Look Ma' what I found on my harddisk ...

 o New faster syscalls for 2.1.x, too
 o Upgrade to 2.1.89.

Don't try to run this.  It's flaky as hell.  But feel free to debug ...

1 files changed, 215 insertions, 86 deletions

diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index eb56950eb..8300fee67 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -32,13 +32,16 @@
 #include <asm/system.h>
 #include <asm/uaccess.h>
 
+#define CONFIG_NFS_SNAPSHOT 1
 #define NFSDBG_FACILITY		NFSDBG_VFS
 #define NFS_PARANOIA 1
 
+static struct inode * __nfs_fhget(struct super_block *, struct nfs_fattr *);
+
 static void nfs_read_inode(struct inode *);
 static void nfs_put_inode(struct inode *);
 static void nfs_delete_inode(struct inode *);
-static int  nfs_notify_change(struct inode *, struct iattr *);
+static int  nfs_notify_change(struct dentry *, struct iattr *);
 static void nfs_put_super(struct super_block *);
 static int  nfs_statfs(struct super_block *, struct statfs *, int);
 
@@ -138,6 +141,7 @@ nfs_put_super(struct super_block *sb)
 	 */
 	nfs_invalidate_dircache_sb(sb);
 
+	kfree(server->hostname);
 	sb->s_dev = 0;
 	unlock_super(sb);
 	MOD_DEC_USE_COUNT;
@@ -180,15 +184,16 @@ struct super_block *
 nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 {
 	struct nfs_mount_data	*data = (struct nfs_mount_data *) raw_data;
-	struct sockaddr_in	srvaddr;
 	struct nfs_server	*server;
-	struct rpc_timeout	timeparms;
 	struct rpc_xprt		*xprt;
 	struct rpc_clnt		*clnt;
+	struct nfs_fh		*root_fh;
+	struct inode		*root_inode;
 	unsigned int		authflavor;
 	int			tcp;
-	kdev_t			dev = sb->s_dev;
-	struct inode		*root_inode;
+	struct sockaddr_in	srvaddr;
+	struct rpc_timeout	timeparms;
+	struct nfs_fattr	fattr;
 
 	MOD_INC_USE_COUNT;
 	if (!data)
@@ -211,7 +216,6 @@ nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 	lock_super(sb);
 
 	sb->s_magic      = NFS_SUPER_MAGIC;
-	sb->s_dev        = dev;
 	sb->s_op         = &nfs_sops;
 	sb->s_blocksize  = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
 	sb->u.nfs_sb.s_root = data->root;
@@ -223,6 +227,10 @@ nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 	server->acregmax = data->acregmax*HZ;
 	server->acdirmin = data->acdirmin*HZ;
 	server->acdirmax = data->acdirmax*HZ;
+
+	server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
+	if (!server->hostname)
+		goto out_unlock;
 	strcpy(server->hostname, data->hostname);
 
 	/* Which protocol do we use? */
@@ -234,21 +242,19 @@ nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 	timeparms.to_maxval  = tcp? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
 	timeparms.to_exponential = 1;
 
-	/* Choose authentication flavor */
-	if (data->flags & NFS_MOUNT_SECURE) {
-		authflavor = RPC_AUTH_DES;
-	} else if (data->flags & NFS_MOUNT_KERBEROS) {
-		authflavor = RPC_AUTH_KRB;
-	} else {
-		authflavor = RPC_AUTH_UNIX;
-	}
-
 	/* Now create transport and client */
 	xprt = xprt_create_proto(tcp? IPPROTO_TCP : IPPROTO_UDP,
 						&srvaddr, &timeparms);
 	if (xprt == NULL)
 		goto out_no_xprt;
 
+	/* Choose authentication flavor */
+	authflavor = RPC_AUTH_UNIX;
+	if (data->flags & NFS_MOUNT_SECURE)
+		authflavor = RPC_AUTH_DES;
+	else if (data->flags & NFS_MOUNT_KERBEROS)
+		authflavor = RPC_AUTH_KRB;
+
 	clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
 						NFS_VERSION, authflavor);
 	if (clnt == NULL)
@@ -260,23 +266,30 @@ nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 	server->client    = clnt;
 
 	/* Fire up rpciod if not yet running */
-#ifdef RPCIOD_RESULT
-	if (rpciod_up())
+	if (rpciod_up() != 0)
 		goto out_no_iod;
-#else
-	rpciod_up();
-#endif
 
 	/*
 	 * Keep the super block locked while we try to get 
 	 * the root fh attributes.
 	 */
-	root_inode = nfs_fhget(sb, &data->root, NULL);
+	root_fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
+	if (!root_fh)
+		goto out_no_fh;
+	*root_fh = data->root;
+
+	if (nfs_proc_getattr(server, root_fh, &fattr) != 0)
+		goto out_no_fattr;
+
+	root_inode = __nfs_fhget(sb, &fattr);
 	if (!root_inode)
 		goto out_no_root;
 	sb->s_root = d_alloc_root(root_inode, NULL);
 	if (!sb->s_root)
 		goto out_no_root;
+	sb->s_root->d_op = &nfs_dentry_operations;
+	sb->s_root->d_fsdata = root_fh;
+
 	/* We're airborne */
 	unlock_super(sb);
 
@@ -289,14 +302,19 @@ nfs_read_super(struct super_block *sb, void *raw_data, int silent)
 out_no_root:
 	printk("nfs_read_super: get root inode failed\n");
 	iput(root_inode);
+	goto out_free_fh;
+
+out_no_fattr:
+	printk("nfs_read_super: get root fattr failed\n");
+out_free_fh:
+	kfree(root_fh);
+out_no_fh:
 	rpciod_down();
-#ifdef RPCIOD_RESULT
 	goto out_shutdown;
 
 out_no_iod:
-	printk("nfs_read_super: couldn't start rpciod!\n");
+	printk("NFS: couldn't start rpciod!\n");
 out_shutdown:
-#endif
 	rpc_shutdown_client(server->client);
 	goto out_unlock;
 
@@ -307,6 +325,7 @@ out_no_client:
 
 out_no_xprt:
 	printk("NFS: cannot create RPC transport.\n");
+	kfree(server->hostname);
 out_unlock:
 	unlock_super(sb);
 	goto out_fail;
@@ -349,46 +368,39 @@ nfs_statfs(struct super_block *sb, struct statfs *buf, int bufsiz)
 }
 
 /*
- * This is our own version of iget that looks up inodes by file handle
- * instead of inode number.  We use this technique instead of using
- * the vfs read_inode function because there is no way to pass the
- * file handle or current attributes into the read_inode function.
+ * Free all unused dentries in an inode's alias list.
+ *
+ * Subtle note: we have to be very careful not to cause
+ * any IO operations with the stale dentries, as this
+ * could cause file corruption. But since the dentry
+ * count is 0 and all pending IO for a dentry has been
+ * flushed when the count went to 0, we're safe here.
  */
-struct inode *
-nfs_fhget(struct super_block *sb, struct nfs_fh *fhandle,
-				  struct nfs_fattr *fattr)
+void nfs_free_dentries(struct inode *inode)
 {
-	struct nfs_fattr newfattr;
-	int error;
-	struct inode *inode;
-
-	if (!sb) {
-		printk("nfs_fhget: super block is NULL\n");
-		return NULL;
-	}
-	if (!fattr) {
-		error = nfs_proc_getattr(&sb->u.nfs_sb.s_server, fhandle,
-			&newfattr);
-		if (error) {
-			printk("nfs_fhget: getattr error = %d\n", -error);
-			return NULL;
+	struct list_head *tmp, *head = &inode->i_dentry;
+
+restart:
+	tmp = head;
+	while ((tmp = tmp->next) != head) {
+		struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
+		if (!dentry->d_count) {
+printk("nfs_free_dentries: freeing %s/%s, i_count=%d\n",
+dentry->d_parent->d_name.name, dentry->d_name.name, inode->i_count);
+			dget(dentry);
+			d_drop(dentry);
+			dput(dentry);
+			goto restart;
 		}
-		fattr = &newfattr;
-	}
-	if (!(inode = iget(sb, fattr->fileid))) {
-		printk("nfs_fhget: iget failed\n");
-		return NULL;
-	}
-#ifdef NFS_PARANOIA
-if (inode->i_dev != sb->s_dev)
-printk("nfs_fhget: impossible\n");
-#endif
-
-	if (inode->i_ino != fattr->fileid) {
-		printk("nfs_fhget: unexpected inode from iget\n");
-		return inode;
 	}
+}
 
+/*
+ * Fill in inode information from the fattr.
+ */
+static void
+nfs_fill_inode(struct inode *inode, struct nfs_fattr *fattr)
+{
 	/*
 	 * Check whether the mode has been set, as we only want to
 	 * do this once. (We don't allow inodes to change types.)
@@ -418,29 +430,140 @@ printk("nfs_fhget: impossible\n");
 		inode->i_size  = fattr->size;
 		inode->i_mtime = fattr->mtime.seconds;
 		NFS_OLDMTIME(inode) = fattr->mtime.seconds;
-		*NFS_FH(inode) = *fhandle;
 	}
-	if (memcmp(NFS_FH(inode), fhandle, sizeof(struct nfs_fh)))
-		printk("nfs_fhget: fhandle changed!\n");
 	nfs_refresh_inode(inode, fattr);
-	dprintk("NFS: fhget(%x/%ld ct=%d)\n",
-		inode->i_dev, inode->i_ino,
-		inode->i_count);
+}
 
+/*
+ * This is our own version of iget that looks up inodes by file handle
+ * instead of inode number.  We use this technique instead of using
+ * the vfs read_inode function because there is no way to pass the
+ * file handle or current attributes into the read_inode function.
+ *
+ * We provide a special check for NetApp .snapshot directories to avoid
+ * inode aliasing problems. All snapshot inodes are anonymous (unhashed).
+ */
+struct inode *
+nfs_fhget(struct dentry *dentry, struct nfs_fh *fhandle,
+				 struct nfs_fattr *fattr)
+{
+	struct super_block *sb = dentry->d_sb;
+
+	dprintk("NFS: nfs_fhget(%s/%s fileid=%d)\n",
+		dentry->d_parent->d_name.name, dentry->d_name.name,
+		fattr->fileid);
+
+	/* Install the filehandle in the dentry */
+	*((struct nfs_fh *) dentry->d_fsdata) = *fhandle;
+
+#ifdef CONFIG_NFS_SNAPSHOT
+	/*
+	 * Check for NetApp snapshot dentries, and get an 
+	 * unhashed inode to avoid aliasing problems.
+	 */
+	if ((dentry->d_parent->d_inode->u.nfs_i.flags & NFS_IS_SNAPSHOT) ||
+	    (IS_ROOT(dentry->d_parent) && dentry->d_name.len == 9 &&
+	     memcmp(dentry->d_name.name, ".snapshot", 9) == 0)) {
+		struct inode *inode = get_empty_inode();
+		if (!inode)
+			goto out;	
+		inode->i_sb = sb;
+		inode->i_dev = sb->s_dev;
+		inode->i_ino = fattr->fileid;
+		nfs_read_inode(inode);
+		nfs_fill_inode(inode, fattr);
+		inode->u.nfs_i.flags |= NFS_IS_SNAPSHOT;
+		dprintk("NFS: nfs_fhget(snapshot ino=%ld)\n", inode->i_ino);
+	out:
+		return inode;
+	}
+#endif
+	return __nfs_fhget(sb, fattr);
+}
+
+/*
+ * Look up the inode by super block and fattr->fileid.
+ *
+ * Note carefully the special handling of busy inodes (i_count > 1).
+ * With the kernel 2.1.xx dcache all inodes except hard links must
+ * have i_count == 1 after iget(). Otherwise, it indicates that the
+ * server has reused a fileid (i_ino) and we have a stale inode.
+ */
+static struct inode *
+__nfs_fhget(struct super_block *sb, struct nfs_fattr *fattr)
+{
+	struct inode *inode;
+	int max_count;
+
+retry:
+	inode = iget(sb, fattr->fileid);
+	if (!inode)
+		goto out_no_inode;
+	/* N.B. This should be impossible ... */
+	if (inode->i_ino != fattr->fileid)
+		goto out_bad_id;
+
+	/*
+	 * Check for busy inodes, and attempt to get rid of any
+	 * unused local references. If successful, we release the
+	 * inode and try again.
+	 *
+	 * Note that the busy test uses the values in the fattr,
+	 * as the inode may have become a different object.
+	 * (We can probably handle modes changes here, too.)
+	 */
+	max_count = S_ISDIR(fattr->mode) ? 1 : fattr->nlink;
+	if (inode->i_count > max_count) {
+printk("__nfs_fhget: inode %ld busy, i_count=%d, i_nlink=%d\n",
+inode->i_ino, inode->i_count, inode->i_nlink);
+		nfs_free_dentries(inode);
+		if (inode->i_count > max_count) {
+printk("__nfs_fhget: inode %ld still busy, i_count=%d\n",
+inode->i_ino, inode->i_count);
+			if (!list_empty(&inode->i_dentry)) {
+				struct dentry *dentry;
+				dentry = list_entry(inode->i_dentry.next,
+						 struct dentry, d_alias);
+printk("__nfs_fhget: killing %s/%s filehandle\n",
+dentry->d_parent->d_name.name, dentry->d_name.name);
+				memset(dentry->d_fsdata, 0, 
+					sizeof(struct nfs_fh));
+			} else
+				printk("NFS: inode %ld busy, no aliases?\n",
+					inode->i_ino);
+			make_bad_inode(inode);
+			remove_inode_hash(inode);
+		}
+		iput(inode);
+		goto retry;
+	}
+	nfs_fill_inode(inode, fattr);
+	dprintk("NFS: __nfs_fhget(%x/%ld ct=%d)\n",
+		inode->i_dev, inode->i_ino, inode->i_count);
+
+out:
 	return inode;
+
+out_no_inode:
+	printk("__nfs_fhget: iget failed\n");
+	goto out;
+out_bad_id:
+	printk("__nfs_fhget: unexpected inode from iget\n");
+	goto out;
 }
 
 int
-nfs_notify_change(struct inode *inode, struct iattr *attr)
+nfs_notify_change(struct dentry *dentry, struct iattr *attr)
 {
+	struct inode *inode = dentry->d_inode;
+	int error;
 	struct nfs_sattr sattr;
 	struct nfs_fattr fattr;
-	int error;
 
 	/*
 	 * Make sure the inode is up-to-date.
 	 */
-	error = nfs_revalidate(inode);
+	error = nfs_revalidate(dentry);
 	if (error) {
 #ifdef NFS_PARANOIA
 printk("nfs_notify_change: revalidate failed, error=%d\n", error);
@@ -476,7 +599,7 @@ printk("nfs_notify_change: revalidate failed, error=%d\n", error);
 		sattr.atime.useconds = 0;
 	}
 
-	error = nfs_proc_setattr(NFS_SERVER(inode), NFS_FH(inode),
+	error = nfs_proc_setattr(NFS_DSERVER(dentry), NFS_FH(dentry),
 				&sattr, &fattr);
 	if (error)
 		goto out;
@@ -503,9 +626,9 @@ out:
  * Externally visible revalidation function
  */
 int
-nfs_revalidate(struct inode *inode)
+nfs_revalidate(struct dentry *dentry)
 {
-	return nfs_revalidate_inode(NFS_SERVER(inode), inode);
+	return nfs_revalidate_inode(NFS_DSERVER(dentry), dentry);
 }
 
 /*
@@ -513,38 +636,43 @@ nfs_revalidate(struct inode *inode)
  * the cached attributes have to be refreshed.
  */
 int
-_nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
+_nfs_revalidate_inode(struct nfs_server *server, struct dentry *dentry)
 {
-	struct nfs_fattr fattr;
+	struct inode	*inode = dentry->d_inode;
 	int		 status = 0;
+	struct nfs_fattr fattr;
 
 	if (jiffies - NFS_READTIME(inode) < NFS_ATTRTIMEO(inode))
 		goto out;
 
-	dfprintk(PAGECACHE, "NFS: revalidating %x/%ld inode\n",
-			inode->i_dev, inode->i_ino);
-	status = nfs_proc_getattr(server, NFS_FH(inode), &fattr);
+	dfprintk(PAGECACHE, "NFS: revalidating %s/%s, ino=%ld\n",
+		dentry->d_parent->d_name.name, dentry->d_name.name,
+		inode->i_ino);
+	status = nfs_proc_getattr(server, NFS_FH(dentry), &fattr);
 	if (status) {
 #ifdef NFS_PARANOIA
-printk("nfs_revalidate_inode: getattr failed, error=%d\n", status);
+printk("nfs_revalidate_inode: %s/%s getattr failed, ino=%ld, error=%d\n",
+dentry->d_parent->d_name.name, dentry->d_name.name, inode->i_ino, status);
 #endif
-		goto done;
+		goto out;
 	}
 
 	status = nfs_refresh_inode(inode, &fattr);
-	if (status)
-		goto done;
+	if (status) {
+#ifdef NFS_PARANOIA
+printk("nfs_revalidate_inode: %s/%s refresh failed, ino=%ld, error=%d\n",
+dentry->d_parent->d_name.name, dentry->d_name.name, inode->i_ino, status);
+#endif
+		goto out;
+	}
 	if (fattr.mtime.seconds == NFS_OLDMTIME(inode)) {
 		/* Update attrtimeo value */
 		if ((NFS_ATTRTIMEO(inode) <<= 1) > NFS_MAXATTRTIMEO(inode))
 			NFS_ATTRTIMEO(inode) = NFS_MAXATTRTIMEO(inode);
 	}
 	NFS_OLDMTIME(inode) = fattr.mtime.seconds;
-
-done:
-	dfprintk(PAGECACHE,
-		"NFS: inode %x/%ld revalidation complete (status %d).\n",
-				inode->i_dev, inode->i_ino, status);
+	dfprintk(PAGECACHE, "NFS: %s/%s revalidation complete\n",
+		dentry->d_parent->d_name.name, dentry->d_name.name);
 out:
 	return status;
 }
@@ -575,7 +703,8 @@ nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
 		goto out;
 	}
 	if (inode->i_ino != fattr->fileid) {
-		printk("nfs_refresh_inode: inode number mismatch\n");
+		printk("nfs_refresh_inode: mismatch, ino=%ld, fattr=%d\n",
+			inode->i_ino, fattr->fileid);
 		goto out;
 	}