[PATCH 0/2] ext4: fix readdir error with getdents.

[PATCH 0/2] ext4: fix readdir error with getdents.

[Tao Ma]

From: Tao Ma <boyu.mt@taobao.com>

Hi all,
    This patch set just tries to fix the problem when we call getdents
against an inline dir.  As file->f_pos has quite different meaning in
case of dir_index enabled/disabled. So 2 separate patches are created
to resolve these 2 issues in different ways.

Thanks,
Tao

[PATCH 1/2] ext4: fix readdir error in case inline_data+dir_index.

[Tao Ma]

From: Tao Ma <boyu.mt@taobao.com>

Zach reported a problem that if inline data is enabled, we don't
tell the difference between the offset of '.' and '..'. And a
getdents will fail if the user only want to get '.' and what's worse,
if there is a conversion happens when the user calls getdents
many times, he/she may get the same entry twice.

In theroy, a dir block would also fail if it is converted to a
hashed-index based dir since f_pos will become a hash value, not the
real one, but it doesn't happen. And a deep investigation
shows that we uses a hash based solution even for a normal dir if
the dir_index feature is enabled.

So this patch just adds a new htree_inlinedir_to_tree for inline dir,
and if we find that the hash index is supported, we will do like what
we do for a dir block.

Reported-by: Zach Brown <zab@redhat.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
---
 fs/ext4/dir.c    |   20 +++++----
 fs/ext4/ext4.h   |   23 +++++++++++
 fs/ext4/inline.c |  109 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 fs/ext4/namei.c  |   29 +++++---------
 4 files changed, 153 insertions(+), 28 deletions(-)

diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
index d8cd1f0..f8d56e4 100644
--- a/fs/ext4/dir.c
+++ b/fs/ext4/dir.c
@@ -46,7 +46,8 @@ static int is_dx_dir(struct inode *inode)
 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
-	     ((inode->i_size >> sb->s_blocksize_bits) == 1)))
+	     ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
+	     ext4_has_inline_data(inode)))
 		return 1;
 
 	return 0;
@@ -115,14 +116,6 @@ static int ext4_readdir(struct file *filp,
 	int ret = 0;
 	int dir_has_error = 0;
 
-	if (ext4_has_inline_data(inode)) {
-		int has_inline_data = 1;
-		ret = ext4_read_inline_dir(filp, dirent, filldir,
-					   &has_inline_data);
-		if (has_inline_data)
-			return ret;
-	}
-
 	if (is_dx_dir(inode)) {
 		err = ext4_dx_readdir(filp, dirent, filldir);
 		if (err != ERR_BAD_DX_DIR) {
@@ -136,6 +129,15 @@ static int ext4_readdir(struct file *filp,
 		ext4_clear_inode_flag(file_inode(filp),
 				      EXT4_INODE_INDEX);
 	}
+
+	if (ext4_has_inline_data(inode)) {
+		int has_inline_data = 1;
+		ret = ext4_read_inline_dir(filp, dirent, filldir,
+					   &has_inline_data);
+		if (has_inline_data)
+			return ret;
+	}
+
 	stored = 0;
 	offset = filp->f_pos & (sb->s_blocksize - 1);
 
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 3b83cd6..83ba3b4 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -2511,6 +2511,11 @@ extern int ext4_try_create_inline_dir(handle_t *handle,
 extern int ext4_read_inline_dir(struct file *filp,
 				void *dirent, filldir_t filldir,
 				int *has_inline_data);
+extern int htree_inlinedir_to_tree(struct file *dir_file,
+				   struct inode *dir, ext4_lblk_t block,
+				   struct dx_hash_info *hinfo,
+				   __u32 start_hash, __u32 start_minor_hash,
+				   int *has_inline_data);
 extern struct buffer_head *ext4_find_inline_entry(struct inode *dir,
 					const struct qstr *d_name,
 					struct ext4_dir_entry_2 **res_dir,
@@ -2547,6 +2552,24 @@ extern void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 extern int ext4_handle_dirty_dirent_node(handle_t *handle,
 					 struct inode *inode,
 					 struct buffer_head *bh);
+#define S_SHIFT 12
+static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
+	[S_IFREG >> S_SHIFT]	= EXT4_FT_REG_FILE,
+	[S_IFDIR >> S_SHIFT]	= EXT4_FT_DIR,
+	[S_IFCHR >> S_SHIFT]	= EXT4_FT_CHRDEV,
+	[S_IFBLK >> S_SHIFT]	= EXT4_FT_BLKDEV,
+	[S_IFIFO >> S_SHIFT]	= EXT4_FT_FIFO,
+	[S_IFSOCK >> S_SHIFT]	= EXT4_FT_SOCK,
+	[S_IFLNK >> S_SHIFT]	= EXT4_FT_SYMLINK,
+};
+
+static inline void ext4_set_de_type(struct super_block *sb,
+				struct ext4_dir_entry_2 *de,
+				umode_t mode) {
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
+		de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
+}
+
 
 /* symlink.c */
 extern const struct inode_operations ext4_symlink_inode_operations;
diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
index c0fd1a1..abf8b62 100644
--- a/fs/ext4/inline.c
+++ b/fs/ext4/inline.c
@@ -19,7 +19,8 @@
 
 #define EXT4_XATTR_SYSTEM_DATA	"data"
 #define EXT4_MIN_INLINE_DATA_SIZE	((sizeof(__le32) * EXT4_N_BLOCKS))
-#define EXT4_INLINE_DOTDOT_SIZE	4
+#define EXT4_INLINE_DOTDOT_OFFSET	2
+#define EXT4_INLINE_DOTDOT_SIZE		4
 
 int ext4_get_inline_size(struct inode *inode)
 {
@@ -1289,6 +1290,112 @@ out:
 	return ret;
 }
 
+/*
+ * This function fills a red-black tree with information from an
+ * inlined dir.  It returns the number directory entries loaded
+ * into the tree.  If there is an error it is returned in err.
+ */
+int htree_inlinedir_to_tree(struct file *dir_file,
+			    struct inode *dir, ext4_lblk_t block,
+			    struct dx_hash_info *hinfo,
+			    __u32 start_hash, __u32 start_minor_hash,
+			    int *has_inline_data)
+{
+	int err = 0, count = 0;
+	unsigned int parent_ino;
+	int pos;
+	struct ext4_dir_entry_2 *de;
+	struct inode *inode = file_inode(dir_file);
+	int ret, inline_size = 0;
+	struct ext4_iloc iloc;
+	void *dir_buf = NULL;
+	struct ext4_dir_entry_2 fake;
+
+	ret = ext4_get_inode_loc(inode, &iloc);
+	if (ret)
+		return ret;
+
+	down_read(&EXT4_I(inode)->xattr_sem);
+	if (!ext4_has_inline_data(inode)) {
+		up_read(&EXT4_I(inode)->xattr_sem);
+		*has_inline_data = 0;
+		goto out;
+	}
+
+	inline_size = ext4_get_inline_size(inode);
+	dir_buf = kmalloc(inline_size, GFP_NOFS);
+	if (!dir_buf) {
+		ret = -ENOMEM;
+		up_read(&EXT4_I(inode)->xattr_sem);
+		goto out;
+	}
+
+	ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
+	up_read(&EXT4_I(inode)->xattr_sem);
+	if (ret < 0)
+		goto out;
+
+	pos = 0;
+	parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
+	while (pos < inline_size) {
+		/*
+		 * As inlined dir doesn't store any information about '.' and
+		 * only the inode number of '..' is stored, we have to handle
+		 * them differently.
+		 */
+		if (pos == 0) {
+			fake.inode = cpu_to_le32(inode->i_ino);
+			fake.name_len = 1;
+			strcpy(fake.name, ".");
+			fake.rec_len = ext4_rec_len_to_disk(
+						EXT4_DIR_REC_LEN(fake.name_len),
+						inline_size);
+			ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+			de = &fake;
+			pos = EXT4_INLINE_DOTDOT_OFFSET;
+		} else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
+			fake.inode = cpu_to_le32(parent_ino);
+			fake.name_len = 2;
+			strcpy(fake.name, "..");
+			fake.rec_len = ext4_rec_len_to_disk(
+						EXT4_DIR_REC_LEN(fake.name_len),
+						inline_size);
+			ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+			de = &fake;
+			pos = EXT4_INLINE_DOTDOT_SIZE;
+		} else {
+			de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
+			pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
+			if (ext4_check_dir_entry(inode, dir_file, de,
+					 iloc.bh, dir_buf,
+					 inline_size, pos)) {
+				ret = count;
+				goto out;
+			}
+		}
+
+		ext4fs_dirhash(de->name, de->name_len, hinfo);
+		if ((hinfo->hash < start_hash) ||
+		    ((hinfo->hash == start_hash) &&
+		     (hinfo->minor_hash < start_minor_hash)))
+			continue;
+		if (de->inode == 0)
+			continue;
+		err = ext4_htree_store_dirent(dir_file,
+				   hinfo->hash, hinfo->minor_hash, de);
+		if (err) {
+			count = err;
+			goto out;
+		}
+		count++;
+	}
+	ret = count;
+out:
+	kfree(dir_buf);
+	brelse(iloc.bh);
+	return ret;
+}
+
 int ext4_read_inline_dir(struct file *filp,
 			 void *dirent, filldir_t filldir,
 			 int *has_inline_data)
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 3825d6a..a35270a 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -971,6 +971,17 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 			hinfo.hash_version +=
 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+		if (ext4_has_inline_data(dir)) {
+			int has_inline_data = 1;
+			count = htree_inlinedir_to_tree(dir_file, dir, 0,
+							&hinfo, start_hash,
+							start_minor_hash,
+							&has_inline_data);
+			if (has_inline_data) {
+				*next_hash = ~0;
+				return count;
+			}
+		}
 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 					       start_hash, start_minor_hash);
 		*next_hash = ~0;
@@ -1455,24 +1466,6 @@ struct dentry *ext4_get_parent(struct dentry *child)
 	return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
 }
 
-#define S_SHIFT 12
-static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
-	[S_IFREG >> S_SHIFT]	= EXT4_FT_REG_FILE,
-	[S_IFDIR >> S_SHIFT]	= EXT4_FT_DIR,
-	[S_IFCHR >> S_SHIFT]	= EXT4_FT_CHRDEV,
-	[S_IFBLK >> S_SHIFT]	= EXT4_FT_BLKDEV,
-	[S_IFIFO >> S_SHIFT]	= EXT4_FT_FIFO,
-	[S_IFSOCK >> S_SHIFT]	= EXT4_FT_SOCK,
-	[S_IFLNK >> S_SHIFT]	= EXT4_FT_SYMLINK,
-};
-
-static inline void ext4_set_de_type(struct super_block *sb,
-				struct ext4_dir_entry_2 *de,
-				umode_t mode) {
-	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
-		de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
-}

[PATCH 2/2] ext4: fix readdir error in case inline_data+^dir_index.

[Tao Ma]

From: Tao Ma <boyu.mt@taobao.com>

Zach reported a problem that if inline data is enabled, we don't
tell the difference between the offset of '.' and '..'. And a
getdents will fail if the user only want to get '.'. And what's
worse, we may meet with duplicate dir entries as the offset
for inline dir and non-inline one is quite different.

This patch just try to resolve this problem if dir_index
is disabled. In this case, f_pos is the real offset with
the dir block, so for inline dir, we just pretend as if
we are a dir block and returns the offset like a norml
dir block does.

Reported-by: Zach Brown <zab@redhat.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
---
 fs/ext4/inline.c |   69 +++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 51 insertions(+), 18 deletions(-)

diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
index abf8b62..3e2bf87 100644
--- a/fs/ext4/inline.c
+++ b/fs/ext4/inline.c
@@ -1396,6 +1396,14 @@ out:
 	return ret;
 }
 
+/*
+ * So this function is called when the volume is mkfsed with
+ * dir_index disabled. In order to keep f_pos persistent
+ * after we convert from an inlined dir to a blocked based,
+ * we just pretend that we are a normal dir and return the
+ * offset as if '.' and '..' really take place.
+ *
+ */
 int ext4_read_inline_dir(struct file *filp,
 			 void *dirent, filldir_t filldir,
 			 int *has_inline_data)
@@ -1409,6 +1417,7 @@ int ext4_read_inline_dir(struct file *filp,
 	int ret, inline_size = 0;
 	struct ext4_iloc iloc;
 	void *dir_buf = NULL;
+	int dotdot_offset, dotdot_size, extra_offset, extra_size;
 
 	ret = ext4_get_inode_loc(inode, &iloc);
 	if (ret)
@@ -1437,8 +1446,21 @@ int ext4_read_inline_dir(struct file *filp,
 	sb = inode->i_sb;
 	stored = 0;
 	parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
+	offset = filp->f_pos;
+
+	/*
+	 * dotdot_offset and dotdot_size is the real offset and
+	 * size for ".." and "." if the dir is block based while
+	 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
+	 * So we will use extra_offset and extra_size to indicate them
+	 * during the inline dir iteration.
+	 */
+	dotdot_offset = EXT4_DIR_REC_LEN(1);
+	dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
+	extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
+	extra_size = extra_offset + inline_size;
 
-	while (!error && !stored && filp->f_pos < inode->i_size) {
+	while (!error && !stored && filp->f_pos < extra_size) {
 revalidate:
 		/*
 		 * If the version has changed since the last call to
@@ -1447,15 +1469,23 @@ revalidate:
 		 * dir to make sure.
 		 */
 		if (filp->f_version != inode->i_version) {
-			for (i = 0;
-			     i < inode->i_size && i < offset;) {
+			for (i = 0; i < extra_size && i < offset;) {
+				/*
+				 * "." is with offset 0 and
+				 * ".." is dotdot_offset.
+				 */
 				if (!i) {
-					/* skip "." and ".." if needed. */
-					i += EXT4_INLINE_DOTDOT_SIZE;
+					i = dotdot_offset;
+					continue;
+				} else if (i == dotdot_offset) {
+					i = dotdot_size;
 					continue;
 				}
+				/* for other entry, the real offset in
+				 * the buf has to be tuned accordingly.
+				 */
 				de = (struct ext4_dir_entry_2 *)
-					(dir_buf + i);
+					(dir_buf + i - extra_offset);
 				/* It's too expensive to do a full
 				 * dirent test each time round this
 				 * loop, but we do have to test at
@@ -1463,43 +1493,47 @@ revalidate:
 				 * failure will be detected in the
 				 * dirent test below. */
 				if (ext4_rec_len_from_disk(de->rec_len,
-					inline_size) < EXT4_DIR_REC_LEN(1))
+					extra_size) < EXT4_DIR_REC_LEN(1))
 					break;
 				i += ext4_rec_len_from_disk(de->rec_len,
-							    inline_size);
+							    extra_size);
 			}
 			offset = i;
 			filp->f_pos = offset;
 			filp->f_version = inode->i_version;
 		}
 
-		while (!error && filp->f_pos < inode->i_size) {
+		while (!error && filp->f_pos < extra_size) {
 			if (filp->f_pos == 0) {
 				error = filldir(dirent, ".", 1, 0, inode->i_ino,
 						DT_DIR);
 				if (error)
 					break;
 				stored++;
+				filp->f_pos = dotdot_offset;
+				continue;
+			}
 
-				error = filldir(dirent, "..", 2, 0, parent_ino,
-						DT_DIR);
+			if (filp->f_pos == dotdot_offset) {
+				error = filldir(dirent, "..", 2,
+						dotdot_offset,
+						parent_ino, DT_DIR);
 				if (error)
 					break;
 				stored++;
 
-				filp->f_pos = offset = EXT4_INLINE_DOTDOT_SIZE;
+				filp->f_pos = dotdot_size;
 				continue;
 			}
 
-			de = (struct ext4_dir_entry_2 *)(dir_buf + offset);
+			de = (struct ext4_dir_entry_2 *)
+				(dir_buf + filp->f_pos - extra_offset);
 			if (ext4_check_dir_entry(inode, filp, de,
 						 iloc.bh, dir_buf,
-						 inline_size, offset)) {
+						 extra_size, filp->f_pos)) {
 				ret = stored;
 				goto out;
 			}
-			offset += ext4_rec_len_from_disk(de->rec_len,
-							 inline_size);
 			if (le32_to_cpu(de->inode)) {
 				/* We might block in the next section
 				 * if the data destination is
@@ -1522,9 +1556,8 @@ revalidate:
 				stored++;
 			}
 			filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
-							      inline_size);
+							      extra_size);
 		}
-		offset = 0;
 	}
 out:
 	kfree(dir_buf);
-- 
1.7.0.4

Re: [PATCH 1/2] ext4: fix readdir error in case inline_data+dir_index.

[Zach Brown]

On Wed, Apr 10, 2013 at 11:37:07PM +0800, Tao Ma wrote:
> From: Tao Ma <boyu.mt@taobao.com>
> 
> Zach reported a problem that if inline data is enabled, we don't
> tell the difference between the offset of '.' and '..'. And a
> getdents will fail if the user only want to get '.' and what's worse,
> if there is a conversion happens when the user calls getdents
> many times, he/she may get the same entry twice.
> 
> In theroy, a dir block would also fail if it is converted to a
> hashed-index based dir since f_pos will become a hash value, not the
> real one, but it doesn't happen. And a deep investigation
> shows that we uses a hash based solution even for a normal dir if
> the dir_index feature is enabled.
> 
> So this patch just adds a new htree_inlinedir_to_tree for inline dir,
> and if we find that the hash index is supported, we will do like what
> we do for a dir block.

Mmm, I like the sound of that fix.  Nice and simple.

The little test app that was seeing duplicate entries now sees stable
f_pos values and no dupes after the patch.  So I think this works.

Thanks for fixing this.

- z

Re: [PATCH 1/2] ext4: fix readdir error in case inline_data+dir_index.

[Theodore Ts'o]

On Wed, Apr 10, 2013 at 11:37:07PM +0800, Tao Ma wrote:
> From: Tao Ma <boyu.mt@taobao.com>
> 
> Zach reported a problem that if inline data is enabled, we don't
> tell the difference between the offset of '.' and '..'. And a
> getdents will fail if the user only want to get '.' and what's worse,
> if there is a conversion happens when the user calls getdents
> many times, he/she may get the same entry twice.
> 
> In theroy, a dir block would also fail if it is converted to a
> hashed-index based dir since f_pos will become a hash value, not the
> real one, but it doesn't happen. And a deep investigation
> shows that we uses a hash based solution even for a normal dir if
> the dir_index feature is enabled.
> 
> So this patch just adds a new htree_inlinedir_to_tree for inline dir,
> and if we find that the hash index is supported, we will do like what
> we do for a dir block.
> 
> Reported-by: Zach Brown <zab@redhat.com>
> Signed-off-by: Tao Ma <boyu.mt@taobao.com>

Thanks, applied.

					- Ted

Re: [PATCH 2/2] ext4: fix readdir error in case inline_data+^dir_index.

[Theodore Ts'o]

On Wed, Apr 10, 2013 at 11:37:08PM +0800, Tao Ma wrote:
> From: Tao Ma <boyu.mt@taobao.com>
> 
> Zach reported a problem that if inline data is enabled, we don't
> tell the difference between the offset of '.' and '..'. And a
> getdents will fail if the user only want to get '.'. And what's
> worse, we may meet with duplicate dir entries as the offset
> for inline dir and non-inline one is quite different.
> 
> This patch just try to resolve this problem if dir_index
> is disabled. In this case, f_pos is the real offset with
> the dir block, so for inline dir, we just pretend as if
> we are a dir block and returns the offset like a norml
> dir block does.
> 
> Reported-by: Zach Brown <zab@redhat.com>
> Signed-off-by: Tao Ma <boyu.mt@taobao.com>

Thanks, applied.

					- Ted