libfs.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *	fs/libfs.c
   4 *	Library for filesystems writers.
   5 */
   6
   7#include <linux/blkdev.h>
   8#include <linux/export.h>
   9#include <linux/pagemap.h>
  10#include <linux/slab.h>
  11#include <linux/cred.h>
  12#include <linux/mount.h>
  13#include <linux/vfs.h>
  14#include <linux/quotaops.h>
  15#include <linux/mutex.h>
  16#include <linux/namei.h>
  17#include <linux/exportfs.h>
  18#include <linux/iversion.h>
  19#include <linux/writeback.h>
  20#include <linux/buffer_head.h> /* sync_mapping_buffers */
  21#include <linux/fs_context.h>
  22#include <linux/pseudo_fs.h>
  23#include <linux/fsnotify.h>
  24#include <linux/unicode.h>
  25#include <linux/fscrypt.h>
  26#include <linux/pidfs.h>
  27
  28#include <linux/uaccess.h>
  29
  30#include "internal.h"
  31
  32int simple_getattr(struct mnt_idmap *idmap, const struct path *path,
  33		   struct kstat *stat, u32 request_mask,
  34		   unsigned int query_flags)
  35{
  36	struct inode *inode = d_inode(path->dentry);
  37	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
  38	stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
  39	return 0;
  40}
  41EXPORT_SYMBOL(simple_getattr);
  42
  43int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
  44{
  45	u64 id = huge_encode_dev(dentry->d_sb->s_dev);
  46
  47	buf->f_fsid = u64_to_fsid(id);
  48	buf->f_type = dentry->d_sb->s_magic;
  49	buf->f_bsize = PAGE_SIZE;
  50	buf->f_namelen = NAME_MAX;
  51	return 0;
  52}
  53EXPORT_SYMBOL(simple_statfs);
  54
  55/*
  56 * Retaining negative dentries for an in-memory filesystem just wastes
  57 * memory and lookup time: arrange for them to be deleted immediately.
  58 */
  59int always_delete_dentry(const struct dentry *dentry)
  60{
  61	return 1;
  62}
  63EXPORT_SYMBOL(always_delete_dentry);
  64
  65/*
  66 * Lookup the data. This is trivial - if the dentry didn't already
  67 * exist, we know it is negative.  Set d_op to delete negative dentries.
  68 */
  69struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
  70{
  71	if (dentry->d_name.len > NAME_MAX)
  72		return ERR_PTR(-ENAMETOOLONG);
  73	if (!dentry->d_op && !(dentry->d_flags & DCACHE_DONTCACHE)) {
  74		spin_lock(&dentry->d_lock);
  75		dentry->d_flags |= DCACHE_DONTCACHE;
  76		spin_unlock(&dentry->d_lock);
  77	}
  78	if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
  79		return NULL;
  80
  81	d_add(dentry, NULL);
  82	return NULL;
  83}
  84EXPORT_SYMBOL(simple_lookup);
  85
  86int dcache_dir_open(struct inode *inode, struct file *file)
  87{
  88	file->private_data = d_alloc_cursor(file->f_path.dentry);
  89
  90	return file->private_data ? 0 : -ENOMEM;
  91}
  92EXPORT_SYMBOL(dcache_dir_open);
  93
  94int dcache_dir_close(struct inode *inode, struct file *file)
  95{
  96	dput(file->private_data);
  97	return 0;
  98}
  99EXPORT_SYMBOL(dcache_dir_close);
 100
 101/* parent is locked at least shared */
 102/*
 103 * Returns an element of siblings' list.
 104 * We are looking for <count>th positive after <p>; if
 105 * found, dentry is grabbed and returned to caller.
 106 * If no such element exists, NULL is returned.
 107 */
 108static struct dentry *scan_positives(struct dentry *cursor,
 109					struct hlist_node **p,
 110					loff_t count,
 111					struct dentry *last)
 112{
 113	struct dentry *dentry = cursor->d_parent, *found = NULL;
 114
 115	spin_lock(&dentry->d_lock);
 116	while (*p) {
 117		struct dentry *d = hlist_entry(*p, struct dentry, d_sib);
 118		p = &d->d_sib.next;
 119		// we must at least skip cursors, to avoid livelocks
 120		if (d->d_flags & DCACHE_DENTRY_CURSOR)
 121			continue;
 122		if (simple_positive(d) && !--count) {
 123			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
 124			if (simple_positive(d))
 125				found = dget_dlock(d);
 126			spin_unlock(&d->d_lock);
 127			if (likely(found))
 128				break;
 129			count = 1;
 130		}
 131		if (need_resched()) {
 132			if (!hlist_unhashed(&cursor->d_sib))
 133				__hlist_del(&cursor->d_sib);
 134			hlist_add_behind(&cursor->d_sib, &d->d_sib);
 135			p = &cursor->d_sib.next;
 136			spin_unlock(&dentry->d_lock);
 137			cond_resched();
 138			spin_lock(&dentry->d_lock);
 139		}
 140	}
 141	spin_unlock(&dentry->d_lock);
 142	dput(last);
 143	return found;
 144}
 145
 146loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
 147{
 148	struct dentry *dentry = file->f_path.dentry;
 149	switch (whence) {
 150		case 1:
 151			offset += file->f_pos;
 152			fallthrough;
 153		case 0:
 154			if (offset >= 0)
 155				break;
 156			fallthrough;
 157		default:
 158			return -EINVAL;
 159	}
 160	if (offset != file->f_pos) {
 161		struct dentry *cursor = file->private_data;
 162		struct dentry *to = NULL;
 163
 164		inode_lock_shared(dentry->d_inode);
 165
 166		if (offset > 2)
 167			to = scan_positives(cursor, &dentry->d_children.first,
 168					    offset - 2, NULL);
 169		spin_lock(&dentry->d_lock);
 170		hlist_del_init(&cursor->d_sib);
 171		if (to)
 172			hlist_add_behind(&cursor->d_sib, &to->d_sib);
 173		spin_unlock(&dentry->d_lock);
 174		dput(to);
 175
 176		file->f_pos = offset;
 177
 178		inode_unlock_shared(dentry->d_inode);
 179	}
 180	return offset;
 181}
 182EXPORT_SYMBOL(dcache_dir_lseek);
 183
 184/*
 185 * Directory is locked and all positive dentries in it are safe, since
 186 * for ramfs-type trees they can't go away without unlink() or rmdir(),
 187 * both impossible due to the lock on directory.
 188 */
 189
 190int dcache_readdir(struct file *file, struct dir_context *ctx)
 191{
 192	struct dentry *dentry = file->f_path.dentry;
 193	struct dentry *cursor = file->private_data;
 194	struct dentry *next = NULL;
 195	struct hlist_node **p;
 196
 197	if (!dir_emit_dots(file, ctx))
 198		return 0;
 199
 200	if (ctx->pos == 2)
 201		p = &dentry->d_children.first;
 202	else
 203		p = &cursor->d_sib.next;
 204
 205	while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
 206		if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
 207			      d_inode(next)->i_ino,
 208			      fs_umode_to_dtype(d_inode(next)->i_mode)))
 209			break;
 210		ctx->pos++;
 211		p = &next->d_sib.next;
 212	}
 213	spin_lock(&dentry->d_lock);
 214	hlist_del_init(&cursor->d_sib);
 215	if (next)
 216		hlist_add_before(&cursor->d_sib, &next->d_sib);
 217	spin_unlock(&dentry->d_lock);
 218	dput(next);
 219
 220	return 0;
 221}
 222EXPORT_SYMBOL(dcache_readdir);
 223
 224ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
 225{
 226	return -EISDIR;
 227}
 228EXPORT_SYMBOL(generic_read_dir);
 229
 230const struct file_operations simple_dir_operations = {
 231	.open		= dcache_dir_open,
 232	.release	= dcache_dir_close,
 233	.llseek		= dcache_dir_lseek,
 234	.read		= generic_read_dir,
 235	.iterate_shared	= dcache_readdir,
 236	.fsync		= noop_fsync,
 237};
 238EXPORT_SYMBOL(simple_dir_operations);
 239
 240const struct inode_operations simple_dir_inode_operations = {
 241	.lookup		= simple_lookup,
 242};
 243EXPORT_SYMBOL(simple_dir_inode_operations);
 244
 245/* simple_offset_add() never assigns these to a dentry */
 246enum {
 247	DIR_OFFSET_FIRST	= 2,		/* Find first real entry */
 248	DIR_OFFSET_EOD		= S32_MAX,
 249};
 250
 251/* simple_offset_add() allocation range */
 252enum {
 253	DIR_OFFSET_MIN		= DIR_OFFSET_FIRST + 1,
 254	DIR_OFFSET_MAX		= DIR_OFFSET_EOD - 1,
 255};
 256
 257static void offset_set(struct dentry *dentry, long offset)
 258{
 259	dentry->d_fsdata = (void *)offset;
 260}
 261
 262static long dentry2offset(struct dentry *dentry)
 263{
 264	return (long)dentry->d_fsdata;
 265}
 266
 267static struct lock_class_key simple_offset_lock_class;
 268
 269/**
 270 * simple_offset_init - initialize an offset_ctx
 271 * @octx: directory offset map to be initialized
 272 *
 273 */
 274void simple_offset_init(struct offset_ctx *octx)
 275{
 276	mt_init_flags(&octx->mt, MT_FLAGS_ALLOC_RANGE);
 277	lockdep_set_class(&octx->mt.ma_lock, &simple_offset_lock_class);
 278	octx->next_offset = DIR_OFFSET_MIN;
 279}
 280
 281/**
 282 * simple_offset_add - Add an entry to a directory's offset map
 283 * @octx: directory offset ctx to be updated
 284 * @dentry: new dentry being added
 285 *
 286 * Returns zero on success. @octx and the dentry's offset are updated.
 287 * Otherwise, a negative errno value is returned.
 288 */
 289int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
 290{
 291	unsigned long offset;
 292	int ret;
 293
 294	if (dentry2offset(dentry) != 0)
 295		return -EBUSY;
 296
 297	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
 298				 DIR_OFFSET_MAX, &octx->next_offset,
 299				 GFP_KERNEL);
 300	if (unlikely(ret < 0))
 301		return ret == -EBUSY ? -ENOSPC : ret;
 302
 303	offset_set(dentry, offset);
 304	return 0;
 305}
 306
 307static int simple_offset_replace(struct offset_ctx *octx, struct dentry *dentry,
 308				 long offset)
 309{
 310	int ret;
 311
 312	ret = mtree_store(&octx->mt, offset, dentry, GFP_KERNEL);
 313	if (ret)
 314		return ret;
 315	offset_set(dentry, offset);
 316	return 0;
 317}
 318
 319/**
 320 * simple_offset_remove - Remove an entry to a directory's offset map
 321 * @octx: directory offset ctx to be updated
 322 * @dentry: dentry being removed
 323 *
 324 */
 325void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
 326{
 327	long offset;
 328
 329	offset = dentry2offset(dentry);
 330	if (offset == 0)
 331		return;
 332
 333	mtree_erase(&octx->mt, offset);
 334	offset_set(dentry, 0);
 335}
 336
 337/**
 338 * simple_offset_rename - handle directory offsets for rename
 339 * @old_dir: parent directory of source entry
 340 * @old_dentry: dentry of source entry
 341 * @new_dir: parent_directory of destination entry
 342 * @new_dentry: dentry of destination
 343 *
 344 * Caller provides appropriate serialization.
 345 *
 346 * User space expects the directory offset value of the replaced
 347 * (new) directory entry to be unchanged after a rename.
 348 *
 349 * Caller must have grabbed a slot for new_dentry in the maple_tree
 350 * associated with new_dir, even if dentry is negative.
 351 */
 352void simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
 353			  struct inode *new_dir, struct dentry *new_dentry)
 354{
 355	struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
 356	struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
 357	long new_offset = dentry2offset(new_dentry);
 358
 359	if (WARN_ON(!new_offset))
 360		return;
 361
 362	simple_offset_remove(old_ctx, old_dentry);
 363	offset_set(new_dentry, 0);
 364	WARN_ON(simple_offset_replace(new_ctx, old_dentry, new_offset));
 365}
 366
 367/**
 368 * simple_offset_rename_exchange - exchange rename with directory offsets
 369 * @old_dir: parent of dentry being moved
 370 * @old_dentry: dentry being moved
 371 * @new_dir: destination parent
 372 * @new_dentry: destination dentry
 373 *
 374 * This API preserves the directory offset values. Caller provides
 375 * appropriate serialization.
 376 *
 377 * Returns zero on success. Otherwise a negative errno is returned and the
 378 * rename is rolled back.
 379 */
 380int simple_offset_rename_exchange(struct inode *old_dir,
 381				  struct dentry *old_dentry,
 382				  struct inode *new_dir,
 383				  struct dentry *new_dentry)
 384{
 385	struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
 386	struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
 387	long old_index = dentry2offset(old_dentry);
 388	long new_index = dentry2offset(new_dentry);
 389	int ret;
 390
 391	if (WARN_ON(!old_index || !new_index))
 392		return -EINVAL;
 393
 394	ret = mtree_store(&new_ctx->mt, new_index, old_dentry, GFP_KERNEL);
 395	if (WARN_ON(ret))
 396		return ret;
 397
 398	ret = mtree_store(&old_ctx->mt, old_index, new_dentry, GFP_KERNEL);
 399	if (WARN_ON(ret)) {
 400		mtree_store(&new_ctx->mt, new_index, new_dentry, GFP_KERNEL);
 401		return ret;
 402	}
 403
 404	offset_set(old_dentry, new_index);
 405	offset_set(new_dentry, old_index);
 406	simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
 407	return 0;
 408}
 409
 410/**
 411 * simple_offset_destroy - Release offset map
 412 * @octx: directory offset ctx that is about to be destroyed
 413 *
 414 * During fs teardown (eg. umount), a directory's offset map might still
 415 * contain entries. xa_destroy() cleans out anything that remains.
 416 */
 417void simple_offset_destroy(struct offset_ctx *octx)
 418{
 419	mtree_destroy(&octx->mt);
 420}
 421
 422/**
 423 * offset_dir_llseek - Advance the read position of a directory descriptor
 424 * @file: an open directory whose position is to be updated
 425 * @offset: a byte offset
 426 * @whence: enumerator describing the starting position for this update
 427 *
 428 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
 429 *
 430 * Returns the updated read position if successful; otherwise a
 431 * negative errno is returned and the read position remains unchanged.
 432 */
 433static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
 434{
 435	switch (whence) {
 436	case SEEK_CUR:
 437		offset += file->f_pos;
 438		fallthrough;
 439	case SEEK_SET:
 440		if (offset >= 0)
 441			break;
 442		fallthrough;
 443	default:
 444		return -EINVAL;
 445	}
 446
 447	return vfs_setpos(file, offset, LONG_MAX);
 448}
 449
 450static struct dentry *find_positive_dentry(struct dentry *parent,
 451					   struct dentry *dentry,
 452					   bool next)
 453{
 454	struct dentry *found = NULL;
 455
 456	spin_lock(&parent->d_lock);
 457	if (next)
 458		dentry = d_next_sibling(dentry);
 459	else if (!dentry)
 460		dentry = d_first_child(parent);
 461	hlist_for_each_entry_from(dentry, d_sib) {
 462		if (!simple_positive(dentry))
 463			continue;
 464		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 465		if (simple_positive(dentry))
 466			found = dget_dlock(dentry);
 467		spin_unlock(&dentry->d_lock);
 468		if (likely(found))
 469			break;
 470	}
 471	spin_unlock(&parent->d_lock);
 472	return found;
 473}
 474
 475static noinline_for_stack struct dentry *
 476offset_dir_lookup(struct dentry *parent, loff_t offset)
 477{
 478	struct inode *inode = d_inode(parent);
 479	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
 480	struct dentry *child, *found = NULL;
 481
 482	MA_STATE(mas, &octx->mt, offset, offset);
 483
 484	if (offset == DIR_OFFSET_FIRST)
 485		found = find_positive_dentry(parent, NULL, false);
 486	else {
 487		rcu_read_lock();
 488		child = mas_find_rev(&mas, DIR_OFFSET_MIN);
 489		found = find_positive_dentry(parent, child, false);
 490		rcu_read_unlock();
 491	}
 492	return found;
 493}
 494
 495static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
 496{
 497	struct inode *inode = d_inode(dentry);
 498
 499	return dir_emit(ctx, dentry->d_name.name, dentry->d_name.len,
 500			inode->i_ino, fs_umode_to_dtype(inode->i_mode));
 501}
 502
 503static void offset_iterate_dir(struct file *file, struct dir_context *ctx)
 504{
 505	struct dentry *dir = file->f_path.dentry;
 506	struct dentry *dentry;
 507
 508	dentry = offset_dir_lookup(dir, ctx->pos);
 509	if (!dentry)
 510		goto out_eod;
 511	while (true) {
 512		struct dentry *next;
 513
 514		ctx->pos = dentry2offset(dentry);
 515		if (!offset_dir_emit(ctx, dentry))
 516			break;
 517
 518		next = find_positive_dentry(dir, dentry, true);
 519		dput(dentry);
 520
 521		if (!next)
 522			goto out_eod;
 523		dentry = next;
 524	}
 525	dput(dentry);
 526	return;
 527
 528out_eod:
 529	ctx->pos = DIR_OFFSET_EOD;
 530}
 531
 532/**
 533 * offset_readdir - Emit entries starting at offset @ctx->pos
 534 * @file: an open directory to iterate over
 535 * @ctx: directory iteration context
 536 *
 537 * Caller must hold @file's i_rwsem to prevent insertion or removal of
 538 * entries during this call.
 539 *
 540 * On entry, @ctx->pos contains an offset that represents the first entry
 541 * to be read from the directory.
 542 *
 543 * The operation continues until there are no more entries to read, or
 544 * until the ctx->actor indicates there is no more space in the caller's
 545 * output buffer.
 546 *
 547 * On return, @ctx->pos contains an offset that will read the next entry
 548 * in this directory when offset_readdir() is called again with @ctx.
 549 * Caller places this value in the d_off field of the last entry in the
 550 * user's buffer.
 551 *
 552 * Return values:
 553 *   %0 - Complete
 554 */
 555static int offset_readdir(struct file *file, struct dir_context *ctx)
 556{
 557	struct dentry *dir = file->f_path.dentry;
 558
 559	lockdep_assert_held(&d_inode(dir)->i_rwsem);
 560
 561	if (!dir_emit_dots(file, ctx))
 562		return 0;
 563	if (ctx->pos != DIR_OFFSET_EOD)
 564		offset_iterate_dir(file, ctx);
 565	return 0;
 566}
 567
 568const struct file_operations simple_offset_dir_operations = {
 569	.llseek		= offset_dir_llseek,
 570	.iterate_shared	= offset_readdir,
 571	.read		= generic_read_dir,
 572	.fsync		= noop_fsync,
 573};
 574
 575struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
 576{
 577	struct dentry *child = NULL, *d;
 578
 579	spin_lock(&parent->d_lock);
 580	d = prev ? d_next_sibling(prev) : d_first_child(parent);
 581	hlist_for_each_entry_from(d, d_sib) {
 582		if (simple_positive(d)) {
 583			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
 584			if (simple_positive(d))
 585				child = dget_dlock(d);
 586			spin_unlock(&d->d_lock);
 587			if (likely(child))
 588				break;
 589		}
 590	}
 591	spin_unlock(&parent->d_lock);
 592	dput(prev);
 593	return child;
 594}
 595EXPORT_SYMBOL(find_next_child);
 596
 597static void __simple_recursive_removal(struct dentry *dentry,
 598                              void (*callback)(struct dentry *),
 599			      bool locked)
 600{
 601	struct dentry *this = dget(dentry);
 602	while (true) {
 603		struct dentry *victim = NULL, *child;
 604		struct inode *inode = this->d_inode;
 605
 606		inode_lock_nested(inode, I_MUTEX_CHILD);
 607		if (d_is_dir(this))
 608			inode->i_flags |= S_DEAD;
 609		while ((child = find_next_child(this, victim)) == NULL) {
 610			// kill and ascend
 611			// update metadata while it's still locked
 612			inode_set_ctime_current(inode);
 613			clear_nlink(inode);
 614			inode_unlock(inode);
 615			victim = this;
 616			this = this->d_parent;
 617			inode = this->d_inode;
 618			if (!locked || victim != dentry)
 619				inode_lock_nested(inode, I_MUTEX_CHILD);
 620			if (simple_positive(victim)) {
 621				d_invalidate(victim);	// avoid lost mounts
 622				if (callback)
 623					callback(victim);
 624				fsnotify_delete(inode, d_inode(victim), victim);
 625				d_make_discardable(victim);
 626			}
 627			if (victim == dentry) {
 628				inode_set_mtime_to_ts(inode,
 629						      inode_set_ctime_current(inode));
 630				if (d_is_dir(dentry))
 631					drop_nlink(inode);
 632				if (!locked)
 633					inode_unlock(inode);
 634				dput(dentry);
 635				return;
 636			}
 637		}
 638		inode_unlock(inode);
 639		this = child;
 640	}
 641}
 642
 643void simple_recursive_removal(struct dentry *dentry,
 644                              void (*callback)(struct dentry *))
 645{
 646	return __simple_recursive_removal(dentry, callback, false);
 647}
 648EXPORT_SYMBOL(simple_recursive_removal);
 649
 650void simple_remove_by_name(struct dentry *parent, const char *name,
 651                           void (*callback)(struct dentry *))
 652{
 653	struct dentry *dentry;
 654
 655	dentry = lookup_noperm_positive_unlocked(&QSTR(name), parent);
 656	if (!IS_ERR(dentry)) {
 657		simple_recursive_removal(dentry, callback);
 658		dput(dentry);	// paired with lookup_noperm_positive_unlocked()
 659	}
 660}
 661EXPORT_SYMBOL(simple_remove_by_name);
 662
 663/* caller holds parent directory with I_MUTEX_PARENT */
 664void locked_recursive_removal(struct dentry *dentry,
 665                              void (*callback)(struct dentry *))
 666{
 667	return __simple_recursive_removal(dentry, callback, true);
 668}
 669EXPORT_SYMBOL(locked_recursive_removal);
 670
 671static const struct super_operations simple_super_operations = {
 672	.statfs		= simple_statfs,
 673};
 674
 675static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
 676{
 677	struct pseudo_fs_context *ctx = fc->fs_private;
 678	struct inode *root;
 679
 680	s->s_maxbytes = MAX_LFS_FILESIZE;
 681	s->s_blocksize = PAGE_SIZE;
 682	s->s_blocksize_bits = PAGE_SHIFT;
 683	s->s_magic = ctx->magic;
 684	s->s_op = ctx->ops ?: &simple_super_operations;
 685	s->s_export_op = ctx->eops;
 686	s->s_xattr = ctx->xattr;
 687	s->s_time_gran = 1;
 688	s->s_d_flags |= ctx->s_d_flags;
 689	root = new_inode(s);
 690	if (!root)
 691		return -ENOMEM;
 692
 693	/*
 694	 * since this is the first inode, make it number 1. New inodes created
 695	 * after this must take care not to collide with it (by passing
 696	 * max_reserved of 1 to iunique).
 697	 */
 698	root->i_ino = 1;
 699	root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
 700	simple_inode_init_ts(root);
 701	s->s_root = d_make_root(root);
 702	if (!s->s_root)
 703		return -ENOMEM;
 704	set_default_d_op(s, ctx->dops);
 705	return 0;
 706}
 707
 708static int pseudo_fs_get_tree(struct fs_context *fc)
 709{
 710	return get_tree_nodev(fc, pseudo_fs_fill_super);
 711}
 712
 713static void pseudo_fs_free(struct fs_context *fc)
 714{
 715	kfree(fc->fs_private);
 716}
 717
 718static const struct fs_context_operations pseudo_fs_context_ops = {
 719	.free		= pseudo_fs_free,
 720	.get_tree	= pseudo_fs_get_tree,
 721};
 722
 723/*
 724 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
 725 * will never be mountable)
 726 */
 727struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
 728					unsigned long magic)
 729{
 730	struct pseudo_fs_context *ctx;
 731
 732	ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
 733	if (likely(ctx)) {
 734		ctx->magic = magic;
 735		fc->fs_private = ctx;
 736		fc->ops = &pseudo_fs_context_ops;
 737		fc->sb_flags |= SB_NOUSER;
 738		fc->global = true;
 739	}
 740	return ctx;
 741}
 742EXPORT_SYMBOL(init_pseudo);
 743
 744int simple_open(struct inode *inode, struct file *file)
 745{
 746	if (inode->i_private)
 747		file->private_data = inode->i_private;
 748	return 0;
 749}
 750EXPORT_SYMBOL(simple_open);
 751
 752int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
 753{
 754	struct inode *inode = d_inode(old_dentry);
 755
 756	inode_set_mtime_to_ts(dir,
 757			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
 758	inc_nlink(inode);
 759	ihold(inode);
 760	d_make_persistent(dentry, inode);
 761	return 0;
 762}
 763EXPORT_SYMBOL(simple_link);
 764
 765int simple_empty(struct dentry *dentry)
 766{
 767	struct dentry *child;
 768	int ret = 0;
 769
 770	spin_lock(&dentry->d_lock);
 771	hlist_for_each_entry(child, &dentry->d_children, d_sib) {
 772		spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
 773		if (simple_positive(child)) {
 774			spin_unlock(&child->d_lock);
 775			goto out;
 776		}
 777		spin_unlock(&child->d_lock);
 778	}
 779	ret = 1;
 780out:
 781	spin_unlock(&dentry->d_lock);
 782	return ret;
 783}
 784EXPORT_SYMBOL(simple_empty);
 785
 786void __simple_unlink(struct inode *dir, struct dentry *dentry)
 787{
 788	struct inode *inode = d_inode(dentry);
 789
 790	inode_set_mtime_to_ts(dir,
 791			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
 792	drop_nlink(inode);
 793}
 794EXPORT_SYMBOL(__simple_unlink);
 795
 796void __simple_rmdir(struct inode *dir, struct dentry *dentry)
 797{
 798	drop_nlink(d_inode(dentry));
 799	__simple_unlink(dir, dentry);
 800	drop_nlink(dir);
 801}
 802EXPORT_SYMBOL(__simple_rmdir);
 803
 804int simple_unlink(struct inode *dir, struct dentry *dentry)
 805{
 806	__simple_unlink(dir, dentry);
 807	d_make_discardable(dentry);
 808	return 0;
 809}
 810EXPORT_SYMBOL(simple_unlink);
 811
 812int simple_rmdir(struct inode *dir, struct dentry *dentry)
 813{
 814	if (!simple_empty(dentry))
 815		return -ENOTEMPTY;
 816
 817	__simple_rmdir(dir, dentry);
 818	d_make_discardable(dentry);
 819	return 0;
 820}
 821EXPORT_SYMBOL(simple_rmdir);
 822
 823/**
 824 * simple_rename_timestamp - update the various inode timestamps for rename
 825 * @old_dir: old parent directory
 826 * @old_dentry: dentry that is being renamed
 827 * @new_dir: new parent directory
 828 * @new_dentry: target for rename
 829 *
 830 * POSIX mandates that the old and new parent directories have their ctime and
 831 * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
 832 * their ctime updated.
 833 */
 834void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
 835			     struct inode *new_dir, struct dentry *new_dentry)
 836{
 837	struct inode *newino = d_inode(new_dentry);
 838
 839	inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
 840	if (new_dir != old_dir)
 841		inode_set_mtime_to_ts(new_dir,
 842				      inode_set_ctime_current(new_dir));
 843	inode_set_ctime_current(d_inode(old_dentry));
 844	if (newino)
 845		inode_set_ctime_current(newino);
 846}
 847EXPORT_SYMBOL_GPL(simple_rename_timestamp);
 848
 849int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
 850			   struct inode *new_dir, struct dentry *new_dentry)
 851{
 852	bool old_is_dir = d_is_dir(old_dentry);
 853	bool new_is_dir = d_is_dir(new_dentry);
 854
 855	if (old_dir != new_dir && old_is_dir != new_is_dir) {
 856		if (old_is_dir) {
 857			drop_nlink(old_dir);
 858			inc_nlink(new_dir);
 859		} else {
 860			drop_nlink(new_dir);
 861			inc_nlink(old_dir);
 862		}
 863	}
 864	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
 865	return 0;
 866}
 867EXPORT_SYMBOL_GPL(simple_rename_exchange);
 868
 869int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
 870		  struct dentry *old_dentry, struct inode *new_dir,
 871		  struct dentry *new_dentry, unsigned int flags)
 872{
 873	int they_are_dirs = d_is_dir(old_dentry);
 874
 875	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
 876		return -EINVAL;
 877
 878	if (flags & RENAME_EXCHANGE)
 879		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
 880
 881	if (!simple_empty(new_dentry))
 882		return -ENOTEMPTY;
 883
 884	if (d_really_is_positive(new_dentry)) {
 885		simple_unlink(new_dir, new_dentry);
 886		if (they_are_dirs) {
 887			drop_nlink(d_inode(new_dentry));
 888			drop_nlink(old_dir);
 889		}
 890	} else if (they_are_dirs) {
 891		drop_nlink(old_dir);
 892		inc_nlink(new_dir);
 893	}
 894
 895	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
 896	return 0;
 897}
 898EXPORT_SYMBOL(simple_rename);
 899
 900/**
 901 * simple_setattr - setattr for simple filesystem
 902 * @idmap: idmap of the target mount
 903 * @dentry: dentry
 904 * @iattr: iattr structure
 905 *
 906 * Returns 0 on success, -error on failure.
 907 *
 908 * simple_setattr is a simple ->setattr implementation without a proper
 909 * implementation of size changes.
 910 *
 911 * It can either be used for in-memory filesystems or special files
 912 * on simple regular filesystems.  Anything that needs to change on-disk
 913 * or wire state on size changes needs its own setattr method.
 914 */
 915int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 916		   struct iattr *iattr)
 917{
 918	struct inode *inode = d_inode(dentry);
 919	int error;
 920
 921	error = setattr_prepare(idmap, dentry, iattr);
 922	if (error)
 923		return error;
 924
 925	if (iattr->ia_valid & ATTR_SIZE)
 926		truncate_setsize(inode, iattr->ia_size);
 927	setattr_copy(idmap, inode, iattr);
 928	mark_inode_dirty(inode);
 929	return 0;
 930}
 931EXPORT_SYMBOL(simple_setattr);
 932
 933static int simple_read_folio(struct file *file, struct folio *folio)
 934{
 935	folio_zero_range(folio, 0, folio_size(folio));
 936	flush_dcache_folio(folio);
 937	folio_mark_uptodate(folio);
 938	folio_unlock(folio);
 939	return 0;
 940}
 941
 942int simple_write_begin(const struct kiocb *iocb, struct address_space *mapping,
 943			loff_t pos, unsigned len,
 944			struct folio **foliop, void **fsdata)
 945{
 946	struct folio *folio;
 947
 948	folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
 949			mapping_gfp_mask(mapping));
 950	if (IS_ERR(folio))
 951		return PTR_ERR(folio);
 952
 953	*foliop = folio;
 954
 955	if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
 956		size_t from = offset_in_folio(folio, pos);
 957
 958		folio_zero_segments(folio, 0, from,
 959				from + len, folio_size(folio));
 960	}
 961	return 0;
 962}
 963EXPORT_SYMBOL(simple_write_begin);
 964
 965/**
 966 * simple_write_end - .write_end helper for non-block-device FSes
 967 * @iocb: kernel I/O control block
 968 * @mapping: 		"
 969 * @pos: 		"
 970 * @len: 		"
 971 * @copied: 		"
 972 * @folio: 		"
 973 * @fsdata: 		"
 974 *
 975 * simple_write_end does the minimum needed for updating a folio after
 976 * writing is done. It has the same API signature as the .write_end of
 977 * address_space_operations vector. So it can just be set onto .write_end for
 978 * FSes that don't need any other processing. i_rwsem is assumed to be held
 979 * exclusively.
 980 * Block based filesystems should use generic_write_end().
 981 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
 982 * is not called, so a filesystem that actually does store data in .write_inode
 983 * should extend on what's done here with a call to mark_inode_dirty() in the
 984 * case that i_size has changed.
 985 *
 986 * Use *ONLY* with simple_read_folio()
 987 */
 988static int simple_write_end(const struct kiocb *iocb,
 989			    struct address_space *mapping,
 990			    loff_t pos, unsigned len, unsigned copied,
 991			    struct folio *folio, void *fsdata)
 992{
 993	struct inode *inode = folio->mapping->host;
 994	loff_t last_pos = pos + copied;
 995
 996	/* zero the stale part of the folio if we did a short copy */
 997	if (!folio_test_uptodate(folio)) {
 998		if (copied < len) {
 999			size_t from = offset_in_folio(folio, pos);
1000
1001			folio_zero_range(folio, from + copied, len - copied);
1002		}
1003		folio_mark_uptodate(folio);
1004	}
1005	/*
1006	 * No need to use i_size_read() here, the i_size
1007	 * cannot change under us because we hold the i_rwsem.
1008	 */
1009	if (last_pos > inode->i_size)
1010		i_size_write(inode, last_pos);
1011
1012	folio_mark_dirty(folio);
1013	folio_unlock(folio);
1014	folio_put(folio);
1015
1016	return copied;
1017}
1018
1019/*
1020 * Provides ramfs-style behavior: data in the pagecache, but no writeback.
1021 */
1022const struct address_space_operations ram_aops = {
1023	.read_folio	= simple_read_folio,
1024	.write_begin	= simple_write_begin,
1025	.write_end	= simple_write_end,
1026	.dirty_folio	= noop_dirty_folio,
1027};
1028EXPORT_SYMBOL(ram_aops);
1029
1030/*
1031 * the inodes created here are not hashed. If you use iunique to generate
1032 * unique inode values later for this filesystem, then you must take care
1033 * to pass it an appropriate max_reserved value to avoid collisions.
1034 */
1035int simple_fill_super(struct super_block *s, unsigned long magic,
1036		      const struct tree_descr *files)
1037{
1038	struct inode *inode;
1039	struct dentry *dentry;
1040	int i;
1041
1042	s->s_blocksize = PAGE_SIZE;
1043	s->s_blocksize_bits = PAGE_SHIFT;
1044	s->s_magic = magic;
1045	s->s_op = &simple_super_operations;
1046	s->s_time_gran = 1;
1047
1048	inode = new_inode(s);
1049	if (!inode)
1050		return -ENOMEM;
1051	/*
1052	 * because the root inode is 1, the files array must not contain an
1053	 * entry at index 1
1054	 */
1055	inode->i_ino = 1;
1056	inode->i_mode = S_IFDIR | 0755;
1057	simple_inode_init_ts(inode);
1058	inode->i_op = &simple_dir_inode_operations;
1059	inode->i_fop = &simple_dir_operations;
1060	set_nlink(inode, 2);
1061	s->s_root = d_make_root(inode);
1062	if (!s->s_root)
1063		return -ENOMEM;
1064	for (i = 0; !files->name || files->name[0]; i++, files++) {
1065		if (!files->name)
1066			continue;
1067
1068		/* warn if it tries to conflict with the root inode */
1069		if (unlikely(i == 1))
1070			printk(KERN_WARNING "%s: %s passed in a files array"
1071				"with an index of 1!\n", __func__,
1072				s->s_type->name);
1073
1074		dentry = d_alloc_name(s->s_root, files->name);
1075		if (!dentry)
1076			return -ENOMEM;
1077		inode = new_inode(s);
1078		if (!inode) {
1079			dput(dentry);
1080			return -ENOMEM;
1081		}
1082		inode->i_mode = S_IFREG | files->mode;
1083		simple_inode_init_ts(inode);
1084		inode->i_fop = files->ops;
1085		inode->i_ino = i;
1086		d_make_persistent(dentry, inode);
1087		dput(dentry);
1088	}
1089	return 0;
1090}
1091EXPORT_SYMBOL(simple_fill_super);
1092
1093static DEFINE_SPINLOCK(pin_fs_lock);
1094
1095int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
1096{
1097	struct vfsmount *mnt = NULL;
1098	spin_lock(&pin_fs_lock);
1099	if (unlikely(!*mount)) {
1100		spin_unlock(&pin_fs_lock);
1101		mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
1102		if (IS_ERR(mnt))
1103			return PTR_ERR(mnt);
1104		spin_lock(&pin_fs_lock);
1105		if (!*mount)
1106			*mount = mnt;
1107	}
1108	mntget(*mount);
1109	++*count;
1110	spin_unlock(&pin_fs_lock);
1111	mntput(mnt);
1112	return 0;
1113}
1114EXPORT_SYMBOL(simple_pin_fs);
1115
1116void simple_release_fs(struct vfsmount **mount, int *count)
1117{
1118	struct vfsmount *mnt;
1119	spin_lock(&pin_fs_lock);
1120	mnt = *mount;
1121	if (!--*count)
1122		*mount = NULL;
1123	spin_unlock(&pin_fs_lock);
1124	mntput(mnt);
1125}
1126EXPORT_SYMBOL(simple_release_fs);
1127
1128/**
1129 * simple_read_from_buffer - copy data from the buffer to user space
1130 * @to: the user space buffer to read to
1131 * @count: the maximum number of bytes to read
1132 * @ppos: the current position in the buffer
1133 * @from: the buffer to read from
1134 * @available: the size of the buffer
1135 *
1136 * The simple_read_from_buffer() function reads up to @count bytes from the
1137 * buffer @from at offset @ppos into the user space address starting at @to.
1138 *
1139 * On success, the number of bytes read is returned and the offset @ppos is
1140 * advanced by this number, or negative value is returned on error.
1141 **/
1142ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1143				const void *from, size_t available)
1144{
1145	loff_t pos = *ppos;
1146	size_t ret;
1147
1148	if (pos < 0)
1149		return -EINVAL;
1150	if (pos >= available || !count)
1151		return 0;
1152	if (count > available - pos)
1153		count = available - pos;
1154	ret = copy_to_user(to, from + pos, count);
1155	if (ret == count)
1156		return -EFAULT;
1157	count -= ret;
1158	*ppos = pos + count;
1159	return count;
1160}
1161EXPORT_SYMBOL(simple_read_from_buffer);
1162
1163/**
1164 * simple_write_to_buffer - copy data from user space to the buffer
1165 * @to: the buffer to write to
1166 * @available: the size of the buffer
1167 * @ppos: the current position in the buffer
1168 * @from: the user space buffer to read from
1169 * @count: the maximum number of bytes to read
1170 *
1171 * The simple_write_to_buffer() function reads up to @count bytes from the user
1172 * space address starting at @from into the buffer @to at offset @ppos.
1173 *
1174 * On success, the number of bytes written is returned and the offset @ppos is
1175 * advanced by this number, or negative value is returned on error.
1176 **/
1177ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1178		const void __user *from, size_t count)
1179{
1180	loff_t pos = *ppos;
1181	size_t res;
1182
1183	if (pos < 0)
1184		return -EINVAL;
1185	if (pos >= available || !count)
1186		return 0;
1187	if (count > available - pos)
1188		count = available - pos;
1189	res = copy_from_user(to + pos, from, count);
1190	if (res == count)
1191		return -EFAULT;
1192	count -= res;
1193	*ppos = pos + count;
1194	return count;
1195}
1196EXPORT_SYMBOL(simple_write_to_buffer);
1197
1198/**
1199 * memory_read_from_buffer - copy data from the buffer
1200 * @to: the kernel space buffer to read to
1201 * @count: the maximum number of bytes to read
1202 * @ppos: the current position in the buffer
1203 * @from: the buffer to read from
1204 * @available: the size of the buffer
1205 *
1206 * The memory_read_from_buffer() function reads up to @count bytes from the
1207 * buffer @from at offset @ppos into the kernel space address starting at @to.
1208 *
1209 * On success, the number of bytes read is returned and the offset @ppos is
1210 * advanced by this number, or negative value is returned on error.
1211 **/
1212ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1213				const void *from, size_t available)
1214{
1215	loff_t pos = *ppos;
1216
1217	if (pos < 0)
1218		return -EINVAL;
1219	if (pos >= available)
1220		return 0;
1221	if (count > available - pos)
1222		count = available - pos;
1223	memcpy(to, from + pos, count);
1224	*ppos = pos + count;
1225
1226	return count;
1227}
1228EXPORT_SYMBOL(memory_read_from_buffer);
1229
1230/*
1231 * Transaction based IO.
1232 * The file expects a single write which triggers the transaction, and then
1233 * possibly a read which collects the result - which is stored in a
1234 * file-local buffer.
1235 */
1236
1237void simple_transaction_set(struct file *file, size_t n)
1238{
1239	struct simple_transaction_argresp *ar = file->private_data;
1240
1241	BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1242
1243	/*
1244	 * The barrier ensures that ar->size will really remain zero until
1245	 * ar->data is ready for reading.
1246	 */
1247	smp_mb();
1248	ar->size = n;
1249}
1250EXPORT_SYMBOL(simple_transaction_set);
1251
1252char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1253{
1254	struct simple_transaction_argresp *ar;
1255	static DEFINE_SPINLOCK(simple_transaction_lock);
1256
1257	if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1258		return ERR_PTR(-EFBIG);
1259
1260	ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1261	if (!ar)
1262		return ERR_PTR(-ENOMEM);
1263
1264	spin_lock(&simple_transaction_lock);
1265
1266	/* only one write allowed per open */
1267	if (file->private_data) {
1268		spin_unlock(&simple_transaction_lock);
1269		free_page((unsigned long)ar);
1270		return ERR_PTR(-EBUSY);
1271	}
1272
1273	file->private_data = ar;
1274
1275	spin_unlock(&simple_transaction_lock);
1276
1277	if (copy_from_user(ar->data, buf, size))
1278		return ERR_PTR(-EFAULT);
1279
1280	return ar->data;
1281}
1282EXPORT_SYMBOL(simple_transaction_get);
1283
1284ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1285{
1286	struct simple_transaction_argresp *ar = file->private_data;
1287
1288	if (!ar)
1289		return 0;
1290	return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1291}
1292EXPORT_SYMBOL(simple_transaction_read);
1293
1294int simple_transaction_release(struct inode *inode, struct file *file)
1295{
1296	free_page((unsigned long)file->private_data);
1297	return 0;
1298}
1299EXPORT_SYMBOL(simple_transaction_release);
1300
1301/* Simple attribute files */
1302
1303struct simple_attr {
1304	int (*get)(void *, u64 *);
1305	int (*set)(void *, u64);
1306	char get_buf[24];	/* enough to store a u64 and "\n\0" */
1307	char set_buf[24];
1308	void *data;
1309	const char *fmt;	/* format for read operation */
1310	struct mutex mutex;	/* protects access to these buffers */
1311};
1312
1313/* simple_attr_open is called by an actual attribute open file operation
1314 * to set the attribute specific access operations. */
1315int simple_attr_open(struct inode *inode, struct file *file,
1316		     int (*get)(void *, u64 *), int (*set)(void *, u64),
1317		     const char *fmt)
1318{
1319	struct simple_attr *attr;
1320
1321	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1322	if (!attr)
1323		return -ENOMEM;
1324
1325	attr->get = get;
1326	attr->set = set;
1327	attr->data = inode->i_private;
1328	attr->fmt = fmt;
1329	mutex_init(&attr->mutex);
1330
1331	file->private_data = attr;
1332
1333	return nonseekable_open(inode, file);
1334}
1335EXPORT_SYMBOL_GPL(simple_attr_open);
1336
1337int simple_attr_release(struct inode *inode, struct file *file)
1338{
1339	kfree(file->private_data);
1340	return 0;
1341}
1342EXPORT_SYMBOL_GPL(simple_attr_release);	/* GPL-only?  This?  Really? */
1343
1344/* read from the buffer that is filled with the get function */
1345ssize_t simple_attr_read(struct file *file, char __user *buf,
1346			 size_t len, loff_t *ppos)
1347{
1348	struct simple_attr *attr;
1349	size_t size;
1350	ssize_t ret;
1351
1352	attr = file->private_data;
1353
1354	if (!attr->get)
1355		return -EACCES;
1356
1357	ret = mutex_lock_interruptible(&attr->mutex);
1358	if (ret)
1359		return ret;
1360
1361	if (*ppos && attr->get_buf[0]) {
1362		/* continued read */
1363		size = strlen(attr->get_buf);
1364	} else {
1365		/* first read */
1366		u64 val;
1367		ret = attr->get(attr->data, &val);
1368		if (ret)
1369			goto out;
1370
1371		size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
1372				 attr->fmt, (unsigned long long)val);
1373	}
1374
1375	ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
1376out:
1377	mutex_unlock(&attr->mutex);
1378	return ret;
1379}
1380EXPORT_SYMBOL_GPL(simple_attr_read);
1381
1382/* interpret the buffer as a number to call the set function with */
1383static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1384			  size_t len, loff_t *ppos, bool is_signed)
1385{
1386	struct simple_attr *attr;
1387	unsigned long long val;
1388	size_t size;
1389	ssize_t ret;
1390
1391	attr = file->private_data;
1392	if (!attr->set)
1393		return -EACCES;
1394
1395	ret = mutex_lock_interruptible(&attr->mutex);
1396	if (ret)
1397		return ret;
1398
1399	ret = -EFAULT;
1400	size = min(sizeof(attr->set_buf) - 1, len);
1401	if (copy_from_user(attr->set_buf, buf, size))
1402		goto out;
1403
1404	attr->set_buf[size] = '\0';
1405	if (is_signed)
1406		ret = kstrtoll(attr->set_buf, 0, &val);
1407	else
1408		ret = kstrtoull(attr->set_buf, 0, &val);
1409	if (ret)
1410		goto out;
1411	ret = attr->set(attr->data, val);
1412	if (ret == 0)
1413		ret = len; /* on success, claim we got the whole input */
1414out:
1415	mutex_unlock(&attr->mutex);
1416	return ret;
1417}
1418
1419ssize_t simple_attr_write(struct file *file, const char __user *buf,
1420			  size_t len, loff_t *ppos)
1421{
1422	return simple_attr_write_xsigned(file, buf, len, ppos, false);
1423}
1424EXPORT_SYMBOL_GPL(simple_attr_write);
1425
1426ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1427			  size_t len, loff_t *ppos)
1428{
1429	return simple_attr_write_xsigned(file, buf, len, ppos, true);
1430}
1431EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1432
1433/**
1434 * generic_encode_ino32_fh - generic export_operations->encode_fh function
1435 * @inode:   the object to encode
1436 * @fh:      where to store the file handle fragment
1437 * @max_len: maximum length to store there (in 4 byte units)
1438 * @parent:  parent directory inode, if wanted
1439 *
1440 * This generic encode_fh function assumes that the 32 inode number
1441 * is suitable for locating an inode, and that the generation number
1442 * can be used to check that it is still valid.  It places them in the
1443 * filehandle fragment where export_decode_fh expects to find them.
1444 */
1445int generic_encode_ino32_fh(struct inode *inode, __u32 *fh, int *max_len,
1446			    struct inode *parent)
1447{
1448	struct fid *fid = (void *)fh;
1449	int len = *max_len;
1450	int type = FILEID_INO32_GEN;
1451
1452	if (parent && (len < 4)) {
1453		*max_len = 4;
1454		return FILEID_INVALID;
1455	} else if (len < 2) {
1456		*max_len = 2;
1457		return FILEID_INVALID;
1458	}
1459
1460	len = 2;
1461	fid->i32.ino = inode->i_ino;
1462	fid->i32.gen = inode->i_generation;
1463	if (parent) {
1464		fid->i32.parent_ino = parent->i_ino;
1465		fid->i32.parent_gen = parent->i_generation;
1466		len = 4;
1467		type = FILEID_INO32_GEN_PARENT;
1468	}
1469	*max_len = len;
1470	return type;
1471}
1472EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1473
1474/**
1475 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1476 * @sb:		filesystem to do the file handle conversion on
1477 * @fid:	file handle to convert
1478 * @fh_len:	length of the file handle in bytes
1479 * @fh_type:	type of file handle
1480 * @get_inode:	filesystem callback to retrieve inode
1481 *
1482 * This function decodes @fid as long as it has one of the well-known
1483 * Linux filehandle types and calls @get_inode on it to retrieve the
1484 * inode for the object specified in the file handle.
1485 */
1486struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1487		int fh_len, int fh_type, struct inode *(*get_inode)
1488			(struct super_block *sb, u64 ino, u32 gen))
1489{
1490	struct inode *inode = NULL;
1491
1492	if (fh_len < 2)
1493		return NULL;
1494
1495	switch (fh_type) {
1496	case FILEID_INO32_GEN:
1497	case FILEID_INO32_GEN_PARENT:
1498		inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1499		break;
1500	}
1501
1502	return d_obtain_alias(inode);
1503}
1504EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1505
1506/**
1507 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1508 * @sb:		filesystem to do the file handle conversion on
1509 * @fid:	file handle to convert
1510 * @fh_len:	length of the file handle in bytes
1511 * @fh_type:	type of file handle
1512 * @get_inode:	filesystem callback to retrieve inode
1513 *
1514 * This function decodes @fid as long as it has one of the well-known
1515 * Linux filehandle types and calls @get_inode on it to retrieve the
1516 * inode for the _parent_ object specified in the file handle if it
1517 * is specified in the file handle, or NULL otherwise.
1518 */
1519struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1520		int fh_len, int fh_type, struct inode *(*get_inode)
1521			(struct super_block *sb, u64 ino, u32 gen))
1522{
1523	struct inode *inode = NULL;
1524
1525	if (fh_len <= 2)
1526		return NULL;
1527
1528	switch (fh_type) {
1529	case FILEID_INO32_GEN_PARENT:
1530		inode = get_inode(sb, fid->i32.parent_ino,
1531				  (fh_len > 3 ? fid->i32.parent_gen : 0));
1532		break;
1533	}
1534
1535	return d_obtain_alias(inode);
1536}
1537EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1538
1539/**
1540 * __generic_file_fsync - generic fsync implementation for simple filesystems
1541 *
1542 * @file:	file to synchronize
1543 * @start:	start offset in bytes
1544 * @end:	end offset in bytes (inclusive)
1545 * @datasync:	only synchronize essential metadata if true
1546 *
1547 * This is a generic implementation of the fsync method for simple
1548 * filesystems which track all non-inode metadata in the buffers list
1549 * hanging off the address_space structure.
1550 */
1551int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1552				 int datasync)
1553{
1554	struct inode *inode = file->f_mapping->host;
1555	int err;
1556	int ret;
1557
1558	err = file_write_and_wait_range(file, start, end);
1559	if (err)
1560		return err;
1561
1562	inode_lock(inode);
1563	ret = sync_mapping_buffers(inode->i_mapping);
1564	if (!(inode_state_read_once(inode) & I_DIRTY_ALL))
1565		goto out;
1566	if (datasync && !(inode_state_read_once(inode) & I_DIRTY_DATASYNC))
1567		goto out;
1568
1569	err = sync_inode_metadata(inode, 1);
1570	if (ret == 0)
1571		ret = err;
1572
1573out:
1574	inode_unlock(inode);
1575	/* check and advance again to catch errors after syncing out buffers */
1576	err = file_check_and_advance_wb_err(file);
1577	if (ret == 0)
1578		ret = err;
1579	return ret;
1580}
1581EXPORT_SYMBOL(__generic_file_fsync);
1582
1583/**
1584 * generic_file_fsync - generic fsync implementation for simple filesystems
1585 *			with flush
1586 * @file:	file to synchronize
1587 * @start:	start offset in bytes
1588 * @end:	end offset in bytes (inclusive)
1589 * @datasync:	only synchronize essential metadata if true
1590 *
1591 */
1592
1593int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1594		       int datasync)
1595{
1596	struct inode *inode = file->f_mapping->host;
1597	int err;
1598
1599	err = __generic_file_fsync(file, start, end, datasync);
1600	if (err)
1601		return err;
1602	return blkdev_issue_flush(inode->i_sb->s_bdev);
1603}
1604EXPORT_SYMBOL(generic_file_fsync);
1605
1606/**
1607 * generic_check_addressable - Check addressability of file system
1608 * @blocksize_bits:	log of file system block size
1609 * @num_blocks:		number of blocks in file system
1610 *
1611 * Determine whether a file system with @num_blocks blocks (and a
1612 * block size of 2**@blocksize_bits) is addressable by the sector_t
1613 * and page cache of the system.  Return 0 if so and -EFBIG otherwise.
1614 */
1615int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1616{
1617	u64 last_fs_block = num_blocks - 1;
1618	u64 last_fs_page, max_bytes;
1619
1620	if (check_shl_overflow(num_blocks, blocksize_bits, &max_bytes))
1621		return -EFBIG;
1622
1623	last_fs_page = (max_bytes >> PAGE_SHIFT) - 1;
1624
1625	if (unlikely(num_blocks == 0))
1626		return 0;
1627
1628	if (blocksize_bits < 9)
1629		return -EINVAL;
1630
1631	if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1632	    (last_fs_page > (pgoff_t)(~0ULL))) {
1633		return -EFBIG;
1634	}
1635	return 0;
1636}
1637EXPORT_SYMBOL(generic_check_addressable);
1638
1639/*
1640 * No-op implementation of ->fsync for in-memory filesystems.
1641 */
1642int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1643{
1644	return 0;
1645}
1646EXPORT_SYMBOL(noop_fsync);
1647
1648ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1649{
1650	/*
1651	 * iomap based filesystems support direct I/O without need for
1652	 * this callback. However, it still needs to be set in
1653	 * inode->a_ops so that open/fcntl know that direct I/O is
1654	 * generally supported.
1655	 */
1656	return -EINVAL;
1657}
1658EXPORT_SYMBOL_GPL(noop_direct_IO);
1659
1660/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1661void kfree_link(void *p)
1662{
1663	kfree(p);
1664}
1665EXPORT_SYMBOL(kfree_link);
1666
1667struct inode *alloc_anon_inode(struct super_block *s)
1668{
1669	static const struct address_space_operations anon_aops = {
1670		.dirty_folio	= noop_dirty_folio,
1671	};
1672	struct inode *inode = new_inode_pseudo(s);
1673
1674	if (!inode)
1675		return ERR_PTR(-ENOMEM);
1676
1677	inode->i_ino = get_next_ino();
1678	inode->i_mapping->a_ops = &anon_aops;
1679
1680	/*
1681	 * Mark the inode dirty from the very beginning,
1682	 * that way it will never be moved to the dirty
1683	 * list because mark_inode_dirty() will think
1684	 * that it already _is_ on the dirty list.
1685	 */
1686	inode_state_assign_raw(inode, I_DIRTY);
1687	/*
1688	 * Historically anonymous inodes don't have a type at all and
1689	 * userspace has come to rely on this.
1690	 */
1691	inode->i_mode = S_IRUSR | S_IWUSR;
1692	inode->i_uid = current_fsuid();
1693	inode->i_gid = current_fsgid();
1694	inode->i_flags |= S_PRIVATE | S_ANON_INODE;
1695	simple_inode_init_ts(inode);
1696	return inode;
1697}
1698EXPORT_SYMBOL(alloc_anon_inode);
1699
1700/**
1701 * simple_nosetlease - generic helper for prohibiting leases
1702 * @filp: file pointer
1703 * @arg: type of lease to obtain
1704 * @flp: new lease supplied for insertion
1705 * @priv: private data for lm_setup operation
1706 *
1707 * Generic helper for filesystems that do not wish to allow leases to be set.
1708 * All arguments are ignored and it just returns -EINVAL.
1709 */
1710int
1711simple_nosetlease(struct file *filp, int arg, struct file_lease **flp,
1712		  void **priv)
1713{
1714	return -EINVAL;
1715}
1716EXPORT_SYMBOL(simple_nosetlease);
1717
1718/**
1719 * simple_get_link - generic helper to get the target of "fast" symlinks
1720 * @dentry: not used here
1721 * @inode: the symlink inode
1722 * @done: not used here
1723 *
1724 * Generic helper for filesystems to use for symlink inodes where a pointer to
1725 * the symlink target is stored in ->i_link.  NOTE: this isn't normally called,
1726 * since as an optimization the path lookup code uses any non-NULL ->i_link
1727 * directly, without calling ->get_link().  But ->get_link() still must be set,
1728 * to mark the inode_operations as being for a symlink.
1729 *
1730 * Return: the symlink target
1731 */
1732const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1733			    struct delayed_call *done)
1734{
1735	return inode->i_link;
1736}
1737EXPORT_SYMBOL(simple_get_link);
1738
1739const struct inode_operations simple_symlink_inode_operations = {
1740	.get_link = simple_get_link,
1741};
1742EXPORT_SYMBOL(simple_symlink_inode_operations);
1743
1744/*
1745 * Operations for a permanently empty directory.
1746 */
1747static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1748{
1749	return ERR_PTR(-ENOENT);
1750}
1751
1752static int empty_dir_setattr(struct mnt_idmap *idmap,
1753			     struct dentry *dentry, struct iattr *attr)
1754{
1755	return -EPERM;
1756}
1757
1758static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1759{
1760	return -EOPNOTSUPP;
1761}
1762
1763static const struct inode_operations empty_dir_inode_operations = {
1764	.lookup		= empty_dir_lookup,
1765	.setattr	= empty_dir_setattr,
1766	.listxattr	= empty_dir_listxattr,
1767};
1768
1769static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1770{
1771	/* An empty directory has two entries . and .. at offsets 0 and 1 */
1772	return generic_file_llseek_size(file, offset, whence, 2, 2);
1773}
1774
1775static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1776{
1777	dir_emit_dots(file, ctx);
1778	return 0;
1779}
1780
1781static const struct file_operations empty_dir_operations = {
1782	.llseek		= empty_dir_llseek,
1783	.read		= generic_read_dir,
1784	.iterate_shared	= empty_dir_readdir,
1785	.fsync		= noop_fsync,
1786};
1787
1788
1789void make_empty_dir_inode(struct inode *inode)
1790{
1791	set_nlink(inode, 2);
1792	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1793	inode->i_uid = GLOBAL_ROOT_UID;
1794	inode->i_gid = GLOBAL_ROOT_GID;
1795	inode->i_rdev = 0;
1796	inode->i_size = 0;
1797	inode->i_blkbits = PAGE_SHIFT;
1798	inode->i_blocks = 0;
1799
1800	inode->i_op = &empty_dir_inode_operations;
1801	inode->i_opflags &= ~IOP_XATTR;
1802	inode->i_fop = &empty_dir_operations;
1803}
1804
1805bool is_empty_dir_inode(struct inode *inode)
1806{
1807	return (inode->i_fop == &empty_dir_operations) &&
1808		(inode->i_op == &empty_dir_inode_operations);
1809}
1810
1811#if IS_ENABLED(CONFIG_UNICODE)
1812/**
1813 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1814 * @dentry:	dentry whose name we are checking against
1815 * @len:	len of name of dentry
1816 * @str:	str pointer to name of dentry
1817 * @name:	Name to compare against
1818 *
1819 * Return: 0 if names match, 1 if mismatch, or -ERRNO
1820 */
1821int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1822			 const char *str, const struct qstr *name)
1823{
1824	const struct dentry *parent;
1825	const struct inode *dir;
1826	union shortname_store strbuf;
1827	struct qstr qstr;
1828
1829	/*
1830	 * Attempt a case-sensitive match first. It is cheaper and
1831	 * should cover most lookups, including all the sane
1832	 * applications that expect a case-sensitive filesystem.
1833	 *
1834	 * This comparison is safe under RCU because the caller
1835	 * guarantees the consistency between str and len. See
1836	 * __d_lookup_rcu_op_compare() for details.
1837	 */
1838	if (len == name->len && !memcmp(str, name->name, len))
1839		return 0;
1840
1841	parent = READ_ONCE(dentry->d_parent);
1842	dir = READ_ONCE(parent->d_inode);
1843	if (!dir || !IS_CASEFOLDED(dir))
1844		return 1;
1845
1846	qstr.len = len;
1847	qstr.name = str;
1848	/*
1849	 * If the dentry name is stored in-line, then it may be concurrently
1850	 * modified by a rename.  If this happens, the VFS will eventually retry
1851	 * the lookup, so it doesn't matter what ->d_compare() returns.
1852	 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1853	 * string.  Therefore, we have to copy the name into a temporary buffer.
1854	 * As above, len is guaranteed to match str, so the shortname case
1855	 * is exactly when str points to ->d_shortname.
1856	 */
1857	if (qstr.name == dentry->d_shortname.string) {
1858		strbuf = dentry->d_shortname; // NUL is guaranteed to be in there
1859		qstr.name = strbuf.string;
1860		/* prevent compiler from optimizing out the temporary buffer */
1861		barrier();
1862	}
1863
1864	return utf8_strncasecmp(dentry->d_sb->s_encoding, name, &qstr);
1865}
1866EXPORT_SYMBOL(generic_ci_d_compare);
1867
1868/**
1869 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1870 * @dentry:	dentry of the parent directory
1871 * @str:	qstr of name whose hash we should fill in
1872 *
1873 * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1874 */
1875int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1876{
1877	const struct inode *dir = READ_ONCE(dentry->d_inode);
1878	struct super_block *sb = dentry->d_sb;
1879	const struct unicode_map *um = sb->s_encoding;
1880	int ret;
1881
1882	if (!dir || !IS_CASEFOLDED(dir))
1883		return 0;
1884
1885	ret = utf8_casefold_hash(um, dentry, str);
1886	if (ret < 0 && sb_has_strict_encoding(sb))
1887		return -EINVAL;
1888	return 0;
1889}
1890EXPORT_SYMBOL(generic_ci_d_hash);
1891
1892static const struct dentry_operations generic_ci_dentry_ops = {
1893	.d_hash = generic_ci_d_hash,
1894	.d_compare = generic_ci_d_compare,
1895#ifdef CONFIG_FS_ENCRYPTION
1896	.d_revalidate = fscrypt_d_revalidate,
1897#endif
1898};
1899
1900/**
1901 * generic_ci_match() - Match a name (case-insensitively) with a dirent.
1902 * This is a filesystem helper for comparison with directory entries.
1903 * generic_ci_d_compare should be used in VFS' ->d_compare instead.
1904 *
1905 * @parent: Inode of the parent of the dirent under comparison
1906 * @name: name under lookup.
1907 * @folded_name: Optional pre-folded name under lookup
1908 * @de_name: Dirent name.
1909 * @de_name_len: dirent name length.
1910 *
1911 * Test whether a case-insensitive directory entry matches the filename
1912 * being searched.  If @folded_name is provided, it is used instead of
1913 * recalculating the casefold of @name.
1914 *
1915 * Return: > 0 if the directory entry matches, 0 if it doesn't match, or
1916 * < 0 on error.
1917 */
1918int generic_ci_match(const struct inode *parent,
1919		     const struct qstr *name,
1920		     const struct qstr *folded_name,
1921		     const u8 *de_name, u32 de_name_len)
1922{
1923	const struct super_block *sb = parent->i_sb;
1924	const struct unicode_map *um = sb->s_encoding;
1925	struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1926	struct qstr dirent = QSTR_INIT(de_name, de_name_len);
1927	int res = 0;
1928
1929	if (IS_ENCRYPTED(parent)) {
1930		const struct fscrypt_str encrypted_name =
1931			FSTR_INIT((u8 *) de_name, de_name_len);
1932
1933		if (WARN_ON_ONCE(!fscrypt_has_encryption_key(parent)))
1934			return -EINVAL;
1935
1936		decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1937		if (!decrypted_name.name)
1938			return -ENOMEM;
1939		res = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1940						&decrypted_name);
1941		if (res < 0) {
1942			kfree(decrypted_name.name);
1943			return res;
1944		}
1945		dirent.name = decrypted_name.name;
1946		dirent.len = decrypted_name.len;
1947	}
1948
1949	/*
1950	 * Attempt a case-sensitive match first. It is cheaper and
1951	 * should cover most lookups, including all the sane
1952	 * applications that expect a case-sensitive filesystem.
1953	 */
1954
1955	if (dirent.len == name->len &&
1956	    !memcmp(name->name, dirent.name, dirent.len))
1957		goto out;
1958
1959	if (folded_name->name)
1960		res = utf8_strncasecmp_folded(um, folded_name, &dirent);
1961	else
1962		res = utf8_strncasecmp(um, name, &dirent);
1963
1964out:
1965	kfree(decrypted_name.name);
1966	if (res < 0 && sb_has_strict_encoding(sb)) {
1967		pr_err_ratelimited("Directory contains filename that is invalid UTF-8");
1968		return 0;
1969	}
1970	return !res;
1971}
1972EXPORT_SYMBOL(generic_ci_match);
1973#endif
1974
1975#ifdef CONFIG_FS_ENCRYPTION
1976static const struct dentry_operations generic_encrypted_dentry_ops = {
1977	.d_revalidate = fscrypt_d_revalidate,
1978};
1979#endif
1980
1981/**
1982 * generic_set_sb_d_ops - helper for choosing the set of
1983 * filesystem-wide dentry operations for the enabled features
1984 * @sb: superblock to be configured
1985 *
1986 * Filesystems supporting casefolding and/or fscrypt can call this
1987 * helper at mount-time to configure default dentry_operations to the
1988 * best set of dentry operations required for the enabled features.
1989 * The helper must be called after these have been configured, but
1990 * before the root dentry is created.
1991 */
1992void generic_set_sb_d_ops(struct super_block *sb)
1993{
1994#if IS_ENABLED(CONFIG_UNICODE)
1995	if (sb->s_encoding) {
1996		set_default_d_op(sb, &generic_ci_dentry_ops);
1997		return;
1998	}
1999#endif
2000#ifdef CONFIG_FS_ENCRYPTION
2001	if (sb->s_cop) {
2002		set_default_d_op(sb, &generic_encrypted_dentry_ops);
2003		return;
2004	}
2005#endif
2006}
2007EXPORT_SYMBOL(generic_set_sb_d_ops);
2008
2009/**
2010 * inode_maybe_inc_iversion - increments i_version
2011 * @inode: inode with the i_version that should be updated
2012 * @force: increment the counter even if it's not necessary?
2013 *
2014 * Every time the inode is modified, the i_version field must be seen to have
2015 * changed by any observer.
2016 *
2017 * If "force" is set or the QUERIED flag is set, then ensure that we increment
2018 * the value, and clear the queried flag.
2019 *
2020 * In the common case where neither is set, then we can return "false" without
2021 * updating i_version.
2022 *
2023 * If this function returns false, and no other metadata has changed, then we
2024 * can avoid logging the metadata.
2025 */
2026bool inode_maybe_inc_iversion(struct inode *inode, bool force)
2027{
2028	u64 cur, new;
2029
2030	/*
2031	 * The i_version field is not strictly ordered with any other inode
2032	 * information, but the legacy inode_inc_iversion code used a spinlock
2033	 * to serialize increments.
2034	 *
2035	 * We add a full memory barrier to ensure that any de facto ordering
2036	 * with other state is preserved (either implicitly coming from cmpxchg
2037	 * or explicitly from smp_mb if we don't know upfront if we will execute
2038	 * the former).
2039	 *
2040	 * These barriers pair with inode_query_iversion().
2041	 */
2042	cur = inode_peek_iversion_raw(inode);
2043	if (!force && !(cur & I_VERSION_QUERIED)) {
2044		smp_mb();
2045		cur = inode_peek_iversion_raw(inode);
2046	}
2047
2048	do {
2049		/* If flag is clear then we needn't do anything */
2050		if (!force && !(cur & I_VERSION_QUERIED))
2051			return false;
2052
2053		/* Since lowest bit is flag, add 2 to avoid it */
2054		new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
2055	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2056	return true;
2057}
2058EXPORT_SYMBOL(inode_maybe_inc_iversion);
2059
2060/**
2061 * inode_query_iversion - read i_version for later use
2062 * @inode: inode from which i_version should be read
2063 *
2064 * Read the inode i_version counter. This should be used by callers that wish
2065 * to store the returned i_version for later comparison. This will guarantee
2066 * that a later query of the i_version will result in a different value if
2067 * anything has changed.
2068 *
2069 * In this implementation, we fetch the current value, set the QUERIED flag and
2070 * then try to swap it into place with a cmpxchg, if it wasn't already set. If
2071 * that fails, we try again with the newly fetched value from the cmpxchg.
2072 */
2073u64 inode_query_iversion(struct inode *inode)
2074{
2075	u64 cur, new;
2076	bool fenced = false;
2077
2078	/*
2079	 * Memory barriers (implicit in cmpxchg, explicit in smp_mb) pair with
2080	 * inode_maybe_inc_iversion(), see that routine for more details.
2081	 */
2082	cur = inode_peek_iversion_raw(inode);
2083	do {
2084		/* If flag is already set, then no need to swap */
2085		if (cur & I_VERSION_QUERIED) {
2086			if (!fenced)
2087				smp_mb();
2088			break;
2089		}
2090
2091		fenced = true;
2092		new = cur | I_VERSION_QUERIED;
2093	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2094	return cur >> I_VERSION_QUERIED_SHIFT;
2095}
2096EXPORT_SYMBOL(inode_query_iversion);
2097
2098ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
2099		ssize_t direct_written, ssize_t buffered_written)
2100{
2101	struct address_space *mapping = iocb->ki_filp->f_mapping;
2102	loff_t pos = iocb->ki_pos - buffered_written;
2103	loff_t end = iocb->ki_pos - 1;
2104	int err;
2105
2106	/*
2107	 * If the buffered write fallback returned an error, we want to return
2108	 * the number of bytes which were written by direct I/O, or the error
2109	 * code if that was zero.
2110	 *
2111	 * Note that this differs from normal direct-io semantics, which will
2112	 * return -EFOO even if some bytes were written.
2113	 */
2114	if (unlikely(buffered_written < 0)) {
2115		if (direct_written)
2116			return direct_written;
2117		return buffered_written;
2118	}
2119
2120	/*
2121	 * We need to ensure that the page cache pages are written to disk and
2122	 * invalidated to preserve the expected O_DIRECT semantics.
2123	 */
2124	err = filemap_write_and_wait_range(mapping, pos, end);
2125	if (err < 0) {
2126		/*
2127		 * We don't know how much we wrote, so just return the number of
2128		 * bytes which were direct-written
2129		 */
2130		iocb->ki_pos -= buffered_written;
2131		if (direct_written)
2132			return direct_written;
2133		return err;
2134	}
2135	invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
2136	return direct_written + buffered_written;
2137}
2138EXPORT_SYMBOL_GPL(direct_write_fallback);
2139
2140/**
2141 * simple_inode_init_ts - initialize the timestamps for a new inode
2142 * @inode: inode to be initialized
2143 *
2144 * When a new inode is created, most filesystems set the timestamps to the
2145 * current time. Add a helper to do this.
2146 */
2147struct timespec64 simple_inode_init_ts(struct inode *inode)
2148{
2149	struct timespec64 ts = inode_set_ctime_current(inode);
2150
2151	inode_set_atime_to_ts(inode, ts);
2152	inode_set_mtime_to_ts(inode, ts);
2153	return ts;
2154}
2155EXPORT_SYMBOL(simple_inode_init_ts);
2156
2157struct dentry *stashed_dentry_get(struct dentry **stashed)
2158{
2159	struct dentry *dentry;
2160
2161	guard(rcu)();
2162	dentry = rcu_dereference(*stashed);
2163	if (!dentry)
2164		return NULL;
2165	if (IS_ERR(dentry))
2166		return dentry;
2167	if (!lockref_get_not_dead(&dentry->d_lockref))
2168		return NULL;
2169	return dentry;
2170}
2171
2172static struct dentry *prepare_anon_dentry(struct dentry **stashed,
2173					  struct super_block *sb,
2174					  void *data)
2175{
2176	struct dentry *dentry;
2177	struct inode *inode;
2178	const struct stashed_operations *sops = sb->s_fs_info;
2179	int ret;
2180
2181	inode = new_inode_pseudo(sb);
2182	if (!inode) {
2183		sops->put_data(data);
2184		return ERR_PTR(-ENOMEM);
2185	}
2186
2187	inode->i_flags |= S_IMMUTABLE;
2188	inode->i_mode = S_IFREG;
2189	simple_inode_init_ts(inode);
2190
2191	ret = sops->init_inode(inode, data);
2192	if (ret < 0) {
2193		iput(inode);
2194		return ERR_PTR(ret);
2195	}
2196
2197	/* Notice when this is changed. */
2198	WARN_ON_ONCE(!S_ISREG(inode->i_mode));
2199
2200	dentry = d_alloc_anon(sb);
2201	if (!dentry) {
2202		iput(inode);
2203		return ERR_PTR(-ENOMEM);
2204	}
2205
2206	/* Store address of location where dentry's supposed to be stashed. */
2207	dentry->d_fsdata = stashed;
2208
2209	/* @data is now owned by the fs */
2210	d_instantiate(dentry, inode);
2211	return dentry;
2212}
2213
2214struct dentry *stash_dentry(struct dentry **stashed, struct dentry *dentry)
2215{
2216	guard(rcu)();
2217	for (;;) {
2218		struct dentry *old;
2219
2220		/* Assume any old dentry was cleared out. */
2221		old = cmpxchg(stashed, NULL, dentry);
2222		if (likely(!old))
2223			return dentry;
2224
2225		/* Check if somebody else installed a reusable dentry. */
2226		if (lockref_get_not_dead(&old->d_lockref))
2227			return old;
2228
2229		/* There's an old dead dentry there, try to take it over. */
2230		if (likely(try_cmpxchg(stashed, &old, dentry)))
2231			return dentry;
2232	}
2233}
2234
2235/**
2236 * path_from_stashed - create path from stashed or new dentry
2237 * @stashed:    where to retrieve or stash dentry
2238 * @mnt:        mnt of the filesystems to use
2239 * @data:       data to store in inode->i_private
2240 * @path:       path to create
2241 *
2242 * The function tries to retrieve a stashed dentry from @stashed. If the dentry
2243 * is still valid then it will be reused. If the dentry isn't able the function
2244 * will allocate a new dentry and inode. It will then check again whether it
2245 * can reuse an existing dentry in case one has been added in the meantime or
2246 * update @stashed with the newly added dentry.
2247 *
2248 * Special-purpose helper for nsfs and pidfs.
2249 *
2250 * Return: On success zero and on failure a negative error is returned.
2251 */
2252int path_from_stashed(struct dentry **stashed, struct vfsmount *mnt, void *data,
2253		      struct path *path)
2254{
2255	struct dentry *dentry, *res;
2256	const struct stashed_operations *sops = mnt->mnt_sb->s_fs_info;
2257
2258	/* See if dentry can be reused. */
2259	res = stashed_dentry_get(stashed);
2260	if (IS_ERR(res))
2261		return PTR_ERR(res);
2262	if (res) {
2263		sops->put_data(data);
2264		goto make_path;
2265	}
2266
2267	/* Allocate a new dentry. */
2268	dentry = prepare_anon_dentry(stashed, mnt->mnt_sb, data);
2269	if (IS_ERR(dentry))
2270		return PTR_ERR(dentry);
2271
2272	/* Added a new dentry. @data is now owned by the filesystem. */
2273	if (sops->stash_dentry)
2274		res = sops->stash_dentry(stashed, dentry);
2275	else
2276		res = stash_dentry(stashed, dentry);
2277	if (IS_ERR(res)) {
2278		dput(dentry);
2279		return PTR_ERR(res);
2280	}
2281	if (res != dentry)
2282		dput(dentry);
2283
2284make_path:
2285	path->dentry = res;
2286	path->mnt = mntget(mnt);
2287	VFS_WARN_ON_ONCE(path->dentry->d_fsdata != stashed);
2288	VFS_WARN_ON_ONCE(d_inode(path->dentry)->i_private != data);
2289	return 0;
2290}
2291
2292void stashed_dentry_prune(struct dentry *dentry)
2293{
2294	struct dentry **stashed = dentry->d_fsdata;
2295	struct inode *inode = d_inode(dentry);
2296
2297	if (WARN_ON_ONCE(!stashed))
2298		return;
2299
2300	if (!inode)
2301		return;
2302
2303	/*
2304	 * Only replace our own @dentry as someone else might've
2305	 * already cleared out @dentry and stashed their own
2306	 * dentry in there.
2307	 */
2308	cmpxchg(stashed, dentry, NULL);
2309}
2310
2311/**
2312 * simple_start_creating - prepare to create a given name
2313 * @parent: directory in which to prepare to create the name
2314 * @name:   the name to be created
2315 *
2316 * Required lock is taken and a lookup in performed prior to creating an
2317 * object in a directory.  No permission checking is performed.
2318 *
2319 * Returns: a negative dentry on which vfs_create() or similar may
2320 *  be attempted, or an error.
2321 */
2322struct dentry *simple_start_creating(struct dentry *parent, const char *name)
2323{
2324	struct qstr qname = QSTR(name);
2325	int err;
2326
2327	err = lookup_noperm_common(&qname, parent);
2328	if (err)
2329		return ERR_PTR(err);
2330	return start_dirop(parent, &qname, LOOKUP_CREATE | LOOKUP_EXCL);
2331}
2332EXPORT_SYMBOL(simple_start_creating);
2333
2334/* parent must have been held exclusive since simple_start_creating() */
2335void simple_done_creating(struct dentry *child)
2336{
2337	inode_unlock(child->d_parent->d_inode);
2338	dput(child);
2339}
2340EXPORT_SYMBOL(simple_done_creating);