[PATCH 19/20] XFS: allow follow_link to often succeed in RCU-walk.

[NeilBrown <neilb@xxxxxxx>]

If LOOKUP_RCU is set, use GFP_ATOMIC rather than GFP_KERNEL,
and try to get the ilock without blocking.

When these succeed, follow_link() can succeed without dropping
out of RCU-walk.

As xfs_readlink can now races with xfs_fs_evict_inode:
- xfs_readlink must check if the inode is being evicted after
  getting the lock
- xfs_fs_evict_inode cannot assert that the lock is not held.

Signed-off-by: NeilBrown <neilb@xxxxxxx>
---
 fs/xfs/xfs_ioctl.c   |    2 +-
 fs/xfs/xfs_iops.c    |   15 ++++++++++-----
 fs/xfs/xfs_super.c   |    2 --
 fs/xfs/xfs_symlink.c |   15 +++++++++++++--
 fs/xfs/xfs_symlink.h |    2 +-
 5 files changed, 25 insertions(+), 11 deletions(-)

diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
index ac4feae45eb3..29d95a1b76c0 100644
--- a/fs/xfs/xfs_ioctl.c
+++ b/fs/xfs/xfs_ioctl.c
@@ -303,7 +303,7 @@ xfs_readlink_by_handle(
 		goto out_dput;
 	}
 
-	error = xfs_readlink(XFS_I(dentry->d_inode), link);
+	error = xfs_readlink(XFS_I(dentry->d_inode), link, 0);
 	if (error)
 		goto out_kfree;
 	error = readlink_copy(hreq->ohandle, olen, link);
diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index c2c136ef3a50..631b1ec2e650 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -416,15 +416,20 @@ xfs_vn_follow_link(
 	int			flags)
 {
 	char			*link;
-	int			error = -ENOMEM;
+	int			error;
 
-	if (flags & LOOKUP_RCU)
-		return ERR_PTR(-ECHILD);
-	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
+	if (flags & LOOKUP_RCU) {
+		error = -ECHILD;
+		link = kmalloc(MAXPATHLEN+1, GFP_ATOMIC);
+	} else {
+		error = -ENOMEM;
+		link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
+	}
 	if (!link)
 		goto out_err;
 
-	error = xfs_readlink(XFS_I(inode), link);
+	error = xfs_readlink(XFS_I(inode), link,
+			     flags & LOOKUP_RCU);
 	if (unlikely(error))
 		goto out_kfree;
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 3827be14383c..e041fa55912b 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -996,8 +996,6 @@ xfs_fs_evict_inode(
 {
 	xfs_inode_t		*ip = XFS_I(inode);
 
-	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
-
 	trace_xfs_evict_inode(ip);
 
 	truncate_inode_pages_final(&inode->i_data);
diff --git a/fs/xfs/xfs_symlink.c b/fs/xfs/xfs_symlink.c
index 25791df6f638..228987b8e758 100644
--- a/fs/xfs/xfs_symlink.c
+++ b/fs/xfs/xfs_symlink.c
@@ -123,7 +123,8 @@ xfs_readlink_bmap(
 int
 xfs_readlink(
 	struct xfs_inode *ip,
-	char		*link)
+	char		*link,
+	int		rcu)
 {
 	struct xfs_mount *mp = ip->i_mount;
 	xfs_fsize_t	pathlen;
@@ -134,7 +135,15 @@ xfs_readlink(
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return -EIO;
 
-	xfs_ilock(ip, XFS_ILOCK_SHARED);
+	if (rcu) {
+		if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED) == 0)
+			return -ECHILD;
+		if (ip->i_vnode.i_state & (I_FREEING | I_CLEAR)) {
+			xfs_iunlock(ip, XFS_ILOCK_EXCL);
+			return -ECHILD;
+		}
+	} else
+		xfs_ilock(ip, XFS_ILOCK_SHARED);
 
 	pathlen = ip->i_d.di_size;
 	if (!pathlen)
@@ -153,6 +162,8 @@ xfs_readlink(
 	if (ip->i_df.if_flags & XFS_IFINLINE) {
 		memcpy(link, ip->i_df.if_u1.if_data, pathlen);
 		link[pathlen] = '\0';
+	} else if (rcu) {
+		error = -ECHILD;
 	} else {
 		error = xfs_readlink_bmap(ip, link);
 	}
diff --git a/fs/xfs/xfs_symlink.h b/fs/xfs/xfs_symlink.h
index e75245d09116..a71d26643e20 100644
--- a/fs/xfs/xfs_symlink.h
+++ b/fs/xfs/xfs_symlink.h
@@ -21,7 +21,7 @@
 
 int xfs_symlink(struct xfs_inode *dp, struct xfs_name *link_name,
 		const char *target_path, umode_t mode, struct xfs_inode **ipp);
-int xfs_readlink(struct xfs_inode *ip, char *link);
+int xfs_readlink(struct xfs_inode *ip, char *link, int rcu);
 int xfs_inactive_symlink(struct xfs_inode *ip);
 
 #endif /* __XFS_SYMLINK_H */


--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html