From: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Subject: [PATCH] ext4: Fix circular locking dependency with migrate and rm.
Date: Wed,  6 Feb 2008 16:30:04 +0530
Message-ID: <1202295604-9948-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com>
Cc: linux-ext4@vger.kernel.org,
	"Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
To: tytso@mit.edu, cmm@us.ibm.com, jack@suse.cz
Return-path: <linux-ext4-owner@vger.kernel.org>
Received: from e28smtp02.in.ibm.com ([59.145.155.2]:60749 "EHLO
	e28esmtp02.in.ibm.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org
	with ESMTP id S1759207AbYBFLAK (ORCPT
	<rfc822;linux-ext4@vger.kernel.org>); Wed, 6 Feb 2008 06:00:10 -0500
Received: from d28relay04.in.ibm.com (d28relay04.in.ibm.com [9.184.220.61])
	by e28esmtp02.in.ibm.com (8.13.1/8.13.1) with ESMTP id m16B03Wk024750
	for <linux-ext4@vger.kernel.org>; Wed, 6 Feb 2008 16:30:03 +0530
Received: from d28av05.in.ibm.com (d28av05.in.ibm.com [9.184.220.67])
	by d28relay04.in.ibm.com (8.13.8/8.13.8/NCO v8.7) with ESMTP id m16B03g2417938
	for <linux-ext4@vger.kernel.org>; Wed, 6 Feb 2008 16:30:03 +0530
Received: from d28av05.in.ibm.com (loopback [127.0.0.1])
	by d28av05.in.ibm.com (8.13.1/8.13.3) with ESMTP id m16B027i006337
	for <linux-ext4@vger.kernel.org>; Wed, 6 Feb 2008 11:00:03 GMT
Sender: linux-ext4-owner@vger.kernel.org
List-ID: <linux-ext4.vger.kernel.org>

We now take inode->i_mutex lock to prevent any update of the inode i_data
field. Before we switch the inode format we take i_data_sem to prevent
parallel read.

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.24-rc8 #6
-------------------------------------------------------
rm/2401 is trying to acquire lock:
 (&ei->i_data_sem){----}, at: [<c01dca58>] ext4_get_blocks_wrap+0x21/0x108

but task is already holding lock:
 (jbd2_handle){--..}, at: [<c01fc4a7>] jbd2_journal_start+0xd2/0xff

which lock already depends on the new lock.


the existing dependency chain (in reverse order) is:

-> #1 (jbd2_handle){--..}:
       [<c0143a5c>] __lock_acquire+0xa31/0xc1a
       [<c0143cbf>] lock_acquire+0x7a/0x94
       [<c01fc4ca>] jbd2_journal_start+0xf5/0xff
       [<c01e3539>] ext4_journal_start_sb+0x48/0x4a
       [<c01eb980>] ext4_ext_migrate+0x7d/0x535
       [<c01df328>] ext4_ioctl+0x528/0x56c
       [<c0177700>] do_ioctl+0x50/0x67
       [<c017794e>] vfs_ioctl+0x237/0x24a
       [<c0177992>] sys_ioctl+0x31/0x4b
       [<c0104f8a>] sysenter_past_esp+0x5f/0xa5
       [<ffffffff>] 0xffffffff

-> #0 (&ei->i_data_sem){----}:
       [<c014394c>] __lock_acquire+0x921/0xc1a
       [<c0143cbf>] lock_acquire+0x7a/0x94
       [<c044f247>] down_read+0x42/0x79
       [<c01dca58>] ext4_get_blocks_wrap+0x21/0x108
       [<c01dcba1>] ext4_getblk+0x62/0x1c4
       [<c01e0de9>] ext4_find_entry+0x350/0x5b7
       [<c01e200c>] ext4_unlink+0x6e/0x1a4
       [<c017449e>] vfs_unlink+0x49/0x89
       [<c0175f02>] do_unlinkat+0x96/0x12c
       [<c0175fa8>] sys_unlink+0x10/0x12
       [<c0104f8a>] sysenter_past_esp+0x5f/0xa5
       [<ffffffff>] 0xffffffff

other info that might help us debug this:

3 locks held by rm/2401:
 #0:  (&type->i_mutex_dir_key#5/1){--..}, at: [<c0175eca>] do_unlinkat+0x5e/0x12c
 #1:  (&sb->s_type->i_mutex_key#8){--..}, at: [<c017448b>] vfs_unlink+0x36/0x89
 #2:  (jbd2_handle){--..}, at: [<c01fc4a7>] jbd2_journal_start+0xd2/0xff

stack backtrace:
Pid: 2401, comm: rm Not tainted 2.6.24-rc8 #6
 [<c0106017>] show_trace_log_lvl+0x1a/0x2f
 [<c0106893>] show_trace+0x12/0x14
 [<c0106b89>] dump_stack+0x6c/0x72
 [<c0141b26>] print_circular_bug_tail+0x5f/0x68
 [<c014394c>] __lock_acquire+0x921/0xc1a
 [<c0143cbf>] lock_acquire+0x7a/0x94
 [<c044f247>] down_read+0x42/0x79
 [<c01dca58>] ext4_get_blocks_wrap+0x21/0x108
 [<c01dcba1>] ext4_getblk+0x62/0x1c4
 [<c01e0de9>] ext4_find_entry+0x350/0x5b7
 [<c01e200c>] ext4_unlink+0x6e/0x1a4
 [<c017449e>] vfs_unlink+0x49/0x89
 [<c0175f02>] do_unlinkat+0x96/0x12c
 [<c0175fa8>] sys_unlink+0x10/0x12
 [<c0104f8a>] sysenter_past_esp+0x5f/0xa5

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
---
 fs/ext4/migrate.c |  117 +++++++++++++++++++++++++++++++++-------------------
 1 files changed, 74 insertions(+), 43 deletions(-)

diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index 9ee1f7c..8c6c685 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -61,10 +61,9 @@ static int finish_range(handle_t *handle, struct inode *inode,
 		retval = ext4_journal_restart(handle, needed);
 		if (retval)
 			goto err_out;
-	}
-	if (needed) {
+	} else if (needed) {
 		retval = ext4_journal_extend(handle, needed);
-		if (retval != 0) {
+		if (retval) {
 			/*
 			 * IF not able to extend the journal restart the journal
 			 */
@@ -220,6 +219,26 @@ static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 
 }
 
+static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
+{
+	int retval = 0, needed;
+
+	if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
+		return 0;
+	/*
+	 * We are freeing a blocks. During this we touch
+	 * superblock, group descriptor and block bitmap.
+	 * So allocate a credit of 3. We may update
+	 * quota (user and group).
+	 */
+	needed = 3 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
+
+	if (ext4_journal_extend(handle, needed) != 0)
+		retval = ext4_journal_restart(handle, needed);
+
+	return retval;
+}
+
 static int free_dind_blocks(handle_t *handle,
 				struct inode *inode, __le32 i_data)
 {
@@ -234,11 +253,14 @@ static int free_dind_blocks(handle_t *handle,
 
 	tmp_idata = (__le32 *)bh->b_data;
 	for (i = 0; i < max_entries; i++) {
-		if (tmp_idata[i])
+		if (tmp_idata[i]) {
+			extend_credit_for_blkdel(handle, inode);
 			ext4_free_blocks(handle, inode,
 					le32_to_cpu(tmp_idata[i]), 1, 1);
+		}
 	}
 	put_bh(bh);
+	extend_credit_for_blkdel(handle, inode);
 	ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 	return 0;
 }
@@ -267,29 +289,32 @@ static int free_tind_blocks(handle_t *handle,
 		}
 	}
 	put_bh(bh);
+	extend_credit_for_blkdel(handle, inode);
 	ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 	return 0;
 }
 
-static int free_ind_block(handle_t *handle, struct inode *inode)
+static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 {
 	int retval;
-	struct ext4_inode_info *ei = EXT4_I(inode);
 
-	if (ei->i_data[EXT4_IND_BLOCK])
+	/* ei->i_data[EXT4_IND_BLOCK] */
+	if (i_data[0]) {
+		extend_credit_for_blkdel(handle, inode);
 		ext4_free_blocks(handle, inode,
-				le32_to_cpu(ei->i_data[EXT4_IND_BLOCK]), 1, 1);
+				le32_to_cpu(i_data[0]), 1, 1);
+	}
 
-	if (ei->i_data[EXT4_DIND_BLOCK]) {
-		retval = free_dind_blocks(handle, inode,
-						ei->i_data[EXT4_DIND_BLOCK]);
+	/* ei->i_data[EXT4_DIND_BLOCK] */
+	if (i_data[1]) {
+		retval = free_dind_blocks(handle, inode, i_data[1]);
 		if (retval)
 			return retval;
 	}
 
-	if (ei->i_data[EXT4_TIND_BLOCK]) {
-		retval = free_tind_blocks(handle, inode,
-						ei->i_data[EXT4_TIND_BLOCK]);
+	/* ei->i_data[EXT4_TIND_BLOCK] */
+	if (i_data[2]) {
+		retval = free_tind_blocks(handle, inode, i_data[2]);
 		if (retval)
 			return retval;
 	}
@@ -297,15 +322,13 @@ static int free_ind_block(handle_t *handle, struct inode *inode)
 }
 
 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
-				struct inode *tmp_inode, int retval)
+				struct inode *tmp_inode)
 {
+	int retval;
+	__le32	i_data[3];
 	struct ext4_inode_info *ei = EXT4_I(inode);
 	struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 
-	retval = free_ind_block(handle, inode);
-	if (retval)
-		goto err_out;
-
 	/*
 	 * One credit accounted for writing the
 	 * i_data field of the original inode
@@ -317,6 +340,11 @@ static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 			goto err_out;
 	}
 
+	i_data[0] = ei->i_data[EXT4_IND_BLOCK];
+	i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
+	i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
+
+	down_write(&EXT4_I(inode)->i_data_sem);
 	/*
 	 * We have the extent map build with the tmp inode.
 	 * Now copy the i_data across
@@ -336,8 +364,15 @@ static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 	spin_lock(&inode->i_lock);
 	inode->i_blocks += tmp_inode->i_blocks;
 	spin_unlock(&inode->i_lock);
+	up_write(&EXT4_I(inode)->i_data_sem);
 
+	/*
+	 * We mark the inode dirty after, because we decrement the
+	 * i_blocks when freeing the indirect meta-data blocks
+	 */
+	retval = free_ind_block(handle, inode, i_data);
 	ext4_mark_inode_dirty(handle, inode);
+
 err_out:
 	return retval;
 }
@@ -365,6 +400,7 @@ static int free_ext_idx(handle_t *handle, struct inode *inode,
 		}
 	}
 	put_bh(bh);
+	extend_credit_for_blkdel(handle, inode);
 	ext4_free_blocks(handle, inode, block, 1, 1);
 	return retval;
 }
@@ -420,7 +456,6 @@ int ext4_ext_migrate(struct inode *inode, struct file *filp,
 		 */
 		return retval;
 
-	down_write(&EXT4_I(inode)->i_data_sem);
 	handle = ext4_journal_start(inode,
 					EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
@@ -454,13 +489,6 @@ int ext4_ext_migrate(struct inode *inode, struct file *filp,
 	ext4_orphan_add(handle, tmp_inode);
 	ext4_journal_stop(handle);
 
-	ei = EXT4_I(inode);
-	i_data = ei->i_data;
-	memset(&lb, 0, sizeof(lb));
-
-	/* 32 bit block address 4 bytes */
-	max_entries = inode->i_sb->s_blocksize >> 2;
-
 	/*
 	 * start with one credit accounted for
 	 * superblock modification.
@@ -469,7 +497,20 @@ int ext4_ext_migrate(struct inode *inode, struct file *filp,
 	 * trascation that created the inode. Later as and
 	 * when we add extents we extent the journal
 	 */
+	/*
+	 * inode_mutex prevent write and truncate on the file. Read still goes
+	 * through. We take i_data_sem in ext4_ext_swap_inode_data before we
+	 * switch the inode format to prevent read.
+	 */
+	mutex_lock(&(inode->i_mutex));
 	handle = ext4_journal_start(inode, 1);
+
+	ei = EXT4_I(inode);
+	i_data = ei->i_data;
+	memset(&lb, 0, sizeof(lb));
+
+	/* 32 bit block address 4 bytes */
+	max_entries = inode->i_sb->s_blocksize >> 2;
 	for (i = 0; i < EXT4_NDIR_BLOCKS; i++, blk_count++) {
 		if (i_data[i]) {
 			retval = update_extent_range(handle, tmp_inode,
@@ -507,19 +548,6 @@ int ext4_ext_migrate(struct inode *inode, struct file *filp,
 	 */
 	retval = finish_range(handle, tmp_inode, &lb);
 err_out:
-	/*
-	 * We are either freeing extent information or indirect
-	 * blocks. During this we touch superblock, group descriptor
-	 * and block bitmap. Later we mark the tmp_inode dirty
-	 * via ext4_ext_tree_init. So allocate a credit of 4
-	 * We may update quota (user and group).
-	 *
-	 * FIXME!! we may be touching bitmaps in different block groups.
-	 */
-	if (ext4_journal_extend(handle,
-			4 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb)) != 0)
-		ext4_journal_restart(handle,
-				4 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb));
 	if (retval)
 		/*
 		 * Failure case delete the extent information with the
@@ -528,7 +556,11 @@ err_out:
 		free_ext_block(handle, tmp_inode);
 	else
 		retval = ext4_ext_swap_inode_data(handle, inode,
-							tmp_inode, retval);
+							tmp_inode);
+
+	/* We mark the tmp_inode dirty via ext4_ext_tree_init. */
+	if (ext4_journal_extend(handle, 1) != 0)
+		ext4_journal_restart(handle, 1);
 
 	/*
 	 * Mark the tmp_inode as of size zero
@@ -556,8 +588,7 @@ err_out:
 	tmp_inode->i_nlink = 0;
 
 	ext4_journal_stop(handle);
-
-	up_write(&EXT4_I(inode)->i_data_sem);
+	mutex_unlock(&(inode->i_mutex));
 
 	if (tmp_inode)
 		iput(tmp_inode);
-- 
1.5.4.23.gef5b9-dirty