ext4: undo ei->i_da_metadata_calc_len increment if we fail to claim space

RFC PATCH In review Brian Foster 2 participants 3 messages Apr 2, 2012 – Jul 23, 2012 ext4 file system

lore

Share & tools

lore

Brian Foster

author

2012-04-02 15:14 UTC #0 lore raw

RFC PATCH 2/2 ext4: undo ei->i_da_metadata_calc_len increment if we fail to claim space

From: Brian Foster <[email protected]>

If we call ext4_calc_metadata_amount() and then fail to claim the space, a subsequent successful request to a block covered by the same ext2 indirect block will set md_needed to 0, fail to update the associated counters and result in a delayed md block allocation without a reservation. Decrement i_da_metadata_calc_len on failure to ensure the next reservation sets md_needed correctly.

Signed-off-by Brian Foster <[email protected]>

---
 fs/ext4/inode.c |   18 ++++++++++++++++--
 1 files changed, 16 insertions(+), 2 deletions(-)

fs/ext4/inode.c +16 -2

index b58845c..7d2a3c0 100644
@@ -1138,13 +1138,14 @@ repeat:
 	 * We do still charge estimated metadata to the sb though;
 	 * we cannot afford to run out of free blocks.
 	 */
-	if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
+	ret = ext4_claim_free_clusters(sbi, md_needed + 1, 0);
+	if (ret) {
 		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
 		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
 			yield();
 			goto repeat;
 		}
-		return -ENOSPC;
+		goto error;
 	}
 	spin_lock(&ei->i_block_reservation_lock);
 	ei->i_reserved_data_blocks++;

@@ -1152,6 +1153,19 @@ repeat:
 	spin_unlock(&ei->i_block_reservation_lock);
 
 	return 0;       /* success */
+
+error:
+	/*
+	 * We've failed to reserve the space. Undo the effect
+	 * of ext4_calc_metadata_amount() to ensure the next
+	 * attempt correctly accounts for metadata.
+	 */
+	spin_lock(&ei->i_block_reservation_lock);
+	if (ei->i_da_metadata_calc_len)
+		ei->i_da_metadata_calc_len--;
+	spin_unlock(&ei->i_block_reservation_lock);
+
+	return ret;
 }
 
 static void ext4_da_release_space(struct inode *inode, int to_free)

Signature · 1 line

1.7.7.6

Theodore Ts'o

maintainer

2012-07-23 04:18 UTC #1 lore raw

Quoted 11 lines · Brian Foster

From: Brian Foster <[email protected]>
If we call ext4_calc_metadata_amount() and then fail to claim the space, a subsequent successful request to a block covered by the same ext2 indirect block will set md_needed to 0, fail to update the associated counters and result in a delayed md block allocation without a reservation. Decrement i_da_metadata_calc_len on failure to ensure the next reservation sets md_needed correctly.
Signed-off-by Brian Foster <[email protected]>

Hi Brian,

This patch was not quite right. i_data_metadata_calc_len is not the only side effect of ext4_calc_metadata(). We need to undo the side effects not only when exit with an error, but also if we yield and then retry the claim. Also, in the extent case we don't always modify i_data_metadata_calc_len, so just decrementing isn't necessarily going to do the right thing.

So this is the patch which I am testing....

- Ted

commit 19ec0f1fe139a642f688177ffd2f91a1c09f6bc0

Author: Theodore Ts'o <[email protected]>
Date:   Mon Jul 23 00:00:20 2012 -0400

ext4: undo ext4_calc_metadata_amount if we fail to claim space

The function ext4_calc_metadata_amount() has side effects, although it's not obvious from its function name. So if we fail to claim space, regardless of whether we retry to claim the space again, or return an error, we need to undo these side effects.

Otherwise we can end up incorrectly calculating the number of metadata blocks needed for the operation, which was responsible for an xfstests failure for test #271 when using an ext2 file system with delalloc enabled.

Reported-by Brian Foster <[email protected]>

Signed-off-by "Theodore Ts'o" <[email protected]>

Cc [email protected]

fs/ext4/inode.c +21 ~~-11~~

index 25f809d..89b59cb 100644
@@ -1182,6 +1182,17 @@ static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
 	struct ext4_inode_info *ei = EXT4_I(inode);
 	unsigned int md_needed;
 	int ret;
+	ext4_lblk_t save_last_lblock;
+	int save_len;
+
+	/*
+	 * We will charge metadata quota at writeout time; this saves
+	 * us from metadata over-estimation, though we may go over by
+	 * a small amount in the end.  Here we just reserve for data.
+	 */
+	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
+	if (ret)
+		return ret;
 
 	/*
 	 * recalculate the amount of metadata blocks to reserve

@@ -1190,32 +1201,31 @@ static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
 	 */
 repeat:
 	spin_lock(&ei->i_block_reservation_lock);
+	/*
+	 * ext4_calc_metadata_amount() has side effects, which we have
+	 * to be prepared undo if we fail to claim space.
+	 */
+	save_len = ei->i_da_metadata_calc_len;
+	save_last_lblock = ei->i_da_metadata_calc_last_lblock;
 	md_needed = EXT4_NUM_B2C(sbi,
 				 ext4_calc_metadata_amount(inode, lblock));
 	trace_ext4_da_reserve_space(inode, md_needed);
-	spin_unlock(&ei->i_block_reservation_lock);
 
 	/*
-	 * We will charge metadata quota at writeout time; this saves
-	 * us from metadata over-estimation, though we may go over by
-	 * a small amount in the end.  Here we just reserve for data.
-	 */
-	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
-	if (ret)
-		return ret;
-	/*
 	 * We do still charge estimated metadata to the sb though;
 	 * we cannot afford to run out of free blocks.
 	 */
 	if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
-		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
+		ei->i_da_metadata_calc_len = save_len;
+		ei->i_da_metadata_calc_last_lblock = save_last_lblock;
+		spin_unlock(&ei->i_block_reservation_lock);
 		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
 			yield();
 			goto repeat;
 		}
+		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
 		return -ENOSPC;
 	}
-	spin_lock(&ei->i_block_reservation_lock);
 	ei->i_reserved_data_blocks++;
 	ei->i_reserved_meta_blocks += md_needed;
 	spin_unlock(&ei->i_block_reservation_lock);

Brian Foster

author

2012-07-23 15:46 UTC #2 lore raw

Quoted 15 lines · Theodore Ts'o

Quoted 11 lines · Brian Foster
From: Brian Foster <[email protected]>
If we call ext4_calc_metadata_amount() and then fail to claim the space, a subsequent successful request to a block covered by the same ext2 indirect block will set md_needed to 0, fail to update the associated counters and result in a delayed md block allocation without a reservation. Decrement i_da_metadata_calc_len on failure to ensure the next reservation sets md_needed correctly.
Signed-off-by Brian Foster <[email protected]>
Hi Brian,

Quoted 9 lines

This patch was not quite right. i_data_metadata_calc_len is not the only side effect of ext4_calc_metadata(). We need to undo the side effects not only when exit with an error, but also if we yield and then retry the claim. Also, in the extent case we don't always modify i_data_metadata_calc_len, so just decrementing isn't necessarily going to do the right thing.
So this is the patch which I am testing....

Hi Ted,

Ok, your patch makes sense. Thanks for the explanation. FWIW, I ran it through my reproducer a couple times and it passes.

Brian

Quoted 88 lines