[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <155620450825.4720.10730742298364202166.stgit@warthog.procyon.org.uk>
Date: Thu, 25 Apr 2019 16:01:48 +0100
From: David Howells <dhowells@...hat.com>
To: viro@...iv.linux.org.uk
Cc: Jan Harkes <jaharkes@...cmu.edu>, coda@...cmu.edu,
codalist@...a.cs.cmu.edu, dhowells@...hat.com,
linux-afs@...ts.infradead.org, linux-ext4@...r.kernel.org,
linux-ntfs-dev@...ts.sourceforge.net,
linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: [PATCH 1/6] vfs,
coda: Fix the lack of locking in FID replacement inode rehashing
[ver #2]
When coda attempts to recover from disconnected operation, there exists the
possibility of an file created during disconnected operation not being
creatable on the server with the same file identifier (FID). In such a
case, coda_replace_fid() has to rehash the inode and move it between chains
- but, as the comment notes, this really needs some locking.
Fix this by moving the core part of the code to fs/inode.c and providing it
with a set() function akin to iget5(). We can then take the inode cache
lock whilst performing the move.
Reported-by: Al Viro <viro@...iv.linux.org.uk>
Signed-off-by: David Howells <dhowells@...hat.com>
cc: Jan Harkes <jaharkes@...cmu.edu>
cc: coda@...cmu.edu
cc: codalist@...a.cs.cmu.edu
---
fs/coda/cnode.c | 16 ++++++++++------
fs/inode.c | 23 +++++++++++++++++++++++
include/linux/fs.h | 3 +++
3 files changed, 36 insertions(+), 6 deletions(-)
diff --git a/fs/coda/cnode.c b/fs/coda/cnode.c
index 845b5a66952a..a7e9a1b4ba2f 100644
--- a/fs/coda/cnode.c
+++ b/fs/coda/cnode.c
@@ -107,6 +107,15 @@ struct inode *coda_cnode_make(struct CodaFid *fid, struct super_block *sb)
}
+static void coda_reset_inode(struct inode *inode, unsigned long hash, void *data)
+{
+ struct CodaFid *fid = (struct CodaFid *)data;
+ struct coda_inode_info *cii = ITOC(inode);
+
+ cii->c_fid = *fid;
+ inode->i_ino = hash;
+}
+
/* Although we treat Coda file identifiers as immutable, there is one
* special case for files created during a disconnection where they may
* not be globally unique. When an identifier collision is detected we
@@ -123,12 +132,7 @@ void coda_replace_fid(struct inode *inode, struct CodaFid *oldfid,
BUG_ON(!coda_fideq(&cii->c_fid, oldfid));
- /* replace fid and rehash inode */
- /* XXX we probably need to hold some lock here! */
- remove_inode_hash(inode);
- cii->c_fid = *newfid;
- inode->i_ino = hash;
- __insert_inode_hash(inode, hash);
+ rehash_inode(inode, hash, coda_reset_inode, newfid);
}
/* convert a fid to an inode. */
diff --git a/fs/inode.c b/fs/inode.c
index e9d97add2b36..00bb48ca3642 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -501,6 +501,29 @@ void __remove_inode_hash(struct inode *inode)
}
EXPORT_SYMBOL(__remove_inode_hash);
+/**
+ * rehash_inode - Relabel and rehash an inode
+ * @inode: Inode to rehash
+ * @hashval: New hash value (usually inode number) to get
+ * @reset: Callback used to relabel the inode
+ * @data: Opaque data pointer to pass to @reset
+ */
+void rehash_inode(struct inode *inode, unsigned long hashval,
+ void (*reset)(struct inode *inode, unsigned long hashval, void *data),
+ void *data)
+{
+ struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
+
+ spin_lock(&inode_hash_lock);
+ spin_lock(&inode->i_lock);
+ hlist_del_init(&inode->i_hash);
+ reset(inode, hashval, data);
+ hlist_add_head(&inode->i_hash, b);
+ spin_unlock(&inode->i_lock);
+ spin_unlock(&inode_hash_lock);
+}
+EXPORT_SYMBOL(rehash_inode);
+
void clear_inode(struct inode *inode)
{
/*
diff --git a/include/linux/fs.h b/include/linux/fs.h
index dd28e7679089..6442ff08b28c 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -3019,6 +3019,9 @@ static inline void remove_inode_hash(struct inode *inode)
__remove_inode_hash(inode);
}
+extern void rehash_inode(struct inode *, unsigned long,
+ void (*reset)(struct inode *, unsigned long, void *), void *);
+
extern void inode_sb_list_add(struct inode *inode);
#ifdef CONFIG_BLOCK
Powered by blists - more mailing lists