dir.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * fs/f2fs/dir.c
   4 *
   5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6 *             http://www.samsung.com/
   7 */
   8#include <linux/unaligned.h>
   9#include <linux/fs.h>
  10#include <linux/f2fs_fs.h>
  11#include <linux/sched/signal.h>
  12#include <linux/unicode.h>
  13#include "f2fs.h"
  14#include "node.h"
  15#include "acl.h"
  16#include "xattr.h"
  17#include <trace/events/f2fs.h>
  18
  19static inline bool f2fs_should_fallback_to_linear(struct inode *dir)
  20{
  21	struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
  22
  23	switch (F2FS_OPTION(sbi).lookup_mode) {
  24	case LOOKUP_PERF:
  25		return false;
  26	case LOOKUP_COMPAT:
  27		return true;
  28	case LOOKUP_AUTO:
  29		return !sb_no_casefold_compat_fallback(sbi->sb);
  30	}
  31	return false;
  32}
  33
  34#if IS_ENABLED(CONFIG_UNICODE)
  35extern struct kmem_cache *f2fs_cf_name_slab;
  36#endif
  37
  38static unsigned long dir_blocks(struct inode *inode)
  39{
  40	return ((unsigned long long) (i_size_read(inode) + PAGE_SIZE - 1))
  41							>> PAGE_SHIFT;
  42}
  43
  44static unsigned int dir_buckets(unsigned int level, int dir_level)
  45{
  46	if (level + dir_level < MAX_DIR_HASH_DEPTH / 2)
  47		return BIT(level + dir_level);
  48	else
  49		return MAX_DIR_BUCKETS;
  50}
  51
  52static unsigned int bucket_blocks(unsigned int level)
  53{
  54	if (level < MAX_DIR_HASH_DEPTH / 2)
  55		return 2;
  56	else
  57		return 4;
  58}
  59
  60#if IS_ENABLED(CONFIG_UNICODE)
  61/* If @dir is casefolded, initialize @fname->cf_name from @fname->usr_fname. */
  62int f2fs_init_casefolded_name(const struct inode *dir,
  63			      struct f2fs_filename *fname)
  64{
  65	struct super_block *sb = dir->i_sb;
  66	unsigned char *buf;
  67	int len;
  68
  69	if (IS_CASEFOLDED(dir) &&
  70	    !is_dot_dotdot(fname->usr_fname->name, fname->usr_fname->len)) {
  71		buf = f2fs_kmem_cache_alloc(f2fs_cf_name_slab,
  72					    GFP_NOFS, false, F2FS_SB(sb));
  73		if (!buf)
  74			return -ENOMEM;
  75
  76		len = utf8_casefold(sb->s_encoding, fname->usr_fname,
  77				    buf, F2FS_NAME_LEN);
  78		if (len <= 0) {
  79			kmem_cache_free(f2fs_cf_name_slab, buf);
  80			if (sb_has_strict_encoding(sb))
  81				return -EINVAL;
  82			/* fall back to treating name as opaque byte sequence */
  83			return 0;
  84		}
  85		fname->cf_name.name = buf;
  86		fname->cf_name.len = len;
  87	}
  88
  89	return 0;
  90}
  91
  92void f2fs_free_casefolded_name(struct f2fs_filename *fname)
  93{
  94	unsigned char *buf = (unsigned char *)fname->cf_name.name;
  95
  96	if (buf) {
  97		kmem_cache_free(f2fs_cf_name_slab, buf);
  98		fname->cf_name.name = NULL;
  99	}
 100}
 101#endif /* CONFIG_UNICODE */
 102
 103static int __f2fs_setup_filename(const struct inode *dir,
 104				 const struct fscrypt_name *crypt_name,
 105				 struct f2fs_filename *fname)
 106{
 107	int err;
 108
 109	memset(fname, 0, sizeof(*fname));
 110
 111	fname->usr_fname = crypt_name->usr_fname;
 112	fname->disk_name = crypt_name->disk_name;
 113#ifdef CONFIG_FS_ENCRYPTION
 114	fname->crypto_buf = crypt_name->crypto_buf;
 115#endif
 116	if (crypt_name->is_nokey_name) {
 117		/* hash was decoded from the no-key name */
 118		fname->hash = cpu_to_le32(crypt_name->hash);
 119	} else {
 120		err = f2fs_init_casefolded_name(dir, fname);
 121		if (err) {
 122			f2fs_free_filename(fname);
 123			return err;
 124		}
 125		f2fs_hash_filename(dir, fname);
 126	}
 127	return 0;
 128}
 129
 130/*
 131 * Prepare to search for @iname in @dir.  This is similar to
 132 * fscrypt_setup_filename(), but this also handles computing the casefolded name
 133 * and the f2fs dirhash if needed, then packing all the information about this
 134 * filename up into a 'struct f2fs_filename'.
 135 */
 136int f2fs_setup_filename(struct inode *dir, const struct qstr *iname,
 137			int lookup, struct f2fs_filename *fname)
 138{
 139	struct fscrypt_name crypt_name;
 140	int err;
 141
 142	err = fscrypt_setup_filename(dir, iname, lookup, &crypt_name);
 143	if (err)
 144		return err;
 145
 146	return __f2fs_setup_filename(dir, &crypt_name, fname);
 147}
 148
 149/*
 150 * Prepare to look up @dentry in @dir.  This is similar to
 151 * fscrypt_prepare_lookup(), but this also handles computing the casefolded name
 152 * and the f2fs dirhash if needed, then packing all the information about this
 153 * filename up into a 'struct f2fs_filename'.
 154 */
 155int f2fs_prepare_lookup(struct inode *dir, struct dentry *dentry,
 156			struct f2fs_filename *fname)
 157{
 158	struct fscrypt_name crypt_name;
 159	int err;
 160
 161	err = fscrypt_prepare_lookup(dir, dentry, &crypt_name);
 162	if (err)
 163		return err;
 164
 165	return __f2fs_setup_filename(dir, &crypt_name, fname);
 166}
 167
 168void f2fs_free_filename(struct f2fs_filename *fname)
 169{
 170#ifdef CONFIG_FS_ENCRYPTION
 171	kfree(fname->crypto_buf.name);
 172	fname->crypto_buf.name = NULL;
 173#endif
 174	f2fs_free_casefolded_name(fname);
 175}
 176
 177static unsigned long dir_block_index(unsigned int level,
 178				int dir_level, unsigned int idx)
 179{
 180	unsigned long i;
 181	unsigned long bidx = 0;
 182
 183	for (i = 0; i < level; i++)
 184		bidx += mul_u32_u32(dir_buckets(i, dir_level),
 185				    bucket_blocks(i));
 186	bidx += idx * bucket_blocks(level);
 187	return bidx;
 188}
 189
 190static struct f2fs_dir_entry *find_in_block(struct inode *dir,
 191				struct folio *dentry_folio,
 192				const struct f2fs_filename *fname,
 193				int *max_slots,
 194				bool use_hash)
 195{
 196	struct f2fs_dentry_block *dentry_blk;
 197	struct f2fs_dentry_ptr d;
 198
 199	dentry_blk = folio_address(dentry_folio);
 200
 201	make_dentry_ptr_block(dir, &d, dentry_blk);
 202	return f2fs_find_target_dentry(&d, fname, max_slots, use_hash);
 203}
 204
 205static inline int f2fs_match_name(const struct inode *dir,
 206				   const struct f2fs_filename *fname,
 207				   const u8 *de_name, u32 de_name_len)
 208{
 209	struct fscrypt_name f;
 210
 211#if IS_ENABLED(CONFIG_UNICODE)
 212	if (fname->cf_name.name)
 213		return generic_ci_match(dir, fname->usr_fname,
 214					&fname->cf_name,
 215					de_name, de_name_len);
 216
 217#endif
 218	f.usr_fname = fname->usr_fname;
 219	f.disk_name = fname->disk_name;
 220#ifdef CONFIG_FS_ENCRYPTION
 221	f.crypto_buf = fname->crypto_buf;
 222#endif
 223	return fscrypt_match_name(&f, de_name, de_name_len);
 224}
 225
 226struct f2fs_dir_entry *f2fs_find_target_dentry(const struct f2fs_dentry_ptr *d,
 227			const struct f2fs_filename *fname, int *max_slots,
 228			bool use_hash)
 229{
 230	struct f2fs_dir_entry *de;
 231	unsigned long bit_pos = 0;
 232	int max_len = 0;
 233	int res = 0;
 234
 235	if (max_slots)
 236		*max_slots = 0;
 237	while (bit_pos < d->max) {
 238		if (!test_bit_le(bit_pos, d->bitmap)) {
 239			bit_pos++;
 240			max_len++;
 241			continue;
 242		}
 243
 244		de = &d->dentry[bit_pos];
 245
 246		if (unlikely(!de->name_len)) {
 247			bit_pos++;
 248			continue;
 249		}
 250
 251		if (!use_hash || de->hash_code == fname->hash) {
 252			res = f2fs_match_name(d->inode, fname,
 253					      d->filename[bit_pos],
 254					      le16_to_cpu(de->name_len));
 255			if (res < 0)
 256				return ERR_PTR(res);
 257			if (res)
 258				goto found;
 259		}
 260
 261		if (max_slots && max_len > *max_slots)
 262			*max_slots = max_len;
 263		max_len = 0;
 264
 265		bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
 266	}
 267
 268	de = NULL;
 269found:
 270	if (max_slots && max_len > *max_slots)
 271		*max_slots = max_len;
 272	return de;
 273}
 274
 275static struct f2fs_dir_entry *find_in_level(struct inode *dir,
 276					unsigned int level,
 277					const struct f2fs_filename *fname,
 278					struct folio **res_folio,
 279					bool use_hash)
 280{
 281	int s = GET_DENTRY_SLOTS(fname->disk_name.len);
 282	unsigned int nbucket, nblock;
 283	unsigned int bidx, end_block, bucket_no;
 284	struct f2fs_dir_entry *de = NULL;
 285	pgoff_t next_pgofs;
 286	bool room = false;
 287	int max_slots;
 288
 289	nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
 290	nblock = bucket_blocks(level);
 291
 292	bucket_no = use_hash ? le32_to_cpu(fname->hash) % nbucket : 0;
 293
 294start_find_bucket:
 295	bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
 296			       bucket_no);
 297	end_block = bidx + nblock;
 298
 299	while (bidx < end_block) {
 300		/* no need to allocate new dentry pages to all the indices */
 301		struct folio *dentry_folio;
 302		dentry_folio = f2fs_find_data_folio(dir, bidx, &next_pgofs);
 303		if (IS_ERR(dentry_folio)) {
 304			if (PTR_ERR(dentry_folio) == -ENOENT) {
 305				room = true;
 306				bidx = next_pgofs;
 307				continue;
 308			} else {
 309				*res_folio = dentry_folio;
 310				break;
 311			}
 312		}
 313
 314		de = find_in_block(dir, dentry_folio, fname, &max_slots, use_hash);
 315		if (IS_ERR(de)) {
 316			*res_folio = ERR_CAST(de);
 317			de = NULL;
 318			break;
 319		} else if (de) {
 320			*res_folio = dentry_folio;
 321			break;
 322		}
 323
 324		if (max_slots >= s)
 325			room = true;
 326		f2fs_folio_put(dentry_folio, false);
 327
 328		bidx++;
 329	}
 330
 331	if (de)
 332		return de;
 333
 334	if (likely(use_hash)) {
 335		if (room && F2FS_I(dir)->chash != fname->hash) {
 336			F2FS_I(dir)->chash = fname->hash;
 337			F2FS_I(dir)->clevel = level;
 338		}
 339	} else if (++bucket_no < nbucket) {
 340		goto start_find_bucket;
 341	}
 342	return NULL;
 343}
 344
 345struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir,
 346					 const struct f2fs_filename *fname,
 347					 struct folio **res_folio)
 348{
 349	unsigned long npages = dir_blocks(dir);
 350	struct f2fs_dir_entry *de = NULL;
 351	unsigned int max_depth;
 352	unsigned int level;
 353	bool use_hash = true;
 354
 355	*res_folio = NULL;
 356
 357#if IS_ENABLED(CONFIG_UNICODE)
 358start_find_entry:
 359#endif
 360	if (f2fs_has_inline_dentry(dir)) {
 361		de = f2fs_find_in_inline_dir(dir, fname, res_folio, use_hash);
 362		goto out;
 363	}
 364
 365	if (npages == 0)
 366		goto out;
 367
 368	max_depth = F2FS_I(dir)->i_current_depth;
 369	if (unlikely(max_depth > MAX_DIR_HASH_DEPTH)) {
 370		f2fs_warn(F2FS_I_SB(dir), "Corrupted max_depth of %lu: %u",
 371			  dir->i_ino, max_depth);
 372		max_depth = MAX_DIR_HASH_DEPTH;
 373		f2fs_i_depth_write(dir, max_depth);
 374	}
 375
 376	for (level = 0; level < max_depth; level++) {
 377		de = find_in_level(dir, level, fname, res_folio, use_hash);
 378		if (de || IS_ERR(*res_folio))
 379			break;
 380	}
 381
 382out:
 383#if IS_ENABLED(CONFIG_UNICODE)
 384	if (f2fs_should_fallback_to_linear(dir) &&
 385		IS_CASEFOLDED(dir) && !de && use_hash) {
 386		use_hash = false;
 387		goto start_find_entry;
 388	}
 389#endif
 390	/* This is to increase the speed of f2fs_create */
 391	if (!de)
 392		F2FS_I(dir)->task = current;
 393	return de;
 394}
 395
 396/*
 397 * Find an entry in the specified directory with the wanted name.
 398 * It returns the page where the entry was found (as a parameter - res_page),
 399 * and the entry itself. Page is returned mapped and unlocked.
 400 * Entry is guaranteed to be valid.
 401 */
 402struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir,
 403			const struct qstr *child, struct folio **res_folio)
 404{
 405	struct f2fs_dir_entry *de = NULL;
 406	struct f2fs_filename fname;
 407	int err;
 408
 409	err = f2fs_setup_filename(dir, child, 1, &fname);
 410	if (err) {
 411		if (err == -ENOENT)
 412			*res_folio = NULL;
 413		else
 414			*res_folio = ERR_PTR(err);
 415		return NULL;
 416	}
 417
 418	de = __f2fs_find_entry(dir, &fname, res_folio);
 419
 420	f2fs_free_filename(&fname);
 421	return de;
 422}
 423
 424struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct folio **f)
 425{
 426	return f2fs_find_entry(dir, &dotdot_name, f);
 427}
 428
 429ino_t f2fs_inode_by_name(struct inode *dir, const struct qstr *qstr,
 430							struct folio **folio)
 431{
 432	ino_t res = 0;
 433	struct f2fs_dir_entry *de;
 434
 435	de = f2fs_find_entry(dir, qstr, folio);
 436	if (de) {
 437		res = le32_to_cpu(de->ino);
 438		f2fs_folio_put(*folio, false);
 439	}
 440
 441	return res;
 442}
 443
 444void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
 445		struct folio *folio, struct inode *inode)
 446{
 447	enum page_type type = f2fs_has_inline_dentry(dir) ? NODE : DATA;
 448
 449	folio_lock(folio);
 450	f2fs_folio_wait_writeback(folio, type, true, true);
 451	de->ino = cpu_to_le32(inode->i_ino);
 452	de->file_type = fs_umode_to_ftype(inode->i_mode);
 453	folio_mark_dirty(folio);
 454
 455	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
 456	f2fs_mark_inode_dirty_sync(dir, false);
 457	f2fs_folio_put(folio, true);
 458}
 459
 460static void init_dent_inode(struct inode *dir, struct inode *inode,
 461			    const struct f2fs_filename *fname,
 462			    struct folio *ifolio)
 463{
 464	struct f2fs_inode *ri;
 465
 466	if (!fname) /* tmpfile case? */
 467		return;
 468
 469	f2fs_folio_wait_writeback(ifolio, NODE, true, true);
 470
 471	/* copy name info. to this inode folio */
 472	ri = F2FS_INODE(ifolio);
 473	ri->i_namelen = cpu_to_le32(fname->disk_name.len);
 474	memcpy(ri->i_name, fname->disk_name.name, fname->disk_name.len);
 475	if (IS_ENCRYPTED(dir)) {
 476		file_set_enc_name(inode);
 477		/*
 478		 * Roll-forward recovery doesn't have encryption keys available,
 479		 * so it can't compute the dirhash for encrypted+casefolded
 480		 * filenames.  Append it to i_name if possible.  Else, disable
 481		 * roll-forward recovery of the dentry (i.e., make fsync'ing the
 482		 * file force a checkpoint) by setting LOST_PINO.
 483		 */
 484		if (IS_CASEFOLDED(dir)) {
 485			if (fname->disk_name.len + sizeof(f2fs_hash_t) <=
 486			    F2FS_NAME_LEN)
 487				put_unaligned(fname->hash, (f2fs_hash_t *)
 488					&ri->i_name[fname->disk_name.len]);
 489			else
 490				file_lost_pino(inode);
 491		}
 492	}
 493	folio_mark_dirty(ifolio);
 494}
 495
 496void f2fs_do_make_empty_dir(struct inode *inode, struct inode *parent,
 497					struct f2fs_dentry_ptr *d)
 498{
 499	struct fscrypt_str dot = FSTR_INIT(".", 1);
 500	struct fscrypt_str dotdot = FSTR_INIT("..", 2);
 501
 502	/* update dirent of "." */
 503	f2fs_update_dentry(inode->i_ino, inode->i_mode, d, &dot, 0, 0);
 504
 505	/* update dirent of ".." */
 506	f2fs_update_dentry(parent->i_ino, parent->i_mode, d, &dotdot, 0, 1);
 507}
 508
 509static int make_empty_dir(struct inode *inode,
 510		struct inode *parent, struct folio *folio)
 511{
 512	struct folio *dentry_folio;
 513	struct f2fs_dentry_block *dentry_blk;
 514	struct f2fs_dentry_ptr d;
 515
 516	if (f2fs_has_inline_dentry(inode))
 517		return f2fs_make_empty_inline_dir(inode, parent, folio);
 518
 519	dentry_folio = f2fs_get_new_data_folio(inode, folio, 0, true);
 520	if (IS_ERR(dentry_folio))
 521		return PTR_ERR(dentry_folio);
 522
 523	dentry_blk = folio_address(dentry_folio);
 524
 525	make_dentry_ptr_block(NULL, &d, dentry_blk);
 526	f2fs_do_make_empty_dir(inode, parent, &d);
 527
 528	folio_mark_dirty(dentry_folio);
 529	f2fs_folio_put(dentry_folio, true);
 530	return 0;
 531}
 532
 533struct folio *f2fs_init_inode_metadata(struct inode *inode, struct inode *dir,
 534		const struct f2fs_filename *fname, struct folio *dfolio)
 535{
 536	struct folio *folio;
 537	int err;
 538
 539	if (is_inode_flag_set(inode, FI_NEW_INODE)) {
 540		folio = f2fs_new_inode_folio(inode);
 541		if (IS_ERR(folio))
 542			return folio;
 543
 544		if (S_ISDIR(inode->i_mode)) {
 545			/* in order to handle error case */
 546			folio_get(folio);
 547			err = make_empty_dir(inode, dir, folio);
 548			if (err) {
 549				folio_lock(folio);
 550				goto put_error;
 551			}
 552			folio_put(folio);
 553		}
 554
 555		err = f2fs_init_acl(inode, dir, folio, dfolio);
 556		if (err)
 557			goto put_error;
 558
 559		err = f2fs_init_security(inode, dir,
 560					 fname ? fname->usr_fname : NULL,
 561					 folio);
 562		if (err)
 563			goto put_error;
 564
 565		if (IS_ENCRYPTED(inode)) {
 566			err = fscrypt_set_context(inode, folio);
 567			if (err)
 568				goto put_error;
 569		}
 570	} else {
 571		folio = f2fs_get_inode_folio(F2FS_I_SB(dir), inode->i_ino);
 572		if (IS_ERR(folio))
 573			return folio;
 574	}
 575
 576	init_dent_inode(dir, inode, fname, folio);
 577
 578	/*
 579	 * This file should be checkpointed during fsync.
 580	 * We lost i_pino from now on.
 581	 */
 582	if (is_inode_flag_set(inode, FI_INC_LINK)) {
 583		if (!S_ISDIR(inode->i_mode))
 584			file_lost_pino(inode);
 585		/*
 586		 * If link the tmpfile to alias through linkat path,
 587		 * we should remove this inode from orphan list.
 588		 */
 589		if (inode->i_nlink == 0)
 590			f2fs_remove_orphan_inode(F2FS_I_SB(dir), inode->i_ino);
 591		f2fs_i_links_write(inode, true);
 592	}
 593	return folio;
 594
 595put_error:
 596	clear_nlink(inode);
 597	f2fs_update_inode(inode, folio);
 598	f2fs_folio_put(folio, true);
 599	return ERR_PTR(err);
 600}
 601
 602void f2fs_update_parent_metadata(struct inode *dir, struct inode *inode,
 603						unsigned int current_depth)
 604{
 605	if (inode && is_inode_flag_set(inode, FI_NEW_INODE)) {
 606		if (S_ISDIR(inode->i_mode))
 607			f2fs_i_links_write(dir, true);
 608		clear_inode_flag(inode, FI_NEW_INODE);
 609	}
 610	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
 611	f2fs_mark_inode_dirty_sync(dir, false);
 612
 613	if (F2FS_I(dir)->i_current_depth != current_depth)
 614		f2fs_i_depth_write(dir, current_depth);
 615
 616	if (inode && is_inode_flag_set(inode, FI_INC_LINK))
 617		clear_inode_flag(inode, FI_INC_LINK);
 618}
 619
 620int f2fs_room_for_filename(const void *bitmap, int slots, int max_slots)
 621{
 622	int bit_start = 0;
 623	int zero_start, zero_end;
 624next:
 625	zero_start = find_next_zero_bit_le(bitmap, max_slots, bit_start);
 626	if (zero_start >= max_slots)
 627		return max_slots;
 628
 629	zero_end = find_next_bit_le(bitmap, max_slots, zero_start);
 630	if (zero_end - zero_start >= slots)
 631		return zero_start;
 632
 633	bit_start = zero_end + 1;
 634
 635	if (zero_end + 1 >= max_slots)
 636		return max_slots;
 637	goto next;
 638}
 639
 640bool f2fs_has_enough_room(struct inode *dir, struct folio *ifolio,
 641			  const struct f2fs_filename *fname)
 642{
 643	struct f2fs_dentry_ptr d;
 644	unsigned int bit_pos;
 645	int slots = GET_DENTRY_SLOTS(fname->disk_name.len);
 646
 647	make_dentry_ptr_inline(dir, &d, inline_data_addr(dir, ifolio));
 648
 649	bit_pos = f2fs_room_for_filename(d.bitmap, slots, d.max);
 650
 651	return bit_pos < d.max;
 652}
 653
 654void f2fs_update_dentry(nid_t ino, umode_t mode, struct f2fs_dentry_ptr *d,
 655			const struct fscrypt_str *name, f2fs_hash_t name_hash,
 656			unsigned int bit_pos)
 657{
 658	struct f2fs_dir_entry *de;
 659	int slots = GET_DENTRY_SLOTS(name->len);
 660	int i;
 661
 662	de = &d->dentry[bit_pos];
 663	de->hash_code = name_hash;
 664	de->name_len = cpu_to_le16(name->len);
 665	memcpy(d->filename[bit_pos], name->name, name->len);
 666	de->ino = cpu_to_le32(ino);
 667	de->file_type = fs_umode_to_ftype(mode);
 668	for (i = 0; i < slots; i++) {
 669		__set_bit_le(bit_pos + i, (void *)d->bitmap);
 670		/* avoid wrong garbage data for readdir */
 671		if (i)
 672			(de + i)->name_len = 0;
 673	}
 674}
 675
 676int f2fs_add_regular_entry(struct inode *dir, const struct f2fs_filename *fname,
 677			   struct inode *inode, nid_t ino, umode_t mode)
 678{
 679	unsigned int bit_pos;
 680	unsigned int level;
 681	unsigned int current_depth;
 682	unsigned long bidx, block;
 683	unsigned int nbucket, nblock;
 684	struct folio *dentry_folio = NULL;
 685	struct f2fs_dentry_block *dentry_blk = NULL;
 686	struct f2fs_dentry_ptr d;
 687	struct folio *folio = NULL;
 688	int slots, err = 0;
 689
 690	level = 0;
 691	slots = GET_DENTRY_SLOTS(fname->disk_name.len);
 692
 693	current_depth = F2FS_I(dir)->i_current_depth;
 694	if (F2FS_I(dir)->chash == fname->hash) {
 695		level = F2FS_I(dir)->clevel;
 696		F2FS_I(dir)->chash = 0;
 697	}
 698
 699start:
 700	if (time_to_inject(F2FS_I_SB(dir), FAULT_DIR_DEPTH))
 701		return -ENOSPC;
 702
 703	if (unlikely(current_depth == MAX_DIR_HASH_DEPTH))
 704		return -ENOSPC;
 705
 706	/* Increase the depth, if required */
 707	if (level == current_depth)
 708		++current_depth;
 709
 710	nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
 711	nblock = bucket_blocks(level);
 712
 713	bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
 714				(le32_to_cpu(fname->hash) % nbucket));
 715
 716	for (block = bidx; block <= (bidx + nblock - 1); block++) {
 717		dentry_folio = f2fs_get_new_data_folio(dir, NULL, block, true);
 718		if (IS_ERR(dentry_folio))
 719			return PTR_ERR(dentry_folio);
 720
 721		dentry_blk = folio_address(dentry_folio);
 722		bit_pos = f2fs_room_for_filename(&dentry_blk->dentry_bitmap,
 723						slots, NR_DENTRY_IN_BLOCK);
 724		if (bit_pos < NR_DENTRY_IN_BLOCK)
 725			goto add_dentry;
 726
 727		f2fs_folio_put(dentry_folio, true);
 728	}
 729
 730	/* Move to next level to find the empty slot for new dentry */
 731	++level;
 732	goto start;
 733add_dentry:
 734	f2fs_folio_wait_writeback(dentry_folio, DATA, true, true);
 735
 736	if (inode) {
 737		f2fs_down_write(&F2FS_I(inode)->i_sem);
 738		folio = f2fs_init_inode_metadata(inode, dir, fname, NULL);
 739		if (IS_ERR(folio)) {
 740			err = PTR_ERR(folio);
 741			goto fail;
 742		}
 743	}
 744
 745	make_dentry_ptr_block(NULL, &d, dentry_blk);
 746	f2fs_update_dentry(ino, mode, &d, &fname->disk_name, fname->hash,
 747			   bit_pos);
 748
 749	folio_mark_dirty(dentry_folio);
 750
 751	if (inode) {
 752		f2fs_i_pino_write(inode, dir->i_ino);
 753
 754		/* synchronize inode page's data from inode cache */
 755		if (is_inode_flag_set(inode, FI_NEW_INODE))
 756			f2fs_update_inode(inode, folio);
 757
 758		f2fs_folio_put(folio, true);
 759	}
 760
 761	f2fs_update_parent_metadata(dir, inode, current_depth);
 762fail:
 763	if (inode)
 764		f2fs_up_write(&F2FS_I(inode)->i_sem);
 765
 766	f2fs_folio_put(dentry_folio, true);
 767
 768	return err;
 769}
 770
 771int f2fs_add_dentry(struct inode *dir, const struct f2fs_filename *fname,
 772		    struct inode *inode, nid_t ino, umode_t mode)
 773{
 774	int err = -EAGAIN;
 775
 776	if (f2fs_has_inline_dentry(dir)) {
 777		/*
 778		 * Should get i_xattr_sem to keep the lock order:
 779		 * i_xattr_sem -> inode_page lock used by f2fs_setxattr.
 780		 */
 781		f2fs_down_read(&F2FS_I(dir)->i_xattr_sem);
 782		err = f2fs_add_inline_entry(dir, fname, inode, ino, mode);
 783		f2fs_up_read(&F2FS_I(dir)->i_xattr_sem);
 784	}
 785	if (err == -EAGAIN)
 786		err = f2fs_add_regular_entry(dir, fname, inode, ino, mode);
 787
 788	f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
 789	return err;
 790}
 791
 792/*
 793 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 794 * f2fs_unlock_op().
 795 */
 796int f2fs_do_add_link(struct inode *dir, const struct qstr *name,
 797				struct inode *inode, nid_t ino, umode_t mode)
 798{
 799	struct f2fs_filename fname;
 800	struct folio *folio = NULL;
 801	struct f2fs_dir_entry *de = NULL;
 802	int err;
 803
 804	err = f2fs_setup_filename(dir, name, 0, &fname);
 805	if (err)
 806		return err;
 807
 808	/*
 809	 * An immature stackable filesystem shows a race condition between lookup
 810	 * and create. If we have same task when doing lookup and create, it's
 811	 * definitely fine as expected by VFS normally. Otherwise, let's just
 812	 * verify on-disk dentry one more time, which guarantees filesystem
 813	 * consistency more.
 814	 */
 815	if (current != F2FS_I(dir)->task) {
 816		de = __f2fs_find_entry(dir, &fname, &folio);
 817		F2FS_I(dir)->task = NULL;
 818	}
 819	if (de) {
 820		f2fs_folio_put(folio, false);
 821		err = -EEXIST;
 822	} else if (IS_ERR(folio)) {
 823		err = PTR_ERR(folio);
 824	} else {
 825		err = f2fs_add_dentry(dir, &fname, inode, ino, mode);
 826	}
 827	f2fs_free_filename(&fname);
 828	return err;
 829}
 830
 831int f2fs_do_tmpfile(struct inode *inode, struct inode *dir,
 832					struct f2fs_filename *fname)
 833{
 834	struct folio *folio;
 835	int err = 0;
 836
 837	f2fs_down_write(&F2FS_I(inode)->i_sem);
 838	folio = f2fs_init_inode_metadata(inode, dir, fname, NULL);
 839	if (IS_ERR(folio)) {
 840		err = PTR_ERR(folio);
 841		goto fail;
 842	}
 843	f2fs_folio_put(folio, true);
 844
 845	clear_inode_flag(inode, FI_NEW_INODE);
 846	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
 847fail:
 848	f2fs_up_write(&F2FS_I(inode)->i_sem);
 849	return err;
 850}
 851
 852void f2fs_drop_nlink(struct inode *dir, struct inode *inode)
 853{
 854	struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
 855
 856	f2fs_down_write(&F2FS_I(inode)->i_sem);
 857
 858	if (S_ISDIR(inode->i_mode))
 859		f2fs_i_links_write(dir, false);
 860	inode_set_ctime_current(inode);
 861
 862	f2fs_i_links_write(inode, false);
 863	if (S_ISDIR(inode->i_mode)) {
 864		f2fs_i_links_write(inode, false);
 865		f2fs_i_size_write(inode, 0);
 866	}
 867	f2fs_up_write(&F2FS_I(inode)->i_sem);
 868
 869	if (inode->i_nlink == 0)
 870		f2fs_add_orphan_inode(inode);
 871	else
 872		f2fs_release_orphan_inode(sbi);
 873}
 874
 875/*
 876 * It only removes the dentry from the dentry page, corresponding name
 877 * entry in name page does not need to be touched during deletion.
 878 */
 879void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct folio *folio,
 880					struct inode *dir, struct inode *inode)
 881{
 882	struct f2fs_dentry_block *dentry_blk;
 883	unsigned int bit_pos;
 884	int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
 885	pgoff_t index = folio->index;
 886	int i;
 887
 888	f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
 889
 890	if (F2FS_OPTION(F2FS_I_SB(dir)).fsync_mode == FSYNC_MODE_STRICT)
 891		f2fs_add_ino_entry(F2FS_I_SB(dir), dir->i_ino, TRANS_DIR_INO);
 892
 893	if (f2fs_has_inline_dentry(dir))
 894		return f2fs_delete_inline_entry(dentry, folio, dir, inode);
 895
 896	folio_lock(folio);
 897	f2fs_folio_wait_writeback(folio, DATA, true, true);
 898
 899	dentry_blk = folio_address(folio);
 900	bit_pos = dentry - dentry_blk->dentry;
 901	for (i = 0; i < slots; i++)
 902		__clear_bit_le(bit_pos + i, &dentry_blk->dentry_bitmap);
 903
 904	/* Let's check and deallocate this dentry page */
 905	bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
 906			NR_DENTRY_IN_BLOCK,
 907			0);
 908	folio_mark_dirty(folio);
 909
 910	if (bit_pos == NR_DENTRY_IN_BLOCK &&
 911		!f2fs_truncate_hole(dir, index, index + 1)) {
 912		f2fs_clear_page_cache_dirty_tag(folio);
 913		folio_clear_dirty_for_io(folio);
 914		folio_clear_uptodate(folio);
 915		folio_detach_private(folio);
 916
 917		inode_dec_dirty_pages(dir);
 918		f2fs_remove_dirty_inode(dir);
 919	}
 920	f2fs_folio_put(folio, true);
 921
 922	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
 923	f2fs_mark_inode_dirty_sync(dir, false);
 924
 925	if (inode)
 926		f2fs_drop_nlink(dir, inode);
 927}
 928
 929bool f2fs_empty_dir(struct inode *dir)
 930{
 931	unsigned long bidx = 0;
 932	unsigned int bit_pos;
 933	struct f2fs_dentry_block *dentry_blk;
 934	unsigned long nblock = dir_blocks(dir);
 935
 936	if (f2fs_has_inline_dentry(dir))
 937		return f2fs_empty_inline_dir(dir);
 938
 939	while (bidx < nblock) {
 940		pgoff_t next_pgofs;
 941		struct folio *dentry_folio;
 942
 943		dentry_folio = f2fs_find_data_folio(dir, bidx, &next_pgofs);
 944		if (IS_ERR(dentry_folio)) {
 945			if (PTR_ERR(dentry_folio) == -ENOENT) {
 946				bidx = next_pgofs;
 947				continue;
 948			} else {
 949				return false;
 950			}
 951		}
 952
 953		dentry_blk = folio_address(dentry_folio);
 954		if (bidx == 0)
 955			bit_pos = 2;
 956		else
 957			bit_pos = 0;
 958		bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
 959						NR_DENTRY_IN_BLOCK,
 960						bit_pos);
 961
 962		f2fs_folio_put(dentry_folio, false);
 963
 964		if (bit_pos < NR_DENTRY_IN_BLOCK)
 965			return false;
 966
 967		bidx++;
 968	}
 969	return true;
 970}
 971
 972int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
 973			unsigned int start_pos, struct fscrypt_str *fstr)
 974{
 975	unsigned char d_type = DT_UNKNOWN;
 976	unsigned int bit_pos;
 977	struct f2fs_dir_entry *de = NULL;
 978	struct fscrypt_str de_name = FSTR_INIT(NULL, 0);
 979	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
 980	struct blk_plug plug;
 981	bool readdir_ra = sbi->readdir_ra;
 982	bool found_valid_dirent = false;
 983	int err = 0;
 984
 985	bit_pos = ((unsigned long)ctx->pos % d->max);
 986
 987	if (readdir_ra)
 988		blk_start_plug(&plug);
 989
 990	while (bit_pos < d->max) {
 991		bit_pos = find_next_bit_le(d->bitmap, d->max, bit_pos);
 992		if (bit_pos >= d->max)
 993			break;
 994
 995		de = &d->dentry[bit_pos];
 996		if (de->name_len == 0) {
 997			if (found_valid_dirent || !bit_pos) {
 998				f2fs_warn_ratelimited(sbi,
 999					"invalid namelen(0), ino:%u, run fsck to fix.",
1000					le32_to_cpu(de->ino));
1001				set_sbi_flag(sbi, SBI_NEED_FSCK);
1002			}
1003			bit_pos++;
1004			ctx->pos = start_pos + bit_pos;
1005			continue;
1006		}
1007
1008		d_type = fs_ftype_to_dtype(de->file_type);
1009
1010		de_name.name = d->filename[bit_pos];
1011		de_name.len = le16_to_cpu(de->name_len);
1012
1013		/* check memory boundary before moving forward */
1014		bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
1015		if (unlikely(bit_pos > d->max ||
1016				le16_to_cpu(de->name_len) > F2FS_NAME_LEN)) {
1017			f2fs_warn(sbi, "%s: corrupted namelen=%d, run fsck to fix.",
1018				  __func__, le16_to_cpu(de->name_len));
1019			set_sbi_flag(sbi, SBI_NEED_FSCK);
1020			err = -EFSCORRUPTED;
1021			f2fs_handle_error(sbi, ERROR_CORRUPTED_DIRENT);
1022			goto out;
1023		}
1024
1025		if (IS_ENCRYPTED(d->inode)) {
1026			int save_len = fstr->len;
1027
1028			err = fscrypt_fname_disk_to_usr(d->inode,
1029						(u32)le32_to_cpu(de->hash_code),
1030						0, &de_name, fstr);
1031			if (err)
1032				goto out;
1033
1034			de_name = *fstr;
1035			fstr->len = save_len;
1036		}
1037
1038		if (!dir_emit(ctx, de_name.name, de_name.len,
1039					le32_to_cpu(de->ino), d_type)) {
1040			err = 1;
1041			goto out;
1042		}
1043
1044		if (readdir_ra)
1045			f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
1046
1047		ctx->pos = start_pos + bit_pos;
1048		found_valid_dirent = true;
1049	}
1050out:
1051	if (readdir_ra)
1052		blk_finish_plug(&plug);
1053	return err;
1054}
1055
1056static int f2fs_readdir(struct file *file, struct dir_context *ctx)
1057{
1058	struct inode *inode = file_inode(file);
1059	unsigned long npages = dir_blocks(inode);
1060	struct f2fs_dentry_block *dentry_blk = NULL;
1061	struct file_ra_state *ra = &file->f_ra;
1062	loff_t start_pos = ctx->pos;
1063	unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK);
1064	struct f2fs_dentry_ptr d;
1065	struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
1066	int err = 0;
1067
1068	if (IS_ENCRYPTED(inode)) {
1069		err = fscrypt_prepare_readdir(inode);
1070		if (err)
1071			goto out;
1072
1073		err = fscrypt_fname_alloc_buffer(F2FS_NAME_LEN, &fstr);
1074		if (err < 0)
1075			goto out;
1076	}
1077
1078	if (f2fs_has_inline_dentry(inode)) {
1079		err = f2fs_read_inline_dir(file, ctx, &fstr);
1080		goto out_free;
1081	}
1082
1083	for (; n < npages; ctx->pos = n * NR_DENTRY_IN_BLOCK) {
1084		struct folio *dentry_folio;
1085		pgoff_t next_pgofs;
1086
1087		/* allow readdir() to be interrupted */
1088		if (fatal_signal_pending(current)) {
1089			err = -ERESTARTSYS;
1090			goto out_free;
1091		}
1092		cond_resched();
1093
1094		/* readahead for multi pages of dir */
1095		if (npages - n > 1 && !ra_has_index(ra, n))
1096			page_cache_sync_readahead(inode->i_mapping, ra, file, n,
1097				min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES));
1098
1099		dentry_folio = f2fs_find_data_folio(inode, n, &next_pgofs);
1100		if (IS_ERR(dentry_folio)) {
1101			err = PTR_ERR(dentry_folio);
1102			if (err == -ENOENT) {
1103				err = 0;
1104				n = next_pgofs;
1105				continue;
1106			} else {
1107				goto out_free;
1108			}
1109		}
1110
1111		dentry_blk = folio_address(dentry_folio);
1112
1113		make_dentry_ptr_block(inode, &d, dentry_blk);
1114
1115		err = f2fs_fill_dentries(ctx, &d,
1116				n * NR_DENTRY_IN_BLOCK, &fstr);
1117		f2fs_folio_put(dentry_folio, false);
1118		if (err)
1119			break;
1120
1121		n++;
1122	}
1123out_free:
1124	fscrypt_fname_free_buffer(&fstr);
1125out:
1126	trace_f2fs_readdir(inode, start_pos, ctx->pos, err);
1127	return err < 0 ? err : 0;
1128}
1129
1130const struct file_operations f2fs_dir_operations = {
1131	.llseek		= generic_file_llseek,
1132	.read		= generic_read_dir,
1133	.iterate_shared	= f2fs_readdir,
1134	.fsync		= f2fs_sync_file,
1135	.unlocked_ioctl	= f2fs_ioctl,
1136#ifdef CONFIG_COMPAT
1137	.compat_ioctl   = f2fs_compat_ioctl,
1138#endif
1139};