Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/file.c
   4 *
   5 *  Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
   6 *
   7 *  Manage the dynamic fd arrays in the process files_struct.
   8 */
   9
  10#include <linux/syscalls.h>
  11#include <linux/export.h>
  12#include <linux/fs.h>
  13#include <linux/kernel.h>
  14#include <linux/mm.h>
  15#include <linux/sched/signal.h>
  16#include <linux/slab.h>
  17#include <linux/file.h>
  18#include <linux/fdtable.h>
  19#include <linux/bitops.h>
  20#include <linux/spinlock.h>
  21#include <linux/rcupdate.h>
  22#include <linux/close_range.h>
  23#include <net/sock.h>
  24
  25unsigned int sysctl_nr_open __read_mostly = 1024*1024;
  26unsigned int sysctl_nr_open_min = BITS_PER_LONG;
  27/* our min() is unusable in constant expressions ;-/ */
  28#define __const_min(x, y) ((x) < (y) ? (x) : (y))
  29unsigned int sysctl_nr_open_max =
  30	__const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
  31
  32static void __free_fdtable(struct fdtable *fdt)
  33{
  34	kvfree(fdt->fd);
  35	kvfree(fdt->open_fds);
  36	kfree(fdt);
  37}
  38
  39static void free_fdtable_rcu(struct rcu_head *rcu)
  40{
  41	__free_fdtable(container_of(rcu, struct fdtable, rcu));
  42}
  43
  44#define BITBIT_NR(nr)	BITS_TO_LONGS(BITS_TO_LONGS(nr))
  45#define BITBIT_SIZE(nr)	(BITBIT_NR(nr) * sizeof(long))
  46
  47/*
  48 * Copy 'count' fd bits from the old table to the new table and clear the extra
  49 * space if any.  This does not copy the file pointers.  Called with the files
  50 * spinlock held for write.
  51 */
  52static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
  53			    unsigned int count)
  54{
  55	unsigned int cpy, set;
  56
  57	cpy = count / BITS_PER_BYTE;
  58	set = (nfdt->max_fds - count) / BITS_PER_BYTE;
  59	memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
  60	memset((char *)nfdt->open_fds + cpy, 0, set);
  61	memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
  62	memset((char *)nfdt->close_on_exec + cpy, 0, set);
  63
  64	cpy = BITBIT_SIZE(count);
  65	set = BITBIT_SIZE(nfdt->max_fds) - cpy;
  66	memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
  67	memset((char *)nfdt->full_fds_bits + cpy, 0, set);
  68}
  69
  70/*
  71 * Copy all file descriptors from the old table to the new, expanded table and
  72 * clear the extra space.  Called with the files spinlock held for write.
  73 */
  74static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
  75{
  76	size_t cpy, set;
  77
  78	BUG_ON(nfdt->max_fds < ofdt->max_fds);
  79
  80	cpy = ofdt->max_fds * sizeof(struct file *);
  81	set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
  82	memcpy(nfdt->fd, ofdt->fd, cpy);
  83	memset((char *)nfdt->fd + cpy, 0, set);
  84
  85	copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
  86}
  87
  88static struct fdtable * alloc_fdtable(unsigned int nr)
  89{
  90	struct fdtable *fdt;
  91	void *data;
  92
  93	/*
  94	 * Figure out how many fds we actually want to support in this fdtable.
  95	 * Allocation steps are keyed to the size of the fdarray, since it
  96	 * grows far faster than any of the other dynamic data. We try to fit
  97	 * the fdarray into comfortable page-tuned chunks: starting at 1024B
  98	 * and growing in powers of two from there on.
  99	 */
 100	nr /= (1024 / sizeof(struct file *));
 101	nr = roundup_pow_of_two(nr + 1);
 102	nr *= (1024 / sizeof(struct file *));
 103	/*
 104	 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
 105	 * had been set lower between the check in expand_files() and here.  Deal
 106	 * with that in caller, it's cheaper that way.
 107	 *
 108	 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
 109	 * bitmaps handling below becomes unpleasant, to put it mildly...
 110	 */
 111	if (unlikely(nr > sysctl_nr_open))
 112		nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
 113
 114	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
 115	if (!fdt)
 116		goto out;
 117	fdt->max_fds = nr;
 118	data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
 119	if (!data)
 120		goto out_fdt;
 121	fdt->fd = data;
 122
 123	data = kvmalloc(max_t(size_t,
 124				 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
 125				 GFP_KERNEL_ACCOUNT);
 126	if (!data)
 127		goto out_arr;
 128	fdt->open_fds = data;
 129	data += nr / BITS_PER_BYTE;
 130	fdt->close_on_exec = data;
 131	data += nr / BITS_PER_BYTE;
 132	fdt->full_fds_bits = data;
 133
 134	return fdt;
 135
 136out_arr:
 137	kvfree(fdt->fd);
 138out_fdt:
 139	kfree(fdt);
 140out:
 141	return NULL;
 142}
 143
 144/*
 145 * Expand the file descriptor table.
 146 * This function will allocate a new fdtable and both fd array and fdset, of
 147 * the given size.
 148 * Return <0 error code on error; 1 on successful completion.
 149 * The files->file_lock should be held on entry, and will be held on exit.
 150 */
 151static int expand_fdtable(struct files_struct *files, unsigned int nr)
 152	__releases(files->file_lock)
 153	__acquires(files->file_lock)
 154{
 155	struct fdtable *new_fdt, *cur_fdt;
 156
 157	spin_unlock(&files->file_lock);
 158	new_fdt = alloc_fdtable(nr);
 159
 160	/* make sure all __fd_install() have seen resize_in_progress
 161	 * or have finished their rcu_read_lock_sched() section.
 162	 */
 163	if (atomic_read(&files->count) > 1)
 164		synchronize_rcu();
 165
 166	spin_lock(&files->file_lock);
 167	if (!new_fdt)
 168		return -ENOMEM;
 169	/*
 170	 * extremely unlikely race - sysctl_nr_open decreased between the check in
 171	 * caller and alloc_fdtable().  Cheaper to catch it here...
 172	 */
 173	if (unlikely(new_fdt->max_fds <= nr)) {
 174		__free_fdtable(new_fdt);
 175		return -EMFILE;
 176	}
 177	cur_fdt = files_fdtable(files);
 178	BUG_ON(nr < cur_fdt->max_fds);
 179	copy_fdtable(new_fdt, cur_fdt);
 180	rcu_assign_pointer(files->fdt, new_fdt);
 181	if (cur_fdt != &files->fdtab)
 182		call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
 183	/* coupled with smp_rmb() in __fd_install() */
 184	smp_wmb();
 185	return 1;
 186}
 187
 188/*
 189 * Expand files.
 190 * This function will expand the file structures, if the requested size exceeds
 191 * the current capacity and there is room for expansion.
 192 * Return <0 error code on error; 0 when nothing done; 1 when files were
 193 * expanded and execution may have blocked.
 194 * The files->file_lock should be held on entry, and will be held on exit.
 195 */
 196static int expand_files(struct files_struct *files, unsigned int nr)
 197	__releases(files->file_lock)
 198	__acquires(files->file_lock)
 199{
 200	struct fdtable *fdt;
 201	int expanded = 0;
 202
 203repeat:
 204	fdt = files_fdtable(files);
 205
 206	/* Do we need to expand? */
 207	if (nr < fdt->max_fds)
 208		return expanded;
 209
 210	/* Can we expand? */
 211	if (nr >= sysctl_nr_open)
 212		return -EMFILE;
 213
 214	if (unlikely(files->resize_in_progress)) {
 215		spin_unlock(&files->file_lock);
 216		expanded = 1;
 217		wait_event(files->resize_wait, !files->resize_in_progress);
 218		spin_lock(&files->file_lock);
 219		goto repeat;
 220	}
 221
 222	/* All good, so we try */
 223	files->resize_in_progress = true;
 224	expanded = expand_fdtable(files, nr);
 225	files->resize_in_progress = false;
 226
 227	wake_up_all(&files->resize_wait);
 228	return expanded;
 229}
 230
 231static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
 232{
 233	__set_bit(fd, fdt->close_on_exec);
 234}
 235
 236static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
 237{
 238	if (test_bit(fd, fdt->close_on_exec))
 239		__clear_bit(fd, fdt->close_on_exec);
 240}
 241
 242static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
 243{
 244	__set_bit(fd, fdt->open_fds);
 245	fd /= BITS_PER_LONG;
 246	if (!~fdt->open_fds[fd])
 247		__set_bit(fd, fdt->full_fds_bits);
 248}
 249
 250static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
 251{
 252	__clear_bit(fd, fdt->open_fds);
 253	__clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
 254}
 255
 256static unsigned int count_open_files(struct fdtable *fdt)
 257{
 258	unsigned int size = fdt->max_fds;
 259	unsigned int i;
 260
 261	/* Find the last open fd */
 262	for (i = size / BITS_PER_LONG; i > 0; ) {
 263		if (fdt->open_fds[--i])
 264			break;
 265	}
 266	i = (i + 1) * BITS_PER_LONG;
 267	return i;
 268}
 269
 270static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds)
 271{
 272	unsigned int count;
 273
 274	count = count_open_files(fdt);
 275	if (max_fds < NR_OPEN_DEFAULT)
 276		max_fds = NR_OPEN_DEFAULT;
 277	return min(count, max_fds);
 278}
 279
 280/*
 281 * Allocate a new files structure and copy contents from the
 282 * passed in files structure.
 283 * errorp will be valid only when the returned files_struct is NULL.
 284 */
 285struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
 286{
 287	struct files_struct *newf;
 288	struct file **old_fds, **new_fds;
 289	unsigned int open_files, i;
 290	struct fdtable *old_fdt, *new_fdt;
 291
 292	*errorp = -ENOMEM;
 293	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 294	if (!newf)
 295		goto out;
 296
 297	atomic_set(&newf->count, 1);
 298
 299	spin_lock_init(&newf->file_lock);
 300	newf->resize_in_progress = false;
 301	init_waitqueue_head(&newf->resize_wait);
 302	newf->next_fd = 0;
 303	new_fdt = &newf->fdtab;
 304	new_fdt->max_fds = NR_OPEN_DEFAULT;
 305	new_fdt->close_on_exec = newf->close_on_exec_init;
 306	new_fdt->open_fds = newf->open_fds_init;
 307	new_fdt->full_fds_bits = newf->full_fds_bits_init;
 308	new_fdt->fd = &newf->fd_array[0];
 309
 310	spin_lock(&oldf->file_lock);
 311	old_fdt = files_fdtable(oldf);
 312	open_files = sane_fdtable_size(old_fdt, max_fds);
 313
 314	/*
 315	 * Check whether we need to allocate a larger fd array and fd set.
 316	 */
 317	while (unlikely(open_files > new_fdt->max_fds)) {
 318		spin_unlock(&oldf->file_lock);
 319
 320		if (new_fdt != &newf->fdtab)
 321			__free_fdtable(new_fdt);
 322
 323		new_fdt = alloc_fdtable(open_files - 1);
 324		if (!new_fdt) {
 325			*errorp = -ENOMEM;
 326			goto out_release;
 327		}
 328
 329		/* beyond sysctl_nr_open; nothing to do */
 330		if (unlikely(new_fdt->max_fds < open_files)) {
 331			__free_fdtable(new_fdt);
 332			*errorp = -EMFILE;
 333			goto out_release;
 334		}
 335
 336		/*
 337		 * Reacquire the oldf lock and a pointer to its fd table
 338		 * who knows it may have a new bigger fd table. We need
 339		 * the latest pointer.
 340		 */
 341		spin_lock(&oldf->file_lock);
 342		old_fdt = files_fdtable(oldf);
 343		open_files = sane_fdtable_size(old_fdt, max_fds);
 344	}
 345
 346	copy_fd_bitmaps(new_fdt, old_fdt, open_files);
 347
 348	old_fds = old_fdt->fd;
 349	new_fds = new_fdt->fd;
 350
 351	for (i = open_files; i != 0; i--) {
 352		struct file *f = *old_fds++;
 353		if (f) {
 354			get_file(f);
 355		} else {
 356			/*
 357			 * The fd may be claimed in the fd bitmap but not yet
 358			 * instantiated in the files array if a sibling thread
 359			 * is partway through open().  So make sure that this
 360			 * fd is available to the new process.
 361			 */
 362			__clear_open_fd(open_files - i, new_fdt);
 363		}
 364		rcu_assign_pointer(*new_fds++, f);
 365	}
 366	spin_unlock(&oldf->file_lock);
 367
 368	/* clear the remainder */
 369	memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
 370
 371	rcu_assign_pointer(newf->fdt, new_fdt);
 372
 373	return newf;
 374
 375out_release:
 376	kmem_cache_free(files_cachep, newf);
 377out:
 378	return NULL;
 379}
 380
 381static struct fdtable *close_files(struct files_struct * files)
 382{
 383	/*
 384	 * It is safe to dereference the fd table without RCU or
 385	 * ->file_lock because this is the last reference to the
 386	 * files structure.
 387	 */
 388	struct fdtable *fdt = rcu_dereference_raw(files->fdt);
 389	unsigned int i, j = 0;
 390
 391	for (;;) {
 392		unsigned long set;
 393		i = j * BITS_PER_LONG;
 394		if (i >= fdt->max_fds)
 395			break;
 396		set = fdt->open_fds[j++];
 397		while (set) {
 398			if (set & 1) {
 399				struct file * file = xchg(&fdt->fd[i], NULL);
 400				if (file) {
 401					filp_close(file, files);
 402					cond_resched();
 403				}
 404			}
 405			i++;
 406			set >>= 1;
 407		}
 408	}
 409
 410	return fdt;
 411}
 412
 413struct files_struct *get_files_struct(struct task_struct *task)
 414{
 415	struct files_struct *files;
 416
 417	task_lock(task);
 418	files = task->files;
 419	if (files)
 420		atomic_inc(&files->count);
 421	task_unlock(task);
 422
 423	return files;
 424}
 425
 426void put_files_struct(struct files_struct *files)
 427{
 428	if (atomic_dec_and_test(&files->count)) {
 429		struct fdtable *fdt = close_files(files);
 430
 431		/* free the arrays if they are not embedded */
 432		if (fdt != &files->fdtab)
 433			__free_fdtable(fdt);
 434		kmem_cache_free(files_cachep, files);
 435	}
 436}
 437
 438void reset_files_struct(struct files_struct *files)
 439{
 440	struct task_struct *tsk = current;
 441	struct files_struct *old;
 442
 443	old = tsk->files;
 444	task_lock(tsk);
 445	tsk->files = files;
 446	task_unlock(tsk);
 447	put_files_struct(old);
 448}
 449
 450void exit_files(struct task_struct *tsk)
 451{
 452	struct files_struct * files = tsk->files;
 453
 454	if (files) {
 455		task_lock(tsk);
 456		tsk->files = NULL;
 457		task_unlock(tsk);
 458		put_files_struct(files);
 459	}
 460}
 461
 462struct files_struct init_files = {
 463	.count		= ATOMIC_INIT(1),
 464	.fdt		= &init_files.fdtab,
 465	.fdtab		= {
 466		.max_fds	= NR_OPEN_DEFAULT,
 467		.fd		= &init_files.fd_array[0],
 468		.close_on_exec	= init_files.close_on_exec_init,
 469		.open_fds	= init_files.open_fds_init,
 470		.full_fds_bits	= init_files.full_fds_bits_init,
 471	},
 472	.file_lock	= __SPIN_LOCK_UNLOCKED(init_files.file_lock),
 473	.resize_wait	= __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
 474};
 475
 476static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
 477{
 478	unsigned int maxfd = fdt->max_fds;
 479	unsigned int maxbit = maxfd / BITS_PER_LONG;
 480	unsigned int bitbit = start / BITS_PER_LONG;
 481
 482	bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
 483	if (bitbit > maxfd)
 484		return maxfd;
 485	if (bitbit > start)
 486		start = bitbit;
 487	return find_next_zero_bit(fdt->open_fds, maxfd, start);
 488}
 489
 490/*
 491 * allocate a file descriptor, mark it busy.
 492 */
 493int __alloc_fd(struct files_struct *files,
 494	       unsigned start, unsigned end, unsigned flags)
 495{
 496	unsigned int fd;
 497	int error;
 498	struct fdtable *fdt;
 499
 500	spin_lock(&files->file_lock);
 501repeat:
 502	fdt = files_fdtable(files);
 503	fd = start;
 504	if (fd < files->next_fd)
 505		fd = files->next_fd;
 506
 507	if (fd < fdt->max_fds)
 508		fd = find_next_fd(fdt, fd);
 509
 510	/*
 511	 * N.B. For clone tasks sharing a files structure, this test
 512	 * will limit the total number of files that can be opened.
 513	 */
 514	error = -EMFILE;
 515	if (fd >= end)
 516		goto out;
 517
 518	error = expand_files(files, fd);
 519	if (error < 0)
 520		goto out;
 521
 522	/*
 523	 * If we needed to expand the fs array we
 524	 * might have blocked - try again.
 525	 */
 526	if (error)
 527		goto repeat;
 528
 529	if (start <= files->next_fd)
 530		files->next_fd = fd + 1;
 531
 532	__set_open_fd(fd, fdt);
 533	if (flags & O_CLOEXEC)
 534		__set_close_on_exec(fd, fdt);
 535	else
 536		__clear_close_on_exec(fd, fdt);
 537	error = fd;
 538#if 1
 539	/* Sanity check */
 540	if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
 541		printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
 542		rcu_assign_pointer(fdt->fd[fd], NULL);
 543	}
 544#endif
 545
 546out:
 547	spin_unlock(&files->file_lock);
 548	return error;
 549}
 550
 551static int alloc_fd(unsigned start, unsigned flags)
 552{
 553	return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
 554}
 555
 556int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
 557{
 558	return __alloc_fd(current->files, 0, nofile, flags);
 559}
 560
 561int get_unused_fd_flags(unsigned flags)
 562{
 563	return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
 564}
 565EXPORT_SYMBOL(get_unused_fd_flags);
 566
 567static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 568{
 569	struct fdtable *fdt = files_fdtable(files);
 570	__clear_open_fd(fd, fdt);
 571	if (fd < files->next_fd)
 572		files->next_fd = fd;
 573}
 574
 575void put_unused_fd(unsigned int fd)
 576{
 577	struct files_struct *files = current->files;
 578	spin_lock(&files->file_lock);
 579	__put_unused_fd(files, fd);
 580	spin_unlock(&files->file_lock);
 581}
 582
 583EXPORT_SYMBOL(put_unused_fd);
 584
 585/*
 586 * Install a file pointer in the fd array.
 587 *
 588 * The VFS is full of places where we drop the files lock between
 589 * setting the open_fds bitmap and installing the file in the file
 590 * array.  At any such point, we are vulnerable to a dup2() race
 591 * installing a file in the array before us.  We need to detect this and
 592 * fput() the struct file we are about to overwrite in this case.
 593 *
 594 * It should never happen - if we allow dup2() do it, _really_ bad things
 595 * will follow.
 596 *
 597 * NOTE: __fd_install() variant is really, really low-level; don't
 598 * use it unless you are forced to by truly lousy API shoved down
 599 * your throat.  'files' *MUST* be either current->files or obtained
 600 * by get_files_struct(current) done by whoever had given it to you,
 601 * or really bad things will happen.  Normally you want to use
 602 * fd_install() instead.
 603 */
 604
 605void __fd_install(struct files_struct *files, unsigned int fd,
 606		struct file *file)
 607{
 608	struct fdtable *fdt;
 609
 610	rcu_read_lock_sched();
 611
 612	if (unlikely(files->resize_in_progress)) {
 613		rcu_read_unlock_sched();
 614		spin_lock(&files->file_lock);
 615		fdt = files_fdtable(files);
 616		BUG_ON(fdt->fd[fd] != NULL);
 617		rcu_assign_pointer(fdt->fd[fd], file);
 618		spin_unlock(&files->file_lock);
 619		return;
 620	}
 621	/* coupled with smp_wmb() in expand_fdtable() */
 622	smp_rmb();
 623	fdt = rcu_dereference_sched(files->fdt);
 624	BUG_ON(fdt->fd[fd] != NULL);
 625	rcu_assign_pointer(fdt->fd[fd], file);
 626	rcu_read_unlock_sched();
 627}
 628
 629/*
 630 * This consumes the "file" refcount, so callers should treat it
 631 * as if they had called fput(file).
 632 */
 633void fd_install(unsigned int fd, struct file *file)
 634{
 635	__fd_install(current->files, fd, file);
 636}
 637
 638EXPORT_SYMBOL(fd_install);
 639
 640static struct file *pick_file(struct files_struct *files, unsigned fd)
 
 
 
 641{
 642	struct file *file = NULL;
 643	struct fdtable *fdt;
 644
 645	spin_lock(&files->file_lock);
 646	fdt = files_fdtable(files);
 647	if (fd >= fdt->max_fds)
 648		goto out_unlock;
 649	file = fdt->fd[fd];
 650	if (!file)
 651		goto out_unlock;
 652	rcu_assign_pointer(fdt->fd[fd], NULL);
 653	__put_unused_fd(files, fd);
 
 
 654
 655out_unlock:
 656	spin_unlock(&files->file_lock);
 657	return file;
 658}
 659
 660/*
 661 * The same warnings as for __alloc_fd()/__fd_install() apply here...
 662 */
 663int __close_fd(struct files_struct *files, unsigned fd)
 664{
 665	struct file *file;
 666
 667	file = pick_file(files, fd);
 668	if (!file)
 669		return -EBADF;
 670
 671	return filp_close(file, files);
 672}
 673EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
 674
 675/**
 676 * __close_range() - Close all file descriptors in a given range.
 677 *
 678 * @fd:     starting file descriptor to close
 679 * @max_fd: last file descriptor to close
 680 *
 681 * This closes a range of file descriptors. All file descriptors
 682 * from @fd up to and including @max_fd are closed.
 683 */
 684int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
 685{
 686	unsigned int cur_max;
 687	struct task_struct *me = current;
 688	struct files_struct *cur_fds = me->files, *fds = NULL;
 689
 690	if (flags & ~CLOSE_RANGE_UNSHARE)
 691		return -EINVAL;
 692
 693	if (fd > max_fd)
 694		return -EINVAL;
 695
 696	rcu_read_lock();
 697	cur_max = files_fdtable(cur_fds)->max_fds;
 698	rcu_read_unlock();
 699
 700	/* cap to last valid index into fdtable */
 701	cur_max--;
 702
 703	if (flags & CLOSE_RANGE_UNSHARE) {
 704		int ret;
 705		unsigned int max_unshare_fds = NR_OPEN_MAX;
 706
 707		/*
 708		 * If the requested range is greater than the current maximum,
 709		 * we're closing everything so only copy all file descriptors
 710		 * beneath the lowest file descriptor.
 711		 */
 712		if (max_fd >= cur_max)
 713			max_unshare_fds = fd;
 714
 715		ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
 716		if (ret)
 717			return ret;
 718
 719		/*
 720		 * We used to share our file descriptor table, and have now
 721		 * created a private one, make sure we're using it below.
 722		 */
 723		if (fds)
 724			swap(cur_fds, fds);
 725	}
 726
 727	max_fd = min(max_fd, cur_max);
 728	while (fd <= max_fd) {
 729		struct file *file;
 730
 731		file = pick_file(cur_fds, fd++);
 732		if (!file)
 733			continue;
 734
 735		filp_close(file, cur_fds);
 736		cond_resched();
 737	}
 738
 739	if (fds) {
 740		/*
 741		 * We're done closing the files we were supposed to. Time to install
 742		 * the new file descriptor table and drop the old one.
 743		 */
 744		task_lock(me);
 745		me->files = cur_fds;
 746		task_unlock(me);
 747		put_files_struct(fds);
 748	}
 749
 750	return 0;
 751}
 752
 753/*
 754 * variant of __close_fd that gets a ref on the file for later fput.
 755 * The caller must ensure that filp_close() called on the file, and then
 756 * an fput().
 757 */
 758int __close_fd_get_file(unsigned int fd, struct file **res)
 759{
 760	struct files_struct *files = current->files;
 761	struct file *file;
 762	struct fdtable *fdt;
 763
 764	spin_lock(&files->file_lock);
 765	fdt = files_fdtable(files);
 766	if (fd >= fdt->max_fds)
 767		goto out_unlock;
 768	file = fdt->fd[fd];
 769	if (!file)
 770		goto out_unlock;
 771	rcu_assign_pointer(fdt->fd[fd], NULL);
 772	__put_unused_fd(files, fd);
 773	spin_unlock(&files->file_lock);
 774	get_file(file);
 775	*res = file;
 776	return 0;
 777
 778out_unlock:
 779	spin_unlock(&files->file_lock);
 780	*res = NULL;
 781	return -ENOENT;
 782}
 783
 784void do_close_on_exec(struct files_struct *files)
 785{
 786	unsigned i;
 787	struct fdtable *fdt;
 788
 789	/* exec unshares first */
 790	spin_lock(&files->file_lock);
 791	for (i = 0; ; i++) {
 792		unsigned long set;
 793		unsigned fd = i * BITS_PER_LONG;
 794		fdt = files_fdtable(files);
 795		if (fd >= fdt->max_fds)
 796			break;
 797		set = fdt->close_on_exec[i];
 798		if (!set)
 799			continue;
 800		fdt->close_on_exec[i] = 0;
 801		for ( ; set ; fd++, set >>= 1) {
 802			struct file *file;
 803			if (!(set & 1))
 804				continue;
 805			file = fdt->fd[fd];
 806			if (!file)
 807				continue;
 808			rcu_assign_pointer(fdt->fd[fd], NULL);
 809			__put_unused_fd(files, fd);
 810			spin_unlock(&files->file_lock);
 811			filp_close(file, files);
 812			cond_resched();
 813			spin_lock(&files->file_lock);
 814		}
 815
 816	}
 817	spin_unlock(&files->file_lock);
 818}
 819
 820static struct file *__fget_files(struct files_struct *files, unsigned int fd,
 821				 fmode_t mask, unsigned int refs)
 822{
 
 823	struct file *file;
 824
 825	rcu_read_lock();
 826loop:
 827	file = fcheck_files(files, fd);
 828	if (file) {
 829		/* File object ref couldn't be taken.
 830		 * dup2() atomicity guarantee is the reason
 831		 * we loop to catch the new file (or NULL pointer)
 832		 */
 833		if (file->f_mode & mask)
 834			file = NULL;
 835		else if (!get_file_rcu_many(file, refs))
 836			goto loop;
 837	}
 838	rcu_read_unlock();
 839
 840	return file;
 841}
 842
 843static inline struct file *__fget(unsigned int fd, fmode_t mask,
 844				  unsigned int refs)
 845{
 846	return __fget_files(current->files, fd, mask, refs);
 847}
 848
 849struct file *fget_many(unsigned int fd, unsigned int refs)
 850{
 851	return __fget(fd, FMODE_PATH, refs);
 852}
 853
 854struct file *fget(unsigned int fd)
 855{
 856	return __fget(fd, FMODE_PATH, 1);
 857}
 858EXPORT_SYMBOL(fget);
 859
 860struct file *fget_raw(unsigned int fd)
 861{
 862	return __fget(fd, 0, 1);
 863}
 864EXPORT_SYMBOL(fget_raw);
 865
 866struct file *fget_task(struct task_struct *task, unsigned int fd)
 867{
 868	struct file *file = NULL;
 869
 870	task_lock(task);
 871	if (task->files)
 872		file = __fget_files(task->files, fd, 0, 1);
 873	task_unlock(task);
 874
 875	return file;
 876}
 877
 878/*
 879 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 880 *
 881 * You can use this instead of fget if you satisfy all of the following
 882 * conditions:
 883 * 1) You must call fput_light before exiting the syscall and returning control
 884 *    to userspace (i.e. you cannot remember the returned struct file * after
 885 *    returning to userspace).
 886 * 2) You must not call filp_close on the returned struct file * in between
 887 *    calls to fget_light and fput_light.
 888 * 3) You must not clone the current task in between the calls to fget_light
 889 *    and fput_light.
 890 *
 891 * The fput_needed flag returned by fget_light should be passed to the
 892 * corresponding fput_light.
 893 */
 894static unsigned long __fget_light(unsigned int fd, fmode_t mask)
 895{
 896	struct files_struct *files = current->files;
 897	struct file *file;
 898
 899	if (atomic_read(&files->count) == 1) {
 900		file = __fcheck_files(files, fd);
 901		if (!file || unlikely(file->f_mode & mask))
 902			return 0;
 903		return (unsigned long)file;
 904	} else {
 905		file = __fget(fd, mask, 1);
 906		if (!file)
 907			return 0;
 908		return FDPUT_FPUT | (unsigned long)file;
 909	}
 910}
 911unsigned long __fdget(unsigned int fd)
 912{
 913	return __fget_light(fd, FMODE_PATH);
 914}
 915EXPORT_SYMBOL(__fdget);
 916
 917unsigned long __fdget_raw(unsigned int fd)
 918{
 919	return __fget_light(fd, 0);
 920}
 921
 922unsigned long __fdget_pos(unsigned int fd)
 923{
 924	unsigned long v = __fdget(fd);
 925	struct file *file = (struct file *)(v & ~3);
 926
 927	if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
 928		if (file_count(file) > 1) {
 929			v |= FDPUT_POS_UNLOCK;
 930			mutex_lock(&file->f_pos_lock);
 931		}
 932	}
 933	return v;
 934}
 935
 936void __f_unlock_pos(struct file *f)
 937{
 938	mutex_unlock(&f->f_pos_lock);
 939}
 940
 941/*
 942 * We only lock f_pos if we have threads or if the file might be
 943 * shared with another process. In both cases we'll have an elevated
 944 * file count (done either by fdget() or by fork()).
 945 */
 946
 947void set_close_on_exec(unsigned int fd, int flag)
 948{
 949	struct files_struct *files = current->files;
 950	struct fdtable *fdt;
 951	spin_lock(&files->file_lock);
 952	fdt = files_fdtable(files);
 953	if (flag)
 954		__set_close_on_exec(fd, fdt);
 955	else
 956		__clear_close_on_exec(fd, fdt);
 957	spin_unlock(&files->file_lock);
 958}
 959
 960bool get_close_on_exec(unsigned int fd)
 961{
 962	struct files_struct *files = current->files;
 963	struct fdtable *fdt;
 964	bool res;
 965	rcu_read_lock();
 966	fdt = files_fdtable(files);
 967	res = close_on_exec(fd, fdt);
 968	rcu_read_unlock();
 969	return res;
 970}
 971
 972static int do_dup2(struct files_struct *files,
 973	struct file *file, unsigned fd, unsigned flags)
 974__releases(&files->file_lock)
 975{
 976	struct file *tofree;
 977	struct fdtable *fdt;
 978
 979	/*
 980	 * We need to detect attempts to do dup2() over allocated but still
 981	 * not finished descriptor.  NB: OpenBSD avoids that at the price of
 982	 * extra work in their equivalent of fget() - they insert struct
 983	 * file immediately after grabbing descriptor, mark it larval if
 984	 * more work (e.g. actual opening) is needed and make sure that
 985	 * fget() treats larval files as absent.  Potentially interesting,
 986	 * but while extra work in fget() is trivial, locking implications
 987	 * and amount of surgery on open()-related paths in VFS are not.
 988	 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
 989	 * deadlocks in rather amusing ways, AFAICS.  All of that is out of
 990	 * scope of POSIX or SUS, since neither considers shared descriptor
 991	 * tables and this condition does not arise without those.
 992	 */
 993	fdt = files_fdtable(files);
 994	tofree = fdt->fd[fd];
 995	if (!tofree && fd_is_open(fd, fdt))
 996		goto Ebusy;
 997	get_file(file);
 998	rcu_assign_pointer(fdt->fd[fd], file);
 999	__set_open_fd(fd, fdt);
1000	if (flags & O_CLOEXEC)
1001		__set_close_on_exec(fd, fdt);
1002	else
1003		__clear_close_on_exec(fd, fdt);
1004	spin_unlock(&files->file_lock);
1005
1006	if (tofree)
1007		filp_close(tofree, files);
1008
1009	return fd;
1010
1011Ebusy:
1012	spin_unlock(&files->file_lock);
1013	return -EBUSY;
1014}
1015
1016int replace_fd(unsigned fd, struct file *file, unsigned flags)
1017{
1018	int err;
1019	struct files_struct *files = current->files;
1020
1021	if (!file)
1022		return __close_fd(files, fd);
1023
1024	if (fd >= rlimit(RLIMIT_NOFILE))
1025		return -EBADF;
1026
1027	spin_lock(&files->file_lock);
1028	err = expand_files(files, fd);
1029	if (unlikely(err < 0))
1030		goto out_unlock;
1031	return do_dup2(files, file, fd, flags);
1032
1033out_unlock:
1034	spin_unlock(&files->file_lock);
1035	return err;
1036}
1037
1038/**
1039 * __receive_fd() - Install received file into file descriptor table
1040 *
1041 * @fd: fd to install into (if negative, a new fd will be allocated)
1042 * @file: struct file that was received from another process
1043 * @ufd: __user pointer to write new fd number to
1044 * @o_flags: the O_* flags to apply to the new fd entry
1045 *
1046 * Installs a received file into the file descriptor table, with appropriate
1047 * checks and count updates. Optionally writes the fd number to userspace, if
1048 * @ufd is non-NULL.
1049 *
1050 * This helper handles its own reference counting of the incoming
1051 * struct file.
1052 *
1053 * Returns newly install fd or -ve on error.
1054 */
1055int __receive_fd(int fd, struct file *file, int __user *ufd, unsigned int o_flags)
1056{
1057	int new_fd;
1058	int error;
1059
1060	error = security_file_receive(file);
1061	if (error)
1062		return error;
1063
1064	if (fd < 0) {
1065		new_fd = get_unused_fd_flags(o_flags);
1066		if (new_fd < 0)
1067			return new_fd;
1068	} else {
1069		new_fd = fd;
1070	}
1071
1072	if (ufd) {
1073		error = put_user(new_fd, ufd);
1074		if (error) {
1075			if (fd < 0)
1076				put_unused_fd(new_fd);
1077			return error;
1078		}
1079	}
1080
1081	if (fd < 0) {
1082		fd_install(new_fd, get_file(file));
1083	} else {
1084		error = replace_fd(new_fd, file, o_flags);
1085		if (error)
1086			return error;
1087	}
1088
1089	/* Bump the sock usage counts, if any. */
1090	__receive_sock(file);
1091	return new_fd;
1092}
1093
1094static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1095{
1096	int err = -EBADF;
1097	struct file *file;
1098	struct files_struct *files = current->files;
1099
1100	if ((flags & ~O_CLOEXEC) != 0)
1101		return -EINVAL;
1102
1103	if (unlikely(oldfd == newfd))
1104		return -EINVAL;
1105
1106	if (newfd >= rlimit(RLIMIT_NOFILE))
1107		return -EBADF;
1108
1109	spin_lock(&files->file_lock);
1110	err = expand_files(files, newfd);
1111	file = fcheck(oldfd);
1112	if (unlikely(!file))
1113		goto Ebadf;
1114	if (unlikely(err < 0)) {
1115		if (err == -EMFILE)
1116			goto Ebadf;
1117		goto out_unlock;
1118	}
1119	return do_dup2(files, file, newfd, flags);
1120
1121Ebadf:
1122	err = -EBADF;
1123out_unlock:
1124	spin_unlock(&files->file_lock);
1125	return err;
1126}
1127
1128SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1129{
1130	return ksys_dup3(oldfd, newfd, flags);
1131}
1132
1133SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1134{
1135	if (unlikely(newfd == oldfd)) { /* corner case */
1136		struct files_struct *files = current->files;
1137		int retval = oldfd;
1138
1139		rcu_read_lock();
1140		if (!fcheck_files(files, oldfd))
1141			retval = -EBADF;
1142		rcu_read_unlock();
1143		return retval;
1144	}
1145	return ksys_dup3(oldfd, newfd, 0);
1146}
1147
1148SYSCALL_DEFINE1(dup, unsigned int, fildes)
1149{
1150	int ret = -EBADF;
1151	struct file *file = fget_raw(fildes);
1152
1153	if (file) {
1154		ret = get_unused_fd_flags(0);
1155		if (ret >= 0)
1156			fd_install(ret, file);
1157		else
1158			fput(file);
1159	}
1160	return ret;
 
 
 
 
 
1161}
1162
1163int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1164{
1165	int err;
1166	if (from >= rlimit(RLIMIT_NOFILE))
1167		return -EINVAL;
1168	err = alloc_fd(from, flags);
1169	if (err >= 0) {
1170		get_file(file);
1171		fd_install(err, file);
1172	}
1173	return err;
1174}
1175
1176int iterate_fd(struct files_struct *files, unsigned n,
1177		int (*f)(const void *, struct file *, unsigned),
1178		const void *p)
1179{
1180	struct fdtable *fdt;
1181	int res = 0;
1182	if (!files)
1183		return 0;
1184	spin_lock(&files->file_lock);
1185	for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1186		struct file *file;
1187		file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1188		if (!file)
1189			continue;
1190		res = f(p, file, n);
1191		if (res)
1192			break;
1193	}
1194	spin_unlock(&files->file_lock);
1195	return res;
1196}
1197EXPORT_SYMBOL(iterate_fd);
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/file.c
   4 *
   5 *  Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
   6 *
   7 *  Manage the dynamic fd arrays in the process files_struct.
   8 */
   9
  10#include <linux/syscalls.h>
  11#include <linux/export.h>
  12#include <linux/fs.h>
 
  13#include <linux/mm.h>
  14#include <linux/sched/signal.h>
  15#include <linux/slab.h>
  16#include <linux/file.h>
  17#include <linux/fdtable.h>
  18#include <linux/bitops.h>
  19#include <linux/spinlock.h>
  20#include <linux/rcupdate.h>
 
 
  21
  22unsigned int sysctl_nr_open __read_mostly = 1024*1024;
  23unsigned int sysctl_nr_open_min = BITS_PER_LONG;
  24/* our min() is unusable in constant expressions ;-/ */
  25#define __const_min(x, y) ((x) < (y) ? (x) : (y))
  26unsigned int sysctl_nr_open_max =
  27	__const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
  28
  29static void __free_fdtable(struct fdtable *fdt)
  30{
  31	kvfree(fdt->fd);
  32	kvfree(fdt->open_fds);
  33	kfree(fdt);
  34}
  35
  36static void free_fdtable_rcu(struct rcu_head *rcu)
  37{
  38	__free_fdtable(container_of(rcu, struct fdtable, rcu));
  39}
  40
  41#define BITBIT_NR(nr)	BITS_TO_LONGS(BITS_TO_LONGS(nr))
  42#define BITBIT_SIZE(nr)	(BITBIT_NR(nr) * sizeof(long))
  43
  44/*
  45 * Copy 'count' fd bits from the old table to the new table and clear the extra
  46 * space if any.  This does not copy the file pointers.  Called with the files
  47 * spinlock held for write.
  48 */
  49static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
  50			    unsigned int count)
  51{
  52	unsigned int cpy, set;
  53
  54	cpy = count / BITS_PER_BYTE;
  55	set = (nfdt->max_fds - count) / BITS_PER_BYTE;
  56	memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
  57	memset((char *)nfdt->open_fds + cpy, 0, set);
  58	memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
  59	memset((char *)nfdt->close_on_exec + cpy, 0, set);
  60
  61	cpy = BITBIT_SIZE(count);
  62	set = BITBIT_SIZE(nfdt->max_fds) - cpy;
  63	memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
  64	memset((char *)nfdt->full_fds_bits + cpy, 0, set);
  65}
  66
  67/*
  68 * Copy all file descriptors from the old table to the new, expanded table and
  69 * clear the extra space.  Called with the files spinlock held for write.
  70 */
  71static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
  72{
  73	unsigned int cpy, set;
  74
  75	BUG_ON(nfdt->max_fds < ofdt->max_fds);
  76
  77	cpy = ofdt->max_fds * sizeof(struct file *);
  78	set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
  79	memcpy(nfdt->fd, ofdt->fd, cpy);
  80	memset((char *)nfdt->fd + cpy, 0, set);
  81
  82	copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
  83}
  84
  85static struct fdtable * alloc_fdtable(unsigned int nr)
  86{
  87	struct fdtable *fdt;
  88	void *data;
  89
  90	/*
  91	 * Figure out how many fds we actually want to support in this fdtable.
  92	 * Allocation steps are keyed to the size of the fdarray, since it
  93	 * grows far faster than any of the other dynamic data. We try to fit
  94	 * the fdarray into comfortable page-tuned chunks: starting at 1024B
  95	 * and growing in powers of two from there on.
  96	 */
  97	nr /= (1024 / sizeof(struct file *));
  98	nr = roundup_pow_of_two(nr + 1);
  99	nr *= (1024 / sizeof(struct file *));
 100	/*
 101	 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
 102	 * had been set lower between the check in expand_files() and here.  Deal
 103	 * with that in caller, it's cheaper that way.
 104	 *
 105	 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
 106	 * bitmaps handling below becomes unpleasant, to put it mildly...
 107	 */
 108	if (unlikely(nr > sysctl_nr_open))
 109		nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
 110
 111	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
 112	if (!fdt)
 113		goto out;
 114	fdt->max_fds = nr;
 115	data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
 116	if (!data)
 117		goto out_fdt;
 118	fdt->fd = data;
 119
 120	data = kvmalloc(max_t(size_t,
 121				 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
 122				 GFP_KERNEL_ACCOUNT);
 123	if (!data)
 124		goto out_arr;
 125	fdt->open_fds = data;
 126	data += nr / BITS_PER_BYTE;
 127	fdt->close_on_exec = data;
 128	data += nr / BITS_PER_BYTE;
 129	fdt->full_fds_bits = data;
 130
 131	return fdt;
 132
 133out_arr:
 134	kvfree(fdt->fd);
 135out_fdt:
 136	kfree(fdt);
 137out:
 138	return NULL;
 139}
 140
 141/*
 142 * Expand the file descriptor table.
 143 * This function will allocate a new fdtable and both fd array and fdset, of
 144 * the given size.
 145 * Return <0 error code on error; 1 on successful completion.
 146 * The files->file_lock should be held on entry, and will be held on exit.
 147 */
 148static int expand_fdtable(struct files_struct *files, unsigned int nr)
 149	__releases(files->file_lock)
 150	__acquires(files->file_lock)
 151{
 152	struct fdtable *new_fdt, *cur_fdt;
 153
 154	spin_unlock(&files->file_lock);
 155	new_fdt = alloc_fdtable(nr);
 156
 157	/* make sure all __fd_install() have seen resize_in_progress
 158	 * or have finished their rcu_read_lock_sched() section.
 159	 */
 160	if (atomic_read(&files->count) > 1)
 161		synchronize_rcu();
 162
 163	spin_lock(&files->file_lock);
 164	if (!new_fdt)
 165		return -ENOMEM;
 166	/*
 167	 * extremely unlikely race - sysctl_nr_open decreased between the check in
 168	 * caller and alloc_fdtable().  Cheaper to catch it here...
 169	 */
 170	if (unlikely(new_fdt->max_fds <= nr)) {
 171		__free_fdtable(new_fdt);
 172		return -EMFILE;
 173	}
 174	cur_fdt = files_fdtable(files);
 175	BUG_ON(nr < cur_fdt->max_fds);
 176	copy_fdtable(new_fdt, cur_fdt);
 177	rcu_assign_pointer(files->fdt, new_fdt);
 178	if (cur_fdt != &files->fdtab)
 179		call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
 180	/* coupled with smp_rmb() in __fd_install() */
 181	smp_wmb();
 182	return 1;
 183}
 184
 185/*
 186 * Expand files.
 187 * This function will expand the file structures, if the requested size exceeds
 188 * the current capacity and there is room for expansion.
 189 * Return <0 error code on error; 0 when nothing done; 1 when files were
 190 * expanded and execution may have blocked.
 191 * The files->file_lock should be held on entry, and will be held on exit.
 192 */
 193static int expand_files(struct files_struct *files, unsigned int nr)
 194	__releases(files->file_lock)
 195	__acquires(files->file_lock)
 196{
 197	struct fdtable *fdt;
 198	int expanded = 0;
 199
 200repeat:
 201	fdt = files_fdtable(files);
 202
 203	/* Do we need to expand? */
 204	if (nr < fdt->max_fds)
 205		return expanded;
 206
 207	/* Can we expand? */
 208	if (nr >= sysctl_nr_open)
 209		return -EMFILE;
 210
 211	if (unlikely(files->resize_in_progress)) {
 212		spin_unlock(&files->file_lock);
 213		expanded = 1;
 214		wait_event(files->resize_wait, !files->resize_in_progress);
 215		spin_lock(&files->file_lock);
 216		goto repeat;
 217	}
 218
 219	/* All good, so we try */
 220	files->resize_in_progress = true;
 221	expanded = expand_fdtable(files, nr);
 222	files->resize_in_progress = false;
 223
 224	wake_up_all(&files->resize_wait);
 225	return expanded;
 226}
 227
 228static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
 229{
 230	__set_bit(fd, fdt->close_on_exec);
 231}
 232
 233static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
 234{
 235	if (test_bit(fd, fdt->close_on_exec))
 236		__clear_bit(fd, fdt->close_on_exec);
 237}
 238
 239static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
 240{
 241	__set_bit(fd, fdt->open_fds);
 242	fd /= BITS_PER_LONG;
 243	if (!~fdt->open_fds[fd])
 244		__set_bit(fd, fdt->full_fds_bits);
 245}
 246
 247static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
 248{
 249	__clear_bit(fd, fdt->open_fds);
 250	__clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
 251}
 252
 253static unsigned int count_open_files(struct fdtable *fdt)
 254{
 255	unsigned int size = fdt->max_fds;
 256	unsigned int i;
 257
 258	/* Find the last open fd */
 259	for (i = size / BITS_PER_LONG; i > 0; ) {
 260		if (fdt->open_fds[--i])
 261			break;
 262	}
 263	i = (i + 1) * BITS_PER_LONG;
 264	return i;
 265}
 266
 
 
 
 
 
 
 
 
 
 
 267/*
 268 * Allocate a new files structure and copy contents from the
 269 * passed in files structure.
 270 * errorp will be valid only when the returned files_struct is NULL.
 271 */
 272struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 273{
 274	struct files_struct *newf;
 275	struct file **old_fds, **new_fds;
 276	unsigned int open_files, i;
 277	struct fdtable *old_fdt, *new_fdt;
 278
 279	*errorp = -ENOMEM;
 280	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 281	if (!newf)
 282		goto out;
 283
 284	atomic_set(&newf->count, 1);
 285
 286	spin_lock_init(&newf->file_lock);
 287	newf->resize_in_progress = false;
 288	init_waitqueue_head(&newf->resize_wait);
 289	newf->next_fd = 0;
 290	new_fdt = &newf->fdtab;
 291	new_fdt->max_fds = NR_OPEN_DEFAULT;
 292	new_fdt->close_on_exec = newf->close_on_exec_init;
 293	new_fdt->open_fds = newf->open_fds_init;
 294	new_fdt->full_fds_bits = newf->full_fds_bits_init;
 295	new_fdt->fd = &newf->fd_array[0];
 296
 297	spin_lock(&oldf->file_lock);
 298	old_fdt = files_fdtable(oldf);
 299	open_files = count_open_files(old_fdt);
 300
 301	/*
 302	 * Check whether we need to allocate a larger fd array and fd set.
 303	 */
 304	while (unlikely(open_files > new_fdt->max_fds)) {
 305		spin_unlock(&oldf->file_lock);
 306
 307		if (new_fdt != &newf->fdtab)
 308			__free_fdtable(new_fdt);
 309
 310		new_fdt = alloc_fdtable(open_files - 1);
 311		if (!new_fdt) {
 312			*errorp = -ENOMEM;
 313			goto out_release;
 314		}
 315
 316		/* beyond sysctl_nr_open; nothing to do */
 317		if (unlikely(new_fdt->max_fds < open_files)) {
 318			__free_fdtable(new_fdt);
 319			*errorp = -EMFILE;
 320			goto out_release;
 321		}
 322
 323		/*
 324		 * Reacquire the oldf lock and a pointer to its fd table
 325		 * who knows it may have a new bigger fd table. We need
 326		 * the latest pointer.
 327		 */
 328		spin_lock(&oldf->file_lock);
 329		old_fdt = files_fdtable(oldf);
 330		open_files = count_open_files(old_fdt);
 331	}
 332
 333	copy_fd_bitmaps(new_fdt, old_fdt, open_files);
 334
 335	old_fds = old_fdt->fd;
 336	new_fds = new_fdt->fd;
 337
 338	for (i = open_files; i != 0; i--) {
 339		struct file *f = *old_fds++;
 340		if (f) {
 341			get_file(f);
 342		} else {
 343			/*
 344			 * The fd may be claimed in the fd bitmap but not yet
 345			 * instantiated in the files array if a sibling thread
 346			 * is partway through open().  So make sure that this
 347			 * fd is available to the new process.
 348			 */
 349			__clear_open_fd(open_files - i, new_fdt);
 350		}
 351		rcu_assign_pointer(*new_fds++, f);
 352	}
 353	spin_unlock(&oldf->file_lock);
 354
 355	/* clear the remainder */
 356	memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
 357
 358	rcu_assign_pointer(newf->fdt, new_fdt);
 359
 360	return newf;
 361
 362out_release:
 363	kmem_cache_free(files_cachep, newf);
 364out:
 365	return NULL;
 366}
 367
 368static struct fdtable *close_files(struct files_struct * files)
 369{
 370	/*
 371	 * It is safe to dereference the fd table without RCU or
 372	 * ->file_lock because this is the last reference to the
 373	 * files structure.
 374	 */
 375	struct fdtable *fdt = rcu_dereference_raw(files->fdt);
 376	unsigned int i, j = 0;
 377
 378	for (;;) {
 379		unsigned long set;
 380		i = j * BITS_PER_LONG;
 381		if (i >= fdt->max_fds)
 382			break;
 383		set = fdt->open_fds[j++];
 384		while (set) {
 385			if (set & 1) {
 386				struct file * file = xchg(&fdt->fd[i], NULL);
 387				if (file) {
 388					filp_close(file, files);
 389					cond_resched();
 390				}
 391			}
 392			i++;
 393			set >>= 1;
 394		}
 395	}
 396
 397	return fdt;
 398}
 399
 400struct files_struct *get_files_struct(struct task_struct *task)
 401{
 402	struct files_struct *files;
 403
 404	task_lock(task);
 405	files = task->files;
 406	if (files)
 407		atomic_inc(&files->count);
 408	task_unlock(task);
 409
 410	return files;
 411}
 412
 413void put_files_struct(struct files_struct *files)
 414{
 415	if (atomic_dec_and_test(&files->count)) {
 416		struct fdtable *fdt = close_files(files);
 417
 418		/* free the arrays if they are not embedded */
 419		if (fdt != &files->fdtab)
 420			__free_fdtable(fdt);
 421		kmem_cache_free(files_cachep, files);
 422	}
 423}
 424
 425void reset_files_struct(struct files_struct *files)
 426{
 427	struct task_struct *tsk = current;
 428	struct files_struct *old;
 429
 430	old = tsk->files;
 431	task_lock(tsk);
 432	tsk->files = files;
 433	task_unlock(tsk);
 434	put_files_struct(old);
 435}
 436
 437void exit_files(struct task_struct *tsk)
 438{
 439	struct files_struct * files = tsk->files;
 440
 441	if (files) {
 442		task_lock(tsk);
 443		tsk->files = NULL;
 444		task_unlock(tsk);
 445		put_files_struct(files);
 446	}
 447}
 448
 449struct files_struct init_files = {
 450	.count		= ATOMIC_INIT(1),
 451	.fdt		= &init_files.fdtab,
 452	.fdtab		= {
 453		.max_fds	= NR_OPEN_DEFAULT,
 454		.fd		= &init_files.fd_array[0],
 455		.close_on_exec	= init_files.close_on_exec_init,
 456		.open_fds	= init_files.open_fds_init,
 457		.full_fds_bits	= init_files.full_fds_bits_init,
 458	},
 459	.file_lock	= __SPIN_LOCK_UNLOCKED(init_files.file_lock),
 460	.resize_wait	= __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
 461};
 462
 463static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
 464{
 465	unsigned int maxfd = fdt->max_fds;
 466	unsigned int maxbit = maxfd / BITS_PER_LONG;
 467	unsigned int bitbit = start / BITS_PER_LONG;
 468
 469	bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
 470	if (bitbit > maxfd)
 471		return maxfd;
 472	if (bitbit > start)
 473		start = bitbit;
 474	return find_next_zero_bit(fdt->open_fds, maxfd, start);
 475}
 476
 477/*
 478 * allocate a file descriptor, mark it busy.
 479 */
 480int __alloc_fd(struct files_struct *files,
 481	       unsigned start, unsigned end, unsigned flags)
 482{
 483	unsigned int fd;
 484	int error;
 485	struct fdtable *fdt;
 486
 487	spin_lock(&files->file_lock);
 488repeat:
 489	fdt = files_fdtable(files);
 490	fd = start;
 491	if (fd < files->next_fd)
 492		fd = files->next_fd;
 493
 494	if (fd < fdt->max_fds)
 495		fd = find_next_fd(fdt, fd);
 496
 497	/*
 498	 * N.B. For clone tasks sharing a files structure, this test
 499	 * will limit the total number of files that can be opened.
 500	 */
 501	error = -EMFILE;
 502	if (fd >= end)
 503		goto out;
 504
 505	error = expand_files(files, fd);
 506	if (error < 0)
 507		goto out;
 508
 509	/*
 510	 * If we needed to expand the fs array we
 511	 * might have blocked - try again.
 512	 */
 513	if (error)
 514		goto repeat;
 515
 516	if (start <= files->next_fd)
 517		files->next_fd = fd + 1;
 518
 519	__set_open_fd(fd, fdt);
 520	if (flags & O_CLOEXEC)
 521		__set_close_on_exec(fd, fdt);
 522	else
 523		__clear_close_on_exec(fd, fdt);
 524	error = fd;
 525#if 1
 526	/* Sanity check */
 527	if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
 528		printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
 529		rcu_assign_pointer(fdt->fd[fd], NULL);
 530	}
 531#endif
 532
 533out:
 534	spin_unlock(&files->file_lock);
 535	return error;
 536}
 537
 538static int alloc_fd(unsigned start, unsigned flags)
 539{
 540	return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
 541}
 542
 
 
 
 
 
 543int get_unused_fd_flags(unsigned flags)
 544{
 545	return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
 546}
 547EXPORT_SYMBOL(get_unused_fd_flags);
 548
 549static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 550{
 551	struct fdtable *fdt = files_fdtable(files);
 552	__clear_open_fd(fd, fdt);
 553	if (fd < files->next_fd)
 554		files->next_fd = fd;
 555}
 556
 557void put_unused_fd(unsigned int fd)
 558{
 559	struct files_struct *files = current->files;
 560	spin_lock(&files->file_lock);
 561	__put_unused_fd(files, fd);
 562	spin_unlock(&files->file_lock);
 563}
 564
 565EXPORT_SYMBOL(put_unused_fd);
 566
 567/*
 568 * Install a file pointer in the fd array.
 569 *
 570 * The VFS is full of places where we drop the files lock between
 571 * setting the open_fds bitmap and installing the file in the file
 572 * array.  At any such point, we are vulnerable to a dup2() race
 573 * installing a file in the array before us.  We need to detect this and
 574 * fput() the struct file we are about to overwrite in this case.
 575 *
 576 * It should never happen - if we allow dup2() do it, _really_ bad things
 577 * will follow.
 578 *
 579 * NOTE: __fd_install() variant is really, really low-level; don't
 580 * use it unless you are forced to by truly lousy API shoved down
 581 * your throat.  'files' *MUST* be either current->files or obtained
 582 * by get_files_struct(current) done by whoever had given it to you,
 583 * or really bad things will happen.  Normally you want to use
 584 * fd_install() instead.
 585 */
 586
 587void __fd_install(struct files_struct *files, unsigned int fd,
 588		struct file *file)
 589{
 590	struct fdtable *fdt;
 591
 592	rcu_read_lock_sched();
 593
 594	if (unlikely(files->resize_in_progress)) {
 595		rcu_read_unlock_sched();
 596		spin_lock(&files->file_lock);
 597		fdt = files_fdtable(files);
 598		BUG_ON(fdt->fd[fd] != NULL);
 599		rcu_assign_pointer(fdt->fd[fd], file);
 600		spin_unlock(&files->file_lock);
 601		return;
 602	}
 603	/* coupled with smp_wmb() in expand_fdtable() */
 604	smp_rmb();
 605	fdt = rcu_dereference_sched(files->fdt);
 606	BUG_ON(fdt->fd[fd] != NULL);
 607	rcu_assign_pointer(fdt->fd[fd], file);
 608	rcu_read_unlock_sched();
 609}
 610
 
 
 
 
 611void fd_install(unsigned int fd, struct file *file)
 612{
 613	__fd_install(current->files, fd, file);
 614}
 615
 616EXPORT_SYMBOL(fd_install);
 617
 618/*
 619 * The same warnings as for __alloc_fd()/__fd_install() apply here...
 620 */
 621int __close_fd(struct files_struct *files, unsigned fd)
 622{
 623	struct file *file;
 624	struct fdtable *fdt;
 625
 626	spin_lock(&files->file_lock);
 627	fdt = files_fdtable(files);
 628	if (fd >= fdt->max_fds)
 629		goto out_unlock;
 630	file = fdt->fd[fd];
 631	if (!file)
 632		goto out_unlock;
 633	rcu_assign_pointer(fdt->fd[fd], NULL);
 634	__put_unused_fd(files, fd);
 635	spin_unlock(&files->file_lock);
 636	return filp_close(file, files);
 637
 638out_unlock:
 639	spin_unlock(&files->file_lock);
 640	return -EBADF;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 641}
 642EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
 643
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 644/*
 645 * variant of __close_fd that gets a ref on the file for later fput
 
 
 646 */
 647int __close_fd_get_file(unsigned int fd, struct file **res)
 648{
 649	struct files_struct *files = current->files;
 650	struct file *file;
 651	struct fdtable *fdt;
 652
 653	spin_lock(&files->file_lock);
 654	fdt = files_fdtable(files);
 655	if (fd >= fdt->max_fds)
 656		goto out_unlock;
 657	file = fdt->fd[fd];
 658	if (!file)
 659		goto out_unlock;
 660	rcu_assign_pointer(fdt->fd[fd], NULL);
 661	__put_unused_fd(files, fd);
 662	spin_unlock(&files->file_lock);
 663	get_file(file);
 664	*res = file;
 665	return filp_close(file, files);
 666
 667out_unlock:
 668	spin_unlock(&files->file_lock);
 669	*res = NULL;
 670	return -ENOENT;
 671}
 672
 673void do_close_on_exec(struct files_struct *files)
 674{
 675	unsigned i;
 676	struct fdtable *fdt;
 677
 678	/* exec unshares first */
 679	spin_lock(&files->file_lock);
 680	for (i = 0; ; i++) {
 681		unsigned long set;
 682		unsigned fd = i * BITS_PER_LONG;
 683		fdt = files_fdtable(files);
 684		if (fd >= fdt->max_fds)
 685			break;
 686		set = fdt->close_on_exec[i];
 687		if (!set)
 688			continue;
 689		fdt->close_on_exec[i] = 0;
 690		for ( ; set ; fd++, set >>= 1) {
 691			struct file *file;
 692			if (!(set & 1))
 693				continue;
 694			file = fdt->fd[fd];
 695			if (!file)
 696				continue;
 697			rcu_assign_pointer(fdt->fd[fd], NULL);
 698			__put_unused_fd(files, fd);
 699			spin_unlock(&files->file_lock);
 700			filp_close(file, files);
 701			cond_resched();
 702			spin_lock(&files->file_lock);
 703		}
 704
 705	}
 706	spin_unlock(&files->file_lock);
 707}
 708
 709static struct file *__fget(unsigned int fd, fmode_t mask, unsigned int refs)
 
 710{
 711	struct files_struct *files = current->files;
 712	struct file *file;
 713
 714	rcu_read_lock();
 715loop:
 716	file = fcheck_files(files, fd);
 717	if (file) {
 718		/* File object ref couldn't be taken.
 719		 * dup2() atomicity guarantee is the reason
 720		 * we loop to catch the new file (or NULL pointer)
 721		 */
 722		if (file->f_mode & mask)
 723			file = NULL;
 724		else if (!get_file_rcu_many(file, refs))
 725			goto loop;
 726	}
 727	rcu_read_unlock();
 728
 729	return file;
 730}
 731
 
 
 
 
 
 
 732struct file *fget_many(unsigned int fd, unsigned int refs)
 733{
 734	return __fget(fd, FMODE_PATH, refs);
 735}
 736
 737struct file *fget(unsigned int fd)
 738{
 739	return __fget(fd, FMODE_PATH, 1);
 740}
 741EXPORT_SYMBOL(fget);
 742
 743struct file *fget_raw(unsigned int fd)
 744{
 745	return __fget(fd, 0, 1);
 746}
 747EXPORT_SYMBOL(fget_raw);
 748
 
 
 
 
 
 
 
 
 
 
 
 
 749/*
 750 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 751 *
 752 * You can use this instead of fget if you satisfy all of the following
 753 * conditions:
 754 * 1) You must call fput_light before exiting the syscall and returning control
 755 *    to userspace (i.e. you cannot remember the returned struct file * after
 756 *    returning to userspace).
 757 * 2) You must not call filp_close on the returned struct file * in between
 758 *    calls to fget_light and fput_light.
 759 * 3) You must not clone the current task in between the calls to fget_light
 760 *    and fput_light.
 761 *
 762 * The fput_needed flag returned by fget_light should be passed to the
 763 * corresponding fput_light.
 764 */
 765static unsigned long __fget_light(unsigned int fd, fmode_t mask)
 766{
 767	struct files_struct *files = current->files;
 768	struct file *file;
 769
 770	if (atomic_read(&files->count) == 1) {
 771		file = __fcheck_files(files, fd);
 772		if (!file || unlikely(file->f_mode & mask))
 773			return 0;
 774		return (unsigned long)file;
 775	} else {
 776		file = __fget(fd, mask, 1);
 777		if (!file)
 778			return 0;
 779		return FDPUT_FPUT | (unsigned long)file;
 780	}
 781}
 782unsigned long __fdget(unsigned int fd)
 783{
 784	return __fget_light(fd, FMODE_PATH);
 785}
 786EXPORT_SYMBOL(__fdget);
 787
 788unsigned long __fdget_raw(unsigned int fd)
 789{
 790	return __fget_light(fd, 0);
 791}
 792
 793unsigned long __fdget_pos(unsigned int fd)
 794{
 795	unsigned long v = __fdget(fd);
 796	struct file *file = (struct file *)(v & ~3);
 797
 798	if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
 799		if (file_count(file) > 1) {
 800			v |= FDPUT_POS_UNLOCK;
 801			mutex_lock(&file->f_pos_lock);
 802		}
 803	}
 804	return v;
 805}
 806
 807void __f_unlock_pos(struct file *f)
 808{
 809	mutex_unlock(&f->f_pos_lock);
 810}
 811
 812/*
 813 * We only lock f_pos if we have threads or if the file might be
 814 * shared with another process. In both cases we'll have an elevated
 815 * file count (done either by fdget() or by fork()).
 816 */
 817
 818void set_close_on_exec(unsigned int fd, int flag)
 819{
 820	struct files_struct *files = current->files;
 821	struct fdtable *fdt;
 822	spin_lock(&files->file_lock);
 823	fdt = files_fdtable(files);
 824	if (flag)
 825		__set_close_on_exec(fd, fdt);
 826	else
 827		__clear_close_on_exec(fd, fdt);
 828	spin_unlock(&files->file_lock);
 829}
 830
 831bool get_close_on_exec(unsigned int fd)
 832{
 833	struct files_struct *files = current->files;
 834	struct fdtable *fdt;
 835	bool res;
 836	rcu_read_lock();
 837	fdt = files_fdtable(files);
 838	res = close_on_exec(fd, fdt);
 839	rcu_read_unlock();
 840	return res;
 841}
 842
 843static int do_dup2(struct files_struct *files,
 844	struct file *file, unsigned fd, unsigned flags)
 845__releases(&files->file_lock)
 846{
 847	struct file *tofree;
 848	struct fdtable *fdt;
 849
 850	/*
 851	 * We need to detect attempts to do dup2() over allocated but still
 852	 * not finished descriptor.  NB: OpenBSD avoids that at the price of
 853	 * extra work in their equivalent of fget() - they insert struct
 854	 * file immediately after grabbing descriptor, mark it larval if
 855	 * more work (e.g. actual opening) is needed and make sure that
 856	 * fget() treats larval files as absent.  Potentially interesting,
 857	 * but while extra work in fget() is trivial, locking implications
 858	 * and amount of surgery on open()-related paths in VFS are not.
 859	 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
 860	 * deadlocks in rather amusing ways, AFAICS.  All of that is out of
 861	 * scope of POSIX or SUS, since neither considers shared descriptor
 862	 * tables and this condition does not arise without those.
 863	 */
 864	fdt = files_fdtable(files);
 865	tofree = fdt->fd[fd];
 866	if (!tofree && fd_is_open(fd, fdt))
 867		goto Ebusy;
 868	get_file(file);
 869	rcu_assign_pointer(fdt->fd[fd], file);
 870	__set_open_fd(fd, fdt);
 871	if (flags & O_CLOEXEC)
 872		__set_close_on_exec(fd, fdt);
 873	else
 874		__clear_close_on_exec(fd, fdt);
 875	spin_unlock(&files->file_lock);
 876
 877	if (tofree)
 878		filp_close(tofree, files);
 879
 880	return fd;
 881
 882Ebusy:
 883	spin_unlock(&files->file_lock);
 884	return -EBUSY;
 885}
 886
 887int replace_fd(unsigned fd, struct file *file, unsigned flags)
 888{
 889	int err;
 890	struct files_struct *files = current->files;
 891
 892	if (!file)
 893		return __close_fd(files, fd);
 894
 895	if (fd >= rlimit(RLIMIT_NOFILE))
 896		return -EBADF;
 897
 898	spin_lock(&files->file_lock);
 899	err = expand_files(files, fd);
 900	if (unlikely(err < 0))
 901		goto out_unlock;
 902	return do_dup2(files, file, fd, flags);
 903
 904out_unlock:
 905	spin_unlock(&files->file_lock);
 906	return err;
 907}
 908
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 909static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
 910{
 911	int err = -EBADF;
 912	struct file *file;
 913	struct files_struct *files = current->files;
 914
 915	if ((flags & ~O_CLOEXEC) != 0)
 916		return -EINVAL;
 917
 918	if (unlikely(oldfd == newfd))
 919		return -EINVAL;
 920
 921	if (newfd >= rlimit(RLIMIT_NOFILE))
 922		return -EBADF;
 923
 924	spin_lock(&files->file_lock);
 925	err = expand_files(files, newfd);
 926	file = fcheck(oldfd);
 927	if (unlikely(!file))
 928		goto Ebadf;
 929	if (unlikely(err < 0)) {
 930		if (err == -EMFILE)
 931			goto Ebadf;
 932		goto out_unlock;
 933	}
 934	return do_dup2(files, file, newfd, flags);
 935
 936Ebadf:
 937	err = -EBADF;
 938out_unlock:
 939	spin_unlock(&files->file_lock);
 940	return err;
 941}
 942
 943SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
 944{
 945	return ksys_dup3(oldfd, newfd, flags);
 946}
 947
 948SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 949{
 950	if (unlikely(newfd == oldfd)) { /* corner case */
 951		struct files_struct *files = current->files;
 952		int retval = oldfd;
 953
 954		rcu_read_lock();
 955		if (!fcheck_files(files, oldfd))
 956			retval = -EBADF;
 957		rcu_read_unlock();
 958		return retval;
 959	}
 960	return ksys_dup3(oldfd, newfd, 0);
 961}
 962
 963int ksys_dup(unsigned int fildes)
 964{
 965	int ret = -EBADF;
 966	struct file *file = fget_raw(fildes);
 967
 968	if (file) {
 969		ret = get_unused_fd_flags(0);
 970		if (ret >= 0)
 971			fd_install(ret, file);
 972		else
 973			fput(file);
 974	}
 975	return ret;
 976}
 977
 978SYSCALL_DEFINE1(dup, unsigned int, fildes)
 979{
 980	return ksys_dup(fildes);
 981}
 982
 983int f_dupfd(unsigned int from, struct file *file, unsigned flags)
 984{
 985	int err;
 986	if (from >= rlimit(RLIMIT_NOFILE))
 987		return -EINVAL;
 988	err = alloc_fd(from, flags);
 989	if (err >= 0) {
 990		get_file(file);
 991		fd_install(err, file);
 992	}
 993	return err;
 994}
 995
 996int iterate_fd(struct files_struct *files, unsigned n,
 997		int (*f)(const void *, struct file *, unsigned),
 998		const void *p)
 999{
1000	struct fdtable *fdt;
1001	int res = 0;
1002	if (!files)
1003		return 0;
1004	spin_lock(&files->file_lock);
1005	for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1006		struct file *file;
1007		file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1008		if (!file)
1009			continue;
1010		res = f(p, file, n);
1011		if (res)
1012			break;
1013	}
1014	spin_unlock(&files->file_lock);
1015	return res;
1016}
1017EXPORT_SYMBOL(iterate_fd);