[PATCH 2/2] sysfs: fix race condition around sd->s_dentry, take#2

[Tejun Heo <htejun@gmail.com>]

Allowing attribute and symlink dentries to be reclaimed means
sd->s_dentry can change dynamically.  However, updates to the field
are unsynchronized leading to race conditions.  This patch adds
sysfs_lock and use it to synchronize updates to sd->s_dentry.

Due to the locking around ->d_iput, the check in sysfs_drop_dentry()
is complex.  sysfs_lock only protect sd->s_dentry pointer itself.  The
validity of the dentry is protected by dcache_lock, so whether dentry
is alive or not can only be tested while holding both locks.

This is minimal backport of sysfs_drop_dentry() rewrite in devel
branch.

DONT APPLY JUST YET
---
Moving sysfs_drop_dentry() and sysfs_put() calls out of mutex isn't
necessary, so this is the minimal one but there shouldn't be
any difference functionality-wise.

 fs/sysfs/dir.c   |   22 ++++++++++++++++++++--
 fs/sysfs/inode.c |   18 +++++++++++++++++-
 fs/sysfs/sysfs.h |    1 +
 3 files changed, 38 insertions(+), 3 deletions(-)

Index: work/fs/sysfs/dir.c
===================================================================
--- work.orig/fs/sysfs/dir.c
+++ work/fs/sysfs/dir.c
@@ -13,14 +13,26 @@
 #include "sysfs.h"
 
 DECLARE_RWSEM(sysfs_rename_sem);
+spinlock_t sysfs_lock = SPIN_LOCK_UNLOCKED;
 
 static void sysfs_d_iput(struct dentry * dentry, struct inode * inode)
 {
 	struct sysfs_dirent * sd = dentry->d_fsdata;
 
 	if (sd) {
-		BUG_ON(sd->s_dentry != dentry);
-		sd->s_dentry = NULL;
+		/* sd->s_dentry is protected with sysfs_lock.  This
+		 * allows sysfs_drop_dentry() to dereference it.
+		 */
+		spin_lock(&sysfs_lock);
+
+		/* The dentry might have been deleted or another
+		 * lookup could have happened updating sd->s_dentry to
+		 * point the new dentry.  Ignore if it isn't pointing
+		 * to this dentry.
+		 */
+		if (sd->s_dentry == dentry)
+			sd->s_dentry = NULL;
+		spin_unlock(&sysfs_lock);
 		sysfs_put(sd);
 	}
 	iput(inode);
@@ -238,7 +250,10 @@ static int sysfs_attach_attr(struct sysf
         }
 
 	dentry->d_fsdata = sysfs_get(sd);
+	/* protect sd->s_dentry against sysfs_d_iput */
+	spin_lock(&sysfs_lock);
 	sd->s_dentry = dentry;
+	spin_unlock(&sysfs_lock);
 	error = sysfs_create(dentry, (attr->mode & S_IALLUGO) | S_IFREG, init);
 	if (error) {
 		sysfs_put(sd);
@@ -260,7 +275,10 @@ static int sysfs_attach_link(struct sysf
 	int err = 0;
 
 	dentry->d_fsdata = sysfs_get(sd);
+	/* protect sd->s_dentry against sysfs_d_iput */
+	spin_lock(&sysfs_lock);
 	sd->s_dentry = dentry;
+	spin_unlock(&sysfs_lock);
 	err = sysfs_create(dentry, S_IFLNK|S_IRWXUGO, init_symlink);
 	if (!err) {
 		dentry->d_op = &sysfs_dentry_ops;
Index: work/fs/sysfs/inode.c
===================================================================
--- work.orig/fs/sysfs/inode.c
+++ work/fs/sysfs/inode.c
@@ -244,9 +244,23 @@ static inline void orphan_all_buffers(st
  */
 void sysfs_drop_dentry(struct sysfs_dirent * sd, struct dentry * parent)
 {
-	struct dentry * dentry = sd->s_dentry;
+	struct dentry *dentry = NULL;
 	struct inode *inode;
 
+	/* We're not holding a reference to ->s_dentry dentry but the
+	 * field will stay valid as long as sysfs_lock is held.
+	 */
+	spin_lock(&sysfs_lock);
+	spin_lock(&dcache_lock);
+
+	/* dget dentry if it's still alive */
+	if (sd->s_dentry && sd->s_dentry->d_inode)
+		dentry = dget_locked(sd->s_dentry);
+
+	spin_unlock(&dcache_lock);
+	spin_unlock(&sysfs_lock);
+
+	/* drop dentry */
 	if (dentry) {
 		spin_lock(&dcache_lock);
 		spin_lock(&dentry->d_lock);
@@ -266,6 +280,8 @@ void sysfs_drop_dentry(struct sysfs_dire
 			spin_unlock(&dentry->d_lock);
 			spin_unlock(&dcache_lock);
 		}
+
+		dput(dentry);
 	}
 }
 
Index: work/fs/sysfs/sysfs.h
===================================================================
--- work.orig/fs/sysfs/sysfs.h
+++ work/fs/sysfs/sysfs.h
@@ -32,6 +32,7 @@ extern const unsigned char * sysfs_get_n
 extern void sysfs_drop_dentry(struct sysfs_dirent *sd, struct dentry *parent);
 extern int sysfs_setattr(struct dentry *dentry, struct iattr *iattr);
 
+extern spinlock_t sysfs_lock;
 extern struct rw_semaphore sysfs_rename_sem;
 extern struct super_block * sysfs_sb;
 extern const struct file_operations sysfs_dir_operations;
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/