[PATCH] lockdep: annotate NTFS locking rules

NTFS uses lots of type-opaque objects which acquire their true identity
runtime - so the lock validator needs to be helped in a couple of places to
figure out object types.

Many thanks to Anton Altaparmakov for giving lots of explanations about NTFS
locking rules.

Has no effect on non-lockdep kernels.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Anton Altaparmakov <aia21@cantab.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 64 insertions, 0 deletions

diff --git a/fs/ntfs/inode.c b/fs/ntfs/inode.c
index 4c86b7e1d1eb..d313f356e66a 100644
--- a/fs/ntfs/inode.c
+++ b/fs/ntfs/inode.c
@@ -367,6 +367,12 @@ static void ntfs_destroy_extent_inode(ntfs_inode *ni)
 	kmem_cache_free(ntfs_inode_cache, ni);
 }
 
+/*
+ * The attribute runlist lock has separate locking rules from the
+ * normal runlist lock, so split the two lock-classes:
+ */
+static struct lock_class_key attr_list_rl_lock_class;
+
 /**
  * __ntfs_init_inode - initialize ntfs specific part of an inode
  * @sb:		super block of mounted volume
@@ -394,6 +400,8 @@ void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
 	ni->attr_list_size = 0;
 	ni->attr_list = NULL;
 	ntfs_init_runlist(&ni->attr_list_rl);
+	lockdep_set_class(&ni->attr_list_rl.lock,
+				&attr_list_rl_lock_class);
 	ni->itype.index.bmp_ino = NULL;
 	ni->itype.index.block_size = 0;
 	ni->itype.index.vcn_size = 0;
@@ -405,6 +413,13 @@ void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
 	ni->ext.base_ntfs_ino = NULL;
 }
 
+/*
+ * Extent inodes get MFT-mapped in a nested way, while the base inode
+ * is still mapped. Teach this nesting to the lock validator by creating
+ * a separate class for nested inode's mrec_lock's:
+ */
+static struct lock_class_key extent_inode_mrec_lock_key;
+
 inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
 		unsigned long mft_no)
 {
@@ -413,6 +428,7 @@ inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
 	ntfs_debug("Entering.");
 	if (likely(ni != NULL)) {
 		__ntfs_init_inode(sb, ni);
+		lockdep_set_class(&ni->mrec_lock, &extent_inode_mrec_lock_key);
 		ni->mft_no = mft_no;
 		ni->type = AT_UNUSED;
 		ni->name = NULL;
@@ -1722,6 +1738,15 @@ err_out:
 	return err;
 }
 
+/*
+ * The MFT inode has special locking, so teach the lock validator
+ * about this by splitting off the locking rules of the MFT from
+ * the locking rules of other inodes. The MFT inode can never be
+ * accessed from the VFS side (or even internally), only by the
+ * map_mft functions.
+ */
+static struct lock_class_key mft_ni_runlist_lock_key, mft_ni_mrec_lock_key;
+
 /**
  * ntfs_read_inode_mount - special read_inode for mount time use only
  * @vi:		inode to read
@@ -2148,6 +2173,14 @@ int ntfs_read_inode_mount(struct inode *vi)
 	ntfs_attr_put_search_ctx(ctx);
 	ntfs_debug("Done.");
 	ntfs_free(m);
+
+	/*
+	 * Split the locking rules of the MFT inode from the
+	 * locking rules of other inodes:
+	 */
+	lockdep_set_class(&ni->runlist.lock, &mft_ni_runlist_lock_key);
+	lockdep_set_class(&ni->mrec_lock, &mft_ni_mrec_lock_key);
+
 	return 0;
 
 em_put_err_out:
diff --git a/fs/ntfs/super.c b/fs/ntfs/super.c
index 0e14acea3f8b..74e0ee8fce72 100644
--- a/fs/ntfs/super.c
+++ b/fs/ntfs/super.c
@@ -1724,6 +1724,14 @@ upcase_failed:
 	return FALSE;
 }
 
+/*
+ * The lcn and mft bitmap inodes are NTFS-internal inodes with
+ * their own special locking rules:
+ */
+static struct lock_class_key
+	lcnbmp_runlist_lock_key, lcnbmp_mrec_lock_key,
+	mftbmp_runlist_lock_key, mftbmp_mrec_lock_key;
+
 /**
  * load_system_files - open the system files using normal functions
  * @vol:	ntfs super block describing device whose system files to load
@@ -1780,6 +1788,10 @@ static BOOL load_system_files(ntfs_volume *vol)
 		ntfs_error(sb, "Failed to load $MFT/$BITMAP attribute.");
 		goto iput_mirr_err_out;
 	}
+	lockdep_set_class(&NTFS_I(vol->mftbmp_ino)->runlist.lock,
+			   &mftbmp_runlist_lock_key);
+	lockdep_set_class(&NTFS_I(vol->mftbmp_ino)->mrec_lock,
+			   &mftbmp_mrec_lock_key);
 	/* Read upcase table and setup @vol->upcase and @vol->upcase_len. */
 	if (!load_and_init_upcase(vol))
 		goto iput_mftbmp_err_out;
@@ -1802,6 +1814,11 @@ static BOOL load_system_files(ntfs_volume *vol)
 			iput(vol->lcnbmp_ino);
 		goto bitmap_failed;
 	}
+	lockdep_set_class(&NTFS_I(vol->lcnbmp_ino)->runlist.lock,
+			   &lcnbmp_runlist_lock_key);
+	lockdep_set_class(&NTFS_I(vol->lcnbmp_ino)->mrec_lock,
+			   &lcnbmp_mrec_lock_key);
+
 	NInoSetSparseDisabled(NTFS_I(vol->lcnbmp_ino));
 	if ((vol->nr_clusters + 7) >> 3 > i_size_read(vol->lcnbmp_ino)) {
 		iput(vol->lcnbmp_ino);
@@ -2743,6 +2760,17 @@ static int ntfs_fill_super(struct super_block *sb, void *opt, const int silent)
 	struct inode *tmp_ino;
 	int blocksize, result;
 
+	/*
+	 * We do a pretty difficult piece of bootstrap by reading the
+	 * MFT (and other metadata) from disk into memory. We'll only
+	 * release this metadata during umount, so the locking patterns
+	 * observed during bootstrap do not count. So turn off the
+	 * observation of locking patterns (strictly for this context
+	 * only) while mounting NTFS. [The validator is still active
+	 * otherwise, even for this context: it will for example record
+	 * lock class registrations.]
+	 */
+	lockdep_off();
 	ntfs_debug("Entering.");
 #ifndef NTFS_RW
 	sb->s_flags |= MS_RDONLY;
@@ -2754,6 +2782,7 @@ static int ntfs_fill_super(struct super_block *sb, void *opt, const int silent)
 		if (!silent)
 			ntfs_error(sb, "Allocation of NTFS volume structure "
 					"failed. Aborting mount...");
+		lockdep_on();
 		return -ENOMEM;
 	}
 	/* Initialize ntfs_volume structure. */
@@ -2940,6 +2969,7 @@ static int ntfs_fill_super(struct super_block *sb, void *opt, const int silent)
 		mutex_unlock(&ntfs_lock);
 		sb->s_export_op = &ntfs_export_ops;
 		lock_kernel();
+		lockdep_on();
 		return 0;
 	}
 	ntfs_error(sb, "Failed to allocate root directory.");
@@ -3059,6 +3089,7 @@ err_out_now:
 	sb->s_fs_info = NULL;
 	kfree(vol);
 	ntfs_debug("Failed, returning -EINVAL.");
+	lockdep_on();
 	return -EINVAL;
 }