2020-11-18 04:04:47

by Kaixu Xia

Subject: [PATCH v2] btrfs: return EAGAIN when receiving a pending signal in the defrag loops

From: Kaixu Xia <[email protected]>

The variable ret is overwritten by the following variable defrag_count.
Actually the code should return EAGAIN when receiving a pending signal
in the defrag loops.

Reported-by: Tosk Robot <[email protected]>
Signed-off-by: Kaixu Xia <[email protected]>
---
v2
-return EAGAIN instead of remove the EAGAIN error.

fs/btrfs/ioctl.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 69a384145dc6..6f13db6d30bd 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1519,7 +1519,7 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
if (btrfs_defrag_cancelled(fs_info)) {
btrfs_debug(fs_info, "defrag_file cancelled");
ret = -EAGAIN;
- break;
+ goto out_ra;
}

if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
--
2.20.0

2020-11-18 09:56:38

by Anand Jain

Subject: Re: [PATCH v2] btrfs: return EAGAIN when receiving a pending signal in the defrag loops

On 18/11/20 12:02 pm, [email protected] wrote:
> From: Kaixu Xia <[email protected]>
>
> The variable ret is overwritten by the following variable defrag_count.
> Actually the code should return EAGAIN when receiving a pending signal
> in the defrag loops.
>
> Reported-by: Tosk Robot <[email protected]>
> Signed-off-by: Kaixu Xia <[email protected]>
> ---
> v2
> -return EAGAIN instead of remove the EAGAIN error.

Sorry I might have missed in v1. Why was EAGAIN needed here?
Return of defrag_count rather makes sense to me as of now.

Thanks, Anand

>
> fs/btrfs/ioctl.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> index 69a384145dc6..6f13db6d30bd 100644
> --- a/fs/btrfs/ioctl.c
> +++ b/fs/btrfs/ioctl.c
> @@ -1519,7 +1519,7 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
> if (btrfs_defrag_cancelled(fs_info)) {
> btrfs_debug(fs_info, "defrag_file cancelled");
> ret = -EAGAIN;
> - break;
> + goto out_ra;
> }
>
> if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
>

2020-11-23 19:21:56

by David Sterba

Subject: Re: [PATCH v2] btrfs: return EAGAIN when receiving a pending signal in the defrag loops

On Wed, Nov 18, 2020 at 12:02:36PM +0800, [email protected] wrote:
> From: Kaixu Xia <[email protected]>
>
> The variable ret is overwritten by the following variable defrag_count.
> Actually the code should return EAGAIN when receiving a pending signal
> in the defrag loops.

This lacks explanation why is EAGAIN supposed to be the right return
value. This is about semantics of the FITRIM ioctl, changing that would
affect userspace applications.