xref: /openbmc/linux/fs/f2fs/dir.c (revision 310d9531)
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 <asm/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  
19  #if IS_ENABLED(CONFIG_UNICODE)
20  extern struct kmem_cache *f2fs_cf_name_slab;
21  #endif
22  
23  static unsigned long dir_blocks(struct inode *inode)
24  {
25  	return ((unsigned long long) (i_size_read(inode) + PAGE_SIZE - 1))
26  							>> PAGE_SHIFT;
27  }
28  
29  static unsigned int dir_buckets(unsigned int level, int dir_level)
30  {
31  	if (level + dir_level < MAX_DIR_HASH_DEPTH / 2)
32  		return BIT(level + dir_level);
33  	else
34  		return MAX_DIR_BUCKETS;
35  }
36  
37  static unsigned int bucket_blocks(unsigned int level)
38  {
39  	if (level < MAX_DIR_HASH_DEPTH / 2)
40  		return 2;
41  	else
42  		return 4;
43  }
44  
45  /* If @dir is casefolded, initialize @fname->cf_name from @fname->usr_fname. */
46  int f2fs_init_casefolded_name(const struct inode *dir,
47  			      struct f2fs_filename *fname)
48  {
49  #if IS_ENABLED(CONFIG_UNICODE)
50  	struct super_block *sb = dir->i_sb;
51  
52  	if (IS_CASEFOLDED(dir) &&
53  	    !is_dot_dotdot(fname->usr_fname->name, fname->usr_fname->len)) {
54  		fname->cf_name.name = f2fs_kmem_cache_alloc(f2fs_cf_name_slab,
55  					GFP_NOFS, false, F2FS_SB(sb));
56  		if (!fname->cf_name.name)
57  			return -ENOMEM;
58  		fname->cf_name.len = utf8_casefold(sb->s_encoding,
59  						   fname->usr_fname,
60  						   fname->cf_name.name,
61  						   F2FS_NAME_LEN);
62  		if ((int)fname->cf_name.len <= 0) {
63  			kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
64  			fname->cf_name.name = NULL;
65  			if (sb_has_strict_encoding(sb))
66  				return -EINVAL;
67  			/* fall back to treating name as opaque byte sequence */
68  		}
69  	}
70  #endif
71  	return 0;
72  }
73  
74  static int __f2fs_setup_filename(const struct inode *dir,
75  				 const struct fscrypt_name *crypt_name,
76  				 struct f2fs_filename *fname)
77  {
78  	int err;
79  
80  	memset(fname, 0, sizeof(*fname));
81  
82  	fname->usr_fname = crypt_name->usr_fname;
83  	fname->disk_name = crypt_name->disk_name;
84  #ifdef CONFIG_FS_ENCRYPTION
85  	fname->crypto_buf = crypt_name->crypto_buf;
86  #endif
87  	if (crypt_name->is_nokey_name) {
88  		/* hash was decoded from the no-key name */
89  		fname->hash = cpu_to_le32(crypt_name->hash);
90  	} else {
91  		err = f2fs_init_casefolded_name(dir, fname);
92  		if (err) {
93  			f2fs_free_filename(fname);
94  			return err;
95  		}
96  		f2fs_hash_filename(dir, fname);
97  	}
98  	return 0;
99  }
100  
101  /*
102   * Prepare to search for @iname in @dir.  This is similar to
103   * fscrypt_setup_filename(), but this also handles computing the casefolded name
104   * and the f2fs dirhash if needed, then packing all the information about this
105   * filename up into a 'struct f2fs_filename'.
106   */
107  int f2fs_setup_filename(struct inode *dir, const struct qstr *iname,
108  			int lookup, struct f2fs_filename *fname)
109  {
110  	struct fscrypt_name crypt_name;
111  	int err;
112  
113  	err = fscrypt_setup_filename(dir, iname, lookup, &crypt_name);
114  	if (err)
115  		return err;
116  
117  	return __f2fs_setup_filename(dir, &crypt_name, fname);
118  }
119  
120  /*
121   * Prepare to look up @dentry in @dir.  This is similar to
122   * fscrypt_prepare_lookup(), but this also handles computing the casefolded name
123   * and the f2fs dirhash if needed, then packing all the information about this
124   * filename up into a 'struct f2fs_filename'.
125   */
126  int f2fs_prepare_lookup(struct inode *dir, struct dentry *dentry,
127  			struct f2fs_filename *fname)
128  {
129  	struct fscrypt_name crypt_name;
130  	int err;
131  
132  	err = fscrypt_prepare_lookup(dir, dentry, &crypt_name);
133  	if (err)
134  		return err;
135  
136  	return __f2fs_setup_filename(dir, &crypt_name, fname);
137  }
138  
139  void f2fs_free_filename(struct f2fs_filename *fname)
140  {
141  #ifdef CONFIG_FS_ENCRYPTION
142  	kfree(fname->crypto_buf.name);
143  	fname->crypto_buf.name = NULL;
144  #endif
145  #if IS_ENABLED(CONFIG_UNICODE)
146  	if (fname->cf_name.name) {
147  		kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
148  		fname->cf_name.name = NULL;
149  	}
150  #endif
151  }
152  
153  static unsigned long dir_block_index(unsigned int level,
154  				int dir_level, unsigned int idx)
155  {
156  	unsigned long i;
157  	unsigned long bidx = 0;
158  
159  	for (i = 0; i < level; i++)
160  		bidx += mul_u32_u32(dir_buckets(i, dir_level),
161  				    bucket_blocks(i));
162  	bidx += idx * bucket_blocks(level);
163  	return bidx;
164  }
165  
166  static struct f2fs_dir_entry *find_in_block(struct inode *dir,
167  				struct page *dentry_page,
168  				const struct f2fs_filename *fname,
169  				int *max_slots)
170  {
171  	struct f2fs_dentry_block *dentry_blk;
172  	struct f2fs_dentry_ptr d;
173  
174  	dentry_blk = (struct f2fs_dentry_block *)page_address(dentry_page);
175  
176  	make_dentry_ptr_block(dir, &d, dentry_blk);
177  	return f2fs_find_target_dentry(&d, fname, max_slots);
178  }
179  
180  #if IS_ENABLED(CONFIG_UNICODE)
181  /*
182   * Test whether a case-insensitive directory entry matches the filename
183   * being searched for.
184   *
185   * Returns 1 for a match, 0 for no match, and -errno on an error.
186   */
187  static int f2fs_match_ci_name(const struct inode *dir, const struct qstr *name,
188  			       const u8 *de_name, u32 de_name_len)
189  {
190  	const struct super_block *sb = dir->i_sb;
191  	const struct unicode_map *um = sb->s_encoding;
192  	struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
193  	struct qstr entry = QSTR_INIT(de_name, de_name_len);
194  	int res;
195  
196  	if (IS_ENCRYPTED(dir)) {
197  		const struct fscrypt_str encrypted_name =
198  			FSTR_INIT((u8 *)de_name, de_name_len);
199  
200  		if (WARN_ON_ONCE(!fscrypt_has_encryption_key(dir)))
201  			return -EINVAL;
202  
203  		decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
204  		if (!decrypted_name.name)
205  			return -ENOMEM;
206  		res = fscrypt_fname_disk_to_usr(dir, 0, 0, &encrypted_name,
207  						&decrypted_name);
208  		if (res < 0)
209  			goto out;
210  		entry.name = decrypted_name.name;
211  		entry.len = decrypted_name.len;
212  	}
213  
214  	res = utf8_strncasecmp_folded(um, name, &entry);
215  	/*
216  	 * In strict mode, ignore invalid names.  In non-strict mode,
217  	 * fall back to treating them as opaque byte sequences.
218  	 */
219  	if (res < 0 && !sb_has_strict_encoding(sb)) {
220  		res = name->len == entry.len &&
221  				memcmp(name->name, entry.name, name->len) == 0;
222  	} else {
223  		/* utf8_strncasecmp_folded returns 0 on match */
224  		res = (res == 0);
225  	}
226  out:
227  	kfree(decrypted_name.name);
228  	return res;
229  }
230  #endif /* CONFIG_UNICODE */
231  
232  static inline int f2fs_match_name(const struct inode *dir,
233  				   const struct f2fs_filename *fname,
234  				   const u8 *de_name, u32 de_name_len)
235  {
236  	struct fscrypt_name f;
237  
238  #if IS_ENABLED(CONFIG_UNICODE)
239  	if (fname->cf_name.name) {
240  		struct qstr cf = FSTR_TO_QSTR(&fname->cf_name);
241  
242  		return f2fs_match_ci_name(dir, &cf, de_name, de_name_len);
243  	}
244  #endif
245  	f.usr_fname = fname->usr_fname;
246  	f.disk_name = fname->disk_name;
247  #ifdef CONFIG_FS_ENCRYPTION
248  	f.crypto_buf = fname->crypto_buf;
249  #endif
250  	return fscrypt_match_name(&f, de_name, de_name_len);
251  }
252  
253  struct f2fs_dir_entry *f2fs_find_target_dentry(const struct f2fs_dentry_ptr *d,
254  			const struct f2fs_filename *fname, int *max_slots)
255  {
256  	struct f2fs_dir_entry *de;
257  	unsigned long bit_pos = 0;
258  	int max_len = 0;
259  	int res = 0;
260  
261  	if (max_slots)
262  		*max_slots = 0;
263  	while (bit_pos < d->max) {
264  		if (!test_bit_le(bit_pos, d->bitmap)) {
265  			bit_pos++;
266  			max_len++;
267  			continue;
268  		}
269  
270  		de = &d->dentry[bit_pos];
271  
272  		if (unlikely(!de->name_len)) {
273  			bit_pos++;
274  			continue;
275  		}
276  
277  		if (de->hash_code == fname->hash) {
278  			res = f2fs_match_name(d->inode, fname,
279  					      d->filename[bit_pos],
280  					      le16_to_cpu(de->name_len));
281  			if (res < 0)
282  				return ERR_PTR(res);
283  			if (res)
284  				goto found;
285  		}
286  
287  		if (max_slots && max_len > *max_slots)
288  			*max_slots = max_len;
289  		max_len = 0;
290  
291  		bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
292  	}
293  
294  	de = NULL;
295  found:
296  	if (max_slots && max_len > *max_slots)
297  		*max_slots = max_len;
298  	return de;
299  }
300  
301  static struct f2fs_dir_entry *find_in_level(struct inode *dir,
302  					unsigned int level,
303  					const struct f2fs_filename *fname,
304  					struct page **res_page)
305  {
306  	int s = GET_DENTRY_SLOTS(fname->disk_name.len);
307  	unsigned int nbucket, nblock;
308  	unsigned int bidx, end_block;
309  	struct page *dentry_page;
310  	struct f2fs_dir_entry *de = NULL;
311  	pgoff_t next_pgofs;
312  	bool room = false;
313  	int max_slots;
314  
315  	nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
316  	nblock = bucket_blocks(level);
317  
318  	bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
319  			       le32_to_cpu(fname->hash) % nbucket);
320  	end_block = bidx + nblock;
321  
322  	while (bidx < end_block) {
323  		/* no need to allocate new dentry pages to all the indices */
324  		dentry_page = f2fs_find_data_page(dir, bidx, &next_pgofs);
325  		if (IS_ERR(dentry_page)) {
326  			if (PTR_ERR(dentry_page) == -ENOENT) {
327  				room = true;
328  				bidx = next_pgofs;
329  				continue;
330  			} else {
331  				*res_page = dentry_page;
332  				break;
333  			}
334  		}
335  
336  		de = find_in_block(dir, dentry_page, fname, &max_slots);
337  		if (IS_ERR(de)) {
338  			*res_page = ERR_CAST(de);
339  			de = NULL;
340  			break;
341  		} else if (de) {
342  			*res_page = dentry_page;
343  			break;
344  		}
345  
346  		if (max_slots >= s)
347  			room = true;
348  		f2fs_put_page(dentry_page, 0);
349  
350  		bidx++;
351  	}
352  
353  	if (!de && room && F2FS_I(dir)->chash != fname->hash) {
354  		F2FS_I(dir)->chash = fname->hash;
355  		F2FS_I(dir)->clevel = level;
356  	}
357  
358  	return de;
359  }
360  
361  struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir,
362  					 const struct f2fs_filename *fname,
363  					 struct page **res_page)
364  {
365  	unsigned long npages = dir_blocks(dir);
366  	struct f2fs_dir_entry *de = NULL;
367  	unsigned int max_depth;
368  	unsigned int level;
369  
370  	*res_page = NULL;
371  
372  	if (f2fs_has_inline_dentry(dir)) {
373  		de = f2fs_find_in_inline_dir(dir, fname, res_page);
374  		goto out;
375  	}
376  
377  	if (npages == 0)
378  		goto out;
379  
380  	max_depth = F2FS_I(dir)->i_current_depth;
381  	if (unlikely(max_depth > MAX_DIR_HASH_DEPTH)) {
382  		f2fs_warn(F2FS_I_SB(dir), "Corrupted max_depth of %lu: %u",
383  			  dir->i_ino, max_depth);
384  		max_depth = MAX_DIR_HASH_DEPTH;
385  		f2fs_i_depth_write(dir, max_depth);
386  	}
387  
388  	for (level = 0; level < max_depth; level++) {
389  		de = find_in_level(dir, level, fname, res_page);
390  		if (de || IS_ERR(*res_page))
391  			break;
392  	}
393  out:
394  	/* This is to increase the speed of f2fs_create */
395  	if (!de)
396  		F2FS_I(dir)->task = current;
397  	return de;
398  }
399  
400  /*
401   * Find an entry in the specified directory with the wanted name.
402   * It returns the page where the entry was found (as a parameter - res_page),
403   * and the entry itself. Page is returned mapped and unlocked.
404   * Entry is guaranteed to be valid.
405   */
406  struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir,
407  			const struct qstr *child, struct page **res_page)
408  {
409  	struct f2fs_dir_entry *de = NULL;
410  	struct f2fs_filename fname;
411  	int err;
412  
413  	err = f2fs_setup_filename(dir, child, 1, &fname);
414  	if (err) {
415  		if (err == -ENOENT)
416  			*res_page = NULL;
417  		else
418  			*res_page = ERR_PTR(err);
419  		return NULL;
420  	}
421  
422  	de = __f2fs_find_entry(dir, &fname, res_page);
423  
424  	f2fs_free_filename(&fname);
425  	return de;
426  }
427  
428  struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct page **p)
429  {
430  	return f2fs_find_entry(dir, &dotdot_name, p);
431  }
432  
433  ino_t f2fs_inode_by_name(struct inode *dir, const struct qstr *qstr,
434  							struct page **page)
435  {
436  	ino_t res = 0;
437  	struct f2fs_dir_entry *de;
438  
439  	de = f2fs_find_entry(dir, qstr, page);
440  	if (de) {
441  		res = le32_to_cpu(de->ino);
442  		f2fs_put_page(*page, 0);
443  	}
444  
445  	return res;
446  }
447  
448  void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
449  		struct page *page, struct inode *inode)
450  {
451  	enum page_type type = f2fs_has_inline_dentry(dir) ? NODE : DATA;
452  
453  	lock_page(page);
454  	f2fs_wait_on_page_writeback(page, type, true, true);
455  	de->ino = cpu_to_le32(inode->i_ino);
456  	de->file_type = fs_umode_to_ftype(inode->i_mode);
457  	set_page_dirty(page);
458  
459  	dir->i_mtime = inode_set_ctime_current(dir);
460  	f2fs_mark_inode_dirty_sync(dir, false);
461  	f2fs_put_page(page, 1);
462  }
463  
464  static void init_dent_inode(struct inode *dir, struct inode *inode,
465  			    const struct f2fs_filename *fname,
466  			    struct page *ipage)
467  {
468  	struct f2fs_inode *ri;
469  
470  	if (!fname) /* tmpfile case? */
471  		return;
472  
473  	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
474  
475  	/* copy name info. to this inode page */
476  	ri = F2FS_INODE(ipage);
477  	ri->i_namelen = cpu_to_le32(fname->disk_name.len);
478  	memcpy(ri->i_name, fname->disk_name.name, fname->disk_name.len);
479  	if (IS_ENCRYPTED(dir)) {
480  		file_set_enc_name(inode);
481  		/*
482  		 * Roll-forward recovery doesn't have encryption keys available,
483  		 * so it can't compute the dirhash for encrypted+casefolded
484  		 * filenames.  Append it to i_name if possible.  Else, disable
485  		 * roll-forward recovery of the dentry (i.e., make fsync'ing the
486  		 * file force a checkpoint) by setting LOST_PINO.
487  		 */
488  		if (IS_CASEFOLDED(dir)) {
489  			if (fname->disk_name.len + sizeof(f2fs_hash_t) <=
490  			    F2FS_NAME_LEN)
491  				put_unaligned(fname->hash, (f2fs_hash_t *)
492  					&ri->i_name[fname->disk_name.len]);
493  			else
494  				file_lost_pino(inode);
495  		}
496  	}
497  	set_page_dirty(ipage);
498  }
499  
500  void f2fs_do_make_empty_dir(struct inode *inode, struct inode *parent,
501  					struct f2fs_dentry_ptr *d)
502  {
503  	struct fscrypt_str dot = FSTR_INIT(".", 1);
504  	struct fscrypt_str dotdot = FSTR_INIT("..", 2);
505  
506  	/* update dirent of "." */
507  	f2fs_update_dentry(inode->i_ino, inode->i_mode, d, &dot, 0, 0);
508  
509  	/* update dirent of ".." */
510  	f2fs_update_dentry(parent->i_ino, parent->i_mode, d, &dotdot, 0, 1);
511  }
512  
513  static int make_empty_dir(struct inode *inode,
514  		struct inode *parent, struct page *page)
515  {
516  	struct page *dentry_page;
517  	struct f2fs_dentry_block *dentry_blk;
518  	struct f2fs_dentry_ptr d;
519  
520  	if (f2fs_has_inline_dentry(inode))
521  		return f2fs_make_empty_inline_dir(inode, parent, page);
522  
523  	dentry_page = f2fs_get_new_data_page(inode, page, 0, true);
524  	if (IS_ERR(dentry_page))
525  		return PTR_ERR(dentry_page);
526  
527  	dentry_blk = page_address(dentry_page);
528  
529  	make_dentry_ptr_block(NULL, &d, dentry_blk);
530  	f2fs_do_make_empty_dir(inode, parent, &d);
531  
532  	set_page_dirty(dentry_page);
533  	f2fs_put_page(dentry_page, 1);
534  	return 0;
535  }
536  
537  struct page *f2fs_init_inode_metadata(struct inode *inode, struct inode *dir,
538  			const struct f2fs_filename *fname, struct page *dpage)
539  {
540  	struct page *page;
541  	int err;
542  
543  	if (is_inode_flag_set(inode, FI_NEW_INODE)) {
544  		page = f2fs_new_inode_page(inode);
545  		if (IS_ERR(page))
546  			return page;
547  
548  		if (S_ISDIR(inode->i_mode)) {
549  			/* in order to handle error case */
550  			get_page(page);
551  			err = make_empty_dir(inode, dir, page);
552  			if (err) {
553  				lock_page(page);
554  				goto put_error;
555  			}
556  			put_page(page);
557  		}
558  
559  		err = f2fs_init_acl(inode, dir, page, dpage);
560  		if (err)
561  			goto put_error;
562  
563  		err = f2fs_init_security(inode, dir,
564  					 fname ? fname->usr_fname : NULL, page);
565  		if (err)
566  			goto put_error;
567  
568  		if (IS_ENCRYPTED(inode)) {
569  			err = fscrypt_set_context(inode, page);
570  			if (err)
571  				goto put_error;
572  		}
573  	} else {
574  		page = f2fs_get_node_page(F2FS_I_SB(dir), inode->i_ino);
575  		if (IS_ERR(page))
576  			return page;
577  	}
578  
579  	init_dent_inode(dir, inode, fname, page);
580  
581  	/*
582  	 * This file should be checkpointed during fsync.
583  	 * We lost i_pino from now on.
584  	 */
585  	if (is_inode_flag_set(inode, FI_INC_LINK)) {
586  		if (!S_ISDIR(inode->i_mode))
587  			file_lost_pino(inode);
588  		/*
589  		 * If link the tmpfile to alias through linkat path,
590  		 * we should remove this inode from orphan list.
591  		 */
592  		if (inode->i_nlink == 0)
593  			f2fs_remove_orphan_inode(F2FS_I_SB(dir), inode->i_ino);
594  		f2fs_i_links_write(inode, true);
595  	}
596  	return page;
597  
598  put_error:
599  	clear_nlink(inode);
600  	f2fs_update_inode(inode, page);
601  	f2fs_put_page(page, 1);
602  	return ERR_PTR(err);
603  }
604  
605  void f2fs_update_parent_metadata(struct inode *dir, struct inode *inode,
606  						unsigned int current_depth)
607  {
608  	if (inode && is_inode_flag_set(inode, FI_NEW_INODE)) {
609  		if (S_ISDIR(inode->i_mode))
610  			f2fs_i_links_write(dir, true);
611  		clear_inode_flag(inode, FI_NEW_INODE);
612  	}
613  	dir->i_mtime = inode_set_ctime_current(dir);
614  	f2fs_mark_inode_dirty_sync(dir, false);
615  
616  	if (F2FS_I(dir)->i_current_depth != current_depth)
617  		f2fs_i_depth_write(dir, current_depth);
618  
619  	if (inode && is_inode_flag_set(inode, FI_INC_LINK))
620  		clear_inode_flag(inode, FI_INC_LINK);
621  }
622  
623  int f2fs_room_for_filename(const void *bitmap, int slots, int max_slots)
624  {
625  	int bit_start = 0;
626  	int zero_start, zero_end;
627  next:
628  	zero_start = find_next_zero_bit_le(bitmap, max_slots, bit_start);
629  	if (zero_start >= max_slots)
630  		return max_slots;
631  
632  	zero_end = find_next_bit_le(bitmap, max_slots, zero_start);
633  	if (zero_end - zero_start >= slots)
634  		return zero_start;
635  
636  	bit_start = zero_end + 1;
637  
638  	if (zero_end + 1 >= max_slots)
639  		return max_slots;
640  	goto next;
641  }
642  
643  bool f2fs_has_enough_room(struct inode *dir, struct page *ipage,
644  			  const struct f2fs_filename *fname)
645  {
646  	struct f2fs_dentry_ptr d;
647  	unsigned int bit_pos;
648  	int slots = GET_DENTRY_SLOTS(fname->disk_name.len);
649  
650  	make_dentry_ptr_inline(dir, &d, inline_data_addr(dir, ipage));
651  
652  	bit_pos = f2fs_room_for_filename(d.bitmap, slots, d.max);
653  
654  	return bit_pos < d.max;
655  }
656  
657  void f2fs_update_dentry(nid_t ino, umode_t mode, struct f2fs_dentry_ptr *d,
658  			const struct fscrypt_str *name, f2fs_hash_t name_hash,
659  			unsigned int bit_pos)
660  {
661  	struct f2fs_dir_entry *de;
662  	int slots = GET_DENTRY_SLOTS(name->len);
663  	int i;
664  
665  	de = &d->dentry[bit_pos];
666  	de->hash_code = name_hash;
667  	de->name_len = cpu_to_le16(name->len);
668  	memcpy(d->filename[bit_pos], name->name, name->len);
669  	de->ino = cpu_to_le32(ino);
670  	de->file_type = fs_umode_to_ftype(mode);
671  	for (i = 0; i < slots; i++) {
672  		__set_bit_le(bit_pos + i, (void *)d->bitmap);
673  		/* avoid wrong garbage data for readdir */
674  		if (i)
675  			(de + i)->name_len = 0;
676  	}
677  }
678  
679  int f2fs_add_regular_entry(struct inode *dir, const struct f2fs_filename *fname,
680  			   struct inode *inode, nid_t ino, umode_t mode)
681  {
682  	unsigned int bit_pos;
683  	unsigned int level;
684  	unsigned int current_depth;
685  	unsigned long bidx, block;
686  	unsigned int nbucket, nblock;
687  	struct page *dentry_page = NULL;
688  	struct f2fs_dentry_block *dentry_blk = NULL;
689  	struct f2fs_dentry_ptr d;
690  	struct page *page = NULL;
691  	int slots, err = 0;
692  
693  	level = 0;
694  	slots = GET_DENTRY_SLOTS(fname->disk_name.len);
695  
696  	current_depth = F2FS_I(dir)->i_current_depth;
697  	if (F2FS_I(dir)->chash == fname->hash) {
698  		level = F2FS_I(dir)->clevel;
699  		F2FS_I(dir)->chash = 0;
700  	}
701  
702  start:
703  	if (time_to_inject(F2FS_I_SB(dir), FAULT_DIR_DEPTH))
704  		return -ENOSPC;
705  
706  	if (unlikely(current_depth == MAX_DIR_HASH_DEPTH))
707  		return -ENOSPC;
708  
709  	/* Increase the depth, if required */
710  	if (level == current_depth)
711  		++current_depth;
712  
713  	nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
714  	nblock = bucket_blocks(level);
715  
716  	bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
717  				(le32_to_cpu(fname->hash) % nbucket));
718  
719  	for (block = bidx; block <= (bidx + nblock - 1); block++) {
720  		dentry_page = f2fs_get_new_data_page(dir, NULL, block, true);
721  		if (IS_ERR(dentry_page))
722  			return PTR_ERR(dentry_page);
723  
724  		dentry_blk = page_address(dentry_page);
725  		bit_pos = f2fs_room_for_filename(&dentry_blk->dentry_bitmap,
726  						slots, NR_DENTRY_IN_BLOCK);
727  		if (bit_pos < NR_DENTRY_IN_BLOCK)
728  			goto add_dentry;
729  
730  		f2fs_put_page(dentry_page, 1);
731  	}
732  
733  	/* Move to next level to find the empty slot for new dentry */
734  	++level;
735  	goto start;
736  add_dentry:
737  	f2fs_wait_on_page_writeback(dentry_page, DATA, true, true);
738  
739  	if (inode) {
740  		f2fs_down_write(&F2FS_I(inode)->i_sem);
741  		page = f2fs_init_inode_metadata(inode, dir, fname, NULL);
742  		if (IS_ERR(page)) {
743  			err = PTR_ERR(page);
744  			goto fail;
745  		}
746  	}
747  
748  	make_dentry_ptr_block(NULL, &d, dentry_blk);
749  	f2fs_update_dentry(ino, mode, &d, &fname->disk_name, fname->hash,
750  			   bit_pos);
751  
752  	set_page_dirty(dentry_page);
753  
754  	if (inode) {
755  		f2fs_i_pino_write(inode, dir->i_ino);
756  
757  		/* synchronize inode page's data from inode cache */
758  		if (is_inode_flag_set(inode, FI_NEW_INODE))
759  			f2fs_update_inode(inode, page);
760  
761  		f2fs_put_page(page, 1);
762  	}
763  
764  	f2fs_update_parent_metadata(dir, inode, current_depth);
765  fail:
766  	if (inode)
767  		f2fs_up_write(&F2FS_I(inode)->i_sem);
768  
769  	f2fs_put_page(dentry_page, 1);
770  
771  	return err;
772  }
773  
774  int f2fs_add_dentry(struct inode *dir, const struct f2fs_filename *fname,
775  		    struct inode *inode, nid_t ino, umode_t mode)
776  {
777  	int err = -EAGAIN;
778  
779  	if (f2fs_has_inline_dentry(dir)) {
780  		/*
781  		 * Should get i_xattr_sem to keep the lock order:
782  		 * i_xattr_sem -> inode_page lock used by f2fs_setxattr.
783  		 */
784  		f2fs_down_read(&F2FS_I(dir)->i_xattr_sem);
785  		err = f2fs_add_inline_entry(dir, fname, inode, ino, mode);
786  		f2fs_up_read(&F2FS_I(dir)->i_xattr_sem);
787  	}
788  	if (err == -EAGAIN)
789  		err = f2fs_add_regular_entry(dir, fname, inode, ino, mode);
790  
791  	f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
792  	return err;
793  }
794  
795  /*
796   * Caller should grab and release a rwsem by calling f2fs_lock_op() and
797   * f2fs_unlock_op().
798   */
799  int f2fs_do_add_link(struct inode *dir, const struct qstr *name,
800  				struct inode *inode, nid_t ino, umode_t mode)
801  {
802  	struct f2fs_filename fname;
803  	struct page *page = NULL;
804  	struct f2fs_dir_entry *de = NULL;
805  	int err;
806  
807  	err = f2fs_setup_filename(dir, name, 0, &fname);
808  	if (err)
809  		return err;
810  
811  	/*
812  	 * An immature stackable filesystem shows a race condition between lookup
813  	 * and create. If we have same task when doing lookup and create, it's
814  	 * definitely fine as expected by VFS normally. Otherwise, let's just
815  	 * verify on-disk dentry one more time, which guarantees filesystem
816  	 * consistency more.
817  	 */
818  	if (current != F2FS_I(dir)->task) {
819  		de = __f2fs_find_entry(dir, &fname, &page);
820  		F2FS_I(dir)->task = NULL;
821  	}
822  	if (de) {
823  		f2fs_put_page(page, 0);
824  		err = -EEXIST;
825  	} else if (IS_ERR(page)) {
826  		err = PTR_ERR(page);
827  	} else {
828  		err = f2fs_add_dentry(dir, &fname, inode, ino, mode);
829  	}
830  	f2fs_free_filename(&fname);
831  	return err;
832  }
833  
834  int f2fs_do_tmpfile(struct inode *inode, struct inode *dir,
835  					struct f2fs_filename *fname)
836  {
837  	struct page *page;
838  	int err = 0;
839  
840  	f2fs_down_write(&F2FS_I(inode)->i_sem);
841  	page = f2fs_init_inode_metadata(inode, dir, fname, NULL);
842  	if (IS_ERR(page)) {
843  		err = PTR_ERR(page);
844  		goto fail;
845  	}
846  	f2fs_put_page(page, 1);
847  
848  	clear_inode_flag(inode, FI_NEW_INODE);
849  	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
850  fail:
851  	f2fs_up_write(&F2FS_I(inode)->i_sem);
852  	return err;
853  }
854  
855  void f2fs_drop_nlink(struct inode *dir, struct inode *inode)
856  {
857  	struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
858  
859  	f2fs_down_write(&F2FS_I(inode)->i_sem);
860  
861  	if (S_ISDIR(inode->i_mode))
862  		f2fs_i_links_write(dir, false);
863  	inode_set_ctime_current(inode);
864  
865  	f2fs_i_links_write(inode, false);
866  	if (S_ISDIR(inode->i_mode)) {
867  		f2fs_i_links_write(inode, false);
868  		f2fs_i_size_write(inode, 0);
869  	}
870  	f2fs_up_write(&F2FS_I(inode)->i_sem);
871  
872  	if (inode->i_nlink == 0)
873  		f2fs_add_orphan_inode(inode);
874  	else
875  		f2fs_release_orphan_inode(sbi);
876  }
877  
878  /*
879   * It only removes the dentry from the dentry page, corresponding name
880   * entry in name page does not need to be touched during deletion.
881   */
882  void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct page *page,
883  					struct inode *dir, struct inode *inode)
884  {
885  	struct	f2fs_dentry_block *dentry_blk;
886  	unsigned int bit_pos;
887  	int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
888  	int i;
889  
890  	f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
891  
892  	if (F2FS_OPTION(F2FS_I_SB(dir)).fsync_mode == FSYNC_MODE_STRICT)
893  		f2fs_add_ino_entry(F2FS_I_SB(dir), dir->i_ino, TRANS_DIR_INO);
894  
895  	if (f2fs_has_inline_dentry(dir))
896  		return f2fs_delete_inline_entry(dentry, page, dir, inode);
897  
898  	lock_page(page);
899  	f2fs_wait_on_page_writeback(page, DATA, true, true);
900  
901  	dentry_blk = page_address(page);
902  	bit_pos = dentry - dentry_blk->dentry;
903  	for (i = 0; i < slots; i++)
904  		__clear_bit_le(bit_pos + i, &dentry_blk->dentry_bitmap);
905  
906  	/* Let's check and deallocate this dentry page */
907  	bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
908  			NR_DENTRY_IN_BLOCK,
909  			0);
910  	set_page_dirty(page);
911  
912  	if (bit_pos == NR_DENTRY_IN_BLOCK &&
913  		!f2fs_truncate_hole(dir, page->index, page->index + 1)) {
914  		f2fs_clear_page_cache_dirty_tag(page);
915  		clear_page_dirty_for_io(page);
916  		ClearPageUptodate(page);
917  		clear_page_private_all(page);
918  
919  		inode_dec_dirty_pages(dir);
920  		f2fs_remove_dirty_inode(dir);
921  	}
922  	f2fs_put_page(page, 1);
923  
924  	dir->i_mtime = inode_set_ctime_current(dir);
925  	f2fs_mark_inode_dirty_sync(dir, false);
926  
927  	if (inode)
928  		f2fs_drop_nlink(dir, inode);
929  }
930  
931  bool f2fs_empty_dir(struct inode *dir)
932  {
933  	unsigned long bidx = 0;
934  	struct page *dentry_page;
935  	unsigned int bit_pos;
936  	struct f2fs_dentry_block *dentry_blk;
937  	unsigned long nblock = dir_blocks(dir);
938  
939  	if (f2fs_has_inline_dentry(dir))
940  		return f2fs_empty_inline_dir(dir);
941  
942  	while (bidx < nblock) {
943  		pgoff_t next_pgofs;
944  
945  		dentry_page = f2fs_find_data_page(dir, bidx, &next_pgofs);
946  		if (IS_ERR(dentry_page)) {
947  			if (PTR_ERR(dentry_page) == -ENOENT) {
948  				bidx = next_pgofs;
949  				continue;
950  			} else {
951  				return false;
952  			}
953  		}
954  
955  		dentry_blk = page_address(dentry_page);
956  		if (bidx == 0)
957  			bit_pos = 2;
958  		else
959  			bit_pos = 0;
960  		bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
961  						NR_DENTRY_IN_BLOCK,
962  						bit_pos);
963  
964  		f2fs_put_page(dentry_page, 0);
965  
966  		if (bit_pos < NR_DENTRY_IN_BLOCK)
967  			return false;
968  
969  		bidx++;
970  	}
971  	return true;
972  }
973  
974  int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
975  			unsigned int start_pos, struct fscrypt_str *fstr)
976  {
977  	unsigned char d_type = DT_UNKNOWN;
978  	unsigned int bit_pos;
979  	struct f2fs_dir_entry *de = NULL;
980  	struct fscrypt_str de_name = FSTR_INIT(NULL, 0);
981  	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
982  	struct blk_plug plug;
983  	bool readdir_ra = sbi->readdir_ra;
984  	bool found_valid_dirent = false;
985  	int err = 0;
986  
987  	bit_pos = ((unsigned long)ctx->pos % d->max);
988  
989  	if (readdir_ra)
990  		blk_start_plug(&plug);
991  
992  	while (bit_pos < d->max) {
993  		bit_pos = find_next_bit_le(d->bitmap, d->max, bit_pos);
994  		if (bit_pos >= d->max)
995  			break;
996  
997  		de = &d->dentry[bit_pos];
998  		if (de->name_len == 0) {
999  			if (found_valid_dirent || !bit_pos) {
1000  				f2fs_warn_ratelimited(sbi,
1001  					"invalid namelen(0), ino:%u, run fsck to fix.",
1002  					le32_to_cpu(de->ino));
1003  				set_sbi_flag(sbi, SBI_NEED_FSCK);
1004  			}
1005  			bit_pos++;
1006  			ctx->pos = start_pos + bit_pos;
1007  			continue;
1008  		}
1009  
1010  		d_type = fs_ftype_to_dtype(de->file_type);
1011  
1012  		de_name.name = d->filename[bit_pos];
1013  		de_name.len = le16_to_cpu(de->name_len);
1014  
1015  		/* check memory boundary before moving forward */
1016  		bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
1017  		if (unlikely(bit_pos > d->max ||
1018  				le16_to_cpu(de->name_len) > F2FS_NAME_LEN)) {
1019  			f2fs_warn(sbi, "%s: corrupted namelen=%d, run fsck to fix.",
1020  				  __func__, le16_to_cpu(de->name_len));
1021  			set_sbi_flag(sbi, SBI_NEED_FSCK);
1022  			err = -EFSCORRUPTED;
1023  			f2fs_handle_error(sbi, ERROR_CORRUPTED_DIRENT);
1024  			goto out;
1025  		}
1026  
1027  		if (IS_ENCRYPTED(d->inode)) {
1028  			int save_len = fstr->len;
1029  
1030  			err = fscrypt_fname_disk_to_usr(d->inode,
1031  						(u32)le32_to_cpu(de->hash_code),
1032  						0, &de_name, fstr);
1033  			if (err)
1034  				goto out;
1035  
1036  			de_name = *fstr;
1037  			fstr->len = save_len;
1038  		}
1039  
1040  		if (!dir_emit(ctx, de_name.name, de_name.len,
1041  					le32_to_cpu(de->ino), d_type)) {
1042  			err = 1;
1043  			goto out;
1044  		}
1045  
1046  		if (readdir_ra)
1047  			f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
1048  
1049  		ctx->pos = start_pos + bit_pos;
1050  		found_valid_dirent = true;
1051  	}
1052  out:
1053  	if (readdir_ra)
1054  		blk_finish_plug(&plug);
1055  	return err;
1056  }
1057  
1058  static int f2fs_readdir(struct file *file, struct dir_context *ctx)
1059  {
1060  	struct inode *inode = file_inode(file);
1061  	unsigned long npages = dir_blocks(inode);
1062  	struct f2fs_dentry_block *dentry_blk = NULL;
1063  	struct page *dentry_page = NULL;
1064  	struct file_ra_state *ra = &file->f_ra;
1065  	loff_t start_pos = ctx->pos;
1066  	unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK);
1067  	struct f2fs_dentry_ptr d;
1068  	struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
1069  	int err = 0;
1070  
1071  	if (IS_ENCRYPTED(inode)) {
1072  		err = fscrypt_prepare_readdir(inode);
1073  		if (err)
1074  			goto out;
1075  
1076  		err = fscrypt_fname_alloc_buffer(F2FS_NAME_LEN, &fstr);
1077  		if (err < 0)
1078  			goto out;
1079  	}
1080  
1081  	if (f2fs_has_inline_dentry(inode)) {
1082  		err = f2fs_read_inline_dir(file, ctx, &fstr);
1083  		goto out_free;
1084  	}
1085  
1086  	for (; n < npages; ctx->pos = n * NR_DENTRY_IN_BLOCK) {
1087  		pgoff_t next_pgofs;
1088  
1089  		/* allow readdir() to be interrupted */
1090  		if (fatal_signal_pending(current)) {
1091  			err = -ERESTARTSYS;
1092  			goto out_free;
1093  		}
1094  		cond_resched();
1095  
1096  		/* readahead for multi pages of dir */
1097  		if (npages - n > 1 && !ra_has_index(ra, n))
1098  			page_cache_sync_readahead(inode->i_mapping, ra, file, n,
1099  				min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES));
1100  
1101  		dentry_page = f2fs_find_data_page(inode, n, &next_pgofs);
1102  		if (IS_ERR(dentry_page)) {
1103  			err = PTR_ERR(dentry_page);
1104  			if (err == -ENOENT) {
1105  				err = 0;
1106  				n = next_pgofs;
1107  				continue;
1108  			} else {
1109  				goto out_free;
1110  			}
1111  		}
1112  
1113  		dentry_blk = page_address(dentry_page);
1114  
1115  		make_dentry_ptr_block(inode, &d, dentry_blk);
1116  
1117  		err = f2fs_fill_dentries(ctx, &d,
1118  				n * NR_DENTRY_IN_BLOCK, &fstr);
1119  		if (err) {
1120  			f2fs_put_page(dentry_page, 0);
1121  			break;
1122  		}
1123  
1124  		f2fs_put_page(dentry_page, 0);
1125  
1126  		n++;
1127  	}
1128  out_free:
1129  	fscrypt_fname_free_buffer(&fstr);
1130  out:
1131  	trace_f2fs_readdir(inode, start_pos, ctx->pos, err);
1132  	return err < 0 ? err : 0;
1133  }
1134  
1135  const struct file_operations f2fs_dir_operations = {
1136  	.llseek		= generic_file_llseek,
1137  	.read		= generic_read_dir,
1138  	.iterate_shared	= f2fs_readdir,
1139  	.fsync		= f2fs_sync_file,
1140  	.unlocked_ioctl	= f2fs_ioctl,
1141  #ifdef CONFIG_COMPAT
1142  	.compat_ioctl   = f2fs_compat_ioctl,
1143  #endif
1144  };
1145