[PATCH] Btrfs: fix truncate down when no_holes feature is enabled

[PATCH] Btrfs: fix truncate down when no_holes feature is enabled

[Liu Bo]

For such a file mapping,

[0-4k][hole][8k-12k]

In NO_HOLES mode, we don't have the [hole] extent any more.
Commit c1aa45759e90 ("Btrfs: fix shrinking truncate when the no_holes feature is enabled")
 fixed disk isize not being updated in NO_HOLES mode when data is not flushed.

However, even if data has been flushed, we can still have trouble
in updating disk isize since we updated disk isize to 'start' of
the last evicted extent.

Also add a ASSERT for better catching (for developers only).

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
---
 fs/btrfs/inode.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 2b790bd..603dd492 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -4483,8 +4483,19 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 		if (found_type > min_type) {
 			del_item = 1;
 		} else {
-			if (item_end < new_size)
+			if (item_end < new_size) {
+				/*
+				 * With NO_HOLES mode, for the following mapping
+				 *
+				 * [0-4k][hole][8k-12k]
+				 *
+				 * if truncating isize down to 6k, it ends up
+				 * isize being 8k.
+				 */
+				if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
+					last_size = new_size;
 				break;
+			}
 			if (found_key.offset >= new_size)
 				del_item = 1;
 			else
@@ -4666,6 +4677,9 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 			btrfs_abort_transaction(trans, ret);
 	}
 error:
+	if (err == 0)
+		ASSERT(last_size == new_size);
+
 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
 		btrfs_ordered_update_i_size(inode, last_size, NULL);
 
-- 
2.5.5

Re: [PATCH] Btrfs: fix truncate down when no_holes feature is enabled

[Chris Mason]

On 11/11/2016 05:27 PM, Liu Bo wrote:
> For such a file mapping,
>
> [0-4k][hole][8k-12k]
>
> In NO_HOLES mode, we don't have the [hole] extent any more.
> Commit c1aa45759e90 ("Btrfs: fix shrinking truncate when the no_holes feature is enabled")
>  fixed disk isize not being updated in NO_HOLES mode when data is not flushed.
>
> However, even if data has been flushed, we can still have trouble
> in updating disk isize since we updated disk isize to 'start' of
> the last evicted extent.
>
> Also add a ASSERT for better catching (for developers only).
>
> Signed-off-by: Liu Bo <bo.li.liu@oracle.com>

Thanks Liu!

Reviewed-by: Chris Mason <clm@fb.com>

Re: [PATCH] Btrfs: fix truncate down when no_holes feature is enabled

[Liu Bo]

On Tue, Nov 22, 2016 at 02:13:21PM -0500, Chris Mason wrote:
> On 11/11/2016 05:27 PM, Liu Bo wrote:
> > For such a file mapping,
> > 
> > [0-4k][hole][8k-12k]
> > 
> > In NO_HOLES mode, we don't have the [hole] extent any more.
> > Commit c1aa45759e90 ("Btrfs: fix shrinking truncate when the no_holes feature is enabled")
> >  fixed disk isize not being updated in NO_HOLES mode when data is not flushed.
> > 
> > However, even if data has been flushed, we can still have trouble
> > in updating disk isize since we updated disk isize to 'start' of
> > the last evicted extent.
> > 
> > Also add a ASSERT for better catching (for developers only).
> > 
> > Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
> 
> Thanks Liu!
> 
> Reviewed-by: Chris Mason <clm@fb.com>

The added ASSERT() can cause generic/068 to crash because I didn't
filter out the 'log tree' case, I'm testing a V2 patch.

Thanks,

-liubo

Re: [PATCH] Btrfs: fix truncate down when no_holes feature is enabled

[David Sterba]

On Tue, Nov 22, 2016 at 12:08:47PM -0800, Liu Bo wrote:
> On Tue, Nov 22, 2016 at 02:13:21PM -0500, Chris Mason wrote:
> > On 11/11/2016 05:27 PM, Liu Bo wrote:
> > > For such a file mapping,
> > > 
> > > [0-4k][hole][8k-12k]
> > > 
> > > In NO_HOLES mode, we don't have the [hole] extent any more.
> > > Commit c1aa45759e90 ("Btrfs: fix shrinking truncate when the no_holes feature is enabled")
> > >  fixed disk isize not being updated in NO_HOLES mode when data is not flushed.
> > > 
> > > However, even if data has been flushed, we can still have trouble
> > > in updating disk isize since we updated disk isize to 'start' of
> > > the last evicted extent.
> > > 
> > > Also add a ASSERT for better catching (for developers only).
> > > 
> > > Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
> > 
> > Thanks Liu!
> > 
> > Reviewed-by: Chris Mason <clm@fb.com>
> 
> The added ASSERT() can cause generic/068 to crash because I didn't
> filter out the 'log tree' case, I'm testing a V2 patch.

I've hit the assert with btrfs/003, once and not deterministically in a VM.