff = dt->dt_ff[fd]) == NULL) { 1788 KASSERT(fd >= NDFDFILE); 1789 continue; 1790 } 1791 KASSERT(fd >= NDFDFILE || 1792 ff == (fdfile_t *)fdp->fd_dfdfile[fd]); 1793 if (ff->ff_file == NULL) 1794 continue; 1795 if (ff->ff_exclose) { 1796 /* 1797 * We need a reference to close the file. 1798 * No other threads can see the fdfile_t at 1799 * this point, so don't bother locking. 1800 */ 1801 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0); 1802 ff->ff_refcnt++; 1803 fd_close(fd); 1804 } 1805 } 1806 } 1807 1808 /* 1809 * Sets descriptor owner. If the owner is a process, 'pgid' 1810 * is set to positive value, process ID. If the owner is process group, 1811 * 'pgid' is set to -pg_id. 1812 */ 1813 int 1814 fsetown(pid_t *pgid, u_long cmd, const void *data) 1815 { 1816 pid_t id = *(const pid_t *)data; 1817 int error; 1818 1819 if (id <= INT_MIN) 1820 return EINVAL; 1821 1822 switch (cmd) { 1823 case TIOCSPGRP: 1824 if (id < 0) 1825 return EINVAL; 1826 id = -id; 1827 break; 1828 default: 1829 break; 1830 } 1831 if (id > 0) { 1832 mutex_enter(&proc_lock); 1833 error = proc_find(id) ? 0 : ESRCH; 1834 mutex_exit(&proc_lock); 1835 } else if (id < 0) { 1836 error = pgid_in_session(curproc, -id); 1837 } else { 1838 error = 0; 1839 } 1840 if (!error) { 1841 *pgid = id; 1842 } 1843 return error; 1844 } 1845 1846 void 1847 fd_set_exclose(struct lwp *l, int fd, bool exclose) 1848 { 1849 filedesc_t *fdp = l->l_fd; 1850 fdfile_t *ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd]; 1851 1852 ff->ff_exclose = exclose; 1853 if (exclose) 1854 fdp->fd_exclose = true; 1855 } 1856 1857 /* 1858 * Return descriptor owner information. If the value is positive, 1859 * it's process ID. If it's negative, it's process group ID and 1860 * needs the sign removed before use. 1861 */ 1862 int 1863 fgetown(pid_t pgid, u_long cmd, void *data) 1864 { 1865 1866 switch (cmd) { 1867 case TIOCGPGRP: 1868 KASSERT(pgid > INT_MIN); 1869 *(int *)data = -pgid; 1870 break; 1871 default: 1872 *(int *)data = pgid; 1873 break; 1874 } 1875 return 0; 1876 } 1877 1878 /* 1879 * Send signal to descriptor owner, either process or process group. 1880 */ 1881 void 1882 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata) 1883 { 1884 ksiginfo_t ksi; 1885 1886 KASSERT(!cpu_intr_p()); 1887 1888 if (pgid == 0) { 1889 return; 1890 } 1891 1892 KSI_INIT(&ksi); 1893 ksi.ksi_signo = signo; 1894 ksi.ksi_code = code; 1895 ksi.ksi_band = band; 1896 1897 mutex_enter(&proc_lock); 1898 if (pgid > 0) { 1899 struct proc *p1; 1900 1901 p1 = proc_find(pgid); 1902 if (p1 != NULL) { 1903 kpsignal(p1, &ksi, fdescdata); 1904 } 1905 } else { 1906 struct pgrp *pgrp; 1907 1908 KASSERT(pgid < 0 && pgid > INT_MIN); 1909 pgrp = pgrp_find(-pgid); 1910 if (pgrp != NULL) { 1911 kpgsignal(pgrp, &ksi, fdescdata, 0); 1912 } 1913 } 1914 mutex_exit(&proc_lock); 1915 } 1916 1917 int 1918 fd_clone(file_t *fp, unsigned fd, int flag, const struct fileops *fops, 1919 void *data) 1920 { 1921 1922 fp->f_flag = flag & FMASK; 1923 fd_set_exclose(curlwp, fd, (flag & O_CLOEXEC) != 0); 1924 fp->f_type = DTYPE_MISC; 1925 fp->f_ops = fops; 1926 fp->f_data = data; 1927 curlwp->l_dupfd = fd; 1928 fd_affix(curproc, fp, fd); 1929 1930 return EMOVEFD; 1931 } 1932 1933 int 1934 fnullop_fcntl(file_t *fp, u_int cmd, void *data) 1935 { 1936 1937 if (cmd == F_SETFL) 1938 return 0; 1939 1940 return EOPNOTSUPP; 1941 } 1942 1943 int 1944 fnullop_poll(file_t *fp, int which) 1945 { 1946 1947 return 0; 1948 } 1949 1950 int 1951 fnullop_kqfilter(file_t *fp, struct knote *kn) 1952 { 1953 1954 return EOPNOTSUPP; 1955 } 1956 1957 void 1958 fnullop_restart(file_t *fp) 1959 { 1960 1961 } 1962 1963 int 1964 fbadop_read(file_t *fp, off_t *offset, struct uio *uio, 1965 kauth_cred_t cred, int flags) 1966 { 1967 1968 return EOPNOTSUPP; 1969 } 1970 1971 int 1972 fbadop_write(file_t *fp, off_t *offset, struct uio *uio, 1973 kauth_cred_t cred, int flags) 1974 { 1975 1976 return EOPNOTSUPP; 1977 } 1978 1979 int 1980 fbadop_ioctl(file_t *fp, u_long com, void *data) 1981 { 1982 1983 return EOPNOTSUPP; 1984 } 1985 1986 int 1987 fbadop_stat(file_t *fp, struct stat *sb) 1988 { 1989 1990 return EOPNOTSUPP; 1991 } 1992 1993 int 1994 fbadop_close(file_t *fp) 1995 { 1996 1997 return EOPNOTSUPP; 1998 } 1999 2000 /* 2001 * sysctl routines pertaining to file descriptors 2002 */ 2003 2004 /* Initialized in sysctl_init() for now... */ 2005 extern kmutex_t sysctl_file_marker_lock; 2006 static u_int sysctl_file_marker = 1; 2007 2008 /* 2009 * Expects to be called with proc_lock and sysctl_file_marker_lock locked. 2010 */ 2011 static void 2012 sysctl_file_marker_reset(void) 2013 { 2014 struct proc *p; 2015 2016 PROCLIST_FOREACH(p, &allproc) { 2017 struct filedesc *fd = p->p_fd; 2018 fdtab_t *dt; 2019 u_int i; 2020 2021 mutex_enter(&fd->fd_lock); 2022 dt = fd->fd_dt; 2023 for (i = 0; i < dt->dt_nfiles; i++) { 2024 struct file *fp; 2025 fdfile_t *ff; 2026 2027 if ((ff = dt->dt_ff[i]) == NULL) { 2028 continue; 2029 } 2030 if ((fp = atomic_load_consume(&ff->ff_file)) == NULL) { 2031 continue; 2032 } 2033 fp->f_marker = 0; 2034 } 2035 mutex_exit(&fd->fd_lock); 2036 } 2037 } 2038 2039 /* 2040 * sysctl helper routine for kern.file pseudo-subtree. 2041 */ 2042 static int 2043 sysctl_kern_file(SYSCTLFN_ARGS) 2044 { 2045 const bool allowaddr = get_expose_address(curproc); 2046 struct filelist flist; 2047 int error; 2048 size_t buflen; 2049 struct file *fp, fbuf; 2050 char *start, *where; 2051 struct proc *p; 2052 2053 start = where = oldp; 2054 buflen = *oldlenp; 2055 2056 if (where == NULL) { 2057 /* 2058 * overestimate by 10 files 2059 */ 2060 *oldlenp = sizeof(filehead) + (nfiles + 10) * 2061 sizeof(struct file); 2062 return 0; 2063 } 2064 2065 /* 2066 * first sysctl_copyout filehead 2067 */ 2068 if (buflen < sizeof(filehead)) { 2069 *oldlenp = 0; 2070 return 0; 2071 } 2072 sysctl_unlock(); 2073 if (allowaddr) { 2074 memcpy(&flist, &filehead, sizeof(flist)); 2075 } else { 2076 memset(&flist, 0, sizeof(flist)); 2077 } 2078 error = sysctl_copyout(l, &flist, where, sizeof(flist)); 2079 if (error) { 2080 sysctl_relock(); 2081 return error; 2082 } 2083 buflen -= sizeof(flist); 2084 where += sizeof(flist); 2085 2086 /* 2087 * followed by an array of file structures 2088 */ 2089 mutex_enter(&sysctl_file_marker_lock); 2090 mutex_enter(&proc_lock); 2091 PROCLIST_FOREACH(p, &allproc) { 2092 struct filedesc *fd; 2093 fdtab_t *dt; 2094 u_int i; 2095 2096 if (p->p_stat == SIDL) { 2097 /* skip embryonic processes */ 2098 continue; 2099 } 2100 mutex_enter(p->p_lock); 2101 error = kauth_authorize_process(l->l_cred, 2102 KAUTH_PROCESS_CANSEE, p, 2103 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), 2104 NULL, NULL); 2105 mutex_exit(p->p_lock); 2106 if (error != 0) { 2107 /* 2108 * Don't leak kauth retval if we're silently 2109 * skipping this entry. 2110 */ 2111 error = 0; 2112 continue; 2113 } 2114 2115 /* 2116 * Grab a hold on the process. 2117 */ 2118 if (!rw_tryenter(&p->p_reflock, RW_READER)) { 2119 continue; 2120 } 2121 mutex_exit(&proc_lock); 2122 2123 fd = p->p_fd; 2124 mutex_enter(&fd->fd_lock); 2125 dt = fd->fd_dt; 2126 for (i = 0; i < dt->dt_nfiles; i++) { 2127 fdfile_t *ff; 2128 2129 if ((ff = dt->dt_ff[i]) == NULL) { 2130 continue; 2131 } 2132 if ((fp = atomic_load_consume(&ff->ff_file)) == NULL) { 2133 continue; 2134 } 2135 2136 mutex_enter(&fp->f_lock); 2137 2138 if ((fp->f_count == 0) || 2139 (fp->f_marker == sysctl_file_marker)) { 2140 mutex_exit(&fp->f_lock); 2141 continue; 2142 } 2143 2144 /* Check that we have enough space. */ 2145 if (buflen < sizeof(struct file)) { 2146 *oldlenp = where - start; 2147 mutex_exit(&fp->f_lock); 2148 error = ENOMEM; 2149 break; 2150 } 2151 2152 fill_file(&fbuf, fp); 2153 mutex_exit(&fp->f_lock); 2154 error = sysctl_copyout(l, &fbuf, where, sizeof(fbuf)); 2155 if (error) { 2156 break; 2157 } 2158 buflen -= sizeof(struct file); 2159 where += sizeof(struct file); 2160 2161 fp->f_marker = sysctl_file_marker; 2162 } 2163 mutex_exit(&fd->fd_lock); 2164 2165 /* 2166 * Release reference to process. 2167 */ 2168 mutex_enter(&proc_lock); 2169 rw_exit(&p->p_reflock); 2170 2171 if (error) 2172 break; 2173 } 2174 2175 sysctl_file_marker++; 2176 /* Reset all markers if wrapped. */ 2177 if (sysctl_file_marker == 0) { 2178 sysctl_file_marker_reset(); 2179 sysctl_file_marker++; 2180 } 2181 2182 mutex_exit(&proc_lock); 2183 mutex_exit(&sysctl_file_marker_lock); 2184 2185 *oldlenp = where - start; 2186 sysctl_relock(); 2187 return error; 2188 } 2189 2190 /* 2191 * sysctl helper function for kern.file2 2192 */ 2193 static int 2194 sysctl_kern_file2(SYSCTLFN_ARGS) 2195 { 2196 struct proc *p; 2197 struct file *fp; 2198 struct filedesc *fd; 2199 struct kinfo_file kf; 2200 char *dp; 2201 u_int i, op; 2202 size_t len, needed, elem_size, out_size; 2203 int error, arg, elem_count; 2204 fdfile_t *ff; 2205 fdtab_t *dt; 2206 2207 if (namelen == 1 && name[0] == CTL_QUERY) 2208 return sysctl_query(SYSCTLFN_CALL(rnode)); 2209 2210 if (namelen != 4) 2211 return EINVAL; 2212 2213 error = 0; 2214 dp = oldp; 2215 len = (oldp != NULL) ? *oldlenp : 0; 2216 op = name[0]; 2217 arg = name[1]; 2218 elem_size = name[2]; 2219 elem_count = name[3]; 2220 out_size = MIN(sizeof(kf), elem_size); 2221 needed = 0; 2222 2223 if (elem_size < 1 || elem_count < 0) 2224 return EINVAL; 2225 2226 switch (op) { 2227 case KERN_FILE_BYFILE: 2228 case KERN_FILE_BYPID: 2229 /* 2230 * We're traversing the process list in both cases; the BYFILE 2231 * case does additional work of keeping track of files already 2232 * looked at. 2233 */ 2234 2235 /* doesn't use arg so it must be zero */ 2236 if ((op == KERN_FILE_BYFILE) && (arg != 0)) 2237 return EINVAL; 2238 2239 if ((op == KERN_FILE_BYPID) && (arg < -1)) 2240 /* -1 means all processes */ 2241 return EINVAL; 2242 2243 sysctl_unlock(); 2244 if (op == KERN_FILE_BYFILE) 2245 mutex_enter(&sysctl_file_marker_lock); 2246 mutex_enter(&proc_lock); 2247 PROCLIST_FOREACH(p, &allproc) { 2248 if (p->p_stat == SIDL) { 2249 /* skip embryonic processes */ 2250 continue; 2251 } 2252 if (arg > 0 && p->p_pid != arg) { 2253 /* pick only the one we want */ 2254 /* XXX want 0 to mean "kernel files" */ 2255 continue; 2256 } 2257 mutex_enter(p->p_lock); 2258 error = kauth_authorize_process(l->l_cred, 2259 KAUTH_PROCESS_CANSEE, p, 2260 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), 2261 NULL, NULL); 2262 mutex_exit(p->p_lock); 2263 if (error != 0) { 2264 /* 2265 * Don't leak kauth retval if we're silently 2266 * skipping this entry. 2267 */ 2268 error = 0; 2269 continue; 2270 } 2271 2272 /* 2273 * Grab a hold on the process. 2274 */ 2275 if (!rw_tryenter(&p->p_reflock, RW_READER)) { 2276 continue; 2277 } 2278 mutex_exit(&proc_lock); 2279 2280 fd = p->p_fd; 2281 mutex_enter(&fd->fd_lock); 2282 dt = fd->fd_dt; 2283 for (i = 0; i < dt->dt_nfiles; i++) { 2284 if ((ff = dt->dt_ff[i]) == NULL) { 2285 continue; 2286 } 2287 if ((fp = atomic_load_consume(&ff->ff_file)) == 2288 NULL) { 2289 continue; 2290 } 2291 2292 if ((op == KERN_FILE_BYFILE) && 2293 (fp->f_marker == sysctl_file_marker)) { 2294 continue; 2295 } 2296 if (len >= elem_size && elem_count > 0) { 2297 mutex_enter(&fp->f_lock); 2298 fill_file2(&kf, fp, ff, i, p->p_pid); 2299 mutex_exit(&fp->f_lock); 2300 mutex_exit(&fd->fd_lock); 2301 error = sysctl_copyout(l, 2302 &kf, dp, out_size); 2303 mutex_enter(&fd->fd_lock); 2304 if (error) 2305 break; 2306 dp += elem_size; 2307 len -= elem_size; 2308 } 2309 if (op == KERN_FILE_BYFILE) 2310 fp->f_marker = sysctl_file_marker; 2311 needed += elem_size; 2312 if (elem_count > 0 && elem_count != INT_MAX) 2313 elem_count--; 2314 } 2315 mutex_exit(&fd->fd_lock); 2316 2317 /* 2318 * Release reference to process. 2319 */ 2320 mutex_enter(&proc_lock); 2321 rw_exit(&p->p_reflock); 2322 } 2323 if (op == KERN_FILE_BYFILE) { 2324 sysctl_file_marker++; 2325 2326 /* Reset all markers if wrapped. */ 2327 if (sysctl_file_marker == 0) { 2328 sysctl_file_marker_reset(); 2329 sysctl_file_marker++; 2330 } 2331 } 2332 mutex_exit(&proc_lock); 2333 if (op == KERN_FILE_BYFILE) 2334 mutex_exit(&sysctl_file_marker_lock); 2335 sysctl_relock(); 2336 break; 2337 default: 2338 return EINVAL; 2339 } 2340 2341 if (oldp == NULL) 2342 needed += KERN_FILESLOP * elem_size; 2343 *oldlenp = needed; 2344 2345 return error; 2346 } 2347 2348 static void 2349 fill_file(struct file *fp, const struct file *fpsrc) 2350 { 2351 const bool allowaddr = get_expose_address(curproc); 2352 2353 memset(fp, 0, sizeof(*fp)); 2354 2355 fp->f_offset = fpsrc->f_offset; 2356 COND_SET_PTR(fp->f_cred, fpsrc->f_cred, allowaddr); 2357 COND_SET_CPTR(fp->f_ops, fpsrc->f_ops, allowaddr); 2358 COND_SET_STRUCT(fp->f_undata, fpsrc->f_undata, allowaddr); 2359 COND_SET_STRUCT(fp->f_list, fpsrc->f_list, allowaddr); 2360 fp->f_flag = fpsrc->f_flag; 2361 fp->f_marker = fpsrc->f_marker; 2362 fp->f_type = fpsrc->f_type; 2363 fp->f_advice = fpsrc->f_advice; 2364 fp->f_count = fpsrc->f_count; 2365 fp->f_msgcount = fpsrc->f_msgcount; 2366 fp->f_unpcount = fpsrc->f_unpcount; 2367 COND_SET_STRUCT(fp->f_unplist, fpsrc->f_unplist, allowaddr); 2368 } 2369 2370 static void 2371 fill_file2(struct kinfo_file *kp, const file_t *fp, const fdfile_t *ff, 2372 int i, pid_t pid) 2373 { 2374 const bool allowaddr = get_expose_address(curproc); 2375 2376 memset(kp, 0, sizeof(*kp)); 2377 2378 COND_SET_VALUE(kp->ki_fileaddr, PTRTOUINT64(fp), allowaddr); 2379 kp->ki_flag = fp->f_flag; 2380 kp->ki_iflags = 0; 2381 kp->ki_ftype = fp->f_type; 2382 kp->ki_count = fp->f_count; 2383 kp->ki_msgcount = fp->f_msgcount; 2384 COND_SET_VALUE(kp->ki_fucred, PTRTOUINT64(fp->f_cred), allowaddr); 2385 kp->ki_fuid = kauth_cred_geteuid(fp->f_cred); 2386 kp->ki_fgid = kauth_cred_getegid(fp->f_cred); 2387 COND_SET_VALUE(kp->ki_fops, PTRTOUINT64(fp->f_ops), allowaddr); 2388 kp->ki_foffset = fp->f_offset; 2389 COND_SET_VALUE(kp->ki_fdata, PTRTOUINT64(fp->f_data), allowaddr); 2390 2391 /* vnode information to glue this file to something */ 2392 if (fp->f_type == DTYPE_VNODE) { 2393 struct vnode *vp = fp->f_vnode; 2394 2395 COND_SET_VALUE(kp->ki_vun, PTRTOUINT64(vp->v_un.vu_socket), 2396 allowaddr); 2397 kp->ki_vsize = vp->v_size; 2398 kp->ki_vtype = vp->v_type; 2399 kp->ki_vtag = vp->v_tag; 2400 COND_SET_VALUE(kp->ki_vdata, PTRTOUINT64(vp->v_data), 2401 allowaddr); 2402 } 2403 2404 /* process information when retrieved via KERN_FILE_BYPID */ 2405 if (ff != NULL) { 2406 kp->ki_pid = pid; 2407 kp->ki_fd = i; 2408 kp->ki_ofileflags = ff->ff_exclose; 2409 kp->ki_usecount = ff->ff_refcnt; 2410 } 2411 } 2412