[PATCH] Fix BUG: KCSAN: data-race in fat16_ent_get / fat16_ent_put

From: Daniel Yang
Date: Mon Oct 28 2024 - 16:34:35 EST


Issue is that fat_free() calls fat_get_cluster() and fat_free_clusters()
at the same time. If the same fatent gets modified, it can lead to a
race condition where fat16_ent_put() and fat16_ent_get() will read/write
conflict on fatent->u.ent16_p.

To fix: add critical sections in fat_free() on the offending function
calls so that they can't both be running at the same time. Since the
critical sections are short, a spinlock is used since the overhead is
not that high.

Signed-off-by: Daniel Yang <danielyangkang@xxxxxxxxx>
Reported-by: syzbot+3999cae1c2d59c2cc8b9@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=3999cae1c2d59c2cc8b9
---
fs/fat/file.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/fs/fat/file.c b/fs/fat/file.c
index e887e9ab7..d7ae152a9 100644
--- a/fs/fat/file.c
+++ b/fs/fat/file.c
@@ -7,6 +7,7 @@
* regular file handling primitives for fat-based filesystems
*/

+#include "linux/spinlock.h"
#include <linux/capability.h>
#include <linux/module.h>
#include <linux/compat.h>
@@ -306,6 +307,9 @@ static long fat_fallocate(struct file *file, int mode,
return err;
}

+/* Prevent data race in fat_free. */
+static DEFINE_SPINLOCK(cluster_lock);
+
/* Free all clusters after the skip'th cluster. */
static int fat_free(struct inode *inode, int skip)
{
@@ -343,7 +347,10 @@ static int fat_free(struct inode *inode, int skip)
struct fat_entry fatent;
int ret, fclus, dclus;

+ /* Ensure fat_get_cluster isn't called while freeing. */
+ spin_lock(&cluster_lock);
ret = fat_get_cluster(inode, skip - 1, &fclus, &dclus);
+ spin_unlock(&cluster_lock);
if (ret < 0)
return ret;
else if (ret == FAT_ENT_EOF)
@@ -373,7 +380,12 @@ static int fat_free(struct inode *inode, int skip)
inode->i_blocks = skip << (MSDOS_SB(sb)->cluster_bits - 9);

/* Freeing the remained cluster chain */
- return fat_free_clusters(inode, free_start);
+ int ret;
+
+ spin_lock(&cluster_lock);
+ ret = fat_free_clusters(inode, free_start);
+ spin_unlock(&cluster_lock);
+ return ret;
}

void fat_truncate_blocks(struct inode *inode, loff_t offset)
--
2.39.2