xfs
[Top] [All Lists]

Re: [PATCH 1/3] Implement generic freeze feature

To: "Christoph Hellwig" <hch@xxxxxxxxxxxxx>
Subject: Re: [PATCH 1/3] Implement generic freeze feature
From: "Takashi Sato" <t-sato@xxxxxxxxxxxxx>
Date: Fri, 29 Aug 2008 18:37:28 +0900
Cc: "Andrew Morton" <akpm@xxxxxxxxxxxxxxxxxxxx>, <linux-fsdevel@xxxxxxxxxxxxxxx>, <dm-devel@xxxxxxxxxx>, <viro@xxxxxxxxxxxxxxxxxx>, <linux-ext4@xxxxxxxxxxxxxxx>, <xfs@xxxxxxxxxxx>, "Christoph Hellwig" <hch@xxxxxxxxxxxxx>, <axboe@xxxxxxxxx>, <mtk.manpages@xxxxxxxxxxxxxx>, <linux-kernel@xxxxxxxxxxxxxxx>
In-reply-to: <20080822181458.GA15469@infradead.org>
References: <20080818212819t-sato@mail.jp.nec.com> <20080822181458.GA15469@infradead.org>
Sender: xfs-bounce@xxxxxxxxxxx
Hi,

Christoph Hellwig wrote:
On Mon, Aug 18, 2008 at 09:28:19PM +0900, Takashi Sato wrote:
+    down(&bdev->bd_freeze_sem);
+    bdev->bd_freeze_count++;
+    if (bdev->bd_freeze_count > 1) {
+            sb = get_super(bdev);
+            drop_super(sb);
+            up(&bdev->bd_freeze_sem);
+            return sb;
+    }
+
     down(&bdev->bd_mount_sem);

Now you have a reference counter of freezes which actually is pretty sensible, but also needs some documentation. What I don't understand here at all is why you do the get_super/drop_super in the already frozen case.

Even if the filesystem has already been frozen, the superblock
should be returned. Because a caller should recognize the success of
freeze_bdev() and call thaw_bdev() to decrease the reference count. But I will remove drop_super() as it should be called in thaw_bdev().



Now that the freeze_count has replaced one of the uses of bd_mount_sem you should also replace the other use in the unmount path by simply checking for the freez_count and abort if it's set. To do so you'll need to hold the bd_mount_sem over the whole unmount operation to prevent new frezes from coming in.

In the original implementation,
unmount is protected by s_umount(semaphore),
not bd_mount_sem. So, unmount task waits for unfreeze.
I think this original behavior shouldn't be changed, so the existing s_umount lock is better.



As others noted it should be a mutex and not a semaphore.

As you said, we should use the mutex. I will replace it.


 /*
+ * ioctl_freeze - Freeze the filesystem.
+ *
+ * @filp:   target file
+ *
+ * Call freeze_bdev() to freeze the filesystem.
+ */
+static int ioctl_freeze(struct file *filp)

This is not quite kerneldcoc format, which would ne a /** as commnt start. But I don't think the comment is actually needed, it's a pretty obvious file scope function. (Same commnt also applies to ioctl_thaw)

I will remove these comments.


+    struct super_block *sb = filp->f_path.dentry->d_inode->i_sb;
+
+    if (!capable(CAP_SYS_ADMIN))
+            return -EPERM;
+
+    /* If filesystem doesn't support freeze feature, return. */
+    if (sb->s_op->write_super_lockfs == NULL)
+            return -EOPNOTSUPP;
+
+    /* If a regular file or a directory isn't specified, return. */
+    if (sb->s_bdev == NULL)
+            return -EINVAL;

I don't understand this commnt. What you are checking is that the filesystem has a non-NULL s_bdev, which implies a not blockdevice-backed filesystem.

I will fix the comment as : " If a blockdevice-backed filesystem isn't specified, return."

Cheers, Takashi


<Prev in Thread] Current Thread [Next in Thread>