Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Mar 24-27, 2025, special US time zones
Register
Loading...
v3.5.6
 
  1/*
  2 *  linux/fs/ext4/dir.c
  3 *
  4 * Copyright (C) 1992, 1993, 1994, 1995
  5 * Remy Card (card@masi.ibp.fr)
  6 * Laboratoire MASI - Institut Blaise Pascal
  7 * Universite Pierre et Marie Curie (Paris VI)
  8 *
  9 *  from
 10 *
 11 *  linux/fs/minix/dir.c
 12 *
 13 *  Copyright (C) 1991, 1992  Linus Torvalds
 14 *
 15 *  ext4 directory handling functions
 16 *
 17 *  Big-endian to little-endian byte-swapping/bitmaps by
 18 *        David S. Miller (davem@caip.rutgers.edu), 1995
 19 *
 20 * Hash Tree Directory indexing (c) 2001  Daniel Phillips
 21 *
 22 */
 23
 24#include <linux/fs.h>
 25#include <linux/jbd2.h>
 26#include <linux/buffer_head.h>
 27#include <linux/slab.h>
 28#include <linux/rbtree.h>
 
 29#include "ext4.h"
 
 30
 31static unsigned char ext4_filetype_table[] = {
 32	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 33};
 34
 35static int ext4_dx_readdir(struct file *filp,
 36			   void *dirent, filldir_t filldir);
 37
 38static unsigned char get_dtype(struct super_block *sb, int filetype)
 39{
 40	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
 41	    (filetype >= EXT4_FT_MAX))
 42		return DT_UNKNOWN;
 43
 44	return (ext4_filetype_table[filetype]);
 45}
 46
 47/**
 
 
 
 48 * Check if the given dir-inode refers to an htree-indexed directory
 49 * (or a directory which chould potentially get coverted to use htree
 50 * indexing).
 51 *
 52 * Return 1 if it is a dx dir, 0 if not
 53 */
 54static int is_dx_dir(struct inode *inode)
 55{
 56	struct super_block *sb = inode->i_sb;
 57
 58	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 59		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
 60	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
 61	     ((inode->i_size >> sb->s_blocksize_bits) == 1)))
 
 62		return 1;
 63
 64	return 0;
 65}
 66
 67/*
 68 * Return 0 if the directory entry is OK, and 1 if there is a problem
 69 *
 70 * Note: this is the opposite of what ext2 and ext3 historically returned...
 
 
 
 71 */
 72int __ext4_check_dir_entry(const char *function, unsigned int line,
 73			   struct inode *dir, struct file *filp,
 74			   struct ext4_dir_entry_2 *de,
 75			   struct buffer_head *bh,
 76			   unsigned int offset)
 77{
 78	const char *error_msg = NULL;
 79	const int rlen = ext4_rec_len_from_disk(de->rec_len,
 80						dir->i_sb->s_blocksize);
 81
 82	if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
 83		error_msg = "rec_len is smaller than minimal";
 84	else if (unlikely(rlen % 4 != 0))
 85		error_msg = "rec_len % 4 != 0";
 86	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
 87		error_msg = "rec_len is too small for name_len";
 88	else if (unlikely(((char *) de - bh->b_data) + rlen >
 89			  dir->i_sb->s_blocksize))
 90		error_msg = "directory entry across blocks";
 91	else if (unlikely(le32_to_cpu(de->inode) >
 92			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
 93		error_msg = "inode out of bounds";
 94	else
 95		return 0;
 96
 97	if (filp)
 98		ext4_error_file(filp, function, line, bh->b_blocknr,
 99				"bad entry in directory: %s - offset=%u(%u), "
100				"inode=%u, rec_len=%d, name_len=%d",
101				error_msg, (unsigned) (offset % bh->b_size),
102				offset, le32_to_cpu(de->inode),
103				rlen, de->name_len);
104	else
105		ext4_error_inode(dir, function, line, bh->b_blocknr,
106				"bad entry in directory: %s - offset=%u(%u), "
107				"inode=%u, rec_len=%d, name_len=%d",
108				error_msg, (unsigned) (offset % bh->b_size),
109				offset, le32_to_cpu(de->inode),
110				rlen, de->name_len);
111
112	return 1;
113}
114
115static int ext4_readdir(struct file *filp,
116			 void *dirent, filldir_t filldir)
117{
118	int error = 0;
119	unsigned int offset;
120	int i, stored;
121	struct ext4_dir_entry_2 *de;
122	int err;
123	struct inode *inode = filp->f_path.dentry->d_inode;
124	struct super_block *sb = inode->i_sb;
125	int ret = 0;
126	int dir_has_error = 0;
 
 
 
 
 
 
127
128	if (is_dx_dir(inode)) {
129		err = ext4_dx_readdir(filp, dirent, filldir);
130		if (err != ERR_BAD_DX_DIR) {
131			ret = err;
132			goto out;
133		}
134		/*
135		 * We don't set the inode dirty flag since it's not
136		 * critical that it get flushed back to the disk.
137		 */
138		ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
139				      EXT4_INODE_INDEX);
140	}
141	stored = 0;
142	offset = filp->f_pos & (sb->s_blocksize - 1);
143
144	while (!error && !stored && filp->f_pos < inode->i_size) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
145		struct ext4_map_blocks map;
146		struct buffer_head *bh = NULL;
147
148		map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
 
 
 
 
 
 
149		map.m_len = 1;
150		err = ext4_map_blocks(NULL, inode, &map, 0);
 
 
 
 
 
 
 
 
151		if (err > 0) {
152			pgoff_t index = map.m_pblk >>
153					(PAGE_CACHE_SHIFT - inode->i_blkbits);
154			if (!ra_has_index(&filp->f_ra, index))
155				page_cache_sync_readahead(
156					sb->s_bdev->bd_inode->i_mapping,
157					&filp->f_ra, filp,
158					index, 1);
159			filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
160			bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
 
 
 
 
 
161		}
162
163		/*
164		 * We ignore I/O errors on directories so users have a chance
165		 * of recovering data when there's a bad sector
166		 */
167		if (!bh) {
168			if (!dir_has_error) {
169				EXT4_ERROR_FILE(filp, 0,
170						"directory contains a "
171						"hole at offset %llu",
172					   (unsigned long long) filp->f_pos);
173				dir_has_error = 1;
174			}
175			/* corrupt size?  Maybe no more blocks to read */
176			if (filp->f_pos > inode->i_blocks << 9)
177				break;
178			filp->f_pos += sb->s_blocksize - offset;
179			continue;
180		}
181
182		/* Check the checksum */
183		if (!buffer_verified(bh) &&
184		    !ext4_dirent_csum_verify(inode,
185				(struct ext4_dir_entry *)bh->b_data)) {
186			EXT4_ERROR_FILE(filp, 0, "directory fails checksum "
187					"at offset %llu",
188					(unsigned long long)filp->f_pos);
189			filp->f_pos += sb->s_blocksize - offset;
 
 
190			continue;
191		}
192		set_buffer_verified(bh);
193
194revalidate:
195		/* If the dir block has changed since the last call to
196		 * readdir(2), then we might be pointing to an invalid
197		 * dirent right now.  Scan from the start of the block
198		 * to make sure. */
199		if (filp->f_version != inode->i_version) {
200			for (i = 0; i < sb->s_blocksize && i < offset; ) {
201				de = (struct ext4_dir_entry_2 *)
202					(bh->b_data + i);
203				/* It's too expensive to do a full
204				 * dirent test each time round this
205				 * loop, but we do have to test at
206				 * least that it is non-zero.  A
207				 * failure will be detected in the
208				 * dirent test below. */
209				if (ext4_rec_len_from_disk(de->rec_len,
210					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
211					break;
212				i += ext4_rec_len_from_disk(de->rec_len,
213							    sb->s_blocksize);
214			}
215			offset = i;
216			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
217				| offset;
218			filp->f_version = inode->i_version;
219		}
220
221		while (!error && filp->f_pos < inode->i_size
222		       && offset < sb->s_blocksize) {
223			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
224			if (ext4_check_dir_entry(inode, filp, de,
225						 bh, offset)) {
 
226				/*
227				 * On error, skip the f_pos to the next block
228				 */
229				filp->f_pos = (filp->f_pos |
230						(sb->s_blocksize - 1)) + 1;
231				brelse(bh);
232				ret = stored;
233				goto out;
234			}
235			offset += ext4_rec_len_from_disk(de->rec_len,
236					sb->s_blocksize);
237			if (le32_to_cpu(de->inode)) {
238				/* We might block in the next section
239				 * if the data destination is
240				 * currently swapped out.  So, use a
241				 * version stamp to detect whether or
242				 * not the directory has been modified
243				 * during the copy operation.
244				 */
245				u64 version = filp->f_version;
246
247				error = filldir(dirent, de->name,
248						de->name_len,
249						filp->f_pos,
250						le32_to_cpu(de->inode),
251						get_dtype(sb, de->file_type));
252				if (error)
253					break;
254				if (version != filp->f_version)
255					goto revalidate;
256				stored++;
 
 
 
 
 
 
257			}
258			filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
259						sb->s_blocksize);
260		}
261		offset = 0;
 
262		brelse(bh);
 
 
263	}
264out:
265	return ret;
 
 
 
 
266}
267
268static inline int is_32bit_api(void)
269{
270#ifdef CONFIG_COMPAT
271	return is_compat_task();
272#else
273	return (BITS_PER_LONG == 32);
274#endif
275}
276
277/*
278 * These functions convert from the major/minor hash to an f_pos
279 * value for dx directories
280 *
281 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
282 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
283 * directly on both 32-bit and 64-bit nodes, under such case, neither
284 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
285 */
286static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
287{
288	if ((filp->f_mode & FMODE_32BITHASH) ||
289	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
290		return major >> 1;
291	else
292		return ((__u64)(major >> 1) << 32) | (__u64)minor;
293}
294
295static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
296{
297	if ((filp->f_mode & FMODE_32BITHASH) ||
298	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299		return (pos << 1) & 0xffffffff;
300	else
301		return ((pos >> 32) << 1) & 0xffffffff;
302}
303
304static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
305{
306	if ((filp->f_mode & FMODE_32BITHASH) ||
307	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
308		return 0;
309	else
310		return pos & 0xffffffff;
311}
312
313/*
314 * Return 32- or 64-bit end-of-file for dx directories
315 */
316static inline loff_t ext4_get_htree_eof(struct file *filp)
317{
318	if ((filp->f_mode & FMODE_32BITHASH) ||
319	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
320		return EXT4_HTREE_EOF_32BIT;
321	else
322		return EXT4_HTREE_EOF_64BIT;
323}
324
325
326/*
327 * ext4_dir_llseek() based on generic_file_llseek() to handle both
328 * non-htree and htree directories, where the "offset" is in terms
329 * of the filename hash value instead of the byte offset.
 
 
 
 
330 *
331 * NOTE: offsets obtained *before* ext4_set_inode_flag(dir, EXT4_INODE_INDEX)
332 *       will be invalid once the directory was converted into a dx directory
333 */
334loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin)
335{
336	struct inode *inode = file->f_mapping->host;
337	loff_t ret = -EINVAL;
338	int dx_dir = is_dx_dir(inode);
 
339
340	mutex_lock(&inode->i_mutex);
341
342	/* NOTE: relative offsets with dx directories might not work
343	 *       as expected, as it is difficult to figure out the
344	 *       correct offset between dx hashes */
345
346	switch (origin) {
347	case SEEK_END:
348		if (unlikely(offset > 0))
349			goto out_err; /* not supported for directories */
350
351		/* so only negative offsets are left, does that have a
352		 * meaning for directories at all? */
353		if (dx_dir)
354			offset += ext4_get_htree_eof(file);
355		else
356			offset += inode->i_size;
357		break;
358	case SEEK_CUR:
359		/*
360		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
361		 * position-querying operation.  Avoid rewriting the "same"
362		 * f_pos value back to the file because a concurrent read(),
363		 * write() or lseek() might have altered it
364		 */
365		if (offset == 0) {
366			offset = file->f_pos;
367			goto out_ok;
368		}
369
370		offset += file->f_pos;
371		break;
372	}
373
374	if (unlikely(offset < 0))
375		goto out_err;
376
377	if (!dx_dir) {
378		if (offset > inode->i_sb->s_maxbytes)
379			goto out_err;
380	} else if (offset > ext4_get_htree_eof(file))
381		goto out_err;
382
383	/* Special lock needed here? */
384	if (offset != file->f_pos) {
385		file->f_pos = offset;
386		file->f_version = 0;
387	}
388
389out_ok:
390	ret = offset;
391out_err:
392	mutex_unlock(&inode->i_mutex);
393
394	return ret;
395}
396
397/*
398 * This structure holds the nodes of the red-black tree used to store
399 * the directory entry in hash order.
400 */
401struct fname {
402	__u32		hash;
403	__u32		minor_hash;
404	struct rb_node	rb_hash;
405	struct fname	*next;
406	__u32		inode;
407	__u8		name_len;
408	__u8		file_type;
409	char		name[0];
410};
411
412/*
413 * This functoin implements a non-recursive way of freeing all of the
414 * nodes in the red-black tree.
415 */
416static void free_rb_tree_fname(struct rb_root *root)
417{
418	struct rb_node	*n = root->rb_node;
419	struct rb_node	*parent;
420	struct fname	*fname;
421
422	while (n) {
423		/* Do the node's children first */
424		if (n->rb_left) {
425			n = n->rb_left;
426			continue;
427		}
428		if (n->rb_right) {
429			n = n->rb_right;
430			continue;
431		}
432		/*
433		 * The node has no children; free it, and then zero
434		 * out parent's link to it.  Finally go to the
435		 * beginning of the loop and try to free the parent
436		 * node.
437		 */
438		parent = rb_parent(n);
439		fname = rb_entry(n, struct fname, rb_hash);
440		while (fname) {
441			struct fname *old = fname;
442			fname = fname->next;
443			kfree(old);
444		}
445		if (!parent)
446			*root = RB_ROOT;
447		else if (parent->rb_left == n)
448			parent->rb_left = NULL;
449		else if (parent->rb_right == n)
450			parent->rb_right = NULL;
451		n = parent;
452	}
453}
454
455
456static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
457							   loff_t pos)
458{
459	struct dir_private_info *p;
460
461	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
462	if (!p)
463		return NULL;
464	p->curr_hash = pos2maj_hash(filp, pos);
465	p->curr_minor_hash = pos2min_hash(filp, pos);
466	return p;
467}
468
469void ext4_htree_free_dir_info(struct dir_private_info *p)
470{
471	free_rb_tree_fname(&p->root);
472	kfree(p);
473}
474
475/*
476 * Given a directory entry, enter it into the fname rb tree.
 
 
 
 
477 */
478int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
479			     __u32 minor_hash,
480			     struct ext4_dir_entry_2 *dirent)
 
481{
482	struct rb_node **p, *parent = NULL;
483	struct fname *fname, *new_fn;
484	struct dir_private_info *info;
485	int len;
486
487	info = dir_file->private_data;
488	p = &info->root.rb_node;
489
490	/* Create and allocate the fname structure */
491	len = sizeof(struct fname) + dirent->name_len + 1;
492	new_fn = kzalloc(len, GFP_KERNEL);
493	if (!new_fn)
494		return -ENOMEM;
495	new_fn->hash = hash;
496	new_fn->minor_hash = minor_hash;
497	new_fn->inode = le32_to_cpu(dirent->inode);
498	new_fn->name_len = dirent->name_len;
499	new_fn->file_type = dirent->file_type;
500	memcpy(new_fn->name, dirent->name, dirent->name_len);
501	new_fn->name[dirent->name_len] = 0;
502
503	while (*p) {
504		parent = *p;
505		fname = rb_entry(parent, struct fname, rb_hash);
506
507		/*
508		 * If the hash and minor hash match up, then we put
509		 * them on a linked list.  This rarely happens...
510		 */
511		if ((new_fn->hash == fname->hash) &&
512		    (new_fn->minor_hash == fname->minor_hash)) {
513			new_fn->next = fname->next;
514			fname->next = new_fn;
515			return 0;
516		}
517
518		if (new_fn->hash < fname->hash)
519			p = &(*p)->rb_left;
520		else if (new_fn->hash > fname->hash)
521			p = &(*p)->rb_right;
522		else if (new_fn->minor_hash < fname->minor_hash)
523			p = &(*p)->rb_left;
524		else /* if (new_fn->minor_hash > fname->minor_hash) */
525			p = &(*p)->rb_right;
526	}
527
528	rb_link_node(&new_fn->rb_hash, parent, p);
529	rb_insert_color(&new_fn->rb_hash, &info->root);
530	return 0;
531}
532
533
534
535/*
536 * This is a helper function for ext4_dx_readdir.  It calls filldir
537 * for all entres on the fname linked list.  (Normally there is only
538 * one entry on the linked list, unless there are 62 bit hash collisions.)
539 */
540static int call_filldir(struct file *filp, void *dirent,
541			filldir_t filldir, struct fname *fname)
542{
543	struct dir_private_info *info = filp->private_data;
544	loff_t	curr_pos;
545	struct inode *inode = filp->f_path.dentry->d_inode;
546	struct super_block *sb;
547	int error;
548
549	sb = inode->i_sb;
550
551	if (!fname) {
552		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
553			 "called with null fname?!?", __func__, __LINE__,
554			 inode->i_ino, current->comm);
555		return 0;
556	}
557	curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
558	while (fname) {
559		error = filldir(dirent, fname->name,
560				fname->name_len, curr_pos,
561				fname->inode,
562				get_dtype(sb, fname->file_type));
563		if (error) {
564			filp->f_pos = curr_pos;
565			info->extra_fname = fname;
566			return error;
567		}
568		fname = fname->next;
569	}
570	return 0;
571}
572
573static int ext4_dx_readdir(struct file *filp,
574			 void *dirent, filldir_t filldir)
575{
576	struct dir_private_info *info = filp->private_data;
577	struct inode *inode = filp->f_path.dentry->d_inode;
578	struct fname *fname;
579	int	ret;
580
581	if (!info) {
582		info = ext4_htree_create_dir_info(filp, filp->f_pos);
583		if (!info)
584			return -ENOMEM;
585		filp->private_data = info;
586	}
587
588	if (filp->f_pos == ext4_get_htree_eof(filp))
589		return 0;	/* EOF */
590
591	/* Some one has messed with f_pos; reset the world */
592	if (info->last_pos != filp->f_pos) {
593		free_rb_tree_fname(&info->root);
594		info->curr_node = NULL;
595		info->extra_fname = NULL;
596		info->curr_hash = pos2maj_hash(filp, filp->f_pos);
597		info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
598	}
599
600	/*
601	 * If there are any leftover names on the hash collision
602	 * chain, return them first.
603	 */
604	if (info->extra_fname) {
605		if (call_filldir(filp, dirent, filldir, info->extra_fname))
606			goto finished;
607		info->extra_fname = NULL;
608		goto next_node;
609	} else if (!info->curr_node)
610		info->curr_node = rb_first(&info->root);
611
612	while (1) {
613		/*
614		 * Fill the rbtree if we have no more entries,
615		 * or the inode has changed since we last read in the
616		 * cached entries.
617		 */
618		if ((!info->curr_node) ||
619		    (filp->f_version != inode->i_version)) {
620			info->curr_node = NULL;
621			free_rb_tree_fname(&info->root);
622			filp->f_version = inode->i_version;
623			ret = ext4_htree_fill_tree(filp, info->curr_hash,
624						   info->curr_minor_hash,
625						   &info->next_hash);
626			if (ret < 0)
627				return ret;
628			if (ret == 0) {
629				filp->f_pos = ext4_get_htree_eof(filp);
630				break;
631			}
632			info->curr_node = rb_first(&info->root);
633		}
634
635		fname = rb_entry(info->curr_node, struct fname, rb_hash);
636		info->curr_hash = fname->hash;
637		info->curr_minor_hash = fname->minor_hash;
638		if (call_filldir(filp, dirent, filldir, fname))
639			break;
640	next_node:
641		info->curr_node = rb_next(info->curr_node);
642		if (info->curr_node) {
643			fname = rb_entry(info->curr_node, struct fname,
644					 rb_hash);
645			info->curr_hash = fname->hash;
646			info->curr_minor_hash = fname->minor_hash;
647		} else {
648			if (info->next_hash == ~0) {
649				filp->f_pos = ext4_get_htree_eof(filp);
650				break;
651			}
652			info->curr_hash = info->next_hash;
653			info->curr_minor_hash = 0;
654		}
655	}
656finished:
657	info->last_pos = filp->f_pos;
 
 
 
 
 
 
 
658	return 0;
659}
660
661static int ext4_release_dir(struct inode *inode, struct file *filp)
662{
663	if (filp->private_data)
664		ext4_htree_free_dir_info(filp->private_data);
665
666	return 0;
667}
668
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
669const struct file_operations ext4_dir_operations = {
670	.llseek		= ext4_dir_llseek,
671	.read		= generic_read_dir,
672	.readdir	= ext4_readdir,
673	.unlocked_ioctl = ext4_ioctl,
674#ifdef CONFIG_COMPAT
675	.compat_ioctl	= ext4_compat_ioctl,
676#endif
677	.fsync		= ext4_sync_file,
 
678	.release	= ext4_release_dir,
679};
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/fs/ext4/dir.c
  4 *
  5 * Copyright (C) 1992, 1993, 1994, 1995
  6 * Remy Card (card@masi.ibp.fr)
  7 * Laboratoire MASI - Institut Blaise Pascal
  8 * Universite Pierre et Marie Curie (Paris VI)
  9 *
 10 *  from
 11 *
 12 *  linux/fs/minix/dir.c
 13 *
 14 *  Copyright (C) 1991, 1992  Linus Torvalds
 15 *
 16 *  ext4 directory handling functions
 17 *
 18 *  Big-endian to little-endian byte-swapping/bitmaps by
 19 *        David S. Miller (davem@caip.rutgers.edu), 1995
 20 *
 21 * Hash Tree Directory indexing (c) 2001  Daniel Phillips
 22 *
 23 */
 24
 25#include <linux/fs.h>
 
 26#include <linux/buffer_head.h>
 27#include <linux/slab.h>
 28#include <linux/iversion.h>
 29#include <linux/unicode.h>
 30#include "ext4.h"
 31#include "xattr.h"
 32
 33static int ext4_dx_readdir(struct file *, struct dir_context *);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 34
 35/**
 36 * is_dx_dir() - check if a directory is using htree indexing
 37 * @inode: directory inode
 38 *
 39 * Check if the given dir-inode refers to an htree-indexed directory
 40 * (or a directory which could potentially get converted to use htree
 41 * indexing).
 42 *
 43 * Return 1 if it is a dx dir, 0 if not
 44 */
 45static int is_dx_dir(struct inode *inode)
 46{
 47	struct super_block *sb = inode->i_sb;
 48
 49	if (ext4_has_feature_dir_index(inode->i_sb) &&
 
 50	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
 51	     ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
 52	     ext4_has_inline_data(inode)))
 53		return 1;
 54
 55	return 0;
 56}
 57
 58/*
 59 * Return 0 if the directory entry is OK, and 1 if there is a problem
 60 *
 61 * Note: this is the opposite of what ext2 and ext3 historically returned...
 62 *
 63 * bh passed here can be an inode block or a dir data block, depending
 64 * on the inode inline data flag.
 65 */
 66int __ext4_check_dir_entry(const char *function, unsigned int line,
 67			   struct inode *dir, struct file *filp,
 68			   struct ext4_dir_entry_2 *de,
 69			   struct buffer_head *bh, char *buf, int size,
 70			   unsigned int offset)
 71{
 72	const char *error_msg = NULL;
 73	const int rlen = ext4_rec_len_from_disk(de->rec_len,
 74						dir->i_sb->s_blocksize);
 75
 76	if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
 77		error_msg = "rec_len is smaller than minimal";
 78	else if (unlikely(rlen % 4 != 0))
 79		error_msg = "rec_len % 4 != 0";
 80	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
 81		error_msg = "rec_len is too small for name_len";
 82	else if (unlikely(((char *) de - buf) + rlen > size))
 83		error_msg = "directory entry overrun";
 
 84	else if (unlikely(le32_to_cpu(de->inode) >
 85			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
 86		error_msg = "inode out of bounds";
 87	else
 88		return 0;
 89
 90	if (filp)
 91		ext4_error_file(filp, function, line, bh->b_blocknr,
 92				"bad entry in directory: %s - offset=%u, "
 93				"inode=%u, rec_len=%d, name_len=%d, size=%d",
 94				error_msg, offset, le32_to_cpu(de->inode),
 95				rlen, de->name_len, size);
 
 96	else
 97		ext4_error_inode(dir, function, line, bh->b_blocknr,
 98				"bad entry in directory: %s - offset=%u, "
 99				"inode=%u, rec_len=%d, name_len=%d, size=%d",
100				 error_msg, offset, le32_to_cpu(de->inode),
101				 rlen, de->name_len, size);
 
102
103	return 1;
104}
105
106static int ext4_readdir(struct file *file, struct dir_context *ctx)
 
107{
 
108	unsigned int offset;
109	int i;
110	struct ext4_dir_entry_2 *de;
111	int err;
112	struct inode *inode = file_inode(file);
113	struct super_block *sb = inode->i_sb;
114	struct buffer_head *bh = NULL;
115	struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
116
117	if (IS_ENCRYPTED(inode)) {
118		err = fscrypt_get_encryption_info(inode);
119		if (err && err != -ENOKEY)
120			return err;
121	}
122
123	if (is_dx_dir(inode)) {
124		err = ext4_dx_readdir(file, ctx);
125		if (err != ERR_BAD_DX_DIR) {
126			return err;
 
127		}
128		/*
129		 * We don't set the inode dirty flag since it's not
130		 * critical that it get flushed back to the disk.
131		 */
132		ext4_clear_inode_flag(file_inode(file),
133				      EXT4_INODE_INDEX);
134	}
 
 
135
136	if (ext4_has_inline_data(inode)) {
137		int has_inline_data = 1;
138		err = ext4_read_inline_dir(file, ctx,
139					   &has_inline_data);
140		if (has_inline_data)
141			return err;
142	}
143
144	if (IS_ENCRYPTED(inode)) {
145		err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
146		if (err < 0)
147			return err;
148	}
149
150	while (ctx->pos < inode->i_size) {
151		struct ext4_map_blocks map;
 
152
153		if (fatal_signal_pending(current)) {
154			err = -ERESTARTSYS;
155			goto errout;
156		}
157		cond_resched();
158		offset = ctx->pos & (sb->s_blocksize - 1);
159		map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
160		map.m_len = 1;
161		err = ext4_map_blocks(NULL, inode, &map, 0);
162		if (err == 0) {
163			/* m_len should never be zero but let's avoid
164			 * an infinite loop if it somehow is */
165			if (map.m_len == 0)
166				map.m_len = 1;
167			ctx->pos += map.m_len * sb->s_blocksize;
168			continue;
169		}
170		if (err > 0) {
171			pgoff_t index = map.m_pblk >>
172					(PAGE_SHIFT - inode->i_blkbits);
173			if (!ra_has_index(&file->f_ra, index))
174				page_cache_sync_readahead(
175					sb->s_bdev->bd_inode->i_mapping,
176					&file->f_ra, file,
177					index, 1);
178			file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
179			bh = ext4_bread(NULL, inode, map.m_lblk, 0);
180			if (IS_ERR(bh)) {
181				err = PTR_ERR(bh);
182				bh = NULL;
183				goto errout;
184			}
185		}
186
 
 
 
 
187		if (!bh) {
 
 
 
 
 
 
 
188			/* corrupt size?  Maybe no more blocks to read */
189			if (ctx->pos > inode->i_blocks << 9)
190				break;
191			ctx->pos += sb->s_blocksize - offset;
192			continue;
193		}
194
195		/* Check the checksum */
196		if (!buffer_verified(bh) &&
197		    !ext4_dirblock_csum_verify(inode, bh)) {
198			EXT4_ERROR_FILE(file, 0, "directory fails checksum "
 
199					"at offset %llu",
200					(unsigned long long)ctx->pos);
201			ctx->pos += sb->s_blocksize - offset;
202			brelse(bh);
203			bh = NULL;
204			continue;
205		}
206		set_buffer_verified(bh);
207
 
208		/* If the dir block has changed since the last call to
209		 * readdir(2), then we might be pointing to an invalid
210		 * dirent right now.  Scan from the start of the block
211		 * to make sure. */
212		if (!inode_eq_iversion(inode, file->f_version)) {
213			for (i = 0; i < sb->s_blocksize && i < offset; ) {
214				de = (struct ext4_dir_entry_2 *)
215					(bh->b_data + i);
216				/* It's too expensive to do a full
217				 * dirent test each time round this
218				 * loop, but we do have to test at
219				 * least that it is non-zero.  A
220				 * failure will be detected in the
221				 * dirent test below. */
222				if (ext4_rec_len_from_disk(de->rec_len,
223					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
224					break;
225				i += ext4_rec_len_from_disk(de->rec_len,
226							    sb->s_blocksize);
227			}
228			offset = i;
229			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
230				| offset;
231			file->f_version = inode_query_iversion(inode);
232		}
233
234		while (ctx->pos < inode->i_size
235		       && offset < sb->s_blocksize) {
236			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
237			if (ext4_check_dir_entry(inode, file, de, bh,
238						 bh->b_data, bh->b_size,
239						 offset)) {
240				/*
241				 * On error, skip to the next block
242				 */
243				ctx->pos = (ctx->pos |
244						(sb->s_blocksize - 1)) + 1;
245				break;
 
 
246			}
247			offset += ext4_rec_len_from_disk(de->rec_len,
248					sb->s_blocksize);
249			if (le32_to_cpu(de->inode)) {
250				if (!IS_ENCRYPTED(inode)) {
251					if (!dir_emit(ctx, de->name,
252					    de->name_len,
253					    le32_to_cpu(de->inode),
254					    get_dtype(sb, de->file_type)))
255						goto done;
256				} else {
257					int save_len = fstr.len;
258					struct fscrypt_str de_name =
259							FSTR_INIT(de->name,
260								de->name_len);
261
262					/* Directory is encrypted */
263					err = fscrypt_fname_disk_to_usr(inode,
264						0, 0, &de_name, &fstr);
265					de_name = fstr;
266					fstr.len = save_len;
267					if (err)
268						goto errout;
269					if (!dir_emit(ctx,
270					    de_name.name, de_name.len,
271					    le32_to_cpu(de->inode),
272					    get_dtype(sb, de->file_type)))
273						goto done;
274				}
275			}
276			ctx->pos += ext4_rec_len_from_disk(de->rec_len,
277						sb->s_blocksize);
278		}
279		if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
280			goto done;
281		brelse(bh);
282		bh = NULL;
283		offset = 0;
284	}
285done:
286	err = 0;
287errout:
288	fscrypt_fname_free_buffer(&fstr);
289	brelse(bh);
290	return err;
291}
292
293static inline int is_32bit_api(void)
294{
295#ifdef CONFIG_COMPAT
296	return in_compat_syscall();
297#else
298	return (BITS_PER_LONG == 32);
299#endif
300}
301
302/*
303 * These functions convert from the major/minor hash to an f_pos
304 * value for dx directories
305 *
306 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
307 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
308 * directly on both 32-bit and 64-bit nodes, under such case, neither
309 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
310 */
311static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
312{
313	if ((filp->f_mode & FMODE_32BITHASH) ||
314	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
315		return major >> 1;
316	else
317		return ((__u64)(major >> 1) << 32) | (__u64)minor;
318}
319
320static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
321{
322	if ((filp->f_mode & FMODE_32BITHASH) ||
323	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
324		return (pos << 1) & 0xffffffff;
325	else
326		return ((pos >> 32) << 1) & 0xffffffff;
327}
328
329static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
330{
331	if ((filp->f_mode & FMODE_32BITHASH) ||
332	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
333		return 0;
334	else
335		return pos & 0xffffffff;
336}
337
338/*
339 * Return 32- or 64-bit end-of-file for dx directories
340 */
341static inline loff_t ext4_get_htree_eof(struct file *filp)
342{
343	if ((filp->f_mode & FMODE_32BITHASH) ||
344	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
345		return EXT4_HTREE_EOF_32BIT;
346	else
347		return EXT4_HTREE_EOF_64BIT;
348}
349
350
351/*
352 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
353 * directories, where the "offset" is in terms of the filename hash
354 * value instead of the byte offset.
355 *
356 * Because we may return a 64-bit hash that is well beyond offset limits,
357 * we need to pass the max hash as the maximum allowable offset in
358 * the htree directory case.
359 *
360 * For non-htree, ext4_llseek already chooses the proper max offset.
 
361 */
362static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
363{
364	struct inode *inode = file->f_mapping->host;
 
365	int dx_dir = is_dx_dir(inode);
366	loff_t ret, htree_max = ext4_get_htree_eof(file);
367
368	if (likely(dx_dir))
369		ret = generic_file_llseek_size(file, offset, whence,
370						    htree_max, htree_max);
371	else
372		ret = ext4_llseek(file, offset, whence);
373	file->f_version = inode_peek_iversion(inode) - 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
374	return ret;
375}
376
377/*
378 * This structure holds the nodes of the red-black tree used to store
379 * the directory entry in hash order.
380 */
381struct fname {
382	__u32		hash;
383	__u32		minor_hash;
384	struct rb_node	rb_hash;
385	struct fname	*next;
386	__u32		inode;
387	__u8		name_len;
388	__u8		file_type;
389	char		name[0];
390};
391
392/*
393 * This functoin implements a non-recursive way of freeing all of the
394 * nodes in the red-black tree.
395 */
396static void free_rb_tree_fname(struct rb_root *root)
397{
398	struct fname *fname, *next;
399
400	rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
401		while (fname) {
402			struct fname *old = fname;
403			fname = fname->next;
404			kfree(old);
405		}
406
407	*root = RB_ROOT;
 
 
 
 
 
 
408}
409
410
411static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
412							   loff_t pos)
413{
414	struct dir_private_info *p;
415
416	p = kzalloc(sizeof(*p), GFP_KERNEL);
417	if (!p)
418		return NULL;
419	p->curr_hash = pos2maj_hash(filp, pos);
420	p->curr_minor_hash = pos2min_hash(filp, pos);
421	return p;
422}
423
424void ext4_htree_free_dir_info(struct dir_private_info *p)
425{
426	free_rb_tree_fname(&p->root);
427	kfree(p);
428}
429
430/*
431 * Given a directory entry, enter it into the fname rb tree.
432 *
433 * When filename encryption is enabled, the dirent will hold the
434 * encrypted filename, while the htree will hold decrypted filename.
435 * The decrypted filename is passed in via ent_name.  parameter.
436 */
437int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
438			     __u32 minor_hash,
439			    struct ext4_dir_entry_2 *dirent,
440			    struct fscrypt_str *ent_name)
441{
442	struct rb_node **p, *parent = NULL;
443	struct fname *fname, *new_fn;
444	struct dir_private_info *info;
445	int len;
446
447	info = dir_file->private_data;
448	p = &info->root.rb_node;
449
450	/* Create and allocate the fname structure */
451	len = sizeof(struct fname) + ent_name->len + 1;
452	new_fn = kzalloc(len, GFP_KERNEL);
453	if (!new_fn)
454		return -ENOMEM;
455	new_fn->hash = hash;
456	new_fn->minor_hash = minor_hash;
457	new_fn->inode = le32_to_cpu(dirent->inode);
458	new_fn->name_len = ent_name->len;
459	new_fn->file_type = dirent->file_type;
460	memcpy(new_fn->name, ent_name->name, ent_name->len);
461	new_fn->name[ent_name->len] = 0;
462
463	while (*p) {
464		parent = *p;
465		fname = rb_entry(parent, struct fname, rb_hash);
466
467		/*
468		 * If the hash and minor hash match up, then we put
469		 * them on a linked list.  This rarely happens...
470		 */
471		if ((new_fn->hash == fname->hash) &&
472		    (new_fn->minor_hash == fname->minor_hash)) {
473			new_fn->next = fname->next;
474			fname->next = new_fn;
475			return 0;
476		}
477
478		if (new_fn->hash < fname->hash)
479			p = &(*p)->rb_left;
480		else if (new_fn->hash > fname->hash)
481			p = &(*p)->rb_right;
482		else if (new_fn->minor_hash < fname->minor_hash)
483			p = &(*p)->rb_left;
484		else /* if (new_fn->minor_hash > fname->minor_hash) */
485			p = &(*p)->rb_right;
486	}
487
488	rb_link_node(&new_fn->rb_hash, parent, p);
489	rb_insert_color(&new_fn->rb_hash, &info->root);
490	return 0;
491}
492
493
494
495/*
496 * This is a helper function for ext4_dx_readdir.  It calls filldir
497 * for all entres on the fname linked list.  (Normally there is only
498 * one entry on the linked list, unless there are 62 bit hash collisions.)
499 */
500static int call_filldir(struct file *file, struct dir_context *ctx,
501			struct fname *fname)
502{
503	struct dir_private_info *info = file->private_data;
504	struct inode *inode = file_inode(file);
505	struct super_block *sb = inode->i_sb;
 
 
 
 
506
507	if (!fname) {
508		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
509			 "called with null fname?!?", __func__, __LINE__,
510			 inode->i_ino, current->comm);
511		return 0;
512	}
513	ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
514	while (fname) {
515		if (!dir_emit(ctx, fname->name,
516				fname->name_len,
517				fname->inode,
518				get_dtype(sb, fname->file_type))) {
 
 
519			info->extra_fname = fname;
520			return 1;
521		}
522		fname = fname->next;
523	}
524	return 0;
525}
526
527static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
 
528{
529	struct dir_private_info *info = file->private_data;
530	struct inode *inode = file_inode(file);
531	struct fname *fname;
532	int	ret;
533
534	if (!info) {
535		info = ext4_htree_create_dir_info(file, ctx->pos);
536		if (!info)
537			return -ENOMEM;
538		file->private_data = info;
539	}
540
541	if (ctx->pos == ext4_get_htree_eof(file))
542		return 0;	/* EOF */
543
544	/* Some one has messed with f_pos; reset the world */
545	if (info->last_pos != ctx->pos) {
546		free_rb_tree_fname(&info->root);
547		info->curr_node = NULL;
548		info->extra_fname = NULL;
549		info->curr_hash = pos2maj_hash(file, ctx->pos);
550		info->curr_minor_hash = pos2min_hash(file, ctx->pos);
551	}
552
553	/*
554	 * If there are any leftover names on the hash collision
555	 * chain, return them first.
556	 */
557	if (info->extra_fname) {
558		if (call_filldir(file, ctx, info->extra_fname))
559			goto finished;
560		info->extra_fname = NULL;
561		goto next_node;
562	} else if (!info->curr_node)
563		info->curr_node = rb_first(&info->root);
564
565	while (1) {
566		/*
567		 * Fill the rbtree if we have no more entries,
568		 * or the inode has changed since we last read in the
569		 * cached entries.
570		 */
571		if ((!info->curr_node) ||
572		    !inode_eq_iversion(inode, file->f_version)) {
573			info->curr_node = NULL;
574			free_rb_tree_fname(&info->root);
575			file->f_version = inode_query_iversion(inode);
576			ret = ext4_htree_fill_tree(file, info->curr_hash,
577						   info->curr_minor_hash,
578						   &info->next_hash);
579			if (ret < 0)
580				return ret;
581			if (ret == 0) {
582				ctx->pos = ext4_get_htree_eof(file);
583				break;
584			}
585			info->curr_node = rb_first(&info->root);
586		}
587
588		fname = rb_entry(info->curr_node, struct fname, rb_hash);
589		info->curr_hash = fname->hash;
590		info->curr_minor_hash = fname->minor_hash;
591		if (call_filldir(file, ctx, fname))
592			break;
593	next_node:
594		info->curr_node = rb_next(info->curr_node);
595		if (info->curr_node) {
596			fname = rb_entry(info->curr_node, struct fname,
597					 rb_hash);
598			info->curr_hash = fname->hash;
599			info->curr_minor_hash = fname->minor_hash;
600		} else {
601			if (info->next_hash == ~0) {
602				ctx->pos = ext4_get_htree_eof(file);
603				break;
604			}
605			info->curr_hash = info->next_hash;
606			info->curr_minor_hash = 0;
607		}
608	}
609finished:
610	info->last_pos = ctx->pos;
611	return 0;
612}
613
614static int ext4_dir_open(struct inode * inode, struct file * filp)
615{
616	if (IS_ENCRYPTED(inode))
617		return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
618	return 0;
619}
620
621static int ext4_release_dir(struct inode *inode, struct file *filp)
622{
623	if (filp->private_data)
624		ext4_htree_free_dir_info(filp->private_data);
625
626	return 0;
627}
628
629int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
630		      int buf_size)
631{
632	struct ext4_dir_entry_2 *de;
633	int rlen;
634	unsigned int offset = 0;
635	char *top;
636
637	de = (struct ext4_dir_entry_2 *)buf;
638	top = buf + buf_size;
639	while ((char *) de < top) {
640		if (ext4_check_dir_entry(dir, NULL, de, bh,
641					 buf, buf_size, offset))
642			return -EFSCORRUPTED;
643		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
644		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
645		offset += rlen;
646	}
647	if ((char *) de > top)
648		return -EFSCORRUPTED;
649
650	return 0;
651}
652
653const struct file_operations ext4_dir_operations = {
654	.llseek		= ext4_dir_llseek,
655	.read		= generic_read_dir,
656	.iterate_shared	= ext4_readdir,
657	.unlocked_ioctl = ext4_ioctl,
658#ifdef CONFIG_COMPAT
659	.compat_ioctl	= ext4_compat_ioctl,
660#endif
661	.fsync		= ext4_sync_file,
662	.open		= ext4_dir_open,
663	.release	= ext4_release_dir,
664};
665
666#ifdef CONFIG_UNICODE
667static int ext4_d_compare(const struct dentry *dentry, unsigned int len,
668			  const char *str, const struct qstr *name)
669{
670	struct qstr qstr = {.name = str, .len = len };
671	struct inode *inode = dentry->d_parent->d_inode;
672
673	if (!IS_CASEFOLDED(inode) || !EXT4_SB(inode->i_sb)->s_encoding) {
674		if (len != name->len)
675			return -1;
676		return memcmp(str, name->name, len);
677	}
678
679	return ext4_ci_compare(inode, name, &qstr, false);
680}
681
682static int ext4_d_hash(const struct dentry *dentry, struct qstr *str)
683{
684	const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb);
685	const struct unicode_map *um = sbi->s_encoding;
686	unsigned char *norm;
687	int len, ret = 0;
688
689	if (!IS_CASEFOLDED(dentry->d_inode) || !um)
690		return 0;
691
692	norm = kmalloc(PATH_MAX, GFP_ATOMIC);
693	if (!norm)
694		return -ENOMEM;
695
696	len = utf8_casefold(um, str, norm, PATH_MAX);
697	if (len < 0) {
698		if (ext4_has_strict_mode(sbi))
699			ret = -EINVAL;
700		goto out;
701	}
702	str->hash = full_name_hash(dentry, norm, len);
703out:
704	kfree(norm);
705	return ret;
706}
707
708const struct dentry_operations ext4_dentry_ops = {
709	.d_hash = ext4_d_hash,
710	.d_compare = ext4_d_compare,
711};
712#endif