Skip to contentSkip to conversation

- r-o-bind-mount-clean-up-ocfs2-nlink-handling.patch removed from -mm tree

DISCUSSION Mark Fasheh 2 participants 4 messages Sep 13, 2006 1 message not in the archive OCFS2

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 9 lines · On Tue, Sep 12, 2006 at 10:06:30PM -0700, [email protected] wrote:

The patch titled

r/o bind mounts: clean up OCFS2 nlink handling

has been removed from the -mm tree. Its filename is

r-o-bind-mount-clean-up-ocfs2-nlink-handling.patch

This patch was dropped because git-ocfs2 changes broke it. New patch, please.

Yep, that was very likely due to my dentry vote removal changes.

Dave, how's this one look? I guess I'll leave the same description message below...

Subject: r/o bind mounts: clean up OCFS2 nlink handling
From: Dave Hansen <[email protected]>

OCFS2 does some operations on i_nlink, then reverts them if some of its operations fail to complete. This does not fit in well with the drop_nlink() logic where we expect i_nlink to stay at zero once it gets there.

So, delay all of the nlink operations until we're sure that the operations have completed. Also, introduce a small helper to check whether an inode has proper "unlinkable" i_nlink counts no matter whether it is a directory or regular inode.

This patch is broken out from the others because it does contain some logical changes.

Cleanups to apply against current git-ocfs2 changes by Mark Fasheh.

Cc Dave Hansen <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Al Viro <[email protected]>
Cc Andrew Morton <[email protected]>
fs/ocfs2/namei.c +23 -18
index 4cfc061..4f74cd9 100644
@@ -797,11 +797,23 @@ static int ocfs2_remote_dentry_delete(st
 	return ret;
 }
 
+static inline int inode_is_unlinkable(struct inode *inode)
+{
+	if (S_ISDIR(inode->i_mode)) {
+		if (inode->i_nlink == 2)
+			return 1;
+		return 0;
+	}
+
+	if (inode->i_nlink == 1)
+		return 1;
+	return 0;
+}
+
 static int ocfs2_unlink(struct inode *dir,
 			struct dentry *dentry)
 {
 	int status;
-	unsigned int saved_nlink = 0;
 	struct inode *inode = dentry->d_inode;
 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 	u64 blkno;
@@ -813,6 +825,7 @@ static int ocfs2_unlink(struct inode *di
 	struct buffer_head *dirent_bh = NULL;
 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 	struct buffer_head *orphan_entry_bh = NULL;
+	unsigned int future_nlink;
 
 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 		   dentry->d_name.len, dentry->d_name.name);
@@ -876,15 +889,10 @@ static int ocfs2_unlink(struct inode *di
 		}
 	}
 
-	/* There are still a few steps left until we can consider the
-	 * unlink to have succeeded. Save off nlink here before
-	 * modification so we can set it back in case we hit an issue
-	 * before commit. */
-	saved_nlink = inode->i_nlink;
-	if (S_ISDIR(inode->i_mode))
-		inode->i_nlink = 0;
+	if (S_ISDIR(inode->i_mode) && (inode->i_nlink == 2))
+		future_nlink = 0;
 	else
-		inode->i_nlink--;
+		future_nlink = inode->i_nlink - 1;
 
 	status = ocfs2_remote_dentry_delete(dentry);
 	if (status < 0) {
@@ -894,7 +902,7 @@ static int ocfs2_unlink(struct inode *di
 		goto leave;
 	}
 
-	if (!inode->i_nlink) {
+	if (inode_is_unlinkable(inode)) {
 		status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 						  orphan_name,
 						  &orphan_entry_bh);
@@ -921,7 +929,7 @@ static int ocfs2_unlink(struct inode *di
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
-	if (!inode->i_nlink) {
+	if (inode_is_unlinkable(inode)) {
 		status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 					  orphan_entry_bh);
 		if (status < 0) {
@@ -937,10 +945,10 @@ static int ocfs2_unlink(struct inode *di
 		goto leave;
 	}
 
-	/* We can set nlink on the dinode now. clear the saved version
-	 * so that it doesn't get set later. */
+	if (S_ISDIR(inode->i_mode))
+		drop_nlink(inode);
+	drop_nlink(inode);
 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
-	saved_nlink = 0;
 
 	status = ocfs2_journal_dirty(handle, fe_bh);
 	if (status < 0) {
@@ -949,7 +957,7 @@ static int ocfs2_unlink(struct inode *di
 	}
 
 	if (S_ISDIR(inode->i_mode)) {
-		dir->i_nlink--;
+		drop_nlink(dir);
 		status = ocfs2_mark_inode_dirty(handle, dir,
 						parent_node_bh);
 		if (status < 0) {
@@ -959,9 +967,6 @@ static int ocfs2_unlink(struct inode *di
 	}
 
 leave:
-	if (status < 0 && saved_nlink)
-		inode->i_nlink = saved_nlink;
-
 	if (handle)
 		ocfs2_commit_trans(handle);
 
#2 lore raw
Quoted 7 lines · Mark Fasheh
Quoted 6 lines · On Tue, Sep 12, 2006 at 10:06:30PM -0700, [email protected] wrote:

The patch titled

r/o bind mounts: clean up OCFS2 nlink handling

has been removed from the -mm tree. Its filename is

Quoted 7 lines
Quoted 3 lines

r-o-bind-mount-clean-up-ocfs2-nlink-handling.patch

This patch was dropped because git-ocfs2 changes broke it. New patch, please.

Yep, that was very likely due to my dentry vote removal changes.

Dave, how's this one look? I guess I'll leave the same description message below...

I was _just_ fighting with your git tree to see what was conflicting! You have impeccable timing.

Quoted 22 lines

static int ocfs2_unlink(struct inode *dir,

fs/ocfs2/namei.c +1 -3
 			struct dentry *dentry)
 {
 	int status;
-	unsigned int saved_nlink = 0;
 	struct inode *inode = dentry->d_inode;
 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 	u64 blkno;
@@ -813,6 +825,7 @@ static int ocfs2_unlink(struct inode *di
 	struct buffer_head *dirent_bh = NULL;
 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 	struct buffer_head *orphan_entry_bh = NULL;
+	unsigned int future_nlink;
 
 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 		   dentry->d_name.len, dentry->d_name.name);
@@ -876,15 +889,10 @@ static int ocfs2_unlink(struct inode *di
 		}
 	}
 
-	/* There are still a few steps left until we can consider the
-	 * unlink to have succeeded. Save off nlink here before

fs/ocfs2/namei.c static int ocfs2_unlink(struct inode *di · line 895

Quoted 10 lines
fs/ocfs2/namei.c +3 -6
-	 * modification so we can set it back in case we hit an issue
-	 * before commit. */
-	saved_nlink = inode->i_nlink;
-	if (S_ISDIR(inode->i_mode))
-		inode->i_nlink = 0;
+	if (S_ISDIR(inode->i_mode) && (inode->i_nlink == 2))
+		future_nlink = 0;
 	else
-		inode->i_nlink--;
+		future_nlink = inode->i_nlink - 1;

Now that the vote call is gone, I don't think we even use future_nlink. Can we just kill this entire section?

-- Dave

#3 lore raw
Quoted 2 lines · Dave Hansen

I was _just_ fighting with your git tree to see what was conflicting! You have impeccable timing.

Heh, glad that the timing worked out :)

Quoted 7 lines
Quoted 7 lines
(quoted diff) +0 -7
-	/* There are still a few steps left until we can consider the
-	 * unlink to have succeeded. Save off nlink here before
-	 * modification so we can set it back in case we hit an issue
-	 * before commit. */
-	saved_nlink = inode->i_nlink;
-	if (S_ISDIR(inode->i_mode))
-		inode->i_nlink = 0;
Quoted 8 lines

Review

Quoted 5 lines
(quoted diff) +3 -1
+	if (S_ISDIR(inode->i_mode) && (inode->i_nlink == 2))
+		future_nlink = 0;
 	else
-		inode->i_nlink--;
+		future_nlink = inode->i_nlink - 1;

Now that the vote call is gone, I don't think we even use future_nlink. Can we just kill this entire section?

Yeah, good catch. Lets try this one again...

Subject: r/o bind mounts: clean up OCFS2 nlink handling
From: Dave Hansen <[email protected]>

OCFS2 does some operations on i_nlink, then reverts them if some of its operations fail to complete. This does not fit in well with the drop_nlink() logic where we expect i_nlink to stay at zero once it gets there.

So, delay all of the nlink operations until we're sure that the operations have completed. Also, introduce a small helper to check whether an inode has proper "unlinkable" i_nlink counts no matter whether it is a directory or regular inode.

This patch is broken out from the others because it does contain some logical changes.

Cc Dave Hansen <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Al Viro <[email protected]>
Cc Andrew Morton <[email protected]>
fs/ocfs2/namei.c +19 -20
index 4cfc061..1c3d296 100644
@@ -797,11 +797,23 @@ static int ocfs2_remote_dentry_delete(st
 	return ret;
 }
 
+static inline int inode_is_unlinkable(struct inode *inode)
+{
+	if (S_ISDIR(inode->i_mode)) {
+		if (inode->i_nlink == 2)
+			return 1;
+		return 0;
+	}
+
+	if (inode->i_nlink == 1)
+		return 1;
+	return 0;
+}
+
 static int ocfs2_unlink(struct inode *dir,
 			struct dentry *dentry)
 {
 	int status;
-	unsigned int saved_nlink = 0;
 	struct inode *inode = dentry->d_inode;
 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 	u64 blkno;
@@ -876,16 +888,6 @@ static int ocfs2_unlink(struct inode *di
 		}
 	}
 
-	/* There are still a few steps left until we can consider the
-	 * unlink to have succeeded. Save off nlink here before
-	 * modification so we can set it back in case we hit an issue
-	 * before commit. */
-	saved_nlink = inode->i_nlink;
-	if (S_ISDIR(inode->i_mode))
-		inode->i_nlink = 0;
-	else
-		inode->i_nlink--;
-
 	status = ocfs2_remote_dentry_delete(dentry);
 	if (status < 0) {
 		/* This vote should succeed under all normal
@@ -894,7 +896,7 @@ static int ocfs2_unlink(struct inode *di
 		goto leave;
 	}
 
-	if (!inode->i_nlink) {
+	if (inode_is_unlinkable(inode)) {
 		status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 						  orphan_name,
 						  &orphan_entry_bh);
@@ -921,7 +923,7 @@ static int ocfs2_unlink(struct inode *di
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
-	if (!inode->i_nlink) {
+	if (inode_is_unlinkable(inode)) {
 		status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 					  orphan_entry_bh);
 		if (status < 0) {
@@ -937,10 +939,10 @@ static int ocfs2_unlink(struct inode *di
 		goto leave;
 	}
 
-	/* We can set nlink on the dinode now. clear the saved version
-	 * so that it doesn't get set later. */
+	if (S_ISDIR(inode->i_mode))
+		drop_nlink(inode);
+	drop_nlink(inode);
 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
-	saved_nlink = 0;
 
 	status = ocfs2_journal_dirty(handle, fe_bh);
 	if (status < 0) {
@@ -949,7 +951,7 @@ static int ocfs2_unlink(struct inode *di
 	}
 
 	if (S_ISDIR(inode->i_mode)) {
-		dir->i_nlink--;
+		drop_nlink(dir);
 		status = ocfs2_mark_inode_dirty(handle, dir,
 						parent_node_bh);
 		if (status < 0) {
@@ -959,9 +961,6 @@ static int ocfs2_unlink(struct inode *di
 	}
 
 leave:
-	if (status < 0 && saved_nlink)
-		inode->i_nlink = saved_nlink;
-
 	if (handle)
 		ocfs2_commit_trans(handle);
 
#4 lore raw
Quoted 3 lines · Mark Fasheh
Quoted 2 lines

Now that the vote call is gone, I don't think we even use future_nlink. Can we just kill this entire section?

Yeah, good catch. Lets try this one again...

Looks good to me. Thanks a bunch for updating this one.

-- Dave